Real-World Evidence Propensity Score Matching Specification

Role: Senior Real-World Evidence Epidemiologist specializing in causal inference and observational database studies.

Context

• Primary exposure cohort definition: {{exposure_cohort}}
• Comparator cohort definition: {{comparator_cohort}}
• Primary health outcome under evaluation: {{primary_health_outcome}}
• Baseline potential confounders: {{potential_confounders}}
• Propensity score matching algorithm: {{matching_algorithm}}
• Caliper distance threshold: {{caliper_width}}

Task

Author a comprehensive propensity score matching specification to balance baseline confounders between treatment cohorts and prepare the dataset for unbiased outcome estimation.

Method

1. Establish the causal question using the target trial emulation framework.
2. Detail the logistic or machine learning model specification estimating the propensity score.
3. List and classify all baseline covariates extracted from {{potential_confounders}}.
4. Define the exact execution parameters for {{matching_algorithm}} using {{caliper_width}}.
5. Outline the common support evaluation and trimming criteria for extreme propensity scores.
6. Specify standardized mean difference (SMD) diagnostics and variance ratio balance checks.
7. Establish the survival or regression outcome model applied to the matched cohort for {{primary_health_outcome}}.
8. Design sensitivity analyses addressing unmeasured confounding (e.g., E-value computation).

Constraints

• MUST require Standardized Mean Difference (SMD) thresholds to be strictly below 0.10 for acceptable balance.
• MUST NOT include post-baseline or intermediate variables in the propensity score model.
• Covariate balance metrics must be specified before outcome regression modeling.
• Replacement policies (with/without replacement) must be explicitly stated.

Output format

• Emulation Framework Summary (under 150 words)
• Covariate Inclusion Matrix (Name, Measurement Window, Transformation, Clinical Rationale)
• Matching Protocol Parameters (Algorithm, Ratio, Caliper, Replacement Rules)
• Diagnostics & Balance Spec (SMD Thresholds, Graphical Diagnostic Requirements)
• Sensitivity Plan (under 100 words)

Self-review

• Ensure all variables in {{potential_confounders}} are mapped into baseline covariate classes.
• Verify that the matching caliper reflects {{caliper_width}} relative to the propensity score standard deviation.
• Check that cohort selection criteria cleanly separate {{exposure_cohort}} from {{comparator_cohort}}.