This retrospective study used clinical data from the Medical Information Mart for Intensive Care IV database (MIMIC-IV) (version 2.0) database, a publicly accessible and well-validated critical care resource developed by the Massachusetts Institute of Technology Laboratory for Computational Physiology. MIMIC-IV contains comprehensive, high-quality medical records of patients in intensive care units at Beth Israel Deaconess Medical Center (11). Ying Zhang (ID: 14346643) registered, completed training, and received permission to access the MIMIC-IV database. Informed consent was unnecessary for data collection since all patient data in the database is anonymized to maintain privacy and confidentiality.

Patients with RDH were identified in the MIMIC-IV database using International Classification of Diseases version 9 (ICD-9) and ICD-10 diagnostic codes. Patients with RHD were identified using ICD-9 codes 9-391, 9-393, 9-394, 9-395, 9-3979, 9-3980, 9–39890 and ICD-10 codes 10-I01, 10-I05, 10-I06, 10-I07, 10-I08. Exclusion criteria included: (1) patients younger than 18 years at initial admission; (2) those diagnosed with end-stage renal disease, liver cirrhosis, or cancer; (3) ICU stays shorter than 24 hours; and (4) missing survival data.

Data extraction utilized PostgreSQL 13.7.2 and Navicat Premium 16 with structured query language. The extracted variables were organized into four primary categories: (1) patient demographics, encompassing age and gender; (2) clinical severity scores, such as the Oxford Acute Severity of Illness Score (OASIS), Glasgow Coma Scale (GCS), and Charlson Comorbidity Index (CCI); (3) comorbidities, including respiratory failure, diabetes, hypertension, renal disease, sepsis, and paraplegia; (4) laboratory parameters at Intensive Care Unit (ICU) admission, comprising neutrophils, lymphocytes, monocytes, and platelets counts ( Supplementary Figure S1 ). Follow-up commenced on the date of ICU admission and continued until the date of death.

Nine inflammatory biomarkers were calculated using routine peripheral blood counts. The neutrophil–monocyte index (NM) was defined as neutrophil count × monocyte count; the neutrophil–platelet index (NP) as neutrophil count × platelet count; and the monocyte–platelet index (MP) as monocyte count × platelet count. The neutrophil-to-lymphocyte ratio (NLR), monocyte-to-lymphocyte ratio (MLR), and platelet-to-lymphocyte ratio (PLR) were calculated by dividing the respective cell counts by the lymphocyte count. The systemic immune-inflammation index (SII) was defined as platelet count × neutrophil count/lymphocyte count, and the systemic inflammation response index (SIRI) as neutrophil count × monocyte count/lymphocyte count. The aggregate index of systemic inflammation response (AIRI) was calculated as platelet count × neutrophil count × monocyte count/lymphocyte count ( Supplementary Table S1 ).

The primary outcomes of this study were 30-day ICU all-cause mortality. To ensure data completeness, cases missing three or more of the four key blood cell counts (neutrophils, lymphocytes, monocytes, and platelets) at ICU admission were excluded. Variables with over 20% missing data were excluded from the analysis to minimize potential bias. Variables with less than 20% missing data were imputed using a random forest algorithm through the ‘mice’ package in R software (12) ( Supplementary Table S2 ).

