Evaluating mortality in intensive care units: contribution of competing risks analyses

Area of Science:

• Critical Care Medicine
• Biostatistics
• Epidemiology

Background:

• Traditional survival analysis methods like Kaplan-Meier curves and logistic models have limitations in intensive care unit (ICU) patient survival analysis.
• These methods often misinterpret censoring and do not adequately address competing risks, potentially leading to inaccurate conclusions.
• The cumulative incidence function (CIF) and the Fine and Gray model offer advanced statistical approaches for competing risks scenarios.

Purpose of the Study:

• To evaluate the utility of standard competing risks methods, specifically cumulative incidence function (CIF) curves and the Fine and Gray model, in analyzing survival data of ICU patients.
• To compare the performance of the Fine and Gray model against traditional logistic models in predicting hospital mortality.

Main Methods:

• A cohort of 203 mechanically ventilated cancer patients with acute respiratory failure admitted to a medical ICU over five years was studied.
• Cumulative incidence function (CIF) curves were estimated to represent the probability of hospital death.
• Both Fine and Gray competing risks models and standard logistic models were employed to identify predictors of hospital mortality.

Main Results:

• The cumulative incidence function (CIF) for hospital death was 35.5% by day 14 and 47.8% by day 60.
• Univariate analysis using both Fine and Gray and logistic models identified eight common predictors of mortality.
• Multivariate analysis revealed four shared predictors: autologous stem cell transplantation, absence of congestive heart failure, neurological impairment, and acute respiratory distress syndrome. The Fine and Gray model additionally identified clinically documented pneumonia and logistic organ dysfunction.

Conclusions:

• The Fine and Gray model demonstrates significant value in predicting mortality among intensive care unit (ICU) patients.
• This competing risks model provides a more nuanced understanding of survival by directly modeling time-to-event data and handling censoring appropriately.
• The Fine and Gray model's adaptability allows for extension to analyze non-fatal outcomes, offering broader applicability in critical care research.