The timely and accurate prediction of mortality risk in hospitalized patients remains a critical challenge in clinical practice. Prognostic evaluation is particularly vital for critically ill individuals, as it guides treatment planning and enhances survival rates (Bauer et al., 2022; Kesecioglu et al., 2024). Inflammation is a central driver of disease progression and mortality, acting as both a protective mechanism and a potential exacerbator of adverse outcomes when dysregulated (Leńska-Mieciek et al., 2023). In critically ill patients, excessive inflammation often leads to organ dysfunction and death. The platelet-to-lymphocyte ratio (PLR), neutrophil-to-lymphocyte ratio (NLR), and monocyte-to-lymphocyte ratio (MLR) serve as simple indicators of inflammatory status, reflecting the dynamic balance between immunity and inflammation through routine blood tests, and hold significant value in prognostic evaluation across various diseases (Liao et al., 2024; Han et al., 2024; Cai et al., 2024; Karra et al., 2023; Wang J. et al., 2022; Mureşan et al., 2022; Zhang et al., 2021a; Kang et al., 2021). NLR captures the interplay between neutrophils and lymphocytes. Neutrophils, as primary effectors of acute inflammation, clear pathogens and damaged tissue, but their overactivation can trigger tissue damage and amplify inflammatory responses. Lymphocytes, in contrast, play a critical role in immune regulation. Elevated NLR typically indicates heightened inflammation and impaired immune function, a feature strongly associated with poor prognosis in sepsis, cardiovascular diseases, and malignancies (Capone et al., 2018). In sepsis patients, enhanced NLR may reflect an aggressive inflammatory cascade leading to multi-organ failure (Zhang et al., 2024). PLR reflects the interaction between platelets and lymphocytes. Beyond their role in coagulation, platelets contribute to inflammation by releasing inflammatory mediators and promoting leukocyte recruitment. Elevated PLR may signal systemic inflammation and increased thrombotic risk, particularly evident in cancer and heart failure diseases (Zhou et al., 2022; Balta and Ozturk, 2015). In cancer, platelets can support tumor growth and metastasis, while in heart failure, the dual effects of inflammation and thrombosis exacerbate mortality risk (Zhou et al., 2022). MLR represents the ratio of monocytes to lymphocytes. Monocytes, key players in innate immunity, differentiate into macrophages and dendritic cells (DCs), regulating inflammation and tissue repair. Elevated MLR may indicate enhanced monocyte activity or lymphocyte depletion, commonly observed in infectious and chronic inflammatory conditions, and is associated with disease severity and mortality risk (Cheng et al., 2023; Hua et al., 2023).

Mechanistically, PLR, NLR, and MLR may predict mortality by capturing the body’s reaction to pathological stressors such as infection, tissue injury, or chronic disease burden. In acute conditions like sepsis, elevated NLR and MLR often correlate with non-survival, likely due to uncontrolled inflammation leading to organ failure (Li et al., 2024). In chronic diseases such as cancer, persistent inflammation may induce immune exhaustion, manifested by lymphocyte depletion and elevated PLR, NLR, and MLR, signaling disease progression and increased mortality risk (Cheng et al., 2023; Hua et al., 2023). Additionally, elevated PLR may further influence prognosis by indicating thrombotic risk, such as in cardiovascular events. While sophisticated biomarkers such as procalcitonin, C-reactive protein panels, and interleukin-6 assays provide enhanced identification accuracy for intensive care unit (ICU) mortality risk, their implementation is hindered by high costs (>$50/test) and dependency on advanced laboratory infrastructure, limiting utility in resource-poor settings. In contrast, PLR, NLR, and MLR calculated from standard complete blood counts (CBCs) offer a pragmatic alternative: inexpensive, immediately accessible, and validated across diverse ICU populations with robust associations to outcomes. This selection could be highly recommended in low- and middle-income countries for equitable diagnosis, balancing accuracy with feasibility to inform timely interventions where advanced testing is unavailable.