The Shapiro-Wilk test was used to assess the normality of continuous variables, revealing that most did not follow a normal distribution. Continuous variables were expressed as medians with interquartile ranges and compared between groups using the Wilcoxon rank-sum test. Categorical variables were expressed as frequencies and percentages, and group comparisons were performed using the Chi-square test. Receiver operating characteristic (ROC) curves were generated for each inflammatory biomarker to evaluate their predictive performance for 30-day all-cause mortality, with the optimal cut-off value determined using the Youden index. Comparisons between two ROC curves were performed using both the DeLong test (13). Kaplan–Meier survival curves were used to compare 30-day ICU survival between groups, with the log-rank test evaluating differences. Cox proportional hazards regression models assessed the relationship among different groups, presenting findings as hazard ratios (HRs) with 95% confidence intervals (CIs). The proportional hazards assumption was tested using Schoenfeld residuals function. Clinically relevant and prognosis-associated variables were included in the multivariate Cox regression models. Model 1 was unadjusted; Model 2 was adjusted for age and gender; Model 3 was adjusted for age, gender, OASIS, GCS, and CCI; and Model 4 was further adjusted for age, gender, OASIS, GCS, CCI, respiratory failure, diabetes, hypertension, renal disease, sepsis, and paraplegia. Additionally, a restricted cubic spline regression model with three knots was used to explore the potential nonlinear association between the baseline optimal biomarker and 30-day ICU all-cause mortality. Subgroup analyses were conducted to examine the association between optimal biomarker and mortality within predefined categories such as age, gender, respiratory failure, diabetes, hypertension, renal disease, sepsis, and paraplegia. A two-sided p-value < 0.05 was considered statistically significant. Analyses were performed using R software (v4.2.1).

A total of 1002 critically ill patients diagnosed with RHD were included in the study. The median age was 73.39 years (interquartile range [IQR], 64.43–80.65), with 488 (48.7%) females and 514 (51.3%) males. The 30-day ICU mortality rates were 6.19%. Compared with survivors, non-survivors were significantly older (median: 77.20 vs. 73.09 years, p < 0.001) and had higher severity scores, including OASIS and CCI (p < 0.001 for both). A higher proportion of non-survivors had respiratory failure (37.1% vs. 21.1%, p = 0.007), diabetes (48.4% vs. 30.9%, p = 0.008), renal disease (43.6% vs. 27.6%, p = 0.012), and sepsis (40.3% vs. 12.0%, p < 0.001), while no significant differences were observed in hypertension or paraplegia. Regarding laboratory parameters, non-survivors had significantly lower lymphocyte counts and higher inflammatory indices such as NLR, PLR, SII, SIRI, and AIRI (all p < 0.05). Platelet counts, NP, and MP were also significantly elevated in the non-survivor group. The median length of stay (LOS) in ICU for patients was 2.46 days (IQR: 1.36–4.91 days). Length of ICU stay was longer in non-survivors compared to survivors (median: 4.73 vs. 2.42 days, p = 0.007). The clinicopathological characteristics of the patients included in this study are presented in Table 1 .

Among the nine inflammatory biomarkers, eight (NP, MP, NLR, MLR, PLR, SII, SIRI, and AIRI) showed significantly higher area under the curve (AUC) values (p < 0.05) for predicting 30-day ICU all-cause mortality ( Figures 1A–I ), with PLR exhibiting the highest AUC (0.794). We then compared the AUC of PLR with those of the other seven biomarkers using the DeLong test. PLR demonstrated significantly higher predictive accuracy than NP (p < 0.001), MP (p < 0.001), NLR (p = 0.008), MLR (p < 0.001), SII (p < 0.001), SIRI (p < 0.001), and AIRI (p < 0.001) ( Figures 2A–G ). In addition, we evaluated the prognostic performance of OASIS, GCS, and CCI in predicting 30-day mortality. Both OASIS and CCI showed significantly elevated AUC values (p < 0.05) for 30-day ICU mortality among RHD patients ( Figures 1J–L ). When compared with PLR, the AUC for PLR was significantly higher than that of OASIS (p = 0.047) and CCI (p = 0.020) ( Figures 2H, I ). Subsequent analyses focused on PLR. To normalize its distribution and reduce skewness, PLR values were log-transformed prior to regression modeling.

The baseline characteristics of critically ill patients with RHD, stratified by PLR tertiles, are presented in Table 2 . Patients were divided into three groups based on PLR levels at ICU admission: tertile 1 (T1, n = 333; PLR:1.62-70.10), tertile 2 (T2, n = 333; PLR: 70.10-134.29), and tertile 3 (T3, n = 336; PLR: 134.29-2,418.94). The median PLR values for each tertile were 48.91(36.99, 57.82) in T1, 91.80 (80.75, 109.02) in T2, and 232.74 (167.94, 333.86) in T3. Significant differences were observed among the three groups. Patients in the highest PLR group (T3) were older and had higher CCI scores compared to those in T1 and T2 (p < 0.001). The proportions of respiratory failure, renal disease, and sepsis increased progressively across the tertiles (all p < 0.001). Laboratory results showed that lymphocyte counts decreased, while neutrophil, monocyte, and platelet counts increased with higher PLR levels (all p < 0.001). Composite inflammatory indicators including NLR, MLR, SII, SIRI, and AIRI were significantly elevated in the T3 group (p < 0.001). Notably, the 30-day ICU mortality was 1.20 percent in T1, 2.70 percent in T2, and 14.58 percent in T3 (p < 0.001).

