case study in medical coding

MEDICAL CODING CASE STUDIES

Let our coding case studies & solutions inspire you.

Our Featured Medical Coding Case Studies provide an in-depth review of challenges with coding quality, reimbursements, rework, compliance, etc. Conducting audits, evaluations, and implementation plans are inherent components of any effective solution. If you're searching for medical coding case studies with answers for your organization, then look no further. Learn more about our great clients, our approach, and the effective results obtained.

HEALTHCARE MERGERS & ACQUISITIONS DUE DILIGENCE AUDITS EBOOK

The healthcare industry sees hundreds of facility acquisitions and mergers every year. A critical part of mergers and acquisitions is the due diligence process. This eBook serves as a guide for medical practitioners and healthcare organizations to navigate the complexities of mergers and acquisitions, due diligence audits, and medical practice valuation.

Healthcare Compliance Plan eBook

Hospital decision-makers face complex challenges related to compliance, making it essential to have a well-crafted plan in place. This eBook offers a comprehensive overview of the fundamental elements required for an effective compliance plan, including data governance, risk assessment, education and training, auditing and monitoring, and incident response.

INPATIENT CODER & NEW GRAD DEVELOPMENT PROGRAM CASE STUDY

The Inpatient Coder Development Program is an immersive curriculum that spans over 9-12 months. The program includes practical education, hands-on Inpatient medical coding training, and mentoring from experienced coders. The goal of the program was to address the perceived need for a change in the way medical coder training is approached, bridge the gap between new graduates and facilities, and build new relationships while strengthening existing ones.

Radiology Coding Backlog Case Study

After 15 months coding the encounters in the client's backlog, we found there were only 500 encounters left daily, effectively meeting the client’s target goal of less than 1,000 encounters. During the Radiology Coding Backlog Case Study, our YES Team developed and implemented internal documentation and coding processes to successfully reduce the client’s backlog by 86 percent.

SEVERE MALNUTRITION STANDARDIZATION & IMPROVEMENT CASE STUDY

The Severe Protein Calorie Malnutrition Standardization & Improvement Plan case study presents the best approach to address severe protein-calorie malnutrition coding compliance with the Office of Inspector General (OIG), and the solutions we have designed for documentation review standardization and medical coding quality assurance.

ProFee Coding Quality Case Study

We studied the performance of professional services coding teams for 33 months, from 1QY1 to 3QY3. Our YES Team developed and successfully implemented this coding accuracy improvement plan with outstanding results, accomplishing long-term and scalable increase in CPT Accuracy, DX Accuracy, and Coder Retention.

IMPACT OF AUDITING AND CODER MENTORING PROGRAM ON ACCURACY

No matter how seasoned your coders may be, coder mentoring is a solution that ensures your team continuously performs at the best of their abilities. This case study presents information gathered throughout a 24-month period, with the objective of highlighting the direct impact on coding teams’ coding quality levels (from medium to large health systems), through the implementation of a standardized Coder Mentoring Program.

Our Services

Coding Support

Coder Mentoring

Denials Remediation

Consulting & Staffing

STAY IN THE KNOW: YES HIM & CODING INSIGHTS NEWSLETTER

Real-Life Examples of Medical Coding in Practice

Table of Contents

Medical coding is a crucial aspect of healthcare administration, as it involves the transformation of medical diagnoses, procedures, and healthcare services into alphanumeric codes. These codes are used for various purposes, such as billing, insurance claims, and statistical analysis. In this article, we will explore real-life examples of medical coding in practice through case studies and examine how it is applied in different healthcare settings.

Real-Life Examples of Medical Coding in Practice: Case Studies

Case study 1: hospital billing.

In a hospital setting, medical coding is essential for accurate and efficient billing. Let’s consider a case where a patient undergoes surgery for gallbladder removal. The surgeon will document the procedure details, such as the type of surgery performed, any complications, and the patient’s condition. The medical coder will then assign the appropriate codes to each aspect of the procedure, such as the surgical technique, anesthesia administered, and any additional services provided. These codes are crucial for generating an accurate hospital bill and ensuring proper reimbursement from insurance companies.

Case Study 2: Insurance Claims

Medical coding also plays a vital role in insurance claims processing. For instance, imagine a patient visits a primary care physician for a routine check-up. During the visit, the physician diagnoses the patient with hypertension and prescribes medication. The medical coder will translate the diagnosis and treatment details into specific codes, such as ICD-10 codes for the hypertension diagnosis and HCPCS codes for the prescribed medication. These codes are then used by the healthcare provider to submit an insurance claim for reimbursement. Accurate coding is crucial to avoid claim denials or delays in payment.

Case Study 3: Public Health Surveillance

Medical coding is not limited to billing and insurance purposes; it also plays a significant role in public health surveillance. For instance, during disease outbreaks or pandemics, medical coders input data from patient records into relevant surveillance systems. This data helps public health agencies monitor the spread of diseases, identify trends, and develop appropriate strategies to control and prevent further outbreaks. Accurate coding ensures that accurate and timely information is available to inform public health interventions.

How Medical Coding is Applied in Various Healthcare Settings

Medical coding is applied in a variety of healthcare settings, including hospitals, clinics, private practices, and research institutions. In hospitals, it is used for billing, reimbursement, and quality improvement purposes. In outpatient settings, such as clinics, medical coding is crucial for generating accurate bills and ensuring proper reimbursement from insurance companies. In research institutions, medical coding facilitates the analysis of large healthcare databases, allowing researchers to study disease patterns, treatment outcomes, and healthcare utilization.

Medical coding serves as the backbone of healthcare administration, facilitating the smooth operation of billing, insurance claims, and public health surveillance. Through real-life case studies, we have seen how medical coding is applied in various healthcare settings, ensuring accurate documentation, reimbursement, and analysis of health-related data. The importance of accurate and up-to-date medical coding cannot be overstated, as it enables healthcare providers, insurance companies, and public health agencies to provide high-quality care, streamline administrative processes, and improve overall healthcare outcomes.

• A: Medical coding plays a crucial role in the billing process by translating healthcare services and procedures into standardized codes. These codes are used for accurate invoicing, reimbursement, and financial transactions between healthcare providers and payers.
• A: Certainly. For instance, an incorrect code for a medical procedure may result in underbilling or overbilling, leading to financial discrepancies. This can impact the revenue cycle and may even result in compliance issues.
• A: Medical coders are trained to handle complex cases by using specific coding guidelines and conventions. They may need to assign multiple codes or modifiers to accurately represent the different components of a complex procedure or service.
• A: Accurate medical coding contributes to patient care by ensuring that healthcare providers are properly reimbursed for their services. This, in turn, helps maintain financial stability, allowing providers to invest in technology, training, and resources to enhance patient care.
• A: Coded data is valuable for healthcare organizations in identifying trends, conducting quality assessments, and implementing improvements. For example, analyzing coded data can help identify areas for clinical improvement, resource allocation, and compliance with healthcare standards.

Leave a comment Cancel reply

You must be logged in to post a comment.

+1 (800) 782 1768

• Revenue Cycle Management
• Reimbursement
• Diabetes Awareness Month
• Risk Management
• Patient Retention
• Medical Economics® 100th Anniversary
• Coding and documentation
• Business of Endocrinology
• Physicians Financial News
• Cybersecurity
• Cardiovascular Clinical Consult
• Locum Tenens, brought to you by LocumLife®
• Weight Management
• Business of Women's Health
• Practice Efficiency
• Finance and Wealth
• Remote Patient Monitoring
• Sponsored Webinars
• Medical Technology
• Billing and collections
• Acute Pain Management
• Exclusive Content
• Value-based Care
• Business of Pediatrics
• Concierge Medicine 2.0 by Castle Connolly Private Health Partners
• Practice Growth
• Concierge Medicine
• Business of Cardiology
• Implementing the Topcon Ocular Telehealth Platform
• Malpractice
• Sexual Health
• Chronic Conditions
• Legal and Policy
• Practice Management
• Patient Relations

ICD-10 in practice: Case studies in using the new coding system

Like it or not, ICD-10-CM is coming in October, which means medical practices are going to have to code and document their physicians’ patient encounters with more detail and complexity, or they won’t get paid. To help physicians prepare, we highlight some examples of ICD-10-CM coding in action.

How behavioral health integration benefits primary care

The consumerization of health care: How patient-centricity is key to greater efficiency

How AI threatens the privacy of patient health data

Why medical economics may change significantly if FTC noncompete ban is upheld

Beyond demographics: Understanding the values that drive women's health care choices

2 Commerce Drive Cranbury, NJ 08512

609-716-7777

• Industry News
• Access and Reimbursement
• Law & Malpractice
• Coding & Documentation
• Practice Management
• Patient Engagement & Communications
• Billing & Collections
• Staffing & Salary

Coding case study: Type 2 Diabetes follow-up

What you need to know about coding a follow-up appointment for Type 2 diabetes.

In the medical billing and coding field, getting paid requires accurate documentation and selecting the correct codes. In our Coding Case Studies, we explore the correct coding for a specific condition based on a hypothetical clinical scenario. This scenario involves a patient presenting for a follow-up visit with symptoms of Type 2 Diabetes. See if you can choose the correct codes.

Clinical Scenario

Chief Complaint

Patient, a 52-year-old male, came to the office for follow up of Type 2 diabetes mellitus (T2DM), hyperlipidemia, hypertension, urine micro albumin. Patient reports he was diagnosed with T2DM at age 45. Patient has been on insulin since 2010. Since last visit, patient reports blood sugars are stable.

Current Treatment:

Current diabetic regimen includes Tresiba 36 units daily and Xigduo (5/1000mg) 2 tab daily.

Reviewed medication list – medication compliance is good.

Glucose records reviewed

Blood glucose monitoring is done 0-1 times daily.

The patient denies hypoglycemia or hypoglycemic symptoms, i.e. no dizziness, sweating, confusion or headaches.

Diet / Exercise / Weight:

Patient is overweight but generally follows a healthy diet. Goes to gym two to three times per week.

Diabetic Related Complications:

Neuropathy symptoms : Positive stocking/glove numbness or tingling. No mononeuropathy. No postprandial bloating.

Retinopathy: Up-to-date on routine surveillance. First diagnosed 1/22/2018.

Nephropathy: Positive. Up-to-date on routine surveillance.

Review of Systems

Constitutional/Endocrine/Musculoskeletal: Negative.

Social History

Smoking status: Does not smoke.

Physical Exam

Weight: 271 lb 12.8 oz (123.3 kg)

BMI: Body mass index is 36.86 kg/m²

General: Alert; NAD with normal affect.

Eyes: EOMI; no icterus.

HENT: Atraumatic; oropharynx clear with moist mucous membranes.

Neck: supple, normal size thyroid, no palpable nodules.

Respiratory: Normal respiratory effort.

Cardiovascular: Regular rate & rhythm; no edema.

Musculoskeletal: FROM; no synovitis.

Neurological: reflexes 2+ at biceps, relaxation phase normal; no tremor.

Skin: No rash; no ulcerations.

Diabetic Foot Exam : No Lesions; good pulses.

Type 2 diabetes mellitus with hyperglycemia, with long-term current use of insulin

Type 2 diabetes mellitus with polyneuropathy

Type 2 diabetes mellitus with microalbuminuria, with long-term current use of insulin

Type 2 diabetes, uncontrolled, with retinopathy

Class 2 severe obesity due to excess calories with serious comorbidity and body mass index (BMI) of 36.0 to 36.9 in adult

Hyperlipidemia associated with type 2 diabetes mellitus

Hypertension associated with diabetes

Vitamin D deficiency

Documentation Coding Requirements

When documenting diabetes, include the following:

Due to underlying condition

Drug or chemical induced diabetes mellitus

Diabetes mellitus in pregnancy, childbirth, and the puerperium

With or without complication

With or without coma

Left Right Bilateral

Diagnosis Codes

E11.65 Type 2 diabetes mellitus with hyperglycemia

E11.42 Type 2 diabetes mellitus with diabetic polyneuropathy

E11.29 Type 2 diabetes mellitus with other diabetic kidney complication

E11.319 Type 2 diabetes with unspecified diabetic retinopathy with macular edema

E11.69 Type 2 diabetes mellitus with other specified complication

E11.59 Type 2 diabetes mellitus with other circulatory complications

E66.01 Morbid (severe) obesity due to excess calories

Z68.36 Body mass index (BMI) 36.0-36.9, adult

Z79.4 Long term (current) use of insulin

E55.9 Vitamin D deficiency, unspecified

R80.9 Proteinuria, unspecified

E78.5 Hyperlipidemia, unspecified

Renee Dowling is a compliance auditor at Sansum Clinic, LLC, in Santa Barbara, California.

