Pulmonary arterial hypertension (PAH) is a life-threatening disorder characterized by progressive vascular remodeling and elevated pulmonary artery pressure, associated with substantial morbidity and mortality despite therapeutic advances [1]. As a subtype of precapillary pulmonary hypertension, PAH has an estimated prevalence of 10–50 per 1 million population and represents a minority among pulmonary hypertension etiologies, where left heart disease predominates [2, 3]. Definitive diagnosis requires right heart catheterization demonstrating a mean pulmonary artery pressure $\ge$25 mmHg and a pulmonary arterial wedge pressure $\le$15 mmHg [4]. Although novel pulmonary vasodilators (e.g., endothelin receptor antagonists, phosphodiesterase-5 inhibitors, prostacyclin analogues) have improved outcomes, long-term survival remains suboptimal [2, 3]. Current risk stratification strategies primarily rely on hemodynamic and functional parameters. However, these metrics exhibit limited sensitivity in critically ill patients experiencing dynamic clinical deterioration [5, 6]. The identification of robust biomarkers that integrate inflammatory and thrombotic pathways may therefore enhance prognostic precision in intensive care settings [7, 8].

The platelet-to-lymphocyte ratio (PLR), a composite index derived from routine hematologic measurements, has emerged as a potential prognostic biomarker in cardiovascular diseases, reflecting both thrombotic propensity and systemic inflammation [9, 10]. Elevated PLR has been associated with adverse outcomes in chronic PAH cohorts. However, its predictive value in critically ill patients, particularly regarding nonlinear mortality associations, remains insufficiently explored [11]. Existing evidence is limited by single-center designs and inadequate adjustment for illness severity.

To address these gaps, we conducted a retrospective cohort study using the Medical Information Mart for Intensive Care IV (MIMIC-IV) 3.0 database (2008–2022) to evaluate the prognostic value of PLR in 1512 critically ill PAH patients. A multidimensional analytical framework incorporating survival analysis, restricted cubic spline modeling, and machine learning algorithms was applied to address three critical knowledge gaps: (1) the quantitative association between PLR and 28-day mortality, (2) the presence of nonlinear threshold effects in intensive care unit (ICU) populations, and (3) the stability of PLR’s predictive performance across clinical subgroups.

This investigation addressed critical knowledge gaps in understanding the prognostic implications of PLR in critically ill PAH populations. Specifically, we examined the relationship between PLR levels and clinical outcomes and explored potential nonlinear associations. By employing complementary analytical approaches, we aim to evaluate the consistency of PLR’s predictive capacity across diverse clinical endings, strengthening early risk assessment strategies and informing timely interventions in this vulnerable patient population.

This retrospective cohort study elucidated the nonlinear threshold effects of PLR on 28-day mortality in critically ill adults with PAH, leveraging comprehensive clinical data from the MIMIC-IV 3.0 database. Our findings reveal that elevated PLR levels correlate with adverse hemodynamic and laboratory profiles, including increased heart rate, respiratory rate, and markers of end-organ dysfunction such as elevated total bilirubin and lactate dehydrogenase. These associations are consistent with the established interplay between systemic inflammation, coagulation abnormalities, and disease progression in PAH [17, 18].

Univariate analysis demonstrated a strong association between PLR tertiles and 28-day mortality ($\chi$² = 59.52, p 0.001; Table 1). Multivariable adjustment and restricted cubic spline analysis further identified a nonlinear threshold effect of PLR on mortality risk, with attenuation of the adjusted hazard ratio after controlling for SOFA score and other confounders. These findings suggest that simple group comparisons may overestimate the predictive value of PLR. Although continuous PLR analysis remained statistically significant (p 0.001), its clinical relevance was limited (HR = 1.001, Supplementary Table 9). These findings support adopting threshold-based stratification, which more effectively identifies high-risk patients (T3 HR = 3.064) and provides clear targets for clinical intervention.

The nonlinear relationship between PLR and in-hospital mortality highlights a critical threshold effect. The unadjusted model identified a PLR threshold of 75.92, beyond which mortality risk rose sharply. The loss of significance in the nonlinear association after covariate adjustment (p = 0.126) may reflect mediation by SOFA score, age, and oxygenation. Although covariate adjustment attenuated this association, the persistence of a U-shaped relationship for 28-day all-cause mortality, with a consistent threshold at 67.32, suggests that PLR remains an independent prognostic marker. This dual-phase association implies that both extremes of PLR, whether reflecting severe thrombocytosis with lymphopenia or profound thrombocytopenia, may signify distinct pathophysiological states contributing to mortality [6, 19]. The reduced significance of the nonlinear relationship after adjusting for covariates such as SOFA score, age, and SpO₂ highlights the multifactorial determinants of mortality in PAH; PLR likely interacts with systemic inflammation and organ dysfunction.