Although using PLR, NLR, and MLR in identifying mortality risk has been investigated across various conditions, such as sepsis, cancer, and cardiovascular diseases, their dynamic changes in the terminal phase of hospitalized patients remain inadequately characterized. Most studies focus on static measurements at specific time points or in disease-specific cohorts, with limited exploration of longitudinal trends in critically ill patients approaching death. Furthermore, comparisons with powered control groups, such as patients with chronic conditions like kidney disease, rehabilitation patients, or healthy individuals, are scarce, limiting the understanding of these biomarkers’ specificity in mortality prediction. This study would address these gaps by conducting a large-scale retrospective analysis of PLR, NLR, and MLR trajectories prior to death in hospitalized patients, compared with kidney disease cases, rehabilitation patients, and healthy controls. By elucidating their temporal dynamics, association with mortality events and applicability in discriminating mortality risk, this study aimed to expand these indicators’ clinical utility and provide novel insights into mortality risk evaluation.

Accurate and timely identification of inpatient mortality risk is essential for enabling early intervention and optimizing treatment strategies, thereby advancing personalized medicine. Inflammatory responses significantly influence disease progression and prognosis, with a range of biomarkers available to monitor these processes. Among these, the PLR, NLR, and MLR have emerged as novel inflammatory markers. These ratios are increasingly utilized to assess inflammatory responses and immune status, playing a critical role in disease diagnosis, condition monitoring, and risk evaluation (Kang et al., 2021; Qiu et al., 2023; Tudurachi et al., 2023; Duman et al., 2021; Kulakli et al., 2023; Shen et al., 2017; Song et al., 2022; Lee and Song, 2024).

In this study, we found that the median levels of PLR, NLR, and MLR prior to patient demise were significantly higher than those in the kidney disease, rehabilitation, and healthy control groups. This observation suggests that elevated PLR, NLR, and MLR levels are associated with increased mortality risk in hospitalized patients. The rise in these markers likely reflects an intensified inflammatory state coupled with suppressed immune function, both characteristic of critical illness. Platelets contribute to inflammation beyond their role in coagulation, while lymphocytes regulate immune responses. Consequently, an elevated PLR may indicate both heightened inflammation and immune suppression. Previous studies have reported a progressive decline in platelet counts toward the end of life, with the increase in PLR primarily driven by reduced lymphocyte counts (Zhang et al., 2014). This pattern aligns with our findings and emphasizes the importance of lymphocyte dynamics in critically ill patients. Similarly, PLR has been closely linked to disease severity and prognosis in conditions such as cancer and autoimmune disorders (Smith et al., 2009; Adhyatma and Warli, 2019), reinforcing its prognostic utility across diverse contexts.

Neutrophils, as the body’s first responders to acute inflammation, exhibited a significant increase before death, reflecting a stress response to severe disease states. Concurrently, a relative reduction in lymphocyte counts suggests immune system failure. The NLR, a widely recognized marker of inflammation and immune activity, has been shown to outperform traditional inflammatory biomarkers such as C-reactive protein in various settings (Li et al., 2020; Lorente et al., 2022). Its elevation is consistently associated with poor prognosis in cardiovascular diseases, malignancies, and infectious conditions (Asik, 2021; Yoon and Lee, 2021). For example, Chen et al. reported that elderly patients with acute myocardial infarction and high NLR faced a 3.091-fold increased risk of in-hospital mortality (Chen et al., 2023), a finding that parallels our results across a broader hospitalized population. In contrast, some studies report lower NLR thresholds in specific diseases; for instance, Zahorec et al. suggested that NLR values above 4.0 indicate severe stress in critical care settings (Zahorec, 2021), highlighting potential variability influenced by patient cohorts or disease states.

