Serum Albumin and Albumin-Based Indices for Admission-Time Risk Classification in Hospitalized Pneumonia: An Exploratory Analysis Using a Verified Death-or-ICU-Level-Care Composite Outcome

Background and Objectives: Hematological and immune-inflammation indices (e.g., neutrophil-to-lymphocyte ratio [NLR], systemic immune-inflammation index [SII] and pan-immune-inflammation value [PIV]) and albumin-based indices (e.g., prognostic nutritional index [PNI] and red cell distribution width [RDW]-to-albumin ratio [RAR]) have each been studied extensively as prognostic markers in pneumonia but are rarely compared head-to-head or benchmarked against serum albumin and their own components. We compared 16 indices, with particular attention to whether composite ratios add value over serum albumin alone. Materials and Methods: In this single-centre retrospective cohort of 957 adults hospitalized with pneumonia, indices were recomputed from raw laboratory values with explicit units. The primary outcome was a composite of in-hospital death or ICU-level care, defined as admission to the institutional intensive care unit (ICU) or verified transfer to an ICU at another hospital when no institutional bed was available (118 events, 12.3%); sensitivity analyses restricted to death or institutional ICU admission (94 events) and excluding the transferred patients (n = 933) are reported. Discrimination used the area under the receiver operating characteristic curve (AUC) with DeLong tests and Holm correction; incremental value used multivariable logistic regression with bootstrap internal validation (including index selection), calibration, category-free net reclassification improvement/integrated discrimination improvement (NRI/IDI), restricted cubic splines, multiple imputation and decision-curve analysis. Results: The RDW-to-albumin ratio (RAR; AUC 0.722, 95% confidence interval [CI] 0.675–0.769), the prognostic nutritional index (PNI; 0.721) and serum albumin (0.710) were the best discriminators. After Holm correction, RAR and PNI had significantly higher AUCs than all inflammatory comparators; albumin had significantly higher AUCs than most, but its differences from the lymphocyte-to-monocyte ratio (LMR), systemic inflammation response index (SIRI) and NLR were not significant after adjustment (adjusted p = 0.053–0.074). RAR did not significantly outperform albumin alone (ΔAUC +0.012, 95% CI −0.020 to +0.044, p = 0.46) and PNI did not significantly differ from albumin (p = 0.40). Added to a base model (apparent AUC 0.599), albumin gave 0.728 and RAR 0.731 (optimism-corrected 0.720 and 0.722), versus 0.651 for SII (0.634). No significant difference was detected between base + RAR and base + albumin + RDW (ΔAUC −0.009, 95% CI −0.031 to +0.006). At the cohort-derived Youden threshold of RAR ≥ 5.03, sensitivity was 0.73 and specificity 0.65 (positive likelihood ratio [LR+] 2.1, negative likelihood ratio [LR−] 0.42); this cut-off is exploratory and requires external validation. Results were materially unchanged in both sensitivity analyses. Component variables had no missing values, so a multiple-imputation analysis returned identical estimates; procalcitonin discrimination was low (AUC 0.63). Conclusions: In this exploratory analysis of a verified composite of death or ICU-level care, RAR and PNI showed the highest univariable discrimination but did not significantly outperform serum albumin, and component modelling found no evidence that RAR added discrimination beyond albumin and RDW modelled separately. Albumin-based risk markers require prospective external validation alongside established pneumonia severity scores before clinical use.