Brakenhoff TB, et al. - Using national registry data, researchers conducted an in silico simulation study comparing eight methods that are used during multicenter intervention studies in ICU for adjusting for confounding by disease severity when different disease severity measures (DSMs) are collected across centers. From January 1, 2011, to January 1, 2016, 55,655 ICU admissions were included. They simulated a fictitious treatment variable whose effect on the outcome was confounded by Acute Physiology and Chronic Health Evaluation (APACHE) IV predicted mortality (a common measure for disease severity), to imitate an intervention study with confounding. They noted neglecting differences between DSMs across centers that resulted in large relative bias in treatment effects. Biases were not eliminated with commonly used methods such as two-stage modeling or standardization. On restricting the analysis to centers where APACHE IV scores were available, least biases were achieved with multiple imputation of APACHE IV score. Low relative biases were noted with regression calibration when some centers had APACHE II score available, but not when these had only Simplified Acute Physiology Score II available.