Alon Gorenshtein, Amir Srour, Yosef Adiniaev, Yiftach Barash, Eyal Klang, Oved Daniel
Favourable discharge declined as predicted mortality rose but remained observable at the highest risk levels, and no pre-specified subgroup was bounded below 5%. A general ICU mortality estimate is not on its own evidence that a favourable outcome is absent.
AIM: General ICU severity scores predict hospital mortality across heterogeneous populations, not recovery within one diagnosis. We described favourable discharge across the range of APACHE IVa-predicted hospital mortality after cardiac arrest, and tested whether any bedside subgroup fell below a pre-specified benchmark.
METHODS: Observational cohort of 3,641 first ICU admissions in 161 eICU hospitals with a cardiac-arrest diagnosis documented within 24 hours. The exposure was APACHE IVa-predicted hospital mortality; the outcome was favourable discharge (home or rehabilitation), a functional-outcome proxy. Fourteen bedside subgroups were pre-specified; the benchmark required an upper 95% bound below 5%.
RESULTS: Favourable discharge occurred in 1,039 (28.5%), falling from 78.9% in the lowest predicted-mortality decile to 4.1% in the highest (trend z, -27.2; P < 0.001). Among 2,139 patients with predicted mortality at or above 50%, 301 (14.1%; 95% CI, 12.6 to 15.6) had a favourable discharge; among 281 at or above 90%, 9 (3.2%) did. No subgroup had an upper 95% bound below 5%; the lowest was 8 of 264 (3.0%; 95% CI, 1.3 to 5.9). At a more permissive 10% benchmark, two qualified. Among 1,961 not obeying commands, 263 (13.4%) had a favourable discharge. Discrimination did not differ detectably between unfavourable functional outcome (AUROC 0.789) and in-hospital death (0.774; P = 0.078).
CONCLUSION: Favourable discharge declined as predicted mortality rose but remained observable at the highest risk levels, and no pre-specified subgroup was bounded below 5%. A general ICU mortality estimate is not on its own evidence that a favourable outcome is absent.