Patient-level data were obtained from a previously established single-center registry in Japan previously reported by Masaki and Shigenori (9) limited to records collected between January 2014 and January 2017. The dataset was made publicly available by the original authors as an open-access resource. The present study represents a secondary analysis of this publicly available dataset. The source publication is distributed under the Creative Commons Attribution license, permitting reuse with appropriate credit to the original authors and the source. The original study was approved by the Ethics Review Committee of Miyanomori Memorial Hospital and was conducted in accordance with the Declaration of Helsinki and relevant guidelines and regulations. As current analysis involved only de-identified public data, no additional ethical approval was required.

Eligibility criteria were patients aged ≥50 years with severe dysphagia who received nutrition support by percutaneous endoscopic gastrostomy (PEG) or total parenteral nutrition (TPN) during the study period. Patients were excluded if they had advanced malignancy, if PEG was performed for gastric decompression, or if PEG placement occurred before 2014. One additional case was excluded due to missing BMI-related nutritional measures (n = 1). The final analytic sample comprised 247 participants.

The primary endpoint was all-cause mortality, defined as death from any cause. Survival time was calculated from the date of initiation of nutritional support to the date of death. The secondary endpoint was pneumonia-related mortality, defined as death attributable to pneumonia. Information on cause of death was obtained from hospital medical records and discharge summaries in the original registry. Pneumonia-related mortality was classified when pneumonia was documented by the treating physicians as the primary cause of death based on clinical, laboratory, and radiographic findings. In cases where multiple causes of death were recorded, pneumonia was considered the cause of death only if it was listed as the leading or underlying cause. Deaths with uncertain or insufficient information regarding pneumonia were not classified as pneumonia-related mortality.

Analyses were performed in R (v4.3.1). Categorical data are reported as n (%). Continuous variables are summarized as mean (SD) when approximately symmetric and as median (IQR) otherwise. For descriptive comparisons across outcome groups, ANOVA was used for continuous variables and chi-square tests for categorical variables; Fisher's exact test was substituted where sparse cells occurred.

Associations between BMI-related indices and time-to-death outcomes were evaluated with Cox regression, separately for all-cause mortality and pneumonia-related mortality, yielding hazard ratios (HRs) with 95% confidence intervals (CI). When adiposity estimators derived from age, sex, and body mass index (CUN-BAE, ECORE-BF, and the Deurenberg equation) were included as exposure variables, age, sex, and body mass index were not additionally adjusted for, because these variables are intrinsic components of the estimation formulas; additional adjustment would therefore constitute overadjustment and could introduce multicollinearity. Before fitting multivariable models, covariate collinearity was screened using generalized variance inflation factors (GVIFs) and the condition index, with no concerning collinearity observed (Supplementary Table S2). The proportional hazards assumption was assessed using Schoenfeld residual-based tests (Supplementary Table S3).

Survival patterns across exposure levels were visualized with Kaplan–Meier curves and compared by log-rank tests. Potential departures from linearity were explored using restricted cubic splines (RCS); likelihood ratio testing was used to evaluate the non-linear component (reported as P for non-linearity). RCS models were fitted using three knots placed at the 10th, 50th, and 90th percentiles of the corresponding adiposity estimator distributions, in accordance with commonly used recommendations.

Effect modification was examined through pre-specified subgroup analyses (e.g., age, sex, comorbidity strata) and by adding interaction terms; interaction P values followed recommended methodological practice (23, 24), and subgroup results are presented as forest plots. Predictive performance of CUN-BAE, ECORE-BF, and Deurenberg for 1-, 2-, and 3-year risks of all-cause and pneumonia-related death was compared using time-dependent ROC analysis. Added predictive value beyond a reference model was quantified using net reclassification improvement (NRI) and integrated discrimination improvement (IDI). Continuous NRI, without pre-specified risk categories, was applied to evaluate improvement in risk prediction for time-to-event outcomes.

