Overview of Data and Methods Used

in Pharmacoepidemiology
Hemalkumar B. Mehta, MS, PhD
Johns Hopkins University

1
Contents

► Randomization vs. the real world

► Data sources

► Study designs

► Statistical analysis

► Common statistical software used in pharmacoepidemiology

► Checklist and protocols

2
Randomization vs. the Real World

3
Randomized Trials (Experimental) vs. Real-World Studies
(Observational)

Randomized trials (experimental) Real-world studies (observational)

► Planned and well-controlled experiments ► What happens in everyday clinical practice

► Randomization
► Lacks randomization
► Consistent monitoring and follow-up
► No strict protocol for treatment and follow-up
► Measures efficacy

► Prerequisite for drug approval ► Measures effectiveness

► Gold standard or highest in the hierarchy of ► Can be used to assess use, safety, and
study designs effectiveness in the real world

4
Randomized Controlled Trials Are Not Always Ideal

► Not ideal for studying real-world effectiveness ► Limited to interests of stakeholders who fund
of a drug them

► Not ideal for studies of harms of drugs (not ► Can answer only one question at a time,
usually big enough and with long follow-up) usually

► Difficult to generalize findings to large groups ► Long (multiple sites, multiple investigators, rare
of people patients)

► Hard to look at subgroups with sufficient ► Expensive

power

5
Real-World Studies

► Can describe outcomes of patients receiving care in their usual care setting from their usual clinicians

► Can be designed to answer many questions

► Can sometimes be large without much increased expense over a small study, depending on design

► Can more easily report on safety, harms, adverse events, drug–drug interactions, and discontinuation
of medication

► Need data and analytic tools to answer important questions from real-world data

6
Data Sources

1
 Common Types of Data Sources

► Primary data source, used for

1. Registry data
2. Survey data
3. Prospective cohort studies

► Secondary data source (used for majority of

real-world studies)
4. Administrative claims data
5. Electronic health records data

Image source: Hemalkumar B. Mehta, with the Center for Teaching and Learning, Johns Hopkins Bloomberg School of Public Health. 2
 1. Registry Data

► Organized system that collects clinical and ► Use cases

other data in a standardized format for a ► Characterize natural history of disease
population defined by a particular disease, ► Biomarker discovery
condition, or drug exposure ► Characterize effectiveness and safety
► National statistics
► One example: Surveillance, Epidemiology, and ► Clinical trials
End Result (SEER) registry for cancer ● Selection of study individuals
► Collects cancer incidence and survival data ● Endpoint selection
on US population
► Provides comprehensive cancer statistics in
the US
► Can be linked with other datasets

Sources:
[Link]
[Link]
[Link] 3
 2. Survey Data

► National surveys to collect information on ► One example: Medical Expenditure Panel

health, nutrition, cost, etc. Survey (MEPS)
► Surveys of families and individuals, their
► Several national surveys conducted by: medical providers (doctors, hospitals,
► National Center for Health Statistics (NCHS) pharmacies, etc.), and employers
► Center for Disease Control and Prevention ► Collects information on
(CDC) ● Health services use
► Agency for Healthcare Research and Quality ● Clinical conditions
(AHRQ) ● Prescription drugs
● Cost
► Mostly freely available

Sources:
[Link]
[Link] 4
 3. Prospective Cohort Study

► Enroll a group of people (cohort) and follow ► One example: Framingham Study
them over time to collect health and other ► Began in 1948
information using questionnaires, biological ► Aimed to unravel the underlying causes of
assays, interview, lab measurements heart disease
► Significantly improved our understanding of
► Several prospective cohort studies: heart disease and informed the current
► Framingham Study clinical practice
► Baltimore Longitudinal Study on Aging ► Major public health impacts
► Women’s Health Initiative Study

Sources:
[Link]
[Link]
[Link] 5
Administrative claims data and electronic
medical records data are widely used for
pharmacoepidemiologic research

6
Record
Generation
Process

Source: Schneeweiss S, Avorn J. A review of uses of health care utilization databases for epidemiologic research on therapeutics. J Clin
Epidemiol. 2005;58(4):323-337. doi:10.1016/[Link].2004.10.012 7
Electronic ► Text describing encounters (history
and physical examination findings)
Medical ► Text describing procedures
Records Data ► Text describing behaviors
► Text describing prescriptions written
► From inpatient encounters
► From outpatient encounters
► Laboratory results
► Radiology results (text)
► Problem list
► Active medication list
► Archived medication list
► Provider
► Age, sex, race/ethnicity
► Insurance information

