Validation of a classification system for causes of death in critical care: an assessment of inter-rater reliability

Objective: Trials in critical care have previously used unvalidated systems to classify cause of death. We aimed to provide initial validation of a method to classify cause of death in intensive care unit patients. Design, setting and participants: One hundred case scenarios of patients who died in an ICU were presented online to raters, who were asked to select a proximate and an underlying cause of death for each, using the ICU Deaths Classification and Reason (ICU-DECLARE) system. We evaluated two methods of categorising proximate cause of death (designated Lists A and B) and one method of categorising underlying cause of death. Raters were ICU specialists and research coordinators from Australia, New Zealand and the United Kingdom. Main outcome measures: Inter-rater reliability, as measured by the Fleiss multirater kappa, and the median proportion of raters choosing the most likely diagnosis (defined as the most popular classification choice in each case). Results: Across all raters and cases, for proximate cause of death List A, kappa was 0.54 (95% Cl, 0.49–0.60), and for proximate cause of death List B, kappa was 0.58 (95% Cl, 0.53–0.63). For the underlying cause of death, kappa was 0.48 (95% Cl, 0.44–0.53). The median proportion of raters choosing the most likely diagnosis for proximate cause of death, List A, was 77.5% (interquartile range [IQR], 60.0%–93.8%), and the median proportion choosing the most likely diagnosis for proximate cause of death, List B, was 82.5% (IQR, 60.0%–92.5%). The median proportion choosing the most likely diagnosis for underlying cause was 65.0% (IQR, 50.0%–81.3%). Kappa and median agreement were similar between countries. ICU specialists showed higher kappa and median agreement than research coordinators. Conclusions: The ICU-DECLARE system allowed ICU doctors to classify the proximate cause of death of patients who died in the ICU with substantial reliability.