7 Ways automated coding can help capture reimbursement

Automation can play a large role in your coding strategy.

Avoiding Claims Denials in Your Practice

Physicians Practice® spoke with Terry Blessing III, Senior Vice President of Client Development at VisiQuate, about how practices can work to reduce the likelihood of encountering denied claims.

Denials, coding, prior auths, and your revenue cycle

Navigating payer denials requires the most expertise to handle within revenue cycle management

Document Automation, Data Extraction for Independent Practices

Physicians Practice® spoke with John Harrison, CCO of Concord Technologies, about how documentation technologies can improve workflows in practices.

3 Ways coding automation can help a practice

Are you leaving money on the table because you aren't coding correctly?

The problem with duplicate and mismatched patient records

Take action to ensure accuracy of patient information

2 Commerce Drive Cranbury, NJ 08512

609-716-7777

How to Pass CPC Exam: Free CPC Practice Exam #1 (2023)

Ready to get started? Our free, full-length, interactive CPC mock exam consists of 100 questions across all 17 categories you'll see on the official Certified Professional Coder exam. Click here to start practicing!

Welcome to our first full-length, free CPC practice exam! We've designed this test to be as close to the AAPC exam as possible, to help you get the most realistic and helpful practice and feedback possible.

Read on to learn how to use it for best results, or click here to skip down to the test questions and get started. Or, download the exam as a PDF below:

The CPC Exam has 100 questions from 17 categories

The exam is composed of 100 questions, drawn from the same 17 medical billing and coding categories you will see on the actual exam:

• CPT 10000 - Integumentary System (Skin, nails, hair)
• CPT 20000 - Musculoskeletal System
• CPT 30000 - Respiratory System
• CPT 40000 - Digestive
• CPT 50000 - Reproductive
• CPT 60000 - Nervous system
• Lab pathology (Panels, drug testing, chemistry, transfusions_
• Medical specialties (Immunizations, psychiatry, ophthalmology, cardiovascular, pulmonary, genetics, chemotherapy)
• Medical Terminology
• Anatomy & Physiology
• HCPCS Level II (Medicare-related codes)
• ICD-10-CM Coding
• Coding Guidelines (both ICD & CPT)
• Compliance and Regulations (Medicare, HIPAA, national/local coverage, reimbursement, and RVUs)
• Healthcare Law & Ethics

How long is the CPC exam?

The 2023 CPC exam is 4 hours long, done in 1 sitting.

You can start and stop the exam at any time, and your progress will be automatically saved.

How can I get the best results from this practice exam?

On the real CPC, you'll be either in your own personal space, taking the exam on a video call with the rest of the test-takers, or in a live test center with other test takers. The 2023 exam is 100 questions, done in a single sitting of 4 hours long.

To get the best results from this practice test, treat this full-length practice exam as if it was the real thing. Find a quiet room, and carve out a solid block of four hours where you will have total quiet, just like if you were taking the real exam.

Turn off your phone and let friends and family know that you will be unavailable, because you are taking the time to practice for the CPC.

Finally, try to complete the test in a single sitting. It's okay to take bathroom breaks, just like on the real exam, but try to eat about 2 hours beforehand

Practice tests will help you get used to the length and material of the real CPC

One of the largest reasons students find the CPC exam challenging is because of the length of the exam.

4 hours is a long time to sit and answer test questions at one stretch!

Taking a real, full-length practice test will help you get used to the length of the exam, and help you get your routine right for test day.

Take this practice exam as if it was the real thing, and you won't be surprised on test day - you'll be a seasoned pro!

How many times can I take the practice exam?

As many times as you like!

For those of you who want to do your absolute best, we recommend taking 3-5 different practice exams over 2-3 months.

Taking 3-5 exams will help get you intimately familiar with the structure of the CPC, and let you know what material you need to study. You'll have plenty of opportunity to study in between exams, and on test day, the real CPC will feel like old hat.

Take this practice test as many times as you like, and be sure to check out the rest of our CPC exam preparation materials as well.

Official CPC Practice Exam

Don't forget to use your CPT®, ICD-10-CM, and HCPCS Level II code books!

100 Questions (Suggested time: 4 hours)

Haven't completed the interactive CPC practice exam? Scroll up to complete the test!

Once you've completed, you'll see your score and a breakdown of questions here, to help you see where you did well, and where you could improve.

As always, we're cheering for you!

All done? Don't forget to get feedback!

Congrats on finishing your first CPC practice exam! Every bit of effort you put in is going to help you score better and continue on the path of becoming a medical billing and coding professional.

Once you've finished the test, take five minutes to reflect on what was easy for you, what was challenging this time, and where you could improve.

• What was easy for me:
• What was challenging for me:
• What I could improve next time:

Think about types of questions that were hard, what questions you spent your time on, and any material that you might want to go back and review. Spending a little time doing this while the exam is fresh will help you figure out exactly what to do next.

What's next?

Feel free to use our feedback tool to see where you need to study up to score even better on your next exam!

Feeling ready for more practice? Try our list of 800+ Free CPC Practice Exam Questions , or check out the rest of our 100% free CPC practice exams and mock tests. You got this!

Or: Feeling ready for the real thing? Start picking your date to take the Official Certified Professional Coder Exam and get certified by the AAPC.

We're cheering for you!

Subscribe to Medical Coding Ace

• Skip to main content
• Skip to primary sidebar
• Skip to footer

Free Medical Coding

ICD10CM, HCPCS, PCS, DRG codes lookup for free! Coding news and articles and more!

ICD-10-CM Case Study Example

January 10, 2023 by Vivek

Case Study Example for ICD-10-CM

Here is an example of an ICD-10-CM case study that will help you understand proper coding.

CHIEF COMPLAINT: Follow up on Diabetes.

HISTORY OF PRESENT ILLNESS:

• Established female patient presents today for a re-evaluation of her Diabetes Mellitus Type I.
• Last A1C was done in April with a result of 7.8.
• Patient is doing Accuchecks on a regular basis. Patient is being following a strict diet.
• Patient is taking medications as prescribed and also is having problems with neuropathy.
• The onset was 2 week ago. Status has been worse, gradually. She states it is a burning in her feet.

REVIEW OF SYSTEMS: MUSCULOSKELETAL:

• Negative for, rheumatologic manifestations, back pain, myalgia.
• Positive for burning in her feet.

DIABETIC: Negative for:

• Hypoglycemia
• New infection
• Weight loss
• Weight gain
• Paresthesia
• Shortness of breath
• Dyspnea on exertion
• Claudication symptoms,
• New neurological symptoms
• Post prandial bloating
• Orthostasis
• Vision changes.

MEDICATIONS:

Humulin R, 100U/ml, 45 units AC.

Lantus 100U/ml, 60 units HS.

ASSESSMENT/PLAN:

1) Diabetes Type I) Diabetic Neuralgia. Medications were reviewed with the patient. Poor control-Medications were reviewed with the patient. Clinical lab work was done today. Continue present management. The patient was instructed on 1600 cal ADA diet.

Diabetes is documented as Type I. A combination code is used for Type I Diabetes with Neuropathy. The look-up of Diabetes, Type I, with neuralgia leads to E10.42 Type I Diabetes Mellitus with diabetic polyneuropathy . Because the patient is Type I the guidelines state the code for long-term use of insulin is not reported.

At  Freemedicalcoding.com , our mission is to provide you with free access to the latest medical codes from authentic Government and industry databases.

We provide free access to ICD9, ICD10, HCPCS, PCS, DRG and other medical codes databases. Our site is updated regularly.

We are also committed to providing you with the latest news about the coding industry, as well as access to jobs, education and certification and memberships in various associations. We share information from Medicare, CMS and other leading medical codes news sources.

Disclaimer: All data and news are the property of their respective companies. We are not affiliated with any of the sources we provide information from.  The information is provided on a best-effort basis. We assume no liability for any information shared. Please make sure you validate the information you get from us.

Stay up-to-date with the latest in medical coding and medical coding software

Provistas, Incorporated 10339 E 37th Street, Yuma, AZ 85365

[email protected]

Explore prestigious scientific journals and award and funding opportunities to advance your research.

Access practice-improvement tools, including guidelines, measures, and practice management resources.

Learn to effectively advocate on behalf of neurologists and their patients, and access AAN position and policy statements.

Loading... please wait

We’re experiencing unusually high levels of traffic. Thank you for your patience.

Outpatient Seizures Case Studies

Outpatient e/m case studies.

Compare case studies of seizures patients at varying visit levels to better understand code selection for outpatient office encounters under the revised guidelines for 2021.

12-Year-Old Female with Seizures

Total Time* for Outpatient E/M in 2021

Refer to the following tables for correct code selection when billing based on time for outpatient E/M Services:

*Total time includes non-face-to-face time on the date of service

Case 1: Level 3 ESTABLISHED (99213)

A 12-year-old female is brought by her mother for a follow-up of known complex partial seizure disorder on carbamazepine.

• Pre-Visit: Reviewed prior clinic notes [2 MINS]
• Visit: Performed medically appropriate history and exam. No recent reported seizures since carbamazepine was initiated. No signs of toxicity on exam. She and family are counseled about her diagnosis of epilepsy, seizure precautions. [20 MINS]

Post-Visit:  Document encounter (3 MINS)

View MDM table for Seizures - 99213

case 2: level 4 established (99214)

Returns for a three-month follow-up visit and her mother denies witnessing any seizures and the patient denies side-effects.

• Pre-Visit: Personally reviewed MR images which reported a cortical dysplasia in the left frontal lobe. Her EEG report was read as normal [5 MINS]
• Visit:  The significance of the cortical dysplasia was reviewed, as were images. No symptoms or signs of toxicity on exam. Oxcarbazepine was continued and appropriate surveillance labs were previously obtained and reviewed during the visit [25 MINS]

Post-Visit:  Document encounter [3 MINS]

View MDM table for Seizures - 99214

case 3: level 5 established (99215)

In the past six months, seizures became refractory to two medications, requiring multiple ED visits in the interim. She underwent a presurgical evaluation. You presented her at a multidisciplinary conference where surgery was recommended. 

• Pre-Visit: You review the consensus recommendations, MR, PET, EEG, and neuropsychology report [7 MINS]
• Visit: Discussion of surgical recommendations, treatment alternatives, and risk of sudden unexpected death in epilepsy patients with patient and parents. Referred patient to neurosurgery for formal consultation. You maintain oxcarbazepine, but increase clobazam [25 MINS]

Post-Visit: Place referral and send prescriptions, document encounter [4 MINS]

View MDM table for Seizures - 99215

Disclaimer: The billing and coding information provided by the American Academy of Neurology and its affiliates (collectively, “Academy”) are assessments of clinical information provided as an educational service. The information (1) is not clinical advice; (2) does not account for how private payers cover and reimburse procedures or services*; (3) is not continually updated and may not reflect the most current clinical information (new clinical information may emerge between the time information is developed and when it is published or read); and (4) is not a substitute for the independent professional judgment of the treating provider, who is responsible for correctly coding procedures and services.

Using this information is voluntary. The Academy is providing the information on an “as is” basis and makes no warranty, expressed or implied, regarding the information. The Academy specifically disclaims any warranties of merchantability or fitness for a particular use or purpose. The Academy assumes no responsibility for any injury or damage to persons or property arising out of or related to any use of this information or for any errors or omissions.

