Pre-ICU Hospital Length of Stay and 30-day Mortality in the Very Old: a National Cohort of 315 042 ICU Admissions

Background. Hospital length of stay before intensive care unit (ICU) admission (pre-ICU LOS) is an established prognostic marker for short-term mortality in critically ill patients: longer pre-ICU stays are generally associated with higher subsequent mortality. Whether this association holds uniformly across the adult age range, or is modified by age, is not established. The question is clinically relevant for the very old (≥80 years), an increasing share of ICU populations, in whom triage decisions are particularly consequential. How age modifies the association between pre-ICU LOS and 30-day mortality will be examined in a national cohort of adult ICU patients.

Design and setting. A retrospective cohort study using prospectively collected data from the Swedish Intensive Care Registry (SIR) linked to the Swedish National Patient Registry (NPR) and the Swedish Cause of Death Registry. The study will include all adult (≥18 years) first ICU admissions reported to SIR between 1 January 2005 and 31 December 2016. Records will be linked across registries via the Swedish personal identification number; patients without a valid number are excluded by the data holders before delivery. The study has been approved by the Regional Ethics Committee of Uppsala (DNR 2016/421), which also waived informed consent.

Exposure. Pre-ICU LOS is the time in hospital between hospital and ICU admission, with zero for direct-from-emergency-department admissions. Where missing, pre-ICU LOS will be reconstructed from NPR care episodes adjacent to the ICU episode.

Outcome. All-cause mortality within 30 days of ICU admission, as a binary endpoint. Follow-up is complete by registry design.

Analysis. The primary analysis will be a binary logistic regression on 30-day mortality, allowing for non-linear effects of age, pre-ICU LOS, and the Quan-updated Charlson comorbidity index, with sex as a binary factor. Pairwise interactions between age and pre-ICU LOS, pre-ICU LOS and sex, and age and sex will be included. The adjustment set is chosen a priori from a directed acyclic graph; acute physiology at admission (SAPS3 Box III) is excluded as a plausible post-exposure mediator. Marginal predicted 30-day mortality and within- and between-age risk differences will be estimated by G-computation at a range of clinically meaningful pre-ICU LOS values across age deciles. Uncertainty will be quantified by nonparametric bootstrap with percentile 95 % confidence intervals. Sensitivity analyses will examine the primary findings under categorical specifications of age and pre-ICU LOS, and after adjustment for SAPS3 Box III with multiple imputation of missing values.