Subgroup mortality probability models: are they necessary for specialized intensive care units?

Area of Science:

• Critical Care Medicine
• Health Services Research
• Biostatistics

Background:

• The Mortality Probability Model at intensive care unit admission (MPM0-III) is a key benchmarking tool for Project IMPACT.
• Existing models may exhibit performance variations across different patient case mixes.
• Understanding these variations is crucial for accurate intensive care unit (ICU) performance assessment.

Purpose of the Study:

• To assess the sensitivity of MPM0-III performance to variations in patient case mix.
• To evaluate the impact of specialized subgroup models on ICU performance evaluation.
• To determine the reliability of MPM0-III across diverse ICU patient populations.

Main Methods:

• A large dataset of 124,171 ICU patients was randomly divided into development and validation sets.
• Logistic regression models were developed for six patient subgroups using MPM0-III variables.
• Model performance was evaluated using Hosmer-Lemeshow and receiver operating characteristic statistics, comparing MPM0-III with subgroup models.

Main Results:

• All six specialized subgroup models demonstrated good performance on their validation sets.
• ICU standardized mortality ratios (SMRs) calculated using MPM0-III and subgroup models were nearly identical.
• MPM0-III performance degradation was significant only when patient mix substantially deviated from the original dataset.

Conclusions:

• MPM0-III is recommended as the primary model for ICU benchmarking due to its overall reliability.
• Specialized subgroup models may offer value for evaluating highly specialized ICUs or for research in homogeneous patient populations.