Development and external validation of a multivariate model for predicting pneumonia in patients receiving maintenance hemodialysis: a retrospective study

Introduction

According to statistics, infections are one of the leading causes of hospitalization for patients with maintenance hemodialysis (MHD), second only to cardiovascular disease (Gandra et al., 2021; Guo et al., 2008). Among infections, pneumonia ranks first, with a high incidence rate of 28% within the first year of dialysis treatment (Gandra et al., 2021; Guo et al., 2008). According to statistics, the cumulative probability of pneumonia hospitalization within one year for MHD patients is 9%, and it increases to 36% within five years (Slinin, Foley & Collins, 2006). The one-year and five-year mortality rates for patients with MHD after contracting pneumonia are 45% and 83%, respectively (Slinin, Foley & Collins, 2006).

In current clinical treatment, MHD treatment is the primary method for eliminating metabolites from the blood and stabilizing the internal environment of dialysis patients to extend their survival, despite causing nutrient and trace element loss, immune dysfunction, weakening barrier function against pathogens, and heightened infection risk (Di Pasquale et al., 2019; Dou et al., 2022; Pettigrew, Tanner & Harris, 2021; Toma, Naka & Iseki, 2021; Vanholder & Ringoir, 1993). In addition, dialysis fluid and dialysis cuff catheters used during dialysis treatment, or puncture damage to arteriovenous fistulas, can also be a potential route of bacterial infection (Machiba et al., 2022; Tavakoli et al., 2021; Toma, Naka & Iseki, 2021). In addition, there are many complications such as heart failure (HF), diabetes and coronary heart disease in patients with MHD, which are difficult to control, and also increase the challenge for the prevention of pulmonary infection (Torres et al., 2021). Therefore, early identification of high-risk MHD patients with pulmonary infection, accurate and personalized dialysis and drug treatment plan design is essential.

Forecasting pulmonary infections in MHD patients using existing models poses a challenge. First of all, the previously published predictive models are mainly developed for the general population or chronic kidney disease (CKD) population. They only include routine demographic and laboratory examination characteristics, and do not consider the important characteristics of dialysis treatment (dialysis mode, dialysis frequency, dialysis duration), heart failure (HF) management (cardiac biomarkers, echocardiography) and prognosis in patients with MHD (Deng et al., 2024; Gearhart et al., 2019; Markussen et al., 2024; Shirata et al., 2021). Secondly, the above studies focused more on the prognosis of patients with pneumonia (death or rehospitalization), rather than the early prediction of pulmonary infection (Deng et al., 2024; Gearhart et al., 2019; Markussen et al., 2024; Shirata et al., 2021). Furthermore, the above prediction models were not verified in the independent cohort (Deng et al., 2024; Gearhart et al., 2019; Markussen et al., 2024; Shirata et al., 2021). Pulmonary infection is the primary infectious disease among patients undergoing MHD and a significant prognostic factor; however, the relevant risk factors for pulmonary infections in these patients remain unclear (Torres et al., 2021).

The purpose of this study was to establish and validate a predictive model for pneumonia in maintenance hemodialysis patients to achieve accurate assessment of individual pneumonia risk. The aim is to improve early detection and intervention of pneumonia by healthcare providers, ultimately enhancing the prognosis of patients receiving MHD.

Materials & Methods

Materials

Subjects

This retrospective analysis examined data from 405 patients undergoing MHD at Guangyuan Central Hospital and Nanchong Central Hospital, between January 2018 and November 2024 (Fig. 1).

Inclusion criteria

(1) The age should be over 18 years old; (2) Patients diagnosed with CKD5 receiving dialysis treatment, fulfilling the requirements established by the Kidney Disease Improving Global Outcomes guidelines. In addition, regular hemodialysis must be performed for at least 3 months, frequency ≥ 2 times (weekly); (3) Patients who initially began to receive MHD treatment.

Exclusion criteria

(i) Patients were considered ineligible if they had chronic obstructive pulmonary disease, severe bronchiectasis, a history of malignancy, hepatic conditions, cystic fibrosis, recent hospitalization within the 14 days preceding admission, or were under a palliative approach (i.e., with a life expectancy of less than 2 weeks); (ii) who had received a kidney transplant; (iii) who refused to participate, could not cooperate or had incomplete clinical data.

Results

Multi-model comparison

The seven assessed variables were prioritized through AdaBoost regression, random forest regression, and Lasso regression analysis. The findings suggested that “CHD” was the most crucial predictive factor (Fig. 5). In order to evaluate the improvement of the model, Model 1 (constructed using six variables except ‘CHD’) and Model 2 (including seven variables) were compared.