Monocytes, key players in immune and inflammatory responses, also contribute to mortality prediction. An elevated MLR suggests immune suppression and activation of inflammatory pathways. In malignancies, increased MLR has been associated with tumor immune evasion, possibly driven by the inflammatory microenvironment (Li et al., 2023; Jakubowska et al., 2020; Wang Q. et al., 2022). Our study observed higher MLR levels in patients with malignant tumors, consistent with reports by Wang et al., who linked elevated MLR to worse survival in lung cancer patients (Wang J. et al., 2022). However, discrepancies exist; unlike NLR, MLR’s prognostic value appears less consistent across non-malignant conditions, potentially due to differences in monocyte activation patterns (Citu et al., 2022). This variability underscores the need to contextualize MLR findings within specific disease frameworks.

Our analysis revealed a trend toward higher PLR, NLR, and MLR levels in male patients compared to females, though this difference was not statistically significant. This observation may stem from gender-based differences in hormone levels, immune responses, or physiological functions. Estrogen, known to modulate immune activity, may dampen inflammatory responses in females, as supported by studies on sex-specific immune profiles (Davis and Baber, 2022). Additionally, PLR was significantly elevated in patients over 60 years, suggesting greater sensitivity in older populations. This age-related trend aligns with findings by Duan et al., who noted that PLR’s prognostic accuracy increases with age due to cumulative inflammatory burden (Duan et al., 2023).

A distinctive feature of this study is we achieved PLR, NLR and MLR at multiple time points before death. Four weeks prior to demise, median levels of these markers were elevated compared to other disease groups, though not significantly. This subtle rise may reflect early physiological changes preceding clinical decline. By 1 week before death, the markers increased further, and within the final 48 h, a sharp escalation occurred, likely indicating severe inflammation and immune dysfunction (Wang et al., 2023). This temporal pattern is consistent with palliative care research, where inflammatory markers surge as death nears (Suh et al., 2019). Compared to cross-sectional studies like that of Templeton et al., which focused on single-point NLR measurements in cancer (Templeton et al., 2014), our dynamic approach offers novel insights into biomarker trajectories, enhancing their clinical relevance.

In a subset of 63 hospitalized patients, we observed significant increases in PLR, NLR, and MLR during episodes of infection, exacerbated inflammation, or tumor progression (Zhang et al., 2023). Following anti-inflammatory or targeted treatments, these ratios declined, reflecting control over inflammation. However, this reduction was often transient, with subsequent rises during disease recurrence. This fluctuating pattern mirrors findings by Liu et al., who noted similar biomarker dynamics in sepsis patient post-treatment (Liu et al., 2021). Unlike their study, which primarily focused on NLR, our inclusion of PLR and MLR provides a broader inflammatory profile, potentially enhancing the monitoring of treatment response and disease progression.

ROC curves illustrated middle-to-high discriminating ability for mortality risk, while NLR exhibited the highest AUC, sensitivity, and specificity, positioning its superiority as a good clinical tool. Our NLR cutoff aligns closely with Zhang et al.’s threshold of 3.93 for cerebral venous thrombosis (Zhang et al., 2021b), while our PLR cut-off (154.28) is comparable to their 149.52, suggesting consistency across contexts. Although we did not provide the ROC for the combining PLR, NLR and MLR together to avoid overfitting risk in ROC model, results from Wang et al., demonstrated improved utility in identification of diabetes with combined multi-marker model (Wang et al., 2020). However, we should be cautious to interpret their results.

A significant positive correlation between PLR, NLR, and MLR was observed in patients nearing death, consistent with Wang et al.’s findings in diabetic cohorts (e.g., r = 0.609 for NLR and MLR, P < 0.001) (Wang et al., 2020). This correlation likely reflects synchronized activation of platelets, neutrophils, and monocytes during severe inflammation, contributing to cytokine production and immune defense. While PLR and MLR showed relative stability across death categories, NLR was significantly higher in patients with malignant tumors and severe infections compared to cardiovascular or sudden death cases. This aligns with Hirahara et al.’s use of NLR to assess tumor response (Hirahara et al., 2019) and Liang et al.’s findings on its predictive value in sepsis mortality (Liang and Yu, 2022). However, our results differ from those of Forget et al., who reported less pronounced NLR variation in non-infectious conditions (Forget et al., 2017), possibly due to differences in sample size or disease severity.

