Prediction Model Flags Stubborn Low Albumin in Peritoneal Dialysis Patients

Patients on peritoneal dialysis whose low blood albumin refuses to rise despite nutritional treatment may now be identifiable early, thanks to a risk prediction model developed by nephrology researchers in China. In a retrospective single-center study of 368 peritoneal dialysis patients treated at the Second Hospital of Hebei Medical University between January 2021 and January 2023, the team found that 132 patients, or 35.9 percent, met their criteria for refractory hypoalbuminemia, a stubborn form of low serum albumin that persists for more than a year despite interventions such as whey protein powder supplementation and dietary counseling to increase protein intake. The work, published in Frontiers in Nutrition, combines routine clinical and laboratory variables into a nomogram that the authors report showed strong discrimination and calibration in internal validation.

The clinical stakes are considerable. Hypoalbuminemia is described in the study’s introduction as an independent risk factor for poor prognosis on peritoneal dialysis, strongly associated with cardiovascular disease and dialysis failure. Citing prior research, the authors note that for every ten grams per liter drop in serum albumin, peritoneal dialysis patients face an approximately four-fold increased risk of heart failure and roughly a two-fold increase in the risk of death. A study of 11,477 peritoneal dialysis patients cited in the paper reported a five-year survival rate of only 54 percent for this population. Because observational studies suggest that correcting hypoalbuminemia may improve survival, the researchers focused on the subset of patients in whom correction repeatedly fails, a phenomenon they call refractory hypoalbuminemia that has received far less attention than common hypoalbuminemia.

To define the condition precisely, the team applied three criteria that all had to be met: serum albumin below 35 grams per liter, duration of hypoalbuminemia exceeding one year, and persistence of low albumin after various interventions including whey protein powder and education on increasing dietary protein intake. All patients with hypoalbuminemia at the center received routine nutritional interventions, and group assignment was based on status at the last follow-up. The study included only patients who had been on continuous ambulatory peritoneal dialysis for more than one year, because previous research indicates serum albumin peaks after about one year on dialysis before declining, and the definition requires hypoalbuminemia to persist for more than six months. Patients with a previous history of hepatic decompensation or malignancy, acute infectious disease within three months, follow-up intervals greater than three months, or significant volume overload judged by the treating clinician were excluded.

The data collection was extensive. The researchers gathered demographic information, medical history including diabetes and cardiovascular disease, body mass index, duration of dialysis, use of compound alpha-keto acid tablets, and residual renal function, which was defined as a 24-hour urine volume of at least 200 milliliters per day. Laboratory measures included serum albumin, low-density and high-density lipoprotein cholesterol, triglycerides, the urea clearance index Kt/v, C-reactive protein, and glycated hemoglobin. Peritoneal transport function was assessed using the four-hour dialysate-to-plasma creatinine ratio, or 4hD/Pcr, from the peritoneal equilibration test. In a subset of 95 patients, the team also collected dialysis effluent and calculated the peritoneal protein excretion rate in milligrams per minute from effluent protein concentration, effluent volume, and dwell time.

The statistical analysis identified a clear pattern of risk. In multivariable logistic regression, diabetes mellitus carried an odds ratio of 4.20, elevated C-reactive protein an odds ratio of 1.22 per unit increase, and higher 4hD/Pcr an odds ratio of 4.37, marking all three as independent risk factors for refractory hypoalbuminemia. Conversely, higher LDL cholesterol, higher HDL cholesterol, triglycerides at or above 1.265 millimoles per liter, and higher body mass index acted as protective factors, with the triglyceride cutoff derived from a receiver operating characteristic analysis after the Box-Tidwell method showed the continuous triglyceride variable lacked a linear relationship with the outcome. Baseline comparisons reinforced these associations: the refractory group had markedly lower BMI, with a median of 18.25 versus 25.00 kilograms per square meter, lower triglycerides, lower HDL and LDL cholesterol, and higher C-reactive protein, while diabetes was present in 47 percent of the refractory group versus 14 percent of the rest.

Using LASSO regression, the researchers narrowed the field to seven predictors: diabetes mellitus, triglycerides at or above 1.265 millimoles per liter, HDL cholesterol, C-reactive protein, BMI, LDL cholesterol, and 4hD/Pcr. The final modeling cohort comprised 345 patients, of whom 111, or 32.2 percent, had refractory hypoalbuminemia. The resulting logistic model, visualized as a nomogram, achieved an area under the ROC curve of 0.939 in the modeling cohort with a Brier score of 0.089 and a calibration curve close to a straight line with a slope of one. Internal validation used two complementary approaches: an enhanced bootstrap with 1,000 resamples produced an optimism of only 0.005 and an optimism-corrected AUC of 0.934, while 10-fold stratified cross-validation yielded a mean AUC of 0.930 and a mean Brier score of 0.092. Decision curve analysis suggested the model offered net clinical benefit across a range of high-risk thresholds.

The authors are explicit that both bootstrap resampling and cross-validation are internal validation methods and cannot replace independent external validation. Because an external validation cohort independent at the patient level and with sufficient follow-up could not be constructed from the available dataset, no external validation was performed. This caveat matters for anyone hoping to apply the nomogram elsewhere: the study population came from a single center, and the authors acknowledge that its relative homogeneity in geographic region, ethnicity, socioeconomic background, and treatment protocols may limit generalizability. They also stress that the study identified associations, not causal relationships, and call for external validation in larger, multicenter, prospective studies with non-overlapping patient populations.

Beyond the model itself, the mechanistic findings point toward protein loss as a central driver. Among the 95 patients with effluent samples, the peritoneal protein excretion rate was significantly higher in patients with refractory hypoalbuminemia than in those without, and significantly higher in patients with diabetes than in those without. The excretion rate was also significantly positively correlated with 4hD/Pcr, with a correlation coefficient of 0.6, suggesting that patients with high peritoneal transport status lose more protein across the peritoneal membrane. Previous work cited in the paper estimates that peritoneal dialysis patients lose roughly 4 to 15 grams of protein per day through the dialysate, driven by leakage of plasma proteins through large peritoneal pores. Among the 59 patients with residual renal function whose urine was analyzed, 24-hour urinary protein excretion was also significantly higher in the refractory group, highlighting urinary protein losses as an additional route of depletion.

Diabetes emerged as a particularly consequential factor. The authors hypothesize that optimizing glycemic control may help prevent refractory hypoalbuminemia in peritoneal dialysis patients with diabetes, noting that among diabetic patients, those with refractory hypoalbuminemia had significantly higher glycated hemoglobin levels. However, they caution that their data cannot directly explain why glycemic control was poorer in the refractory group, because they did not systematically collect information on medication adherence, dietary adherence, diabetes education, or socioeconomic status, so the impact of adherence differences cannot be ruled out. Inflammation offers another thread: as C-reactive protein levels rose, the risk of refractory hypoalbuminemia increased significantly, and the authors argue that low-grade inflammation, which suppresses hepatic protein synthesis and enhances catabolism, deserves attention alongside nutritional support in managing these patients.

One pragmatic finding concerns a widely used supplement. After propensity score weighting adjusted for confounders including inflammation, oral compound alpha-keto acid tablets did not affect the occurrence of refractory hypoalbuminemia, with a p-value of 0.375, suggesting that supplementation with these tablets alone may not effectively correct the condition. The authors conclude that management should address protein intake, essential amino acid supplementation, and low-grade inflammation together, noting that malnutrition and inflammation may reinforce each other in a bidirectional relationship. They also acknowledge that volume overload, which they controlled for by excluding patients judged to have significant overload on clinical assessment, may still contribute through mechanisms their center did not measure with objective tools such as bioimpedance analysis. If externally validated, the nomogram, built entirely from variables routinely assessed in clinical practice, could give clinicians an early warning system for identifying which dialysis patients need intensified, individualized intervention before their albumin becomes irretrievably low.

Subject of Research: A retrospective single-center study developing and internally validating a risk prediction model for refractory hypoalbuminemia in peritoneal dialysis patients.

Article Title: Development and internal validation of a risk prediction model for refractory hypoalbuminemia in peritoneal dialysis patients to guide nutritional intervention

Article References: Zhang, L., Duan, D., Lu, S., Wang, M., Wang, Y., & Pei, H. (2026). Development and internal validation of a risk prediction model for refractory hypoalbuminemia in peritoneal dialysis patients to guide nutritional intervention. Frontiers in Nutrition, 13, Article 1950177. https://doi.org/10.3389/fnut.2026.1950177

Image Credits: AI Generated

DOI: 10.3389/fnut.2026.1950177

Keywords: peritoneal dialysis, hypoalbuminemia, risk prediction model, nomogram, diabetes mellitus, C-reactive protein, peritoneal protein excretion, malnutrition, internal validation, clinical nutrition, nephrology, Development