*The Academy recommends always checking private payer policies before rendering procedures or services

• Login / Signup

Medical Coding Tutorial

• What is Medical Coding?
• What are Medical Coding Systems?
• Importance of Medical Coding
• What are Medical Coding Guidelines?
• What is Medical Decision-Making Level?
• What is ICD-10 coding?
• What is CPT Coding?
• What is HCPCS coding?
• What are HIPAA Regulations?
• What is Protected Health Information?
• Anatomy and Physiology Fundamentals
• Human Anatomy: An Overview of Major Body Systems, Organs, and Tissues
• Physiology Basics: Understanding Body Functions for Accurate Medical Coding
• ICD-10-CM Coding Overview
• Code Structure of ICD-10-CM Coding
• Coding Conventions of Medical Coding
• Coding Guidelines for ICD-10-CM
• How to navigate the Code Sets?
• Medical Coding Case Studies: Practice Coding Real-World Scenarios
• Tough Medical Coding Case Studies
• Current Procedural Terminology Coding System
• CPT Coding Categories
• What is Chief Complaint?
• What is History of the Present Illness?
• What are Surgical Modifiers?
• What are Bundled Procedures?
• Evaluation and Management (E/M) Codes in CPT coding system
• CPT Coding: Case Studies
• EM Coding: Case Studies

Inpatient E/M coding Case Studies

• Outpatient E/M Coding Case Studies
• Difference between inpatient and outpatient E/M coding
• What is the HCPCS Coding?
• What is DMPEOS Coding?
• HCPCS Overview
• What is Medical Coding Compliance?
• Ethical Considerations in Medical Coding
• Medical Coding Compliance and Fraud
• Health Insurance Portability and Accountability Act (HIPAA)
• Medical Coding Auditing and Documentation
• What Medical Billing?
• Revenue Cycle Management in Medical Billing
• Medical Coding and Reimbursement Relationship
• Medical Coding Claim Submission
• What are Medical Terminologies?
• Medical Language Basics for Improved Coding Accuracy
• Medical Terminologies: Prefixes, Suffixes, and Root Words

Module 1: Introduction to Medical Coding

Module 2: anatomy and physiology fundamentals, module 3: icd-10-cm coding, module 4: cpt coding, module 5: hcpcs level ii coding, module 6: medical coding compliance and ethics, module 7: medical billing and reimbursement, module 8: medical terminology, module 10: medical coding interview questions & answers, join our community on telegram, join the biggest community of pharma students and professionals..

Join our Telegram Community.

Here are some examples of inpatient Evaluation and Management (E/M) coding scenarios:

Example 1: Inpatient Hospital Visit

Scenario: A 65-year-old male patient with a history of congestive heart failure is admitted to the hospital due to worsening symptoms. The attending physician conducts a detailed history, comprehensive examination, and orders additional cardiac tests. The patient's condition requires frequent monitoring and adjustment of medications during his three-day hospital stay.

In this scenario, the appropriate inpatient E/M code would be:

- CPT code 99233 (Subsequent hospital care, per day, for the evaluation and management of a patient, which requires at least two of these three key components: a detailed history, a detailed examination, and medical decision-making of high complexity).

Example 2: Inpatient Consultation

Scenario: A 50-year-old female patient is admitted to the hospital with a suspected gastrointestinal issue. The gastroenterologist is asked to consult and provide expert advice on the patient's condition. The gastroenterologist reviews the patient's medical history, performs a comprehensive examination, and orders diagnostic tests to determine the cause of her symptoms.

- CPT code 99254 (Initial inpatient consultation for a patient, which requires these three key components: a comprehensive history, a comprehensive examination, and medical decision-making of moderate complexity).

Example 3: Inpatient Critical Care

Scenario: A 30-year-old male patient is admitted to the intensive care unit (ICU) with severe sepsis. The critical care team provides continuous monitoring, advanced life support, and multiple interventions to stabilize the patient's condition.

- CPT code 99291 (Critical care, evaluation, and management of the critically ill or critically injured patient; first 30-74 minutes).

- Additional time-based codes may be used if the critical care provided exceeds 74 minutes:

  - CPT code 99292 (Critical care, each additional 30 minutes).

Example 4: Inpatient Psychiatric Evaluation

Scenario: A 40-year-old female patient is admitted to the psychiatric unit with symptoms of severe depression and suicidal ideation. The psychiatric team conducts a detailed psychiatric evaluation, assesses her mental status, and develops a comprehensive treatment plan.

- CPT code 99222 (Initial hospital care, per day, for the evaluation and management of a patient, which requires these three key components: a comprehensive history, a comprehensive examination, and medical decision-making of low complexity).

Remember that the appropriate inpatient E/M code is determined based on the complexity of medical decision-making, the level of history and examination performed, and the time spent providing care during the hospital stay. Proper documentation is essential to support the chosen E/M code and ensure accurate coding and billing.

Book a FREE live class. NOW!

Fill your details and select a date for your live class

• Covid Case Report
• Virtual Student Services
• Academic Advising & Support
• Academic Calendar
• Academic Catalog
• Adult Education
• Apprenticeships
• Brightspace
• CareerFinder
• CareerNetwork
• Career Planning & Placement
• Class Schedules
• Clubs, Groups & Organizations
• Compliments/Complaints
• Dining & Food Services
• Employment at City Colleges
• Event Calendar
• File a Title IX Complaint
• Net Price Calculator
• Program Finder
• Scholarships
• Student Policies
• Student Portal
• Testing Resources
• Transfer/Articulation
• Curriculum Development
• Employee Portal
• Faculty Forms
• HR Self Service
• Human Resources
• Inspector General
• Library System
• Print Job Request
• Salesforce for Enrollment
• Salesforce for District Office
• Schedule Events & Rooms
• Special Assignment Request
• Supportive Intervention Team
• Taleo Recruiting
• Tenure Documents
• Travel & Expense Reimbursement
• Union Contracts
• Academic Departments
• Board of Trustees
• Colleges & Satellites
• Computer Labs
• Departments
• Employee Directory
• Tutoring Centers
• Contact City Colleges
• Apply for Financial Aid
• Buy my Books
• Submit a Compliment or Complaint
• Pay Tuition
• Register for Classes
• Work At CCC
• Ask a Question
• More…
• Harold Washington
• Harry S Truman
• Kennedy-King
• OIive-Harvey
• Richard J. Daley
• Wilbur Wright
• City Colleges Home

• Important Dates and Deadlines
• Tuition Rates
• Transfer to CCC
• Search Class Schedules
• View the Academic Catalog

Financial Resources

• Paying for College
• Apply for a Scholarship
• Financial Aid
• Get Your CTA Ventra/U-Pass
• Credit Programs
• Earn a Degree or Certificate
• Find a Program
• View Course Offerings
• International Students
• Annual Security Report

Adult Education Programs

• Learn English (Free)
• Finish High School
• Earn Your GED (Free)

Continuing Education Programs

• Professional Development
• Personal Development
• Programs for High School Students
• Take Classes While in High School
• Alternative High School
• Online Learning
• 360º College View

Campus Life

• Get Involved (Clubs, Groups, Orgs,SGA)
• Student Athletics

Dining Options

Academic resources.

• Meet with an Advisor
• Testing Resource Centers
• Buy Your Books
• Locate a Computer Lab
• Technology Support
• Meet with a Tutor

Student Resources

• Registrar / Student Records
• Visit the Wellness Center
• First Year Student Experience
• Access Veterans Services
• Access Center
• Childcare While You Study
• Safety and Security
• Find a Job or Internship
• Prepare to Transfer
• Request a Transcript
• Join the CCC Alumni Association
• Prepare to Graduate
• Continuing Education
• View Class Schedules
• Review Academic & Student Policies
• Transfer Resources
• Star Scholarship Partners
• Credit for Prior Learning (CPL)
• Career Planning and Placement
• Career Center
• College to Careers
• Find Your Career
• Find a Job/Internship

Facts & Figures

• About the College
• Mission Statement
• Facts and Statistics
• Accreditation
• Disclosures
• College Locations
• Media Inquiries
• Community Relations
• Accessibility

Development

• City Colleges Foundation
• Ways to Give

Safety & Security

General contact information.

• General Information
• IT Help Desk
• Business Services Office

For Employers

• Train Your Employees
• Continuing Education Classes
• Hire a CCC Student or Post a Job
• Make a Gift/Donation

Partner with CCC

• Become a Vendor
• Bid Solicitations

Medical Coding

Deshawna hill-burns, mpa, rhia, hit pro cp, medical director, note: prior to enrolling in coursework, please contact your college advisor and/or director of the health information technology (hit), medical coding, and medical billing programs, deshawna hill-burns, [email protected]​ ..

To Apply to Program:

• Complete program prerequisite courses.   
• Submit official  college  transcripts.    
• Calculate admissions cumulative GPA.    
• Apply to Program in Student Portal [my.ccc.edu]

To Learn More About Program and Admissions:  

About Program      

Medi cal Coding Program is intended for individuals seeking entry-level employment as well as for advancement or cross-training opportunities for those who are currently employed in healthcare environment. The students learn ICD-10-CM/PCS and CPT/HCPCS Level II guidelines and apply them while coding various coding scenarios including inpatient records, outpatient visits, operative reports, and other healthcare records. The students utilize 3M encoder software in both basic and advanced coding courses  in order to  master their coding skills and smoothly transition to their entry-level healthcare careers. In addition, the students participate in the planned and supervised clinical affiliations in medical facilities. Our stackable degree path allows students to apply all credits earned in the Medical Billing-Basic Certificate (BC) program to the Medical Coding-Advanced Certificate (AC) program. Upon the completion of the Medical Coding-Advanced Certificate, the students can then pursue a degree towards an Associate in Applied Science in Health Information Technology.                        

Program Mission    

The mission of the Medical Coding Program is to graduate competent and ethical medical coders with the knowledge and the skills satisfying needs of entry-level positions in this field. It is also the department’s mission to support the Malcolm X College mission by offering quality medical coding courses designed to prepare students to demonstrate the skills and behaviors compatible with those of professional medical coders.                    

Certification  

The Medical Coding Program-AC prepares students to take the Certified Professional Coder (CPC) exam administered by the American Academy of Professional Coders (AAPC).                    

Admission Requirements  

• Completion of the Malcolm X College Medical Billing [BC] or  
• Currently enrolled in Malcolm X College Medical Billing program are eligible to apply
• Submit official college transcripts from all colleges attended outside of City Colleges of Chicago to the Malcolm X College Office of Registrar Services.  
• On-campus, in-person i nterview; this will occur in late October  
• Submit one Letter of Recommendation at the interview  
• Once accepted to the program and prior to registration; students must complete the following:  
• Background check  
• Drug screen  
• Provide proof of health insurance  
• Provide current immunization records
• Students need to have transportation to and from their clinical sites     

Submitting Application within Student Portal ( my.ccc.edu )

• Under “My Academic Plan”  
• Within “Academic Planning”  
• Click “Selective Application”

Faculty and Staff

https://www.ccc.edu/colleges/malcolm-x/departments/Pages/Medical-Coding.aspx

Fall credit classes start August 22 Fall ESL & GED/HiSET prep classes start August 26

• Current Student? Register Now!
• New Student? Apply Now!
• Need Help? Get Virtual Support

• Document Portal
• Online Classroom
• Student Portal
• Call Us 702-254-7577
• Request Info

Addressing the Ethical Issues in Medical Billing and Coding

• Medical Billing and Coding
• June 12, 2024

Medical billing and coding are essential components of the healthcare system and are pivotal in reimbursing healthcare providers. However, ethical challenges often arise in this field. 

Medical billing and coding professionals are a crucial link between healthcare providers, insurance companies, and patients. They are responsible for ensuring accuracy, fairness, and timeliness. Addressing these ethical challenges is essential to maintaining trust and upholding the integrity of healthcare practices. 

Below are key considerations and strategies for navigating these challenges effectively:

Table of Contents

Ensuring Accuracy

Confidentiality and data security, transparency in patient care, handling errors, fraudulent practices, fair billing practices.

Mistakes in coding can result in incorrect billing, leading to overcharging patients or underpayment for healthcare providers. It is essential to use accurate codes and resist the temptation to overcode or undercode to meet financial goals. Regular training and careful oversight help maintain high accuracy and ethical compliance standards.

In any medical practice, it is crucial to maintain extremely high levels of confidentiality , especially in departments such as billing and coding. Professionals in this field can access sensitive patient information, which must be safeguarded from unauthorized access or leaks. 

