As the prevalence of long-term diseases such as diabetes mellitus and hypertension continue to increase accompanied with the aging of the population, the incidence of end-stage renal disease (ESRD) has also risen dramatically and has become a widespread public health problem (1–4). Hemodialysis is the most frequently practiced renal replacement therapy for patients with ESRD (5). Nevertheless, patients with chronic kidney disease (CKD), especially those with ESRD, are susceptible to muscular atrophy, muscle sparing, and cachexia because of the need for long-term maintenance hemodialysis (MHD), leading to anemia, immune dysfunction, poor dialysis tolerance, and frequent infections, and even directly affect the quality of patients’ life and survival rate (6–13).

In 2007, the International Society of Renal Nutrition and Metabolism (ISRNM) officially adopted the term of Protein Energy Wasting (PEW) to describe metabolic and nutritional disorders in chronic disease states (14). PEW refers to a state of declining protein and energy stores in the body, characterized by insufficient dietary nutrient and calorie intake (unintentional low daily protein intake (DPI) < 0.80 g/kg per day for at least 2 months for dialysis patients or DPI < 0.60 g/kg per day for patients with CKD stages 2–5), low body mass index (BMI) (BMI < 23 kg/m², unintentional 5% weight loss over 3 months or 10% weight loss over 6 months, and total body fat percentage < 10%), hypoproteinemia (serum albumin < 38 g/L, serum prealbumin < 0.3 g/L, or serum cholesterol < 1 g/L), micro-inflammatory states, and progressive skeletal muscle atrophy (reduced 5% muscle mass over 3 months or 10% over 6 months, or reduction of mid-arm muscle circumference (MAMC) area over 10% about 50th percentile of reference population) (11, 15, 16). It has previously been documented that 18–75% of the population suffers from varying degrees of malnutrition, especially in patients with ESRD in dialysis, and quality of life and mortality in CKD is closely associated with PEW which has been confirmed by reverse epidemiology (9, 17–20).

In 2008, ISRNM recommended a four-component diagnosis of PEW based on low biochemical markers [serum albumin, prealbumin, or total cholesterol (TC)]; generalized adiposity or weight loss; loss of muscle mass; and insufficient protein or energy intake ratios (14). Unfortunately, in clinical practice, these diagnostic criteria are not easily measured, especially the assessment of protein or energy intake. Even worse, this is probable to hamper the early diagnosis of PEW, to the detriment of improving the prognosis of patients.

In recent years, the study involved in predictive models in the medical field has proliferated. Predictive model is a mathematical formula that determines the risk of a particular outcome based on a person’s predictive variables. The models have gained attention for their potential use in personalized medicine, individualized decision-making, and risk stratification (21). As a result, researchers have developed a large number of tools to predict, score risk, etc. To streamline the diagnostic method of PEW, a novel prediction model was submitted by Chen et al. (22). They identified independent risk factors for PEW through univariate and multivariate logistic regression, and combined them with the diagnostic criteria for ISRNM, resulting in the inclusion of seven influencing factors, BMI, gender, albumin, TC, triglyceride (TG), vitamin D, and N-terminal Pro-B-Type Natriuretic Peptide (NT-proBNP), respectively. Each influencing factor was then valued for its level of value according to the degree of its contribution to the outcome variable (occurrence of PEW). A nomogram was constructed. Ultimately, the predictive value of the incidence of PEW can be calculated from the composite score based on nomogram (22). In addition, they proved that the model had good predictive ability through calibration curves and the Receiver Operating Characteristic (ROC) curve, where the area under the curve was 0.851 (95% CI: 0.799–0.904) (22).

Because predictive models typically perform worse in new patients than in the developed population, models should not be recommended for clinical use until external validity has been established (21, 23). However, we would have to demonstrate that the predictive model also has a high PEW prediction accuracy for different real-life populations, in the sense of externally validating it, before the model can be applied to everyday practice (24–28). The purpose of this study is to conduct an independent external validation of this novel predictive model.

In the current study, a total of 622 samples from five different medical centers were collected to externally validate the predictive model for PEW. As seen in our results, the predictive model provided excellent discrimination and calibration, and was able to identify whether a hemodialysis patient was PEW at an earlier stage. We also indicated that the 60% threshold probability had a high capacity to predict the risk of PEW with a high clinical benefit. In our external validation set, the incidence of PEW was 46.14%, which was almost the same as the incidence of PEW in the model development set. Although the AUC was 0.777, which was slightly smaller than the internal validation AUC of 0.85, this result was expected during external validation (22, 23, 33). The possible reasons for this discrepancy are analyzed as follows: first, the sample size of the development cohort of the model is comparatively small; second, the data distribution of the development cohort and the validation cohort is very similar during internal validation, which may lead to overfitting, while the distribution of the population characteristics of the external validation cohort differs from that of the development cohort.

Accompanied with the development of science and technology, predictive models are increasingly used in kidney diseases. The quality and clinical influence of these prediction models have fallen short of their intended potential. One reason for this is that despite the development of many models, only part of them have been externally validated (34), and the field of nephrology is no exception (21). Given the number of forecasting models developed, the proportion of studies that address external validation is small. A quick PubMed search revealed 84,032 studies on predictive modeling, of with only 4,309 (5%) referring to external validation in the title or abstract (21). External validation is needed to assess the repeatability and generalizability of the model (35, 36). Regarding the method of external validation, the score of all patients in the validation cohort is calculated from the predictive model that has been built. It is worth noting that the weights for the indicators are based on the already established model, rather than re-running a multifactorial analysis of the validation cohort, which would otherwise only prove that these variables do remain predictors in the external validation cohort (31).

