Segmented Regression and Difference-in-Difference Methods: Assessing the Impact of Systemic Changes in Health Care

Area of Science:

• Healthcare research methodology
• Clinical practice improvement
• Statistical analysis in medicine

Background:

• Healthcare professionals often need to assess practice changes.
• Before-after study designs are common but prone to bias.
• Cluster randomized trials, while optimal, are frequently impractical.

Purpose of the Study:

• To discuss biases in before-after study designs.
• To present methods for robustly evaluating systematic practice changes.
• To focus on segmented regression of interrupted time series analysis.

Main Methods:

• Discusses biases in before-after designs (confounding, regression to the mean, Hawthorne effect).
• Highlights segmented regression of interrupted time series (ITS) analysis.
• Presents alternative designs: difference-in-difference, stepped wedge, and cluster randomization.
• Emphasizes the need for sufficient time points and confounding variables in ITS.

Main Results:

• Segmented regression of ITS compares pre- and post-intervention trends without a concurrent control.
• Difference-in-difference methods incorporate a concurrent control for stronger inference.
• Methods allow for robust inference on intervention effects, acknowledging limitations.

Conclusions:

• Appropriate designs and analyses can mitigate biases in evaluating practice changes.
• Segmented regression of ITS is a valuable tool when concurrent controls are not feasible.
• The discussed methods, when applied correctly, enable reliable assessment of intervention impact.