Compared with model 1, the 1-year and 2-year DCA curves of model 2 in the modeling set and the external validation set show greater net benefits (Fig. 6). This suggests that the incorporation of “CHD” notably enhanced the model. Therefore, we believe that Model 2 is the best model and further verification is carried out.

The best model verification

The ROC curves of the modeling set and the external validation set were drawn, and the area under the curve of Model 2 was calculated. Over time, the time-AUC curve showed the stability of the prediction model, and the AUC values were maintained at about 0.8 (Fig. 7).

The calibration curve closely follows the calibration chart of the reference line, suggesting the accurate prediction of the model in the whole risk range (Fig. 8).

Kaplan–Meier analysis was used to further verify the effect of risk factors on the occurrence of pneumonia in patients undergoing MHD (Fig. S4). WBC, A/G, LVMI, and Age were categorized according to cut-off values determined through ROC curve analysis.

Discussion

In this retrospective analysis, we created and validated a nomogram prediction model specifically for the risk of pneumonia in patients undergoing MHD. Based on our limited knowledge, this seems to be the first externally validated predictive model for pneumonia in MHD patients. Our results show that the model provides accurate and personalized risk assessment for individuals and has strong clinical practical value.

Compared with general patients, dialysis patients face more complex risk factors for pneumonia. These factors include rapid changes in hemodynamics and electrolyte levels, along with inadequate dialysis treatment. Therefore, prioritizing the identification of individuals at an increased risk of pneumonia-related hospitalization or mortality is essential. Numerous current prediction models do not consider patients undergoing MHD, making them unsuitable for this specific population (Ramirez et al., 2020; Toma, Naka & Iseki, 2021; Vanholder & Ringoir, 1992).

In line with prior research, our model took into account age, WBC, and albumin levels (Deng et al., 2024; Huang et al., 2024; Shirata et al., 2021). Moreover, our model integrated essential variables relevant to current MHD treatment management, including LVEF, LVMI, and URR assessment (Deng et al., 2024; Gearhart et al., 2019; Markussen et al., 2024; Shirata et al., 2021). Kaplan–Meier analysis showed that age > 60.5 years, LVMI ≥ 138.4 g/m², WBC ≥ 6.71 × 10⁹/L, A/G < 1.27, history of diabetes, and CHD were recognized as independent risk factors that elevate the likelihood of pneumonia.

Our results showed an 80% rise in pneumonia risk in patients with CHD and a 64.4% increase in pneumonia risk in patients with diabetes. There seems to be a two-way relationship between pneumonia and CHD. On the one hand, coronary artery disease increases the risk of pneumonia hospitalization (Kim et al., 2021); In contrast, pneumonia might also elevate the risk of acute coronary syndrome, such as myocardial infarction or unstable angina (Corrales-Medina et al., 2013). While pulmonary infections are typically viewed as acute occurrences, there is evidence suggesting that pneumonia is linked to cardiovascular complications that may manifest years later, notably acute coronary events such as left ventricular dysfunction, arrhythmia, ischemia, and infarction, as well as HF (Corrales-Medina et al., 2013). Regarding diabetes, a number of studies have shown that it increases the susceptibility of patients to pulmonary infection, and the infection rate will increase with age, poor blood glucose control, or deterioration of immune function (Critchley et al., 2018; Fazeli Farsani et al., 2015; Visca et al., 2018). The main pathogenesis of diabetes is closely related to immune dysfunction, including chemotaxis, phagocytosis, cytokine release and so on (Erener, 2020). The latest data show that, on the one hand, hypoproteinemia caused by diabetes and hemodialysis can affect the synthesis of immune factors and increase the risk of pulmonary infection (Oliver et al., 2022; Wand et al., 2022); on the other hand, the level of opportunistic Enterobacteriaceae in diabetic patients is high, and the intestinal barrier function is decreased, resulting in an increased risk of pathogen transmission (Anhê et al., 2020; Thaiss et al., 2018). Regarding serous effusion, current studies suggest that it may be the result of local inflammation, volume overload, and metabolic abnormalities caused by uremic toxin accumulation or infection (Ito & Akamatsu, 2024).