This study is the first to establish mortality risk thresholds for PLR (67.32 and 75.92) in critically ill PAH patients. These cutoff values likely represent distinct pathophysiological states. Values below the threshold may reflect preserved lymphocyte function and relative immune homeostasis, potentially contributing to improved survival [20]. Conversely, values exceeding the threshold suggest a vicious cycle of platelet activation and lymphocyte depletion, overlapping with the PLR range observed in PAH-related thromboinflammatory processes and multiorgan failure in critical illness [7, 21].

Machine learning validation further corroborated the predictive value of PLR. The random forest model, incorporating PLR and 14 other clinically relevant features, demonstrated moderate discriminative performance (AUC: 0.759), with PLR-associated risk scores independently predicting a 19.8% increase in mortality per unit rise. This observation aligns with previous evidence implicating PLR as a surrogate marker of immune-thrombotic dysregulation [19], though our study is the first to quantify its threshold effects in PAH. Subgroup analyses reinforced the robustness of PLR’s prognostic value, showing consistent mortality gradients across age, sex, and comorbidity strata without significant interaction effects. The heightened mortality observed in CKD and COPD subgroups, although not significantly modified, may reflect amplified systemic inflammation in these populations [15, 19].

From a methodological perspective, exclusion of collinear variables (e.g., hematocrit) preserved model integrity, with retained predictors (PLR and SOFA score) demonstrating low multicollinearity. The adjusted Cox model achieved a C-index of 0.733, indicating clinically meaningful predictive accuracy, while machine learning-derived risk scores provide an additional framework for personalized mortality risk assessment.

This study has several limitations. First, its retrospective design introduces inherent biases and potential unmeasured confounders. Second, reliance on single-timepoint PLR measurements precluded longitudinal assessment of dynamic changes. The lack of granular PAH-specific parameters, such as right ventricular function metrics, further limits mechanistic insights into the role of PLR in pulmonary vascular remodeling [22, 23]. Additionally, the fixed-time hematologic sampling in MIMIC-IV precluded evaluation of PLR trajectory changes relative to evolving clinical outcomes. Future prospective studies should validate these thresholds in diverse populations with protocol-driven blood sampling and investigate whether targeted interventions that modulate PLR across disease phases can improve outcomes.

PLR emerges as a pragmatic, easily obtainable biomarker for mortality risk stratification in critically ill PAH patients. Identification of specific PLR thresholds (67.32–75.92) offers actionable targets for prognostic assessment, facilitating earlier recognition of high-risk patients. Integration of PLR into existing prognostic models could enhance predictive accuracy and optimize resource allocation in critical care settings, though clinical decisions regarding monitoring or therapeutic strategies should remain individualized and based on comprehensive evaluation.

• PLR demonstrates a U-shaped relationship with 28-day mortality in critical PAH, with risk increasing beyond 67.32 and relatively reduced below this threshold (adjusted HR = 3.787, 95% CI: 2.431–5.898).

• Defined PLR thresholds (67.32–75.92) enable rapid ICU risk stratification using routine admission blood tests.

• Machine learning confirmed the contribution of PLR to mortality prediction (AUC = 0.759), with each unit increase in PLR associated with a 19.8% rise in risk.

• Elevated PLR predicted mortality consistently across age, sex, CKD, and COPD subgroups without significant interaction effects.

The data analyzed in this study comes from the publicly available MIMIC-IV database, but the use of the data has certain restrictions and requires training and permission. The research data can be found at the following URL: https://mimic.mit.edu/.

Conceptualization, DDG and ZGZ; methodology, DDG; software, DDG; validation, DDG, ZGZ; formal analysis, DDG; investigation, DDG; resources, DDG, ZGZ; data curation, DDG; writing—original draft preparation, DDG; visualization, DDG; supervision, ZGZ; project administration, ZGZ. Both authors contributed to revising the manuscript critically for important intellectual content. Both authors have read and agreed to the final published version of the manuscript. Both authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.

This study is a retrospective analysis based on the de-identified Medical Information Mart for Intensive Care IV (MIMIC-IV) database. As the database has already undergone ethical review and obtained informed consent waiver from the Institutional Review Board (IRB) of Beth Israel Deaconess Medical Center, no additional ethical approval or informed consent is required for the present study.

The authors express gratitude to the participants involved in the MIMIC-IV database and all related researchers.

This research received no external funding.

The authors declare no conflict of interest.

Supplementary material associated with this article can be found, in the online version, at https://doi.org/10.31083/BJHM52966.