It is a matter of ethics and a legal requirement to work within laws such as the Health Insurance Portability and Accountability Act ( HIPAA ) and other privacy regulations, ensuring that patient information is kept secure and only shared with appropriate parties. To meet these requirements, it is essential to implement and maintain advanced security measures and provide regular security training.

Mistakes are inevitable in any practice, but how they are corrected matters. This reflects significantly on the ethical considerations of the billing and coding practice. Any instances of overbilling or underbilling should be rectified as soon as possible with as much transparency as possible. Ethical practices are essential in medical billing and coding to ensure errors are identified, reported, and corrected systematically.

It would be a severe ethical violation to engage in fraudulent billing , which could involve billing for services that were not provided or misrepresenting services as medically necessary when they are not. 

These practices could have legal consequences and harm the trust between patients, healthcare providers, and insurance companies. Ethical behavior in this area involves actively preventing fraud and abuse by understanding and following the laws and regulations that pertain to medical billing and coding.

Fair billing involves more than just ensuring that services are coded correctly. It also requires ethical billing practices, such as charging the same rates for similar services provided under identical conditions. 

The ethical challenges associated with medical billing and coding are substantial and significantly impact healthcare delivery. Professionals in this field must promote fairness, transparency, and trust in the healthcare system. 

Ethical billing and coding practices benefit both patients and healthcare providers and are critical for maintaining the integrity of the healthcare system. As the field continues to evolve, an unwavering commitment to ethical practice must exist to support the health and well-being of all involved.

Nancy began her career as a Medical Assistant in 1979. From there, Nancy mastered many other areas of the medical field. She spent 35 years in Ohio, building a successful medical practice with the same Physician until his retirement in… Read Full Bio

Related Article

How to Prepare for a Career in Medical Billing

Best Medical Billing Software of 2024

July 8th, 2024

We Value Your Privacy

• © 2012-2024 NorthwestCareerCollege.edu
• Privacy Policy
• Website & SEO Services by Axilweb

• Skip to main content

Update your browser for the best possible experience

As of January 1st, 2020, Internet Explorer (versions 11 and below) is no longer supported by Evolve. To get the best possible experience using Evolve, we recommend that you use another web browser. For HESI iNet users click here .

• Search Menu

Sign in through your institution

• Advance articles
• Editor's Choice
• Focus Issue Archive
• Open Access Articles
• JAMIA Journal Club
• Author Guidelines
• Submission Site
• Open Access
• Call for Papers
• About Journal of the American Medical Informatics Association
• About the American Medical Informatics Association
• Journals Career Network
• Editorial Board
• Advertising and Corporate Services
• Self-Archiving Policy
• Dispatch Dates
• For Reviewers
• Journals on Oxford Academic
• Books on Oxford Academic

Standardizing imaging findings representation: harnessing Common Data Elements semantics and Fast Healthcare Interoperability Resources structures

• Article contents
• Figures & tables
• Supplementary Data

Ali S Tejani, Brian Bialecki, Kevin O’Donnell, Teri Sippel Schmidt, Marc D Kohli, Tarik Alkasab, Standardizing imaging findings representation: harnessing Common Data Elements semantics and Fast Healthcare Interoperability Resources structures, Journal of the American Medical Informatics Association , 2024;, ocae134, https://doi.org/10.1093/jamia/ocae134

• Permissions Icon Permissions

Designing a framework representing radiology results in a standards-based data structure using joint Radiological Society of North America/American College of Radiology Common Data Elements (CDEs) as the semantic labels on standard structures. This allows radiologist-created report data to integrate with artificial intelligence-generated results for use throughout downstream systems.

We developed a framework modeling radiology findings as Health Level 7 (HL7) Fast Healthcare Interoperability Resources (FHIR) observations using CDE set/element identifiers as standardized semantic labels. This framework deploys CDE identifiers to specify radiology findings and attributes, providing consistent labels for radiology report concepts—diagnoses, recommendations, tabular/quantitative data—with built-in integration with RadLex, SNOMED CT, LOINC, and other ontologies. Observation structures fit within larger HL7 FHIR DiagnosticReport resources, providing output including both nuanced text and structured data.

Labeling radiology findings as discrete data for interchange between systems requires two components: structure and semantics. CDE definitions provide semantic identifiers for findings and their component values. The FHIR observation resource specifies a structure for associating identifiers with radiology findings in the context of reports, with CDE-encoded observations referring to definitions for CDE identifiers in a central repository. The discussion includes an example of encoding pulmonary nodules on a chest CT as CDE-labeled observations, demonstrating the application of this framework to exchange findings throughout the imaging workflow, making imaging data available to downstream clinical systems.

CDE-labeled observations establish a lingua franca for encoding, exchanging, and consuming radiology data at the level of individual findings, facilitating use throughout healthcare systems.

CDE-labeled FHIR observation objects can increase the value of radiology results by facilitating their use throughout patient care.

American Medical Informatics Association members

Personal account.

• Sign in with email/username & password
• Get email alerts
• Save searches
• Purchase content
• Activate your purchase/trial code
• Add your ORCID iD

Institutional access

Sign in with a library card.

• Sign in with username/password
• Recommend to your librarian
• Institutional account management
• Get help with access

Access to content on Oxford Academic is often provided through institutional subscriptions and purchases. If you are a member of an institution with an active account, you may be able to access content in one of the following ways:

IP based access

Typically, access is provided across an institutional network to a range of IP addresses. This authentication occurs automatically, and it is not possible to sign out of an IP authenticated account.

Choose this option to get remote access when outside your institution. Shibboleth/Open Athens technology is used to provide single sign-on between your institution’s website and Oxford Academic.

• Click Sign in through your institution.
• Select your institution from the list provided, which will take you to your institution's website to sign in.
• When on the institution site, please use the credentials provided by your institution. Do not use an Oxford Academic personal account.
• Following successful sign in, you will be returned to Oxford Academic.

If your institution is not listed or you cannot sign in to your institution’s website, please contact your librarian or administrator.

Enter your library card number to sign in. If you cannot sign in, please contact your librarian.

Society Members

Society member access to a journal is achieved in one of the following ways:

Sign in through society site

Many societies offer single sign-on between the society website and Oxford Academic. If you see ‘Sign in through society site’ in the sign in pane within a journal:

• Click Sign in through society site.
• When on the society site, please use the credentials provided by that society. Do not use an Oxford Academic personal account.

If you do not have a society account or have forgotten your username or password, please contact your society.

Sign in using a personal account

Some societies use Oxford Academic personal accounts to provide access to their members. See below.

A personal account can be used to get email alerts, save searches, purchase content, and activate subscriptions.

Some societies use Oxford Academic personal accounts to provide access to their members.

Viewing your signed in accounts

Click the account icon in the top right to:

• View your signed in personal account and access account management features.
• View the institutional accounts that are providing access.

Signed in but can't access content

Oxford Academic is home to a wide variety of products. The institutional subscription may not cover the content that you are trying to access. If you believe you should have access to that content, please contact your librarian.

For librarians and administrators, your personal account also provides access to institutional account management. Here you will find options to view and activate subscriptions, manage institutional settings and access options, access usage statistics, and more.

Short-term Access

To purchase short-term access, please sign in to your personal account above.

Don't already have a personal account? Register

Email alerts

Citing articles via.

• Recommend to your Library

Affiliations

• Online ISSN 1527-974X
• Copyright © 2024 American Medical Informatics Association
• About Oxford Academic
• Publish journals with us
• University press partners
• What we publish
• New features  
• Open access
• Rights and permissions
• Accessibility
• Advertising
• Media enquiries
• Oxford University Press
• Oxford Languages
• University of Oxford

Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide

• Copyright © 2024 Oxford University Press
• Cookie settings
• Cookie policy
• Privacy policy
• Legal notice

This Feature Is Available To Subscribers Only

Sign In or Create an Account

This PDF is available to Subscribers Only

For full access to this pdf, sign in to an existing account, or purchase an annual subscription.

This paper is in the following e-collection/theme issue:

Published on 18.6.2024 in Vol 26 (2024)

Monitoring Adverse Drug Events in Web Forums: Evaluation of a Pipeline and Use Case Study

Authors of this article:

Original Paper

• Pierre Karapetiantz 1 , PhD   ; 
• Bissan Audeh 1 , PhD   ; 
• Akram Redjdal 1 , PhD   ; 
• Théophile Tiffet 2, 3 , MD   ; 
• Cédric Bousquet 1, 2 , PhD, PharmD   ; 
• Marie-Christine Jaulent 1 , PhD  

1 Inserm, Sorbonne Université, université Paris 13, Laboratoire d’informatique médicale et d’ingénierie des connaissances en e-santé, LIMICS, F-75006, Paris, France

2 Service de santé publique et information médicale, CHU de Saint Etienne, 42000 Saint-Etienne, France

3 Institut National de la Santé et de la Recherche Médicale, Université Jean Monnet, SAnté INgéniérie BIOlogie St-Etienne, SAINBIOSE, 42270 Saint-Priest-en-Jarez, France

Corresponding Author:

Marie-Christine Jaulent, PhD

Sorbonne Université

université Paris 13, Laboratoire d’informatique médicale et d’ingénierie des connaissances en e-santé, LIMICS, F-75006

15 rue de l'école de Médecine

Paris, 75006

Phone: 33 144279108

Email: [email protected]

Background: To mitigate safety concerns, regulatory agencies must make informed decisions regarding drug usage and adverse drug events (ADEs). The primary pharmacovigilance data stem from spontaneous reports by health care professionals. However, underreporting poses a notable challenge within the current system. Explorations into alternative sources, including electronic patient records and social media, have been undertaken. Nevertheless, social media’s potential remains largely untapped in real-world scenarios.

Objective: The challenge faced by regulatory agencies in using social media is primarily attributed to the absence of suitable tools to support decision makers. An effective tool should enable access to information via a graphical user interface, presenting data in a user-friendly manner rather than in their raw form. This interface should offer various visualization options, empowering users to choose representations that best convey the data and facilitate informed decision-making. Thus, this study aims to assess the potential of integrating social media into pharmacovigilance and enhancing decision-making with this novel data source. To achieve this, our objective was to develop and assess a pipeline that processes data from the extraction of web forum posts to the generation of indicators and alerts within a visual and interactive environment. The goal was to create a user-friendly tool that enables regulatory authorities to make better-informed decisions effectively.

Methods: To enhance pharmacovigilance efforts, we have devised a pipeline comprising 4 distinct modules, each independently editable, aimed at efficiently analyzing health-related French web forums. These modules were (1) web forums’ posts extraction, (2) web forums’ posts annotation, (3) statistics and signal detection algorithm, and (4) a graphical user interface (GUI). We showcase the efficacy of the GUI through an illustrative case study involving the introduction of the new formula of Levothyrox in France. This event led to a surge in reports to the French regulatory authority.

Results: Between January 1, 2017, and February 28, 2021, a total of 2,081,296 posts were extracted from 23 French web forums. These posts contained 437,192 normalized drug-ADE couples, annotated with the Anatomical Therapeutic Chemical (ATC) Classification and Medical Dictionary for Regulatory Activities (MedDRA). The analysis of the Levothyrox new formula revealed a notable pattern. In August 2017, there was a sharp increase in posts related to this medication on social media platforms, which coincided with a substantial uptick in reports submitted by patients to the national regulatory authority during the same period.

Conclusions: We demonstrated that conducting quantitative analysis using the GUI is straightforward and requires no coding. The results aligned with prior research and also offered potential insights into drug-related matters. Our hypothesis received partial confirmation because the final users were not involved in the evaluation process. Further studies, concentrating on ergonomics and the impact on professionals within regulatory agencies, are imperative for future research endeavors. We emphasized the versatility of our approach and the seamless interoperability between different modules over the performance of individual modules. Specifically, the annotation module was integrated early in the development process and could undergo substantial enhancement by leveraging contemporary techniques rooted in the Transformers architecture. Our pipeline holds potential applications in health surveillance by regulatory agencies or pharmaceutical companies, aiding in the identification of safety concerns. Moreover, it could be used by research teams for retrospective analysis of events.

Introduction

Social media as a complementary data source for pharmacovigilance.