Source: Schneeweiss S, Avorn J. A review of uses of health care utilization databases for epidemiologic research on therapeutics. J Clin
Epidemiol. 2005;58(4):323-337. doi:10.1016/[Link].2004.10.012 8
Administrative ► Unique patient number
Enrollment dates
Data From ►
► Procedure and diagnostic codes
Insurance ► Dates of service
Companies ► Age, sex
► Reimbursed costs
► Sometimes out-of-pocket costs
► Place of service
► Provider type
► Type of admission
► Discharge destination
► Pharmacy claims (drug dispensed
and dose, pill count, day supply,
date)

Source: Schneeweiss S, Avorn J. A review of uses of health care utilization databases for epidemiologic research on therapeutics. J Clin
Epidemiol. 2005;58(4):323-337. doi:10.1016/[Link].2004.10.012 9
4. Administrative Claims Data—1

► Generated by the insurance company when ► Examples of claims datasets:

they are billed by a service provider ► Medicare
► Medicaid
► Separate data files for different health care ► BlueCross/Blue Shield
services ► UnitedHealthcare
► Enrollment
► Inpatient
► Outpatient
► Prescription drug
► Skilled nursing facility

10
4. Administrative Claims Data—2

Strengths Weaknesses

► Records all interactions with health care ► Missing information on nonprescription

provider and captures health care received medications and prescriptions paid out of
across the care continuum pocket
► Complete information on diagnoses, ► Lack of detailed clinical information
procedures, prescription drug dispensing, and
payment ► Lack of information on lifestyle variables and
health behaviors
► Typically, large samples of patients
► Lag time due to adjudication process
► Relatively low cost
► High data quality due to financial transactions

11
 4. Medicare: Administrative Claims Data

► Accessed through a contractor: ► The Standard Analytical Files (SAFs) contain information
ResDAC ([Link]) collected by Medicare to pay for health care services
► Research Identifiable Files— provided to a Medicare beneficiary
contains patient identifiable
► SAFs are available for:
information
► Institutional services
► Limited Data Sets—limited
● Inpatient
patient identifiable information
● Outpatient
► Public Use Files—aggregate level
● Skilled nursing facility
with no patient identifiable
● Hospice
information
● Home health agency
► Noninstitutional services
● Physician
● Durable medical equipment

Source: [Link] 12
5. Electronic Health Records Data—1

► Generated when individual visits health care ► Examples of electronic health records datasets:
provider or facility to seek medical care ► Clinical Practice Research Datalink
► The Health Improvement Network
► Medical records to track information about ► Geisinger
patients ► Veterans Affairs
► Individual visits primary care physician or ► Electronic health records from Johns
specialist for medical conditions Hopkins
► Individual visits emergency room or
hospital for medical conditions
► Not meant for billing or insurance purpose

13
5. Electronic Health Records Data—2

Strengths Weaknesses

► Contains rich data on clinical variables such as ► Variability in quality of data

laboratory results, physiological
measurements, and vitals ► Incomplete data is a common challenge

► Might provide rationale for treatment decision, ► No universal standard for types of data in
depending on quality of free text electronic health record (EHR), resulting in
heterogeneity among EHR systems
► Might capture over-the-counter medication
use
► Unable to determine whether absence of data
► Detailed information provided from hospital is due to care received outside of health care
stays or truly no care was received

14
 5. CPRD: Electronic Health Records Data

► Accessed through Clinical Practice Research ► The CPRD contains anonymized EHR data from
Datalink (CPRD; [Link]) primary care practices:
► Longitudinal primary care records ► Demographic characteristics
► 2,000 primary care practices and includes ► Diagnoses and symptoms
60 million patients ► Drug exposures
● 18 million active patients with at least 20 ► Vaccination history
years of follow-up for 25% of the ► Laboratory tests
patients ► Referrals to hospital and specialist care

► Can be linked with other datasets

Source: [Link] 15
Claims vs. Electronic Health Records Data—1

Information Electronic Health Records Administrative Claims

Age/sex Y Y

Race/ethnicity Possibly Possibly

Socioeconomic data Possibly N (maybe ZIP code level)

Insurance information Y Y

Prescriptions ordered Y N

Pharmacy data (drugs dispensed) Y or N Y

Clinical data: vital signs or point-of-care testing Y N
results
Clinical data: inpatient Y N

Clinical data: outpatient Y N

16
Claims vs. Electronic Health Records Data—2

Information Electronic Health Records Administrative Claims

Spontaneously reported adverse events Y (if sought medical care) Y (if sought medical care)
Diagnoses or procedures coded for payment N Y
Behavioral risk factors and diet Y N
Indicators of procedures having being done Y Y
(laboratory, radiological, procedures)
Results from procedures (echocardiography, Y N
radiology)
Laboratory results Y N
Problem list or summary Y N
Prescriptions ordered Y N
Pharmacy data (drugs dispensed) N Y

17
 Electronic Health Record + Claims Data

► Both types of data can complement each other

► Improve validity
► Broaden the scope of research

► If all data can be linked (complete overlap),

then analysis can be done in linked dataset

► If some data can be linked (partial overlap),

then analysis can be done in claims data and
enhance analyses in the subset with EHR

Image source: Hemalkumar B. Mehta.

Source: Schneeweiss S, Desai RJ, Ball R. A future of data-rich pharmacoepidemiology studies: transitioning to large-scale linked electronic health record + claims data. Am J Epidemiol.
2025;194(2):315-321. doi:10.1093/aje/kwae226 18
 Example of Pharmacoepi Studies

Administrative claims data Electronic health records

► Use of oral anticoagulants among individuals ► Glucagon-like peptide-1 receptor agonists

with cancer and atrial fibrillation in the United (GLP-1) and risk of suicidality among patients
States, 2010–2016 with type 2 diabetes
► Surveillance, Epidemiology, and End Results ► UK Clinical Practice Research Datalink linked
(SEER)–Medicare data to the Hospital Episodes Statistics
► Characterized drug use among individuals ► Assessed the use of GLP-1 and risk of
with cancer and atrial fibrillation suicide

Sources:
Ardeshirrouhanifard S, An H, Goyal RK, et al. Use of oral anticoagulants among individuals with cancer and atrial fibrillation in the United States, 2010-2016. Pharmacotherapy. 2022;42(5):375-386.
doi:10.1002/phar.2679
Shapiro SB, Yin H, Yu OHY, Rej S, Suissa S, Azoulay L. Glucagon-like peptide-1 receptor agonists and risk of suicidality among patients with type 2 diabetes: active comparator, new user cohort
study. BMJ. 2025;388:e080679. Published 2025 Feb 26. doi:10.1136/bmj-2024-080679 19
Study Designs

1
What Are Study Designs?

► Scientific plan to answer a question ► Several study designs

► Similar to a “blueprint” before construction ► High-level overview of 3 study designs
begins 1. Cross-sectional study
2. Cohort study
► Good study design is important to conduct 3. Case–control study
valid pharmacoepidemiology (p’epi) research
► Study design is more important than
analysis
► Poorly designed study cannot be saved by
fancy analysis

2
 Cross-Sectional Study

► Snapshot in time
► Time unit: day, week, month, year

► Describe study population or outcomes in a

single point in time

► For example:
► What percentage of people have diabetes
in 2024?
► What percentage of people are using drugs
for diabetes in 2024?

Image source: Hemalkumar B. Mehta. 3

 Cohort Study

► Follow people forward in time

► Identify people who are exposed to two different

exposures (e.g., drug A vs. drug B) and follow
them over time to observe outcomes

► For example:
► Is drug A better than drug B to reduce the risk
of mortality?
► Does drug A lead to more adverse events than
drug B?

Image source: Hemalkumar B. Mehta.

Source: Pottegård A. Core concepts in pharmacoepidemiology: Fundamentals of the cohort and case-control study designs. Pharmacoepidemiol Drug Saf. 2022;31(8):817-826. doi:10.1002/pds.5482 4
 Case–Control Study

► Follow people backward in time

► Identify people with outcome and without

outcome
► Follow them backward in time to see who is
exposed to something (e.g., drug A vs. drug B)

► For example:
► Does drug A lead to more adverse events than
drug B?
► Useful for rare outcomes

Image source: Hemalkumar B. Mehta.

