Adjusting for Disease Severity Across ICUs in Multicenter Studies.

Area of Science:

• Critical Care Medicine
• Biostatistics
• Health Services Research

Background:

• Multicenter intensive care unit (ICU) studies often face confounding by disease severity.
• Different centers may collect varying disease severity measures, complicating analysis.
• Standard methods for adjusting confounding may be inadequate when disease severity data is heterogeneous.

Purpose of the Study:

• To compare different statistical methods for adjusting confounding by disease severity in multicenter ICU intervention studies.
• To evaluate method performance when diverse disease severity measures (APACHE IV, APACHE II, SAPS II) are used across centers.

Main Methods:

• An in silico simulation study using national registry data from 20 Dutch ICUs (55,655 admissions).
• Simulated a fictitious treatment variable confounded by Acute Physiology and Chronic Health Evaluation IV (APACHE IV) predicted mortality.
• Investigated scenarios with varied availability of APACHE IV, APACHE II, and Simplified Acute Physiology Score II (SAPS II) across centers, applying eight adjustment methods.

Main Results:

• Complete neglect of differing disease severity measures caused substantial bias (10.2%–173.6%).
• No single common methodology effectively eliminated bias across all scenarios.
• In scenarios with mixed APACHE II or SAPS II data, restricting analysis to APACHE IV centers or using multiple imputation of APACHE IV scores yielded the least relative bias.

Conclusions:

• The performance of confounding adjustment methods varies significantly when different disease severity measures are present across study centers.
• Researchers must plan for and incorporate diverse disease measures into statistical analyses for multicenter studies.
• Contingency planning is essential to mitigate bias arising from heterogeneous disease severity data in multicenter ICU research.