One primary mission of regulatory agencies such as the FDA (Food and Drug Administration) or the EMA (European Medicines Agency) is to monitor drug usage and adverse drug events (ADEs) to mitigate the risks associated with drugs within the population. This task entails analyzing diverse data sources, including clinical trials, postmarketing surveillance, spontaneous reporting systems, and published scientific literature. Despite the wealth of available data, some ADEs are not always detected promptly, largely because of underreporting. In France, for instance, underreporting was estimated to range between 78% and 99% from 1997 to 2002 [ 1 ]. To tackle this challenge, several countries have implemented systems allowing patients to report ADEs.

Additional sources for detecting ADEs have been under exploration, such as electronic patient records [ 2 - 4 ] and social media platforms [ 5 - 9 ]. While some argue that social media alone cannot serve as a primary source for signal detection [ 10 ], it can be viewed as a valuable secondary source for monitoring emerging adverse drug reactions or reinforcing signals previously identified through spontaneous reports stored in traditional pharmacovigilance databases [ 11 ]. In a prior study by the authors, patient profiles and reported ADEs found in web forums were compared with those in the French Pharmacovigilance Database (FPVD). The forums tended to represent younger patients, more women, less severe cases, and a higher incidence of psychiatric disorder–related ADEs compared with the FPVD [ 12 ]. Moreover, forums reported a greater number of unexpected ADEs. Over the past decade, several tools for evaluating social media posts have been described in the literature [ 13 ]. Specifically, effective ADE detection in social media necessitates both quantitative and qualitative analyses of data [ 14 ].

Qualitative Approach for Individual Assessment of Posts

Qualitative assessment entails evaluating whether users’ messages contain pertinent information for an assessment akin to a pharmacovigilance case report. This includes details such as the patient’s age and gender, the severity of the case, the expectedness and timeline of the adverse event, time-to-onset, dechallenge (outcome upon drug withdrawal), and rechallenge (outcome upon drug reintroduction). For instance, GlaxoSmithKline Inc. implemented the qualitative approach Insight Explorer, which facilitates the collection of extensive data for causality and quality assessment. Users can input data including personal information (eg, age range, gender) and product details (eg, name, route of administration, duration of use, dosage). This approach was adapted for the WEB-RADR (Recognizing Adverse Drug Reactions) project to manually construct a gold standard of curated patient-authored text [ 15 ].

Quantitative Approach for Monitoring Adverse Drug Events on Social Media

Quantitative evaluation involves analyzing extracted data using descriptive and analytical statistics, such as signal detection and change-point analysis. Numerous projects have been undertaken to monitor ADEs on social media. One of the earliest projects is the PREDOSE (Prescription Drug Abuse Online Surveillance and Epidemiology) project [ 5 ], which investigates the illicit use of pharmaceutical opioids reported in web forums. While the PREDOSE project showcased the potential of leveraging social media for opioid monitoring, notable limitations are the lack of deidentification and signal detection methods. MedWatcher Social, a monitoring platform for health-related web forums, Twitter, and Facebook, represents a prototype application developed in 2014 [ 16 ]. Yeleswarapu et al [ 6 ] outlined a semiautomatic pipeline that applies natural language processing (NLP) tasks to extract ADEs from MEDLINE abstracts and user comments from health-related websites. However, this pipeline was not intended for routine use.

The Domino’s interface [ 17 ], developed in 2018 by the University of Bordeaux in France and funded by the French Medicines Agency (Agence nationale de sécurité du médicament et des produits de santé [ANSM]), was designed to analyze drug misuses in health-related web forums using NLP methods and the summary of product characteristics. Initially tailored for antidepressant drugs, this tool does not primarily focus on ADE surveillance.

Another pipeline, described by Nikfarjam et al in 2019 [ 7 ], used a neural network–based named entity recognition system specifically designed for user-generated content in social media. This platform is dedicated to identifying the association of cutaneous ADEs with cancer therapy drugs. The study focused on a selection of drugs and only examined 8 ADEs.

Magge et al [ 8 ] described a pipeline aimed at the extraction and normalization of adverse drug mentions on Twitter. Their pipeline consisted of an ADE classifier designed to identify tweets mentioning an ADE, which were then mapped to a MedDRA (Medical Dictionary for Regulatory Activities Terminology) code. However, the normalization process was confined to the ADEs present in the training set. Neither Nikfarjam’s nor Magge’s pipeline provides a graphical user interface.

Some private companies also offer tools for analyzing social media for pharmacovigilance purposes. For instance, the DETECT platform was developed as part of a collaborative project in France by Kappa Santé [ 18 ]. This system enabled the labeling of posts with known controlled vocabulary concepts, and signal detection was conducted [ 19 ]. Within the scope of this project, Expert System Company implemented BIOPHARMA Navigator to extract web forum posts, while the Luxid Annotation Server provided web services for the automatic annotation of posts.

An important finding from the studies of the last decade is that while regulatory agencies have begun using data sources beyond spontaneous reports, social media has yet to be fully leveraged in real-world settings due to the immaturity of available solutions. Primarily, these solutions are essentially proofs of concept that lack scalability and are challenging for experts to evaluate routinely, primarily due to the absence of a graphical user interface to present information.

Our aim was to assess the potential of integrating social media into pharmacovigilance and enhancing decision-making with this novel data source. To achieve this, our objective was to develop and assess a pipeline that processes data from the extraction of web forum posts to the generation of indicators and alerts within a visual and interactive environment. The goal was to create a user-friendly tool that enables regulatory authorities to make better-informed decisions effectively.

This article presents the design and implementation of our pipeline dedicated to harnessing posts from social media. In addition, we showcase the use of the pipeline through a specific use case, emphasizing the importance of monitoring drugs in social media to better address patients’ expectations.

The PHARES project (Pharmacovigilance in Social Networks), funded from 2017 to 2019 by the French ANSM, aimed to develop a software suite (a pipeline) enabling pharmacovigilance users to analyze social networks, particularly messages posted on forums. The objective of the pipeline is to facilitate routine use through continuous post extraction and quantitative data analysis from web forums, specifically tailored for the French language.

The pipeline is made up of 4 modules, each referring to its own methods ( Figure 1 ):

The Scraper module, which extracts posts from forums using a previously developed tool, Vigi4Med (V4M) scraper [ 9 ], and produces a comma-separated values (CSV) file filled with the texts extracted.

The Annotation module, which extracts elements of interest from the posts and registers annotations in CSV files, with each line representing an annotation of an ADE or a drug. When a causality relationship is identified, both an ADE and a drug are annotated on the same line.

The Statistical module, which performs quantitative analysis on the annotated posts, generating numerical data, tables, or figures.

The Interface module, which supports query definition and visualization of results.

The methodology used to evaluate the PHARES pipeline involved comparing its performance with existing platforms mentioned above, in accordance with a set of criteria established with prospective PHARES users. The criteria, specific to each module, are as follows:

• General level: focus on ADEs, designed for routine usage.
• Scraper: collects all posts of a selected website, performs deidentification, allows to extract posts from web forums, and is open source.
• Statistics: the temporal evolution of posts or annotations is displayed and a change-point analysis (detecting breakpoints) is possible.
• Signal detection: allows to apply at least one signal detection method, displays the temporal evolution of the proportional reporting ratio (PRR), and allows to perform a logistic regression–based signal detection method.
• Graphical user interface: has an interface for users.

Scraper Module

V4M Scraper is an open-source tool designed for data extraction from web forums [ 9 ]. Its primary functions are optimizing scraping time, filtering out posts primarily focused on advertisements, and structuring the extracted data semantically. The module operates by taking a configuration file as input, which contains the URL of the targeted forum. The algorithm navigates through forum pages and generates resource description framework (RDF) triplets for each extracted element, allowing for potential alignment with external semantic resources. A caching mechanism has been integrated into this tool to maintain a local copy of previously visited pages, thereby avoiding redundant requests to websites for already scraped web pages, particularly in cases of errors or testing, for example. Vigi4Med V4M Scraper was customized for the PHARES project, as indicated by the red elements in Figure S1 in Multimedia Appendix 1 . The database format (Figure S2 in Multimedia Appendix 1 ) was implemented to enhance interaction with the interface. Specifically, the main scraping script was adjusted to produce a simplified tabular format (CSV) of the extracted data and to store these data in a database. This modification aims to facilitate input to the subsequent module of the pipeline (annotation). V4M Scraper was customized to enable a continuous scraping routine, wherein data extracted from web forums are automatically and regularly annotated and registered. A log file was integrated into the scraper structure to maintain a record of the last scraped element. This log file ensures that the daily routine scraping always begins from the last scraped point. An automation tool (crontab) is used to schedule the execution of the pipeline for each forum on a daily basis at a specific time.

A total of 23 public French health-related web forums were selected through a combination of Google searches and from a list of certified health websites provided by the HON Foundation, in collaboration with the French National Health Authority (HAS). The selection criteria included the requirement for websites to be hosted in France, feature a discussion board or space for sharing experiences, and have more than 10 patient contributions. Furthermore, Twitter posts are collected and analyzed by the pipeline. This is achieved using the Twitter API for data collection, followed by employing the same modules used for processing web forum posts.

Annotation Module

Entities corresponding to drugs and pathological conditions in social media were identified and annotated using an NLP pipeline [ 20 ]. Initially, conditional random fields were used to account for global dependencies [ 21 ]. Specifically, the model considers the entire sequence when making predictions for individual tokens. This approach is advantageous for entity extraction tasks, as the presence of an entity in one part of the text can influence the likelihood of other entities in the vicinity. Second, a support vector machine is used to predict the causality relationship between an entity identified as a drug and another entity identified as an ADE. The annotation method used in this module was implemented at an early stage of the pipeline’s design. Currently, the named entity recognition task of this module is undergoing revision to incorporate more recent advancements in NLP algorithms [ 22 - 26 ].

In a third step, the detected annotations were normalized using codes from the MedDRA and the Anatomical Therapeutic Classification (ATC) to ensure they were suitable for signal detection purposes.

MedDRA is an international medical hierarchical terminology comprising 5 levels used to code potential ADEs in pharmacovigilance. The highest level is the system organ class, which is further divided into high-level group terms, then into high-level terms, preferred terms (PTs), and finally lowest level terms. Typically, the PT level is used in pharmacovigilance signal detection.

The ATC classification system is a drug classification used in France for pharmacovigilance purposes. It categorizes the active ingredients of drugs based on the organ system they primarily affect. The classification comprises 5 levels: the anatomical main group (consisting of 14 main groups), the therapeutic subgroup, the therapeutic/pharmacological subgroup, the chemical/therapeutic/pharmacological subgroup, and the chemical substance. Typically, the fifth level (chemical substance) is used in pharmacovigilance signal detection.

The outputs of the annotation module are CSV files with the following variables:

• Concerning the post: forum name, post ID, and date
• Concerning the ADE: verbatim, normalized term, unified medical language system’s concept unique identifier, and MedDRA code
• Concerning the drug: verbatim, normalized term, active ingredient, and ATC code

In these CSV files, each line can consist of either an adverse event (ADE) annotation, a drug annotation, or both when a causality relationship has been identified between the drug and the ADE. Table 1 provides a sample of the database.

In a prior study, we selected posts where at least one ADE associated with 6 drugs (agomelatine, baclofen, duloxetine, exenatide, strontium ranelate, and tetrazepam) had been detected by this algorithm. A manual review revealed that among 5149 posts, 1284 (24.94%) were validated as pharmacovigilance cases [ 12 ]. The fundamental metrics used to assess the performance of the annotation module were precision (P), recall (R), and their harmonic mean F 1 -score. To calculate these metrics, it is necessary to evaluate false negatives for nonrecognition of relevant terms, false positives for irrelevant recognitions, and true positives for correct recognitions. Precision, recall, and F 1 -score are defined as follows:

Precision = (true positive)/(true positive + false positive); recall = (true positive)/(true positive + false negative); F 1 -score = (2 × precision × recall)/(precision + recall) (1)

In the “Results” section, we present a comparison of the performance of the annotation module with the performance of state-of-the-art methods [ 8 , 22 , 25 , 26 ].

a ADE: adverse event.

b MedDRA: Medical Dictionary for Regulatory Activities Terminology.

c ATC: Anatomical Therapeutic Classification.

d No data are available for this slot.

