MultiCenter Interrupted Time Series Analysis: Incorporating Within and Between-Center Heterogeneity

Area of Science:

• Biostatistics
• Epidemiology
• Health Services Research

Background:

• Segmented regression (SR) is standard for interrupted time series (ITS) data but can yield spurious results with aggregated data.
• Multicenter ITS studies often aggregate data, making conventional SR suboptimal.
• Heterogeneity exists both between participants and across study sites in multicenter settings.

Purpose of the Study:

• To develop a robust method for analyzing ITS data that accounts for participant and site heterogeneity.
• To address limitations of conventional SR in aggregated multicenter ITS data.

Main Methods:

• Introduced a weighted segmented regression (wSR) framework incorporating weights for between-participant and between-site variation.
• Empirically compared wSR against conventional SR and a pooled analysis method.
• Utilized data from the multisite Mobility of Vulnerable Elders in Ontario (MOVE-ON) project for comparison.

Main Results:

• wSR yielded the most precise estimates with the narrowest 95% confidence intervals compared to conventional SR.
• The wSR method demonstrated increased statistical power.
• wSR and pooled analysis showed comparable results with ≤4 sites and moderate to high heterogeneity (I² statistic).

Conclusions:

• Accounting for participant-level and site-level variability enhances the precision and accuracy of intervention effect estimates in ITS.
• The proposed wSR method increases statistical power, highlighting the importance of addressing data variability.
• Further simulations are needed to evaluate wSR across diverse scenarios and outcome types.