Robustness checks included: repeating analyses after multiple imputation by chained equations (25) for missing covariates (Supplementary Tables S12, S13); repeating analyses after excluding deaths within 30 days of baseline (Supplementary Tables S14, S15); and additionally adjusting for the Clinical Frailty Scale (CFS; Supplementary Tables S16, S17). Two-sided P values were used, with P < 0.05 indicating statistical significance unless stated otherwise.

Baseline characteristics stratified by survival status are shown in Table 2. In total, 247 patients were included (mean age, 83.0 ± 9.3 years), and 151 (61.1%) were male. Relative to survivors, non-survivors were older and demonstrated a less favorable clinical profile. Regarding nutritional support modality, PEG was less frequently used among non-survivors, who were more often managed with TPN and also had a lower energy intake. The burden of comorbidities was higher in the non-survival group, with greater proportions of cardiovascular disease (CVD), severe dementia, aspiration pneumonia (Asp), ischemic heart disease (IHD), congestive heart failure (CHF), and chronic kidney disease (CKD). Laboratory data were consistent with poorer nutritional reserve and a more pronounced inflammatory state among non-survivors, including lower serum albumin (ALB), total lymphocyte count (TLC), total cholesterol (TC), and hemoglobin, together with higher C-reactive protein (CRP). The remaining variables were comparable between groups. Notably, BMI did not differ between groups; however, BMI-derived adiposity indices (CUN-BAE, ECORE-BF, and the Deurenberg formula) were higher in non-survivors.

As shown in Figure 1, clear survival differences were observed across tertiles of the three BMI-derived adiposity indices. For all-cause mortality (Figures 1A–C), the curves separated in a graded manner, with survival progressively decreasing from the lowest to the highest tertile; between-group differences were significant (CUN-BAE: P = 0.002; ECORE-BF: P = 0.020; Deurenberg: P = 0.002). A concordant pattern was also seen for pneumonia-specific mortality (Figures 1D–F), where participants in higher tertiles consistently exhibited poorer survival (log-rank tests, all P < 0.01). Collectively, higher BMI-derived adiposity estimates were associated with unfavorable outcomes for both endpoints.

Multivariable Cox regression results for all-cause mortality are summarized in Table 3. In the fully adjusted model, all three indices were independently related to mortality risk and demonstrated a dose-graded pattern. When analyzed by tertiles, the highest tertile (T3) showed an approximately twofold higher hazard compared with T1: CUN-BAE (HR of 2.33, 95% CI from 1.40 to 3.89), ECORE-BF (HR of 2.02, 95% CI from 1.22 to 3.32), and the Deurenberg index (HR of 2.30, 95% CI from 1.37 to 3.88), with all comparisons reaching statistical significance (P < 0.01). Trend tests further supported an increasing risk across tertiles (all P for trend ≤ 0.006). These findings indicate that BMI-derived adiposity indices provide consistent and independent prognostic information for all-cause mortality.

Cox regression results for pneumonia-related mortality are presented in Table 4. Relative to all-cause mortality, stronger associations were evident for pneumonia-related death. In the fully adjusted model, the hazard in T3 was markedly elevated compared with T1: CUN-BAE (HR of 4.81, 95% CI from 1.86 to 12.47), ECORE-BF (HR of 3.78, 95% CI from 1.48 to 9.68), and the Deurenberg index (HR of 5.09, 95% CI from 1.82 to 14.20); all were statistically significant (P < 0.01). A monotonic increase in risk was also supported by trend analyses (all P for trend ≤ 0.005). Overall, these BMI-derived indices effectively stratified the risk of pneumonia-related mortality, with a more pronounced gradient than that observed for all-cause death.

RCS analyses are shown in Figure 2. For all-cause mortality (Figures 2A–C), each index was significantly associated with mortality risk (overall association: CUN-BAE, P = 0.001; ECORE-BF, P = 0.021; Deurenberg, P = 0.002), and there was no evidence supporting non-linearity (all P for non-linearity > 0.05), suggesting an approximately linear increase in risk with higher index values.