Statistical Module

This module generates general statistics and diagrams for web forums or Twitter. It provides data such as the number of annotated posts (related to the drug, the ADE, or both), the count of drug-ADE pairs identified, and the distribution of ADEs’ MedDRA-PTs. In addition, a change-point analysis method was used to detect significant changes over time in the mean number of posts mentioning the drug and ADE [ 27 ].

Besides, several statistical signal detection methods were implemented to generate potential signals. Safety signals, which provide information on adverse events that may potentially be caused by a medicine, were further evaluated by pharmacovigilance experts to determine the causal relationship between the medicine and the reported adverse event.

The statistical module implements 3 signal detection methods, including 2 well-known and frequently used disproportionality signal detection methods: the PRR [ 28 ] and the reporting odds ratio (ROR) [ 29 ]. In addition, a complementary method, a logistic regression–based signal detection method known as the class imbalanced subsampling lasso [ 30 ], was used.

PRR and ROR are akin to a relative risk and an odds ratio, respectively. However, they differ in their denominators: as the number of exposed patients is typically unknown in pharmacovigilance databases, the denominator in PRR and ROR calculations is the number of cases reported in the pharmacovigilance database.

PRR and ROR are specific to each drug-ADE pair and can be directly computed from the contingency table ( Table 2 ).

The PRR compares the proportion of an ADE among all the ADEs reported for a specific drug with the same proportion for all other drugs in the database (Equation 2). A PRR significantly greater than 1 suggests that the ADE is more frequently reported for patients taking the drug of interest, while a PRR equal to 1 suggests independence between the 2 variables.

PRR = [a/(a + b)]/[c/(c + d)] (2)

The ROR quantifies the strength of the association between drug administration and the occurrence of the ADE. It represents the ratio of the odds of drug administration when the ADE is present to the odds of drug administration when the ADE is absent (Equation 3). When the 2 events are independent, the ROR equals 1. An ROR significantly greater than 1 suggests that drug administration is associated with the presence of the ADE.

ROR = ad / bc (3)

We considered events over posts for the calculation of disproportionality statistics. If the same drug-ADE pair was identified multiple times within a post, the pair was counted as many times as it occurred in the calculation.

Disproportionality analysis has certain limitations, including the confounding effect resulting from coreported drugs and the masking effect, where the background relative reporting rate of an ADE is distorted by extensive reporting on the ADE with a specific drug or drug group. Caster et al [ 31 ] demonstrated through 2 real case examples how multivariate regression–based approaches can address these issues. Harpaz et al also suggested that logistic regression could be used for safety surveillance [ 32 ]. Initially designed for pharmacovigilance case reports, we hypothesize that they may also be applicable to posts.

The logistic regression model specifically focuses on a particular ADE or a group of ADEs. It involves creating a vector that represents the presence (1) or absence (0) of the ADE of interest in the pharmacovigilance case (in our case, in the post). Additionally, a matrix is generated to represent the administration or nonadministration of all drugs in the database by the patient (1 for administration and 0 for nonadministration). Figure S3 in Multimedia Appendix 1 illustrates an example of using logistic regression. In our case, we assumed that if a drug was annotated in the post, it was taken by the patient. The logistic regression aims to predict the probability of the presence of the ADE (ADE=1) of interest based on the presence of all ( N m ) drugs in the database (Equation 4), where X represents the distribution of the presence/absence of the drugs. The adjusted factors included only concomitant medications, as patient-related factors are often missing in web forums’ posts. Therefore, we did not need to address the impact of missing data, which should be evaluated when necessary.

ln([P(X|ADE=1)]/[P(X|ADE=0)]) = a + b1 × Drug1 + ... + bi × Drug i + .. . + bNm × Drug Nm (4)

The selection of the drugs depends on the parameter b i . If b i <0, the drug i decreases the risk of the ADE, and if b i >0, the drug i increases the risk of the ADE.

Then, 2 sets are defined:

• S 1 : set of n 1 posts with an annotation of the ADEs of interest.
• S 0 : set of n 0 posts without an annotation of the ADEs of interest.

In our case n 0 >> n 1 , indicating a significant imbalance toward posts lacking annotations of the ADEs of interest. To address this issue, we took a subsample with a more favorable ratio of posts with annotated ADEs versus those without. Additionally, to enhance result stability, we conducted multiple draws instead of just one.

In practice, we generated B subsamples. Each subsample was constructed by randomly drawing, with replacement, n 1 posts from S 1 and R posts from S 0 , where R=max(4 n 1 , 4 N m ). The choice of 4 n 1 was inspired by case-control studies, while 4 N m was included to ensure an adequate number of observations considering the multitude of predictors.

We implemented a change-point analysis method described in [ 27 ] to detect whether there was a change in the evolution over time of a chosen statistic, such as the number of a specific drug-ADE pair, the number of ADEs associated with a specific drug, or the number of drugs associated with a specific ADE. The method uses the Cumulative Sum (CUSUM) algorithm to analyze the evolution of statistics over time, comparing current values with the period mean. It identifies breakpoints by calculating the highest difference in statistical values and comparing it with random samples. The process repeats for periods before and after detected breakpoints until no more are found.

User Interface Module

The user interface module facilitates user interaction with the pipeline in a user-friendly manner. The interface comprises a dashboard divided into 2 main parts. The left dark column ( Figure 2 ) serves as a control sidebar, where users can select parameters to filter the data, including the forum, period, drug(s) according to the ATC classification, and ADE(s) according to a level in the MedDRA hierarchy. On the right side of the interface, various visualizations are available, organized into several tabs such as “Forum Statistics” and “Consultation of Posts,” with additional tabs for statistics that become active upon querying.

Before applying a specific query, the interface provides general information about the currently available data ( Figure 2 ), including the total annotated posts since 2017 (n=2,081,296) and total annotations since 2017 (n=2,454,310). In addition, a “Consultation of Tweets” tab (not visible in the figure) displays the total annotated tweets since March 2020 (n=46,153).

Furthermore, several tabs corresponding to different types of statistics, including “Forums Statistics” and “Twitter Statistics,” provide general statistics and diagrams for web forums and Twitter. Examples of these are pie charts showing forum distribution, line charts depicting the evolution of drug and ADE mentions, histograms displaying ADE distribution by system organ class, and line charts illustrating the temporal trend of posts containing the drug and an ADE, as shown in Figures 3 and 4 . The “Annotations Plot” tab displays annotations of drugs and adverse effects selected by the user, along with forum information, PTs, high-level terms, high-level group terms, dates, and hours. The “Logistic Regression” tab allows users to choose parameters for applying logistic regression. In the “Disproportionality” tab, users can choose between the PRR and ROR methods, with the time evolution of the chosen method displayed. The “Change-Point” tab enables analysis of temporal evolution, with identified breakpoints indicated. The “Consultation of Posts” and “Consultation of Tweets” tabs provide details on annotated posts/tweets, including downloadable tables. The statistical module performs calculations based on user queries, updating the interface accordingly. If multiple drugs or adverse events are selected, they are treated as new entities for analysis.

The interface was implemented using the R language and environment (R Foundation) for statistical computing and graphics [ 33 ], leveraging the Shiny package [ 34 ] for development.

Ethical Considerations

A statement by an Institutional Review Board was not required because we used only publicly available data that do not necessitate Institutional Review Board review.

This study complied with the European General Data Protection Regulation (GDPR), which has been in force since 2018 in Europe [ 35 ]. The GDPR enhances the protection of individuals by introducing the right to be informed about the processing of personal data. However, informing each user individually may be impractical. Therefore, the GDPR introduces 2 legal conditions where informed consent is not mandatory, which can be interpreted as supporting the processing of web forum posts for pharmacovigilance (Article 9): “(e) processing relates to personal data which are manifestly made public by the data subject; [. . .] (i) processing is necessary for reasons of public interest in the area of public health, such as [. . .] ensuring high standards of quality and safety of health care and of medicinal products . . ..” The GDPR also requires data processing to “not permit or no longer permits the identification of data subjects” (Article 89). Deidentification was conducted during the extraction of posts from web forums to ensure privacy [ 9 ]. User identifiers in the main RDF file were encrypted using the SHA1 algorithm [ 36 ]. The correspondence between these encrypted identifiers and the original keys is presented in RDF triplets in a separate file, referred to as the “keys file.” Therefore, the only way to retrieve the original authors’ identities is by concatenating the main RDF containing the encrypted data with the keys file, which is kept in a secured location. Moreover, all our data processing was carried out on a secured server with restricted access.

General Results About the Pipeline

The primary outcome of this study is the operational PHARES pipeline itself. Daily extraction and annotation of posts are initiated and imported into the database linked to the user interface. In this paper, the platform’s use will be demonstrated through a specific use case on the analysis of Levothyrox ADE mentions in forums (discussed later). In addition, we conducted a comparative analysis of the PHARES pipeline with the existing platforms mentioned in the “Introduction” section, based on the criteria listed in the “Methods” section.

Of the 10 identified pipelines, half were public and half were private. While 8 out of 10 focused on ADEs, only 4 were designed for routine usage. Five scrapers were open source, and all posts from considered websites were extracted by only 6 of the scrapers (with others extracting posts under certain conditions). Six scraped web forum posts, but only 3 performed deidentification. Additionally, 4 pipelines focused on the French language. A total of 6 pipelines displayed the temporal evolution of the number of posts, but only 1 conducted a change-point analysis. Signal detection methods were performed by only 4 of them, with none displaying the temporal evolution of the PRR nor a logistic regression–based method. Finally, 6 of them had an interface ( Table 3 ).

a PHARES: Pharmacovigilance in Social Networks.

b The X symbol means that the characteristic is missing and the symbol ✓ means the characteristic is fulfilled.

c ADE: adverse drug event.

d PRR: proportional reporting ratio.

e PREDOSE: Prescription Drug Abuse Online Surveillance and Epidemiology.

f ADR-PRISM: Adverse Drug Reaction from Patient Reports in Social Media.

Annotation Module’s Comparison With Up-to-Date State-of-the-Art Methods

We also compared the performance of our annotation process with those of up-to-date state-of-the-art methods ( Table 4 ).

While the annotation module demonstrated good performance for named entity recognition ( F 1 -score=0.886), it remains slightly below the state of the art. Presently, in medical texts, the best performances are achieved by Hussain et al [ 25 ] and Ding et al [ 26 ] for the named entity recognition task, and by Xia [ 22 ] for the relationship extraction task. On Twitter, known for its notably more complex data, Hussain et al [ 25 ] achieved slightly better results than our annotator, while Ding et al [ 26 ] achieved slightly worse results.

a The 2 categories are entity recognition, which is the detection of a drug or ADE mention, and relationship extraction, which is the detection of a relation between a drug and an ADE.

b PHARES: Pharmacovigilance in Social Networks.

c BERT: Bidirectional Encoder Representations from Transformer.

d Not available.

e HAMLE: Historical Awareness Multi-Level Embedding.

f BGRU: Bidirectional Gated Recurrent Unit.

g LSTM: Long-Short-Term-Memory.

Summary of the Result

From January 1, 2017, to February 28, 2021, a total of 2,081,296 posts were extracted from 23 French web forums ( Table 5 ). We obtained 713,057 normalized annotations of drugs, 1,527,004 normalized annotations of ADEs, and 437,192 annotations of normalized drug-ADE couples. The number of posts annotated with at least one normalized drug-ADE couple was equal to 125,279 (6.02%). Table 4 summarizes the number of posts extracted per forum, the publication dates, and the description of the web forum. For 1 forum, the publication dates were not available. A total of 9 were generalist health forums, 3 were specialized for parents of a young baby, 2 for families, 3 for mothers, 2 specialized in thyroid issues, 1 for pregnant women, 1 for women, 1 for parents of a teenager or for teenagers, 1 for sports persons, and 1 specialized in rare diseases.

a Not available.