As no single diagnostic marker or tool is most effective in determining whether a patient was PEW, clinical studies focusing on PEW will necessarily need to incorporate one or more nutrition-related surrogates for the diagnosis. The diagnostic criteria proposed by the 2008 ISRNM are four components: low biochemical markers (serum albumin, prealbumin, or TC); generalized adiposity or weight loss; loss of muscle mass; and inadequate protein or energy intake ratios (14). In 2014, Moreau-Gaudry et al. introduced a new simplified assessment method that used serum creatinine corrected for body surface area (sCr/BSA) as a surrogate for muscle loss over time (37, 38). The new marker has the advantage of being easy to measure and enabling earlier diagnosis of protein depletion, rather than having to wait 3–6 months for muscle mass loss to be detected (37). The study showed that the new PEW-score 2014, which incorporated sCr/BSA, identified a higher proportion of dialysis PEW patients than the PEW-ISRNM 2008. Although the PEW-score 2014 was more clinically relevant as it provided more timely information, its correlation with premature mortality would need to be proven in larger studies, which were not yet available to further prove it, and this score only predicted all-cause mortality in European patients undergoing hemodialysis, with ethnicity, habitus, and social background contributing to significant variations in the nutritional status and parameters of patient populations across countries (37).

Additionally, in 2021, the Geriatric Nutritional Risk Index (GNRI), launched initially as a modified nutritional exposure index for older adults, drew attention in assessing PEW (39). Compared to the Subjective Global Assessment (SGA) and the Malnutrition Inflammation Score (MIS), which require subjective assessment, the GNRI is a brief, objective nutritional measure that only involves two constituents (serum albumin concentration and actual-to-ideal body weight ratio), and it has already as validated as an effective assessment tool for ESRD patients in Asia (39). Beberashvili et al. conducted a comparison of MIS and GNRI for hemodialysis patients and observed that there was less agreement between observers for MIS than for GNRI (40). However, variations in daily energy and protein intake were correlated with the MIS rather than the GNRI. In terms of this study, moreover, it suggests that only MIS is an important risk factor for death, and thus MIS is likely to be a more comprehensive tool than the GNRI (40).

For the diagnosis of PEW, our previous study proposed a novel model for predicting the risk of PEW in adult hemodialysis patients, and its validity was confirmed by internal validation (22). The study included 380 adult hemodialysis patients who had been on continuous dialysis for more than 6 months in the hemodialysis centers of several tertiary hospitals in Shanghai, and incorporated seven indicators as predictors: albumin, TC, TG, BMI, gender, vitamin D, and NT-proBNP (22). In contrast to the diagnostic criteria proposed in 2008, the model added four new independent influences: gender, TG, vitamin D, and NT-proBNP. The study showed that female patients had a higher risk of developing PEW than male patients, which might be related to sex hormones and different adipokines distribution (22). In addition, TG was found to be protective in the development of PEW. Notably, it had been implicated that plasma TG n-3 polyunsaturated fatty acids (PUFAs) were linked to both an inferior level of inflammatory markers and improved nutritional condition in patients with MHD. Furthermore, TG n-6 PUFAs have shown a positive correlation with higher serum albumin levels and increased grip strength (41). For vitamin D, it performs an influential part in the regulation of skeletal muscle metabolism. Low levels of vitamin D notably increase the mortality in MHD patients with PEW (42). Besides, NT-proBNP is a possible independent biomarker for the occurrence of PEW in patients with MHD, probably because it is negatively correlated with the amount of body fat and dramatically increases the incidence of PEW in adult hemodialysis patients (43). Metrics such as discrimination and calibration revealed that the model has good predictive ability and clinical utility. It was more accessible and objective, facilitating early identification and intervention of PEW in MHD patients by clinical physicians. Unfortunately, it was not externally validated to demonstrate a high clinical translation rate.

For the strengths of our study, the main points are as follows: Firstly, our study population is independent of the model development set, which is more heterogeneous and has a larger sample size, making the results more convincing. Secondly, the multicenter design makes the data more extensive, which allows a better assessment of the generalizability of the model. Lastly, we use a combination of temporal and spatial validation, which is more comprehensive and prospective.

Nevertheless, our study also contains limitations. At first, this is a cross-sectional study, which is affected by external factors to some extent, and prospective studies are required to provide more instructive information. Furthermore, previous studies have developed predictive models for PEW in peritoneal dialysis patients (26), and our study could be compared and optimized to allow for a larger population-based prediction of PEW. Finally, the population in our study was from Shanghai. Although bone mineral metabolism, cardiac function are now routinely assessed for complications in dialysis patients and have become hemodialysis standard operating procedure (SOP) in China, it’s important to note that many traditional hemodialysis clinics globally may not routinely measure biomarkers such as vitamin D and NT-proBNP. To address this, we may need to expand the sample size and consider a broader geographic region for further improvements.

This external validation study demonstrated the feasibility of the novel PEW risk prediction model which was previously developed and established by our center. Its diagnostic validity was in high agreement with the PEW diagnostic gold standard proposed by ISRNM in 2008. It has the potential to replace the current gold standard for discriminating PEW in adult hemodialysis patients. It simplifies the gold standard, aiding in early identification and prevention of PEW, and improves long-term prognosis and survival.