Causal Inference Methods for Real-World Evidence

Training Series for Epidemiologists and RWE Data Scientists

A hands-on Python tutorial series for learning causal inference methods applicable to observational studies, pharmacoepidemiology, and real-world evidence (RWE) research.

Overview

These tutorials are designed as training materials for teams working with longitudinal observational data. Each module builds on the previous, progressing from foundational concepts to advanced methods for handling time-varying confounding.

Target Audience:

• Epidemiologists and pharmacoepidemiologists
• RWE/HEOR scientists
• Biostatisticians working with observational data
• Researchers transitioning from clinical trials to RWE

Training Modules

Module	Topic	Key Concepts	Estimated Time
1	Target Trial Emulation	Cloning, ITT effects, sequential trials	45 min
2	Marginal Structural Models	IPTW, time-varying confounding, weighted regression	60 min
3	Parametric G-formula	Counterfactual simulation, treatment strategies	60 min
4	Structural Nested Models	G-estimation, blip functions, effect heterogeneity	75 min

Method Selection Guide

Research Question	Recommended Method
"What is the effect of initiating treatment?"	Target Trial Emulation
"What is the marginal effect of treatment history?"	MSM with IPTW
"What would outcomes be under strategy X vs Y?"	Parametric G-formula
"What is the direct effect at each time point?"	Structural Nested Models
"Does the effect vary by subgroup?"	SNM with effect modification

Dataset

SEQdata.csv - Simulated longitudinal cohort data analogous to claims/EHR databases:

Variable	Description	Type
ID	Patient identifier	Integer (1-300)
time	Time point	Integer (0-59)
eligible	Eligibility flag	Binary
tx_init	Treatment at this time	Binary
outcome	Event indicator	Binary
sex	Sex	Binary
N, L, P	Time-varying confounders	Continuous

Prerequisites

Knowledge

• Basic epidemiology (confounding, bias, study design)
• Regression modeling (linear, logistic)
• Familiarity with causal diagrams (DAGs) helpful but not required

Technical

pip install pandas numpy scikit-learn scipy statsmodels

Quick Start

# Clone the repository
git clone https://github.com/geethamalika/causal-inference-tutorials.git
cd causal-inference-tutorials

# Run tutorials in order
python 01_target_trial_emulation.py
python 02_msm_iptw.py
python 03_gformula.py
python 04_snm_gestimation.py

Key Assumptions (All Methods)

1. No unmeasured confounding (Sequential exchangeability)
2. Positivity (All treatment levels possible for all covariate patterns)
3. Consistency (Well-defined interventions)
4. Correct model specification

Regulatory Context

These methods align with:

• FDA guidance on RWE for regulatory decisions
• ICH E9(R1) addendum on estimands
• EMA guidance on non-interventional studies
• ISPE/ISPOR guidelines for comparative effectiveness

References

Foundational Texts

1. Hernán MA, Robins JM. Causal Inference: What If. Chapman & Hall/CRC, 2020. Free online
2. Rothman KJ, Greenland S, Lash TL. Modern Epidemiology. 3rd ed. Lippincott Williams & Wilkins, 2008.

Method-Specific Papers

• Target Trials: Hernán MA, Robins JM. Using Big Data to Emulate a Target Trial. Am J Epidemiol. 2016.
• MSMs: Robins JM, Hernán MA, Brumback B. Marginal Structural Models. Epidemiology. 2000.
• G-formula: Robins JM. A new approach to causal inference. Computers and Mathematics. 1986.
• SNMs: Vansteelandt S, Joffe M. Structural Nested Models. In: Handbook of Causal Analysis. 2014.

Reporting Guidelines

• STROBE Statement for observational studies
• RECORD Extension for routinely collected data
• ISPE Guidelines for Good Pharmacoepidemiology Practices

Learning Path Recommendations

Week 1: Complete Tutorials 1-2, focus on understanding confounding Week 2: Complete Tutorials 3-4, compare methods on same data Week 3: Apply to your own data, start with simplest applicable method Ongoing: Sensitivity analyses, assumption checking, peer review

Contributing

Contributions welcome! Please submit issues for:

• Bug reports
• Clarification requests
• Additional examples
• Method extensions

License

MIT License - See LICENSE for details.

Acknowledgments

These tutorials draw on the causal inference curriculum developed at Harvard T.H. Chan School of Public Health and the extensive methodological work of James Robins, Miguel Hernán, and colleagues.

---

Developed for training epidemiology and RWE teams in causal inference methods for observational studies.