Use Case: Analysis of Levothyrox ADE Mentions in Forums

To demonstrate the usage of the pipeline, we chose to focus on Levothyrox as a case study. Levothyrox is a drug prescribed in France since 1980 for hypothyroidism and circumstances where it is necessary to limit the thyroid-stimulating hormone. In 2017, a new formula of Levothyrox, differing from the 30-year-old drug at the excipient level (with lactose being replaced by mannitol and citric acid in the new formula), was marketed with widespread media coverage. In parallel, an unexpected increase in notifications of ADEs for this drug was detected. Viard et al [ 37 ] were unable to find any pharmacological rationale to explain that signal. Approximately 32,000 adverse effects were reported by patients in France in 2017, representing 42% of all the ADEs collected yearly [ 38 ]. Most of these notifications concerned the new formulation of Levothyrox and led to the “French Levothyrox crisis.” In 2017, 1664 notifications of ADEs were spontaneously reported by patients to the Pharmacovigilance Center of Nice. Among the 1544 reviewed notifications, 1372 concerned Levothyrox while only 172 concerned other drugs [ 37 ].

In this use case, the study period was from January 1, 2017, to February 28, 2021, and the drugs included were 2 drugs from the “H03AA Thyroid hormones” ATC class: “Levothyroxine sodium” and “associations of levothyroxine and liothyronine.” A total of 17 forums were selected as they included at least one post with information about these drugs. Posts were extracted, annotated, and analyzed through the pipeline from several forums ( Table 6 ). Signal detection methods were applied to an ADE chosen as it frequently appeared with Levothyrox in our data: “tiredness.” A signal can be detected when the lower bound of the 95% CI of the logarithm of the PRR is greater than 0. For logistic regression, we applied the tenth quantile. A total of 11,340 posts contained an annotation concerning the drugs of interest. Figure S4 in Multimedia Appendix 1 illustrates the source and evolution over time of these posts. Out of a total of 50,127 annotations of Levothyrox, they principally originated from the Vivre sans thyroïde forum and were mostly posted in mid-2017 ( Figure 4 , Table 6 ). The results of the statistical analysis were displayed by the user interface.

ADEs annotated with Levothyrox were mainly from system organ classes: general disorders and administration site conditions (29.6%), metabolism and nutrition disorders (11.6%), and endocrine disorders (11.4%). The PTs mostly found in association with Levothyrox are listed in Table 7 . All this information is accessible in the interface module (Figure S5 in Multimedia Appendix 1 ).

We chose the PT “tiredness” for the signal detection analysis. A total of 85,976 posts were annotated with either one of the drugs of interest or the ADE tiredness. Among them, 1841 Levothyrox-tiredness couples were found, mostly in 2017 ( Table 7 ).

Figure 5 illustrates the time evolution of the PRR for the Levothyrox-tiredness couple. Figure S6 in Multimedia Appendix 1 displays the source and evolution over time of French web forums’ posts for this couple. A signal is consistently generated throughout the period as the logarithm of the PRR is always greater than 0.

A total of 11 drugs were found to be associated with tiredness using logistic regression: paclitaxel, pegfilgrastim, Levothyrox, glatiramer acetate, escitalopram ferrous sulfate, the combination of Levothyrox and liothyronine, secukinumab, methotrexate, bismuth potassium, tetracycline, and metronidazole.

Change-point analysis was conducted on the monthly evolution of the number of Levothyrox-ADE couples detected in web forums. Six breakpoints were identified ( Figure 6 ), and 3 of them correlated with an increase in the number of ADEs found with Levothyrox on web forums. These increases occurred in August 2017 and in September and December 2018.

This use case demonstrates that the results obtained through the pipeline, particularly in the context of Levothyrox, align with findings in the literature derived from more traditional data sources such as case reports in pharmacovigilance (see the “Discussion” section). It underscores the potential of leveraging such a pipeline to monitor a drug, not only retrospectively but also in real time using social media. Consequently, PHARES has the capability to potentially uncover new signals in pharmacovigilance.

Principal Findings

To align with our objective, we implemented and evaluated a pipeline that processes data from the extraction of web forum posts to the generation of indicators and alerts within a visual and interactive environment. Through this pipeline, we demonstrated that quantitative analysis can be conducted through the interface without requiring the user to code. We discovered the feasibility of acquiring information akin to the literature regarding a drug’s ADEs, as well as unexpected ADEs and significant event dates related to a drug. This underscores the relevance and utility of such a pipeline.

A conceptual contribution of this research was the proposal of a methodology for designing a pipeline to facilitate pharmacovigilance studies on web forums. This involved describing 4 independent modules and outlining their interactions. Additionally, another contribution was the adaptation of certain pharmacovigilance analysis methods for the examination of data extracted from web forum posts. The logistic regression–based method presented in this article was originally tailored for pharmacovigilance cases to consider co-prescriptions of drugs. We have adapted it to suit the analysis of pharmacovigilance data extracted from web forum posts.

Comparison With Prior Work

The PHARES pipeline offers added value compared with previous pipelines in terms of the criteria set, which reflects an analysis of experts’ needs for routine monitoring of ADEs in social media. Unlike previous approaches, the scrapers used in PHARES routinely perform deidentification, and the inclusion of change-point analysis, the evolution of PRRs over time, and a logistic regression–based signal detection method were previously unavailable. The temporal evolution of the number of posts and a signal detection method are also seldom supported. Designed for routine usage and focused on ADEs, all posts from selected web forums are scraped and deidentified using an open-source scraper.

The period and selected web forums differed between both studies: Audeh et al [ 38 ] covered the period from January 2015 to December 2017, while our study spanned from January 2017 to February 2021. Additionally, Audeh et al [ 38 ] included only 1 web forum specialized in thyroid issues, whereas we incorporated this specific forum along with 16 others. The main ADEs associated with Levothyrox in our study align with those found by Audeh et al [ 38 ] on similar data, albeit without using the interface. In our study, the 10 most frequent symptoms were pain, tiredness, faintness, hypothyroidism, dizziness, insomnia, palpitations, hyperthyroidism, malignant tumor, and anxiety. By contrast, Audeh et al [ 38 ] reported tiredness, weight gain, pain, ganglions, hot flush, chilly, inflammation, faintness, weight loss, and discomfort.

Furthermore, the PHARES pipeline surpasses previous efforts, particularly regarding several criteria. These include the annotation tool, where only 4 pipelines were identified using a French annotator tool. In terms of available statistics, only 1 pipeline met both criteria we identified. Regarding signal detection, among the 3 criteria identified, 5 pipelines matched with only 1, while the remaining 5 matched with none.

In the use case, a notable increase in the number of ADEs associated with Levothyrox was detected using the change-point analysis method a few months after the introduction of the new formula in March 2017, specifically in August 2017. This surge coincided with the initial declaration to the pharmacovigilance network and a petition initiated by patients to reintroduce the former formula in June 2017. We compared these findings with results from a pharmacovigilance study based on spontaneous reporting. Out of 1554 notifications spontaneously addressed by patients to the Pharmacovigilance Center of Nice from January 1, 2017, to December 31, 2017, 1372 were related to the new formula of Levothyrox, representing 7342 ADEs. Our comparison with these data clarified our findings. The 10 most frequently reported ADEs in these notifications closely resembled our own results [ 37 ]. These were asthenia, headache, dizziness, hair loss, insomnia, cramps, weight gain, nausea, muscle pain, and irritability. Consequently, our results demonstrate coherence with the existing literature. This study illustrates the feasibility of identifying the date of significant events related to a drug. However, it is noteworthy that the detection of such events is not necessarily expedited through social media compared with the traditional pharmacovigilance system.

Limitations

The method used in our annotation process was integrated at an early stage during the pipeline’s design. Regarding the identification of drugs and symptoms, our annotation process exhibited the following performances: precision=0.926, recall=0.845, and F 1 -score=0.886 [ 20 ]. Similarly, for discerning the relationship between the drug and the ADEs, the performances were precision=0.683, recall=0.956, and F 1 -score=0.797 [ 20 ]. This study marked the inaugural publication on using NLP methods to identify ADEs in French-language web forums. The annotation process was thus developed using contemporary state-of-the-art methodologies at the time. However, it would now stand to gain from the integration of more recent NLP algorithms for named entity recognition [ 8 , 23 , 24 ]. These newer algorithms offer comparable performances while effectively handling more complex data, thereby enhancing the efficacy of NLP analysis. However, because of our emphasis on the genericity of the approach and the interoperability between the different modules rather than solely focusing on the performance of each module, we opted not to use these algorithms. Nevertheless, contemporary state-of-the-art methods for annotating ADEs from social media posts encompass convolutional neural networks trained on top of pretrained word vectors for sentence-level classification [ 24 ] and transformers using the bidirectional encoder representations from transformers (BERT) language model [ 39 ]. Hussain et al [ 25 ] introduced a multitask neural network based on BERT with hyperparameter optimization capable of sentence classification and named entity recognition. This model achieved performances of precision=0.840, recall=0.861, and F 1 -score=0.896 on the Twitter (X)-TwiMed data set. Additionally, Magge et al [ 8 ] presented a pipeline consisting of 3 BERT neural networks designed to classify sentences, extract named entities, and normalize those entities to their respective MedDRA concepts. The performances of this model were as follows: precision=0.82, recall=0.76, and F 1 -score=0.78 on the SMM4H-2020 data set (Twitter/X). Thanks to our modular design, it will be straightforward to substitute our current annotation process with an enhanced model in the future.

Several limitations should be acknowledged for future work. First, the scraper relies on the HTML structure of web forums, necessitating updates to its configuration files if a forum alters its page design. Additionally, our interface lacks the capability to incorporate alternate identifiers for drugs or ADEs. For instance, patients may commonly refer to the drug “baclofen” as “baclo” on social media platforms. Consequently, the number of posts pertaining to a drug or ADE could potentially be underestimated.

Forums must be selected before query execution to mitigate calculation time. However, selecting forums based on the presence of information related to a particular drug or ADE can introduce bias into signal detection methods, particularly in disproportionality analysis, where the drug-ADE pair may be overrepresented. Another limitation in qualitative analysis of posts is the inability of users to edit annotations or record typical pharmacovigilance qualitative data.

The assumption that all drugs mentioned in a post were consumed simultaneously by the user, as applied in the logistic regression–based method, introduces an evident bias.

One limitation associated with the use of social media data pertains to fraudulent posts. The pseudonymity inherent in these platforms provides malevolent individuals with the opportunity to disseminate false rumors. Additionally, patients might post identical or similar messages across multiple discussion boards, or even multiple times on the same board. Thus, it is crucial to consider these factors to mitigate biases in signal detection.

Perspectives

In the short to medium term, our objectives are updating the annotation module to enhance accuracy, improving the qualitative analysis by enabling users to edit and correct annotations, and expanding the range of signal detection methods available in the statistics module.

This method could indeed be beneficial for identifying potential drug misuse and unknown ADEs [ 40 ]. By categorizing pathological terms found in web forums based on their presence in the summary of product characteristics, we can distinguish between indications, known ADEs, and potential instances of drug misuse or unexpected ADEs. However, it is important to note that considering all pathological terms found in the summary of product characteristics as indications might obscure cases of drug inefficiency. Therefore, a nuanced approach is necessary to ensure comprehensive and accurate analysis.

We next tested our pipeline from the perspective of end users. However, the hypothesis was only partially confirmed, indicating the need for further studies. These studies should include evaluations with ergonomic criteria.

In the long term, our vision is to expand this tool to encompass other languages and themes beyond pharmacovigilance. This includes areas such as drug misuse, the consumption of food supplements, and the use of illegal drugs. French web forums dedicated to recreational drug use already exist, providing a valuable source of data for such endeavors.

Conclusions

Our hypothesis focused on the challenge encountered by regulatory agencies in using social media, primarily because of the lack of appropriate decision-making tools. To tackle this challenge, we devised a pipeline consisting of 4 editable modules aimed at effectively analyzing health-related French web forums for pharmacovigilance purposes. Using this pipeline and its user-friendly interface, we successfully demonstrated the feasibility of conducting quantitative analyses without the need for coding. This approach yielded coherent results and holds the potential to reveal new insights about drugs.

A practical implication of our pipeline is its potential application in health surveillance by regulatory agencies such as the ANSM or pharmaceutical companies. It can be instrumental in detecting issues related to drug safety and efficacy in real time. Furthermore, research teams can leverage this tool to retrospectively analyze events and gain valuable insights into pharmacovigilance trends.