Source: Pottegård A. Core concepts in pharmacoepidemiology: Fundamentals of the cohort and case-control study designs. Pharmacoepidemiol Drug Saf. 2022;31(8):817-826. doi:10.1002/pds.5482 5
Statistical Analysis

1
Need for Statistical Analysis in Real-World Studies

► In randomized controlled trials, experimental design and randomization help lessen the need for
advanced statistical analysis methods

► In real-world studies, there are variety of reasons …

► Why some people receive treatment versus some don’t
► Why some will receive drug A versus drug B

► Need to account for such reasons is important

► First, account at study design level
► Second, account by conducting appropriate statistical analysis

2
Some Statistical Analysis Methods

► Descriptive analysis ► Propensity score analysis

► Characterize the study population ► Matching
► Describe drug use ► Stratification
► Weighting
► Regression analysis ► Covariate adjustment
► Linear regression for continuous outcomes
► Logistic regression for binary outcomes ► Advanced causal inference methods
► Poisson regression for count outcomes ► Instrumental variable analysis
► Cox regression for survival outcomes ► Marginal structural models
► G-computation

3
Software

1
Commonly Used Software

Need software to analyze data in pharmacoepidemiologic research

Software Website Cost

SAS [Link] Free for student

R [Link] Free

STATA [Link] Discounted price for education

SPSS [Link] Discounted price for education

Microsoft Excel [Link] Discounted price for students

SQL Multiple implementations of the language Free

Python [Link] Free

2
Checklist and Protocols

1
 Checklist—It Works!

Sources:
[Link]
[Link] 2
Need for Protocols in Pharmacoepidemiology

► Different guidelines and templates exists for the best conduct of real-world studies

► Overarching goal is to have transparency, reproducibility, and clear communication

► Equator network has several guidelines to enhance the quality and transparency of heath research
([Link]

► Two examples:
► STaRT-RWE—Structured template for planning and reporting on the implementation of real-world
evidence studies
► HARPER—HARmonized Protocol Template to Enhance Reproducibility of hypothesis evaluating
real-world evidence studies on treatment effects

3
 STaRT-RWE

► A public–private consortium developed a Table of contents:

structured template for planning and reporting 1. Administrative information
on the implementation of real-world evidence • Title, objective, protocol registration,
studies of the safety and effectiveness of ethics
treatments 2. Version history
3. Design diagram
► Appendix has a template and also provides 4. Summary of analytics study population
some examples for real-world studies 5. Analysis specification
6. Sensitivity analyses
7. Attrition table
8. Power and sample size calculations
9. Glossary of terminology
10. Abbreviations

Source: Wang SV, Pinheiro S, Hua W, et al. STaRT-RWE: structured template for planning and reporting on the implementation of real world evidence studies. BMJ. 2021;372:m4856. Published 2021
Jan 12. doi:10.1136/bmj.m4856 4
 HARPER

► International Society for Table of contents

Pharmacoepidemiology (ISPE) and 1. Title page
International Society for Pharmacoeconomics 2. Abstract
and Outcomes Research (ISPOR) convened a 3. Amendments and updates
joint task force to create a template for real- 4. Timeline
world studies 5. Rationale and background
6. Research question and objectives
► Builds on existing efforts 7. Research methods
8. Limitations
9. Human subject research
► Sample templates available on Github website
10. Reporting of adverse events
that incorporate key components
11. References
12. Appendices
Sources:
Wang SV, Pottegård A, Crown W, et al. HARmonized Protocol Template to Enhance Reproducibility of hypothesis evaluating real-world evidence studies on treatment effects: A good practices
report of a joint ISPE/ISPOR task force. Pharmacoepidemiol Drug Saf. 2023;32(1):44-55. doi:10.1002/pds.5507
[Link] 5
 Overall Idea

► Guideline or checklist can help in several ways by:

► Improving transparency
► Improving clear communication
► Improving credibility
► Improving reproducibility
► Reducing confusion
► Reducing errors/mistakes
► Increasing acceptance by regulators, decision makers, and payers

Source: [Link] 6
Summary

1
Key Points

► Randomization vs. the real world: Real-world studies can complement randomized controlled trials
for evidence generation

► Data sources: Variety of data sources are available; pick one that fits the need

► Study designs: Good study design is a prerequisite; cross-sectional, cohort, and case–control

► Statistical analysis: Need appropriate analysis methods based on the research question

► Software: Several options to choose from—SAS and R

► Checklist and protocols: Can improve transparency, reproducibility, and clear communication