After comparing to other studies, current study has limitations. First of all, this study is limited by the observational nature which should be highly addressed. Although we performed PSM analysis to make individuals in different groups more comparable, selection and recall biases could not be totally excluded. Then, as a single-center retrospective analysis with a modest sample size, its generalizability is limited. Multi-center studies with larger cohorts are needed to validate these findings. Additionally, confounding factors such as underlying disease heterogeneity, therapeutic interventions, and variability in laboratory equipment may have influenced biomarker levels. Future research should incorporate standardized protocols to mitigate these effects. Thirdly, we evolved biochemistry values in longitudinal manner, but we did not have control groups like the cross-sectional analysis. On one hand, it is quite difficult in clinical practice to achieve the matched control cohorts in different timepoints as expected, on the other hand the shortcoming will relatively impair the translational significance of current results. Thus, we recommend future studies involve PSM principles or target trial emulation (TTE) framework to address this issue and serve as good supplements on current results. TTE represents a robust approach to causal inference in observational studies, emulating the design of a hypothetical randomized controlled trial (RCT) using real-world data to mitigate biases inherent in non-experimental settings (Wang et al., 2025; Lin et al., 2025; Yang et al., 2025). By explicitly defining key trial components such as eligibility criteria, treatment strategies (e.g., interventions targeting PLR/NLR reduction via anti-inflammatory therapies), time zero (baseline assessment), follow-up periods, and outcome ascertainment, TTE ensures that analyses align with the counterfactual framework, thereby enhancing internal validity without requiring prospective randomization. In the context of our study, TTE framework could be applied to simulate a target trial evaluating whether modulating PLR and NLR (as exposures) causally influences mortality risk across patient groups. For instance, using electronic health records or cohort data, we could define interventions versus no interventions with intention-to-treat analyses accounting for time-varying confounders via inverse probability weighting (IPW). The emulation would address potential immortal time bias and time-dependent confounding, which are limitations of our current retrospective design. Beyond TTE, complementary statistical methods for trial emulation include instrumental variable analysis (e.g., leveraging genetic variants as instruments for PLR) and doubly robust estimators, which combine outcome regression with IPW to improve efficiency and robustness against model misspecification (Bidulka et al., 2024; Larsson et al., 2023). These approaches, while computationally intensive, offer avenues for future extensions of our work, particularly in prospective validation studies. Nonetheless, TTE’s structured emulation remains particularly advantageous for modifiable exposures like PLR and NLR, as it facilitates policy-relevant inferences on preventive strategies. Future research should prioritize TTE in longitudinal cohorts to definitively establish causality and guide clinical interventions. Finally, because the kidney disease group, rehabilitation group, and healthy control group were introduced as external controls, potential demographic heterogeneity should be acknowledged (Yang et al., 2024). Although we performed PSM and additional multivariable logistic regression analyses as sensitivity analyses, these approaches reduced, but did not completely remove the heterogeneity and inherent observational biases. To further clarify the sources of heterogeneity, subgroup and stratified analyses would be warranted.

In conclusion, this study demonstrates that elevated PLR, NLR, and MLR are significantly associated with mortality risk in hospitalized patients. Monitoring of these ratios offers a simple, cost-effective mortality risk assessment framework, providing clinicians with actionable insights. However, their interpretation requires consideration on comparable control groups and multi-center powered sample size, given potential influences from observational nature. Future studies should apply more advanced statistical methods, recruit large-scale participants and elucidate the underlying biological processes driving these biomarker changes.