Methods

Index Date In Program Evaluation

In program evaluation, the index date is the line in the sand that separates historical risk data from post-program outcomes. Getting this wrong can invalidate both matching and effect estimates.

Home / Evidence & Performance / Index Date In Program Evaluation

1. Set index dates for the treatment group

For program participants, the index date is tied to first program exposure or first point of eligibility.

  • Clinical start definition: first documented program visit.
  • Eligibility definition: first eligible service date after the qualifying event, when needed for consistency.
  • Administrative rule: restrict claims to dates from eligibility through study-end.

2. Set index dates for the comparison group

Non-participants have no natural start date, so you must assign pseudo-index dates that mirror treatment timing.

  • Create random pseudo-index dates sampled from the treatment group date distribution.
  • Anchor to comparable clinical history, such as days since a prior hospitalization or acute event.
  • Use timeline matching so both groups are compared at the same stage of risk evolution.

This multi-version approach avoids systematic bias from calendar-time effects and care-journey misalignment.

3. Why index dates matter

  • Look-back window: usually 365 days before index date for baseline risk, burden, and utilization.
  • Follow-up window: starts on the index date to capture immediate post-start events.
  • Matching quality: baseline comparability depends on synchronized reference points.

Practical examples

Example A: Patient A has an index date of July 4, 2023. In a 12-month pre/post design, the look-back window is July 4, 2022 through July 3, 2023, and follow-up is July 4, 2023 through July 3, 2024.

Example B: Patient B has an index date of December 25, 2023. In the same design, the look-back window is December 25, 2022 through December 24, 2023, and follow-up is December 25, 2023 through December 24, 2024.

At scale, this requires careful cohort versioning, day-level date normalization, claims truncation rules, and reproducible ETL logic so every member’s windows are applied consistently.

Related methods

Previous: How to evaluate a healthcare program  |  Next: Small details in evaluation

For more granular data, more recent data, or scientific analysis support, please email us.

Back to Evidence & Performance