Among the included basic management indicators of cardiovascular and dialysis treatment, we found that LVMI had a high predictive value (HR: 1.007, 95% CI [1.002–1.019], P = 0.004). In the field of echocardiography, LVMI is often used to assess the health status of cardiac structure (left ventricular size and weight) and function; the increase of LVMI is closely related to heart diseases such as hypertension, coronary heart disease, left ventricular hypertrophy and HF (Heidenreich et al., 2022). The incidence of pneumonia in patients with HF is very high, which is about 2–3 times the expected incidence (Jobs et al., 2018; Shen et al., 2021). Water and sodium retention and toxic accumulation in patients with MHD increase cardiopulmonary function load and tissue structure damage. The decrease of cardiac pumping function in patients with HF is easy to cause pulmonary circulation disorder, which leads to pulmonary congestion and edema, pulmonary dysfunction and local defense function decline, resulting in pulmonary infection (Bartlett et al., 2019); Pulmonary infection can lead to the release of inflammatory factors and respiratory dysfunction in the body. At the same time, myocardial ischemia and hypoxia are aggravated, cardiac load is increased, and HF is aggravated (Mancini & Gibson, 2021). Current research and guidelines emphasize that in the secondary prevention of pulmonary infection, the most critical thing is to ensure that all heart (e.g., HF) and lung (e.g., chronic obstructive pulmonary disease, asthma) complications are treated in accordance with the guidelines, and to restart goal-directed therapy before discharge (Tralhão & Póvoa, 2020; Vaughn et al., 2024).

Standard laboratory tests contributed valuable prognostic information in our models. We found that higher levels of WBC and lower levels of A/G were closely related to the occurrence of pulmonary infection. The results of K-M curve analysis showed that ‘WBC’ with a cut-off value of 6.71 × 10⁹/L and ‘A/G’ with a cut-off value of 1.27 could effectively distinguish the high-risk population of pulmonary infection. In clinical practice, WBC is often used as one of the diagnostic indicators of pulmonary inflammation (WBC 10 × 10⁹/L), while A/G is used as a prognostic indicator of prostate cancer, lymphoma, and rectal cancer, with little attention paid to its prognostic value for pulmonary infection in MHD patients (An et al., 2022; Salciccia et al., 2022). Compared with the previous diagnostic criteria >10,000/µL, the results of this study showed that 6.71 as a cutoff seemed to effectively reduce the missed diagnosis of high-risk MHD patients with pulmonary infection. About A/G, albumin can reflect the nutritional status of the body, globulin can reflect the immune and inflammatory state, their proportion can be divided by albumin in serum total protein minus albumin value to evaluate (Soeters, Wolfe & Shenkin, 2019). The possible reasons for the decrease of the ratio are the decrease of albumin and the abnormal increase of globulin, which indicate the decrease of nutritional level, the hyperfunction of inflammatory state and the chronic inflammatory state (Kadatane et al., 2023; Salciccia et al., 2022). Inflammation in MHD is caused by a variety of mechanisms, including accumulation of pro-inflammatory cytokines due to clearance defects, uremia, oxidative stress, infection, volume overload, and dialysis treatment measures (Kadatane et al., 2023; Zoccali et al., 2017). Long-term dialysis patients have serious loss of nutrients, poor nutritional status, and severe reduction of immune function, so they are more prone to pulmonary infection (Wand et al., 2021). Therefore, the dynamic monitoring of WBC, A/G in patients with MHD should be strengthened. According to the 2019 Global Burden of Disease study, older adults, specifically those over the age of 70, are the demographic most impacted by pneumonia (2020). Our results also showed that age was an independent risk factor for pneumonia in patients with MHD, which was positively correlated with pulmonary infection. The risk of pneumonia increased by 28.8% for every 10-year increase in age (HR: 1.288, 95% CI [1.103–1.502], P = 0.001). However, our study found that MHD patients aged ≥ 60.5 years were at high risk of pneumonia by K-M curve analysis, which was significantly different from the age stratification of the high-risk population in the Global Burden of Disease study. It is further explained that the complex background of MHD patients (toxin accumulation, micro-inflammatory state, nutrient loss, uncontrolled blood pressure, prevalence of HF, etc.) makes the risk population of pneumonia in this population more younger (Ebert et al., 2020; Torres et al., 2021).

In this study, we included more variables related to HF management and dialysis treatment, and further evaluated their correlation with pneumonia. Our model demonstrated outstanding predictive performance, optimizing patient outcomes through the analysis of diverse metrics like ROC curves, calibration plots, and DCA curves. Nevertheless, there are still some limitations in our research. First of all, this study is a retrospective cohort study involving two centers. The clinical data of two hospitals located in different geographical regions were collected. Through the comparison of baseline data between groups, it can be found that there are significant differences between different centers. The differences in disease diagnosis and treatment strategies between different centers may directly affect the results and accuracy of the prediction model. Secondly, despite the external validation of this study, the sample size remains relatively small, potentially restricting the generalizability and statistical reliability of the findings. Future clinical studies are anticipated to utilize a larger sample size to provide additional validation for the model.

Conclusions

Our study established and validated a pneumonia prediction model specifically for patients undergoing MHD. The model developed by showed excellent predictive performance and was a valuable tool for assessing the risk of pneumonia in subgroup patients.

Supplemental Information