A Clinical Prediction Tool for Hospital Mortality in Critically Ill Elderly Patients

Abstract Background Very elderly (80 years of age and above) critically ill patients admitted to Medical ICUs have a high incidence of mortality, prolonged hospital length of stay, and living in a dependent state should they survive. Objective To develop a clinical prediction tool for hospital mortality in order to improve future end-of-life decision making for very elderly patients who are admitted to Canadian ICUs. Design. A prospective, multicenter cohort study. Setting. Data from 1033 very elderly medical patients admitted to 22 Canadian academic and non-academic ICUs. Interventions. A univariate analysis of selected predictors to ascertain prognostic power was performed, followed by multivariable logistic regression to derive the final prediction tool. Main Results. We included 1033 elderly patients in the analyses. Mean age was 84.6 ± 3.5 years, 55% were male, mean APACHE II score 23.1 ± 7.9, SOFA score was 5.3 ± 3.4, median ICU length of stay was 4.1(IQR 6.2) days, median hospital length of stay was 16.2(IQR 25.0) days and ICU mortality and all cause hospital mortality were 27% and 41%, respectively. Important predictors of hospital mortality at the time of ICU admission include: age [85–90 years of age had an odds ratio of hospital mortality of 1.63 (1.04–2.56), > 90 years of age had an odds ratio of hospital mortality of 2.64 (1.27–5.48)], serum creatinine [120–300 had an odds ratio of hospital mortality of 1.57 (1.01–2.44), > 300 had an odds ratio of hospital mortality of 5.29 (2.43–11.51)], GCS [13–14 had an odds ratio of hospital mortality of 2.09 (1.09–3.98), 8–12 had an odds ratio of hospital mortality of 2.31 (1.34–3.97), 4–7 had an odds ratio of hospital mortality of 5.75 (3.02–10.95), 3 had an odds ratio of hospital mortality of 8.97 (3.70–21.74)], and serum pH [< 7.15 had an odds ratio of hospital mortality of 2.44 (1.07–5.60)]. Conclusion We identified high risk characteristics for hospital mortality in the elderly population and developed a Risk Scale that may be used to inform discussions regarding goals of care in the future. Further study is warranted to validate the Risk Scale in other settings and evaluate its impact on clinical decision-making.