Reflection on modern methods: a common error in the segmented regression parameterization of interrupted time-series...

Area of Science:

• Epidemiology
• Biostatistics
• Public Health Research

Background:

• Interrupted time-series (ITS) designs are crucial for causal inference in public health interventions.
• Segmented regression is a common method for analyzing ITS data, modeling pre- and post-intervention trends.
• Existing tutorials and publications often present an incorrect parameterization for segmented regression in ITS analyses.

Purpose of the Study:

• To identify and correct a common error in segmented regression parameterization for interrupted time-series (ITS) designs.
• To provide accurate guidance on modeling post-intervention trends in ITS analyses.
• To prevent erroneous results in public health intervention evaluations.

Main Methods:

• The study critically examines the parameterization of segmented regression for ITS designs.
• It demonstrates how incorrect use of calendar time and intervention indicator variables leads to errors.
• It proposes the correct method using time elapsed since intervention onset for post-intervention trend analysis.

Main Results:

• Using the product of calendar time and a pre-post indicator incorrectly models the post-intervention linear segment.
• This common error results in inaccurate estimation of the level change following an intervention.
• Correct parameterization requires a variable that starts at zero at intervention onset for the post-intervention segment.

Conclusions:

• Accurate segmented regression parameterization is vital for valid causal inference in ITS studies.
• Researchers must use time elapsed since intervention, not study start, for post-intervention trend modeling.
• This clarification aims to improve the rigor and accuracy of public health intervention evaluations using ITS designs.