CONUT score at ICU discharge predicts long-term all-cause mortality risk in elderly survivors of pneumonia-induced sepsis

Background and objectives Although the Controlling Nutritional Status (CONUT) score has been confirmed to be associated with short-term mortality in intensive care unit (ICU) patients with sepsis, evidence on long-term post-ICU mortality remains limited. This study assessed its prognostic performance for long-term all-cause mortality in elderly survivors of pneumonia-induced sepsis and compared it with several immunoinflammatory nutritional indices. Methods From the initial pool of 620 candidates, we screened 510 first-episode sepsis survivors with no sepsis recurrence more than 3 months after ICU discharge. We subsequently established a multicenter ambispective longitudinal cohort of 461 eligible participants for prospective follow-up, with a median follow-up period of 36 months. Using a complete-case approach with ICU discharge as time zero, we assessed the association between CONUT and mortality across the CONUT score tertiles: T1 (score 0–3, n = 162), T2 (4–6, n = 172), and T3 (7–12, n = 127). Sensitivity analyses were conducted to address selection, immortal-time, and landmark biases. Results Long-term risk of all-cause mortality increased stepwise across CONUT tertiles (24.07% vs. 37.21% vs. 58.27%; p < 0.0001). Kaplan–Meier analysis demonstrated significantly lower survival in patients with higher CONUT scores (log-rank, p < 0.0001). Multivariable Cox regression (Model III) revealed that one tertile increase in CONUT was associated with a 23.3% higher mortality risk (hazard ratio (HR) = 1.233, 95% confidence interval (CI) = 1.004–1.515; p for trend = 0.036), while the highest CONUT tertile carried a 51.6% increased mortality risk (HR = 1.516, 95% CI = 1.001–2.296; p = 0.039). Stratified analyses detected gender-related effect heterogeneity ( p for interaction = 0.034). Restricted cubic spline (RCS) modeling confirmed a significant positive dose–response association between CONUT and mortality ( p for overall < 0.001, p for nonlinear = 0.575). Multivariable linear regression evidenced that each one-tertile CONUT increase produced measurable changes in immunoinflammatory nutrition-derived indices (calculated as Model III | β | × tertile interval (3.0), with positive β values indicating increases and negative β values indicating reductions): blood urea nitrogen-to-albumin ratio (BAR) increase of 0.2889 ( β = 0.0963), neutrophil-to-lymphocyte ratio (NLR) increase of 0.5313 ( β = 0.1771), platelet-to-lymphocyte ratio (PLR) increase of 0.2721 ( β = 0.0907), and lymphocyte to high-density lipoprotein ratio (LHR) reduction of 0.126 ( β = −0.0420) (all p < 0.01). Time-dependent AUC analysis showed that CONUT prognostic discrimination was comparable to BAR and superior to NLR, PLR, and LHR over a median 36-month follow-up (DeLong test, all p for ΔAUC < 0.05). Decision curve analysis (DCA) illustrated that the CONUT model provided greater overall net benefit than the other markers. Time-dependent integrated discrimination improvement (IDI), time-dependent continuous net reclassification improvement (NRI), a mutually adjusted Cox model, a bidirectional incremental C-index, and calibration curve analyses indicated that CONUT was a core prognostic marker. BAR further improved prognostic performance when added to the CONUT-based model, whereas NLR, PLR, and LHR added little incremental predictive value. Conclusion The bedside CONUT score measured at ICU discharge may be associated with long-term all-cause mortality risk in older sepsis survivors. Further research is required before these findings can inform nutritional interventions in routine clinical practice.