Acknowledgments

The annotation module was developed by François Morlane-Hondère, Cyril Grouin, Pierre Zweigenbaum, and Leonardo Campillos-Llanos from the Computer Science Laboratory for Mechanics and Engineering Sciences (LIMSI). Code review for the graphical user interface in R language was performed by Stevenn Volant under a contract with the Stat4Decision company. Stat4Decision was not involved in designing the study and writing this article. This work was funded by the Agence nationale de sécurité du médicament et des produits de santé (ANSM) through Convention No. 2016S076 and was supported by a PhD contract with Sorbonne Université.

Data Availability

Our data were extracted from web forums that do not allow data sharing. Thus, as we are not the owners of the data we cannot make the data available. The scrapper we developed to extract these data is open source and can be used to extract data from web forum posts. The tool as well as full documentation (in English and French) of the code and configuration file are available online [ 41 ].

Conflicts of Interest

None declared.

Vigi4Med Scraper structure, PHARES database structure, example of data representation, and source and evolution over time of web forum posts. PHARES: Pharmacovigilance in Social Networks.

• Hazell L, Shakir SAW. Under-reporting of adverse drug reactions : a systematic review. Drug Saf. 2006;29(5):385-396. [ CrossRef ] [ Medline ]
• Liu F, Jagannatha A, Yu H. Towards drug safety surveillance and pharmacovigilance: current progress in detecting medication and adverse drug events from electronic health records. Drug Saf. Jan 2019;42(1):95-97. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Warrer P, Hansen EH, Juhl-Jensen L, Aagaard L. Using text-mining techniques in electronic patient records to identify ADRs from medicine use. Br J Clin Pharmacol. May 2012;73(5):674-684. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Black C, Tagiyeva‐Milne N, Helms P, Moir D. Pharmacovigilance in children: detecting adverse drug reactions in routine electronic healthcare records. A systematic review. Brit J Clinical Pharma. May 28, 2015;80(4):844-854. [ CrossRef ] [ Medline ]
• Cameron D, Smith GA, Daniulaityte R, Sheth AP, Dave D, Chen L, et al. PREDOSE: a semantic web platform for drug abuse epidemiology using social media. Journal of Biomedical Informatics. Dec 2013;46(6):985-997. [ CrossRef ] [ Medline ]
• Yeleswarapu S, Rao A, Joseph T, Saipradeep VG, Srinivasan R. A pipeline to extract drug-adverse event pairs from multiple data sources. BMC Med Inform Decis Mak. Feb 24, 2014;14(1):1-16. [ CrossRef ]
• Nikfarjam A, Ransohoff JD, Callahan A, Jones E, Loew B, Kwong BY, et al. Early detection of adverse drug reactions in social health networks: a natural language processing pipeline for signal detection. JMIR Public Health Surveill. Jun 03, 2019;5(2):e11264. [ CrossRef ] [ Medline ]
• Magge A, Tutubalina E, Miftahutdinov Z, Alimova I, Dirkson A, Verberne S. DeepADEMiner: a deep learning pharmacovigilance pipeline for extraction and normalization of adverse drug event mentions on Twitter. J Am Med Inform Assoc. Sep 18, 2021;28(10):2184-2192. [ CrossRef ] [ Medline ]
• Audeh B, Beigbeder M, Zimmermann A, Jaillon P, Bousquet C. Vigi4Med scraper: a framework for web forum structured data extraction and semantic representation. PLoS One. Jan 25, 2017;12(1):e0169658. [ CrossRef ] [ Medline ]
• Caster O, Dietrich J, Kürzinger ML, Lerch M, Maskell S, Norén GN, et al. Assessment of the utility of social media for broad-ranging statistical signal detection in pharmacovigilance: results from the WEB-RADR project. Drug Saf. Dec 2018;41(12):1355-1369. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Bousquet C, Audeh B, Bellet F, Lillo-Le Louët A. Comment on "Assessment of the utility of social media for broad-ranging statistical signal detection in pharmacovigilance: results from the WEB-RADR project". Drug Saf. Dec 19, 2018;41(12):1371-1373. [ CrossRef ] [ Medline ]
• Karapetiantz P, Bellet F, Audeh B, Lardon J, Leprovost D, Aboukhamis R, et al. Descriptions of adverse drug reactions are less informative in forums than in the French pharmacovigilance database but provide more unexpected reactions. Front Pharmacol. May 1, 2018;9:439-411. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Lardon J, Abdellaoui R, Bellet F, Asfari H, Souvignet J, Texier N, et al. Adverse drug reaction identification and extraction in social media: a scoping review. J Med Internet Res. Jul 10, 2015;17(7):e171. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Karapetiantz P, Audeh B, Faille J, Lillo-Le Louët A, Bousquet C. Qualitative and quantitative analysis of web forums for adverse events detection: "strontium ranelate" case study. Stud Health Technol Inform. Aug 21, 2019;264:964-968. [ CrossRef ] [ Medline ]
• Casperson T, Painter J, Dietrich J. Strategies for distributed curation of social media data for safety and pharmacovigilance. 2016. Presented at: International Conference on Data Science (ICDATA); October 1, 2016:118-124; Barcelona, Spain.
• Freifeld CC. Digital pharmacovigilance: The medwatcher system for monitoring adverse events through automated processing of internet social media and crowdsourcing. OpenBU Libraries. Boston University. OpenBU; 2014. URL: https://open.bu.edu/handle/2144/10995
• Cossin S, Lebrun L, Lobre G, Loustau R, Jouhet V, Griffier R, et al. Romedi: an open data source about French drugs on the semantic web. Stud Health Technol Inform. Aug 21, 2019;264:79-82. [ CrossRef ] [ Medline ]
• Abdellaoui R, Schück S, Texier N, Burgun A. Filtering entities to optimize identification of adverse drug reaction from social media: how can the number of words between entities in the messages help? JMIR Public Health Surveill. Jun 22, 2017;3(2):e36. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Bousquet C, Dahamna B, Guillemin-Lanne S, Darmoni SJ, Faviez C, Huot C, et al. The adverse drug reactions from patient reports in social media project: five major challenges to overcome to operationalize analysis and efficiently support pharmacovigilance process. JMIR Res Protoc. Sep 21, 2017;6(9):e179. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Morlane-Hondère F, Grouin C, Zweigenbaum P. Identification of drug-related medical conditions in social media. 2016. Presented at: The Tenth International Conference on Language Resources and Evaluation (LREC'16); May 2, 2016:2022-2028; Portoroz, Slovenia.
• Lafferty J, McCallum A, Pereira F. Conditional random fields: probabilistic models for segmenting and labeling sequence data. San Francisco, CA. Morgan Kaufmann Publishers; 2001. Presented at: Eighteenth International Conference on Machine Learning (ICML 2001); June 28, 2001 to July 1, 2001:282-289; Williamstown, MA.
• Xia L. Historical profile will tell? A deep learning-based multi-level embedding framework for adverse drug event detection and extraction. Decision Support Systems. Sep 2022;160:113832. [ CrossRef ]
• Yu D, Vydiswaran VGV. An assessment of mentions of adverse drug events on social media with natural language processing: model development and analysis. JMIR Med Inform. Sep 28, 2022;10(9):e38140. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Rezaei Z, Ebrahimpour-Komleh H, Eslami B, Chavoshinejad R, Totonchi M. Adverse drug reaction detection in social media by deep learning methods. Cell J. Oct 2020;22(3):319-324. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Hussain S, Afzal H, Saeed R, Iltaf N, Umair MY. Pharmacovigilance with transformers: a framework to detect adverse drug reactions using BERT fine-tuned with farm. Comput Math Methods Med. 2021;2021:5589829. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Ding P, Zhou X, Zhang X, Wang J, Lei Z. An attentive neural sequence labeling model for adverse drug reactions mentions extraction. IEEE Access. 2018;6:73305-73315. [ CrossRef ]
• Xu Z, Kass-Hout T, Anderson-Smits C, Gray G. Signal detection using change point analysis in postmarket surveillance. Pharmacoepidemiol Drug Saf. Jun 22, 2015;24(6):663-668. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Evans SJW, Waller PC, Davis S. Use of proportional reporting ratios (PRRs) for signal generation from spontaneous adverse drug reaction reports. Pharmacoepidemiol Drug Saf. Dec 10, 2001;10(6):483-486. [ CrossRef ] [ Medline ]
• van Puijenbroek EP, Bate A, Leufkens HGM, Lindquist M, Orre R, Egberts ACG. A comparison of measures of disproportionality for signal detection in spontaneous reporting systems for adverse drug reactions. Pharmacoepidemiol Drug Saf. Feb 06, 2002;11(1):3-10. [ CrossRef ] [ Medline ]
• Ahmed I, Pariente A, Tubert-Bitter P. Class-imbalanced subsampling lasso algorithm for discovering adverse drug reactions. Stat Methods Med Res. Mar 25, 2018;27(3):785-797. [ CrossRef ] [ Medline ]
• Caster O, Norén GN, Madigan D, Bate A. Large‐scale regression‐based pattern discovery: the example of screening the WHO global drug safety database. Statistical Analysis. Jul 20, 2010;3(4):197-208. [ CrossRef ]
• Harpaz R, DuMouchel W, LePendu P, Bauer-Mehren A, Ryan P, Shah NH. Performance of pharmacovigilance signal-detection algorithms for the FDA adverse event reporting system. Clin Pharmacol Ther. Jun 11, 2013;93(6):539-546. [ FREE Full text ] [ CrossRef ] [ Medline ]
• Team R. The R Project for Statistical Computing. R Foundation. URL: http://www.R-project.org/ [accessed 2024-04-26]
• Chang W, Cheng J, Allaire J, Sievert C, Schloerke B, Xie Y. shiny: web application framework for R. Comprehensive R Archive Network. URL: https://CRAN.R-project.org/package=shiny [accessed 2023-01-30]
• Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation) (Text with EEA relevance). EUR-Lex. URL: https://eur-lex.europa.eu/legal-content/en/TXT/?uri=CELEX:32016R0679 [accessed 2024-04-26]
• SHA-1. Wikipedia. 2023. URL: https://en.wikipedia.org/w/index.php?title=SHA-1&oldid=1135933131 [accessed 2023-01-30]
• Viard D, Parassol-Girard N, Romani S, Van Obberghen E, Rocher F, Berriri S, et al. Spontaneous adverse event notifications by patients subsequent to the marketing of a new formulation of Levothyrox amidst a drug media crisis: atypical profile as compared with other drugs. Fundam Clin Pharmacol. Aug 07, 2019;33(4):463-470. [ CrossRef ] [ Medline ]
• Audeh B, Grouin C, Zweigenbaum P, Bousquet C, Jaulent M, Benkhebil M. French Levothyrox® crisis: retrospective analysis of social media. Bogota, Colombia. Springer International Publishing; 2019. Presented at: Conference ISOP - International Society of Pharmacovigilance; October 1, 2019; Bogota, Colombie. URL: https://hal.archives-ouvertes.fr/hal-02411632
• Devlin J, Chang M, Lee K, Toutanova K. BERT: Pre-training of deep bidirectional transformers for language understanding. In: Proceedings of NAACL-HLT 2019. 2019. Presented at: Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies; June 2-7, 2019:4171-4186; Minneapolis, MN. URL: https://aclanthology.org/N19-1423.pdf
• Campillos-Llanos L, Grouin C, Lillo-Le Louët A, Zweigenbaum P. Initial experiments for pharmacovigilance analysis in social media using summaries of product characteristics. Stud Health Technol Inform. Aug 21, 2019;264:60-64. [ CrossRef ] [ Medline ]
• Vigi4Med Scraper. GitHub. URL: https://github.com/bissana/Vigi4Med-Scraper [accessed 2024-04-22]

Abbreviations

Edited by A Mavragani; submitted 01.02.23; peer-reviewed by S Matsuda, L Shang; comments to author 06.07.23; revised version received 20.10.23; accepted 12.03.24; published 18.06.24.

©Pierre Karapetiantz, Bissan Audeh, Akram Redjdal, Théophile Tiffet, Cédric Bousquet, Marie-Christine Jaulent. Originally published in the Journal of Medical Internet Research (https://www.jmir.org), 18.06.2024.

This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included.