Multivariable models using administrative data and biomarkers to adjust for case mix in the ICU

Area of Science:

• Critical Care Medicine
• Biomarkers
• Health Services Research

Background:

• Routinely collected laboratory biomarkers can improve control of confounding from disease severity in non-interventional studies of general intensive care unit (ICU) patients.
• The predictive ability of these biomarkers for both short- and long-term mortality requires evaluation.

Purpose of the Study:

• To evaluate the performance of routinely collected laboratory biomarkers and demographic data in predicting mortality in general ICU patients.
• To compare the predictive accuracy of these variables against established scoring systems like SAPS II.

Main Methods:

• A study involving 5505 general ICU stays evaluated age, sex, Charlson comorbidity index, and ten baseline blood biomarkers.
• Regression models were used to assess the predictive performance for 30-day and 1-year mortality.

Main Results:

• Models using biomarkers showed slightly lower accuracy than the SAPS II score for predicting 30-day and 1-year mortality.
• Age and Charlson comorbidity index were more significant predictors than individual biomarkers; Cystatin C offered a slight improvement over creatinine.
• Multiple imputation for missing biomarker values significantly enhanced model predictive ability.

Conclusions:

• Automatically collected baseline variables, including laboratory biomarkers, demonstrate predictive power for short- and long-term mortality in general ICU patients, comparable to the SAPS II score.
• These findings facilitate improved internal control of confounding in non-interventional studies utilizing administrative data for mortality research.