Similar dose–response patterns were observed for pneumonia-specific mortality (Figures 2D–F), with significant overall associations (all P for overall < 0.01) and a more conspicuous risk elevation toward the upper range of the indices. Taken together, the RCS findings corroborate the Cox regression results and support a continuous, largely linear positive relationship between BMI-derived adiposity indices and mortality risk.

Time-dependent receiver operating characteristic (ROC) analyses were used to quantify the discriminatory performance of CUN-BAE, ECORE-BF, and the Deurenberg index at 1, 2, and 3 years (Figure 3; Supplementary Tables S6, S7). For all three indices, area under the ROC curve (AUC) values increased as the prediction horizon lengthened. With respect to all-cause mortality, the Deurenberg index and CUN-BAE consistently produced higher AUCs than ECORE-BF, with very similar estimates between the two leading indices (3-year AUC: 0.730 and 0.728 vs. 0.676). For pneumonia-related mortality, AUCs were generally higher than those observed for all-cause mortality; at 3 years, CUN-BAE and the Deurenberg index reached AUCs of 0.809 and 0.796, respectively, exceeding that of ECORE-BF (0.755; Supplementary Tables S6, S7).

Discrimination over the 1–3-year window was further evaluated using the concordance index (C-index; Supplementary Tables S8, S9). C-indices ranged from 0.57 to 0.61 across the three measures, with the Deurenberg index and CUN-BAE remaining slightly higher than ECORE-BF at each time point (e.g., at 2 years: 0.609 and 0.605 vs. 0.569), in agreement with the ranking suggested by the ROC analysis. Overall, all three BMI-derived indices demonstrated measurable discriminatory ability, with the Deurenberg index and CUN-BAE showing comparatively better performance, particularly for pneumonia-related mortality.

Using the baseline risk model as the reference, incremental prognostic utility of CUN-BAE, ECORE-BF, and the Deurenberg index was quantified with NRI and IDI (Supplementary Tables S10, S11). For all-cause mortality, adding CUN-BAE yielded a statistically significant gain in both reclassification and discrimination (NRI = 0.468, P < 0.05; IDI = 0.028, P = 0.04). The Deurenberg index also improved reclassification (NRI = 0.475, P < 0.05) with a borderline increase in discrimination (IDI = 0.023, P = 0.06), whereas ECORE-BF did not materially enhance model performance (all P > 0.05). For pneumonia-related mortality, none of the three indices produced significant improvements in NRI or IDI. Overall, CUN-BAE and the Deurenberg index provided additional prognostic information for all-cause mortality beyond the baseline model, while their incremental value for pneumonia-related mortality appeared limited.

A series of sensitivity analyses was conducted to assess the robustness of the primary results and the stability of model specifications (Supplementary Figures S1–S6; Supplementary Tables S12–S17). In subgroup analyses, positive associations between the three BMI-derived adiposity indices and both all-cause and pneumonia-related mortality were broadly consistent across strata defined by age, sex, and major comorbidities, with no evidence of effect modification (all P for interaction >0.05). After addressing missingness via multiple imputation, effect estimates closely aligned with those of the primary analysis: under the fully adjusted model, the highest tertile was associated with increased all-cause mortality risk (CUN-BAE HR = 2.35; ECORE-BF HR = 1.96; Deurenberg HR = 2.31) and showed stronger associations with pneumonia-related mortality (HR = 4.18, 3.28, and 4.29, respectively; Supplementary Tables S12, S13). Exclusion of deaths occurring within 30 days did not change the direction of associations and slightly strengthened effect sizes, particularly for pneumonia-related death (T3 vs. T1: CUN-BAE HR = 6.89; ECORE-BF HR = 6.26; Deurenberg HR = 9.30; Supplementary Tables S14, S15). Additional adjustment for CFS did not materially alter the findings, and all three indices remained significantly associated with mortality outcomes (Supplementary Tables S16, S17). Collectively, these analyses indicate that the observed relationships between BMI-derived adiposity indices and both all-cause and pneumonia-related mortality are internally consistent and robust across analytic strategies.