Kaplan–Meier survival analysis revealed significant differences in 30-day ICU survival among the three PLR tertiles ( Figure 3A ). Patients in the highest PLR group (T3) exhibited the poorest survival probability, while those in the lowest PLR group (T1) had the most favorable outcomes. The survival differences between all pairwise comparisons were statistically significant, with p = 0.045 for T1 versus T2, p < 0.001 for T1 versus T3, and p < 0.001 for T2 versus T3.

The proportional hazards assumption was verified using Schoenfeld residuals, and no significant violations were observed for any covariate or the global test (all p > 0.05) ( Supplementary Table S3 ). Cox regression models demonstrated a strong positive association between PLR and mortality risk among critically ill patients with RHD. When considered as a continuous variable, PLR was consistently correlated with an increased risk of 30-day all-cause mortality across all models. In the fully adjusted model (Model 4), each unit increase in PLR was associated with a 2.53-fold increase in the risk of death (HR = 2.53, 95% CI: 1.87-3.42, p < 0.001). When analyzed as tertiles, patients in the highest PLR group (T3) had a substantially elevated risk compared to those in the lowest tertile (T1). The adjusted hazard ratios (HR) for T2 and T3 in Model 4 were 2.60 (95% confidence Interval[CI]: 1.01-8.62, p = 0.049) and 11.85 (95% CI: 3.98-35.24, p < 0.001), respectively, with a significant linear trend across tertiles (p for trend < 0.001). These associations remained robust after adjusting for demographic characteristics, severity scores, and comorbidities ( Table 3 ), suggesting that PLR is an independent predictor of short-term mortality in RHD patients.

Restricted cubic spline (RCS) analysis revealed a positive, approximately linear relationship between PLR and 30-day all-cause mortality risk in patients with rheumatic heart disease, without significant evidence of non-linearity (p for non-linear = 0.114) ( Figure 3B ). To further explore the potential threshold effect, a two-piecewise Cox proportional hazards regression model was applied. The inflection point was identified at a PLR value of 4.52. Below this threshold, the risk of 30-day mortality increased steeply (HR = 8.64, 95% CI: 1.05-99.45, p = 0.047), while above the threshold, the association remained statistically significant but less pronounced (HR = 2.35, 95% CI: 1.69-3.26, p < 0.001) ( Table 4 ). However, the log-likelihood ratio test comparing the one-line and two-line models did not indicate a significantly better fit for the two-piecewise model (p = 0.366), supporting the plausibility of a linear dose-response relationship between PLR and 30-day mortality.

To assess the consistency of the association between PLR and 30-day all-cause mortality across clinically relevant subpopulations, stratified subgroup analyses were conducted based on age, gender, respiratory failure, diabetes, hypertension, renal disease, sepsis, and paraplegia status ( Figure 4 ). Elevated PLR remained significantly associated with increased 30-day mortality in all subgroups analyzed (all p < 0.01). Specifically, HR for patients aged < 65 years was 3.35 (95% CI: 1.37-8.18, p = 0.008), while for those ≥ 65 years, the HR was 2.45 (95% CI: 1.93-3.13, p < 0.001), with no significant interaction observed (p for interaction = 0.554). Similarly, the association was present in both males (HR = 3.87, 95% CI: 2.53-5.91) and females (HR = 2.13, 95% CI: 1.56-2.90). The strength of association remained robust in subgroups with or without respiratory failure, diabetes, renal disease, or sepsis, and no significant effect modification was identified across any of the tested covariates (all p for interaction > 0.05).