Abstract
Objective
Hyperprolactinemia (HPRL) is a prevalent endocrine disorder in patients with chronic kidney disease (CKD), yet individualized risk prediction tools remain limited. This study aimed to develop and externally validate a nomogram for predicting HPRL risk in patients with CKD.
Methods
In this retrospective multicenter study, 346 patients with CKD were enrolled from three tertiary centers. Patients from Centers 1 and 2 constituted the development cohort (n = 250, 108 HPRL events), whereas patients from Center 3 served as an independent external validation cohort (n = 96, 36 HPRL events). Demographic, clinical, and laboratory characteristics were compared between patients with and without HPRL. Candidate predictors were screened using univariable and LASSO logistic regression, then entered into multivariable logistic regression. Significant variables were incorporated into the final nomogram. Model performance was assessed via AUC, calibration plots, and decision curve analysis (DCA).
Results
Univariable analysis identified age, body mass index (BMI <18.5 or ≥28kg/m2), estimated glomerular filtration rate (eGFR), lupus nephritis, renal arteriosclerosis, comorbid diabetes, renal replacement therapy, dialysis duration, urinary microalbumin (U-Alb), 24-hour urinary total protein (24-UTP), parathyroid hormone (PTH), fasting plasma glucose (FPG), and urinary albumin-to-creatinine ratio (U-ACR) as factors significantly associated with HPRL. Multivariable regression revealed that age, BMI <18.5 or ≥28, reduced eGFR, comorbid diabetes, and dialysis duration were independent predictors (all P < 0.05). The resulting nomogram demonstrated excellent discriminative ability, with an AUC of 0.914 (95% CI, 0.879–0.949; sensitivity, 88.9%; specificity, 79.6%) in the development cohort and 0.896 (95% CI, 0.834–0.958; sensitivity, 77.8%; specificity, 83.3%) in the validation cohort. Calibration curves showed good agreement between predicted and observed outcomes, and DCA suggested potential net benefit across a range of threshold probabilities.
Conclusion
This externally validated nomogram, based on readily available clinical variables, demonstrated good performance for individualized HPRL risk estimation in patients with CKD and may support risk-based prioritization of prolactin assessment and further clinical evaluation.
Keywords: nomogram, hyperprolactinemia, individualized risk prediction, chronic kidney disease
Graphical Abstract
Introduction
Chronic kidney disease (CKD) represents a major global health challenge, with an estimated prevalence of approximately 10% worldwide.1,2 The progressive decline in renal function is accompanied by a broad spectrum of metabolic, cardiovascular, and endocrine disturbances that substantially increase morbidity and mortality.3 Among endocrine abnormalities, hyperprolactinemia (HPRL) is common but not universal. Reported prevalence varies substantially with CKD severity and dialysis status: earlier studies summarized rates from 18.3% in mild renal insufficiency to more than 70% in hemodialysis (HD) or peritoneal dialysis populations, while more recent data have also shown lower estimates in specific cohorts, including 7.25% in predialysis CKD and 43.2% in HD patients.4–6 This variability highlights heterogeneity across disease stage, dialysis exposure, study populations, and laboratory definitions.
HPRL in CKD is primarily attributed to reduced renal clearance and disturbances in dopaminergic inhibition of prolactin secretion.6 Elevated prolactin levels exert numerous adverse physiological effects. In male patients, HPRL contributes to hypogonadism, erectile dysfunction, infertility, and decreased bone mineral density. In female patients, it is associated with menstrual irregularities, galactorrhea, and impaired fertility.7 Beyond reproductive dysfunction, HPRL has been linked to increased cardiovascular risks, heightened inflammatory activity, insulin resistance, anemia, and impaired immune responses, all of which may accelerate CKD progression and worsen patient outcomes.8 For instance, a recent study of 157 CKD patients demonstrated that prolactin correlates significantly with inflammatory markers (TNF-α, IL-6, IL-1β) and with NT-proBNP, a marker of cardiac stress, suggesting that elevated prolactin may contribute to cardiovascular and nutritional alterations in CKD patients. Despite these consequences, HPRL is often overlooked during routine CKD evaluation, partly due to nonspecific symptoms and the heterogeneity of clinical presentations. Early identification of patients at risk for elevated prolactin is therefore essential for timely monitoring and targeted intervention.
Although direct serum prolactin measurement remains the definitive approach for identifying HPRL, routine prolactin screening is not universally incorporated into the clinical evaluation of all patients with CKD. An individualized risk prediction model may therefore provide complementary clinical value by identifying patients at higher risk of HPRL who may benefit from prioritized prolactin testing, closer endocrine assessment, and subsequent monitoring. Such a risk-based approach could facilitate more targeted evaluation of CKD-related endocrine abnormalities while avoiding indiscriminate testing in lower-risk patients.
Nomograms have emerged as reliable, widely used predictive tools in clinical research because they can synthesize multiple predictors into individualized, quantitative risk estimates.9 Compared with traditional risk classification approaches, nomograms offer enhanced predictive accuracy, interpretability, and clinical applicability. In nephrology, several nomogram-based risk models have been developed and validated for different CKD-related outcomes, including progression to advanced CKD stages,10 rapid kidney function decline, cardiovascular events in non-dialysis-dependent CKD, mortality or survival in dialysis populations,11–13 and other comorbidities or complications.14,15 These studies demonstrate the feasibility and clinical value of nomogram models in CKD care.16 However, despite the clinical relevance of HPRL in CKD, to our knowledge, no nomogram-based predictive model has been developed to estimate individualized HPRL risk in the CKD population. This gap limits clinicians’ ability to implement proactive management strategies for this endocrine complication.
In this context, the present multicenter retrospective study aims to develop and validate a nomogram to predict the risk of HPRL in patients with CKD. By integrating readily available demographic, biochemical, and clinical variables, we aim to develop a practical, accurate tool to support individualized risk assessment in clinical settings. We anticipate that this model will offer clinicians a user-friendly instrument for early risk stratification of HPRL, thereby facilitating timely monitoring and intervention and ultimately improving endocrine and cardiovascular management in CKD patients.
Materials and Methods
Study Population
Patients with CKD were retrospectively enrolled between December 2024 and April 2026 from three tertiary medical centers in China. The details are as follows: 155 from Center 1 (Sir Run Run Hospital of Nanjing Medical University), 106 from Center 2 (Jinling Hospital, Affiliated Hospital of Medical School, Nanjing University), and 136 from Center 3 (Yijishan Hospital, the First Affiliated Hospital of Wannan Medical College).
Inclusion Criteria
All participants were diagnosed with CKD according to the Kidney Disease: Improving Global Outcomes (KDIGO) 2024 clinical practice guidelines.
Exclusion Criteria
(1) age < 18 years; (2) incomplete clinical or laboratory data; (3) pregnancy or lactation; (4) acute infection, acute heart failure exacerbation, or major surgery within the preceding 3 months; (5) malignancy, severe hepatic dysfunction, or other endocrine disorders known to markedly influence prolactin levels; (6) use of medications known to affect prolactin secretion; (7) confirmed prolactinoma or pituitary adenoma. These criteria were designed to minimize potential confounding factors that may affect serum prolactin levels and to improve the reliability of the HPRL risk prediction model. All participants provided written informed consent before enrollment, and all study data were anonymized to protect patient confidentiality. The overall study flowchart and schematic diagrams are presented in Figure 1. The same predefined inclusion and exclusion criteria were applied uniformly across all three participating centers.
Figure 1.
Study flowchart.
Model Construction
Patients from Centers 1 and 2 (n = 250), including 108 patients with hyperprolactinemia (HPRL), constituted the development cohort, whereas patients from Center 3 (n = 96) served as an independent external validation cohort.
As this was a retrospective study, no a priori sample size or statistical power calculation was performed, and all eligible patients during the predefined study period were included. Although the final model contained five predictor constructs, categorical coding resulted in eight regression parameters excluding the intercept, corresponding to approximately 13.5 HPRL events per parameter in the development cohort.
Prolactin Measurement and ULN-Based Normalization
Serum PRL concentrations were measured at the three participating centers using locally available, clinically validated immunoassays. Hyperprolactinemia was defined according to the sex-specific reference ranges provided by each laboratory. In Center 1, PRL was quantified using the original Roche Diagnostics reagents (ee electrochemiluminescence immunoassay) on a cobas e 801 analyzer (Roche Diagnostics, Shanghai, China). Center 2 used the Abbott Architect i2000SR automated chemiluminescent immunoassay system (Abbott Diagnostics, Ireland), whereas Center 3 employed Abbott reagents on the Alinity platform using a microparticle chemiluminescent immunoassay. For each assay, the upper limit of normal (ULN) was defined according to the manufacturer’s reference ranges provided in the accompanying instructions for use.
To account for interassay variability across centers, all PRL values were normalized using an ULN-based approach, calculated as the ratio of the measured PRL value to the assay-specific ULN (PRL/ULN). This dimensionless index allowed comparison of PRL levels independent of the analytical platform used.
After ULN-based normalization, the distribution of PRL index showed improved comparability across centers. Sensitivity analysis with center adjustment yielded similar predictor estimates, and the final selected variables remained unchanged.
Definition of HPRL
PRL levels were measured in the clinical laboratories of the three participating tertiary hospitals. Because assay methods and reference ranges differed across centers, HPRL was defined using each laboratory’s sex-specific upper limit of normal ULN. In all centers, HPRL was diagnosed when the measured PRL concentration exceeded the patient’s sex-specific ULN.
To ensure comparability across centers, PRL values were standardized using the ratio of the measured PRL value to the laboratory-specific ULN. For each participant, a normalized PRL index was calculated as:
 |
Participants with a PRL index > 1.0 were classified as having HPRL. This ULN-based normalization enabled harmonization of PRL measurements across the three centers and allowed uniform classification for subsequent statistical analyses.
BMI Classification
BMI was categorized according to the Chinese adult BMI classification as follows: underweight (<18.5 kg/m2), normal weight (18.5–23.9 kg/m2), overweight (24.0–27.9 kg/m2), and obesity (≥28.0 kg/m2), which is widely used in Chinese populations and recommended by national guidelines.17 For statistical modeling, the extreme BMI categories (<18.5 and ≥28 kg/m2) were used as risk groups to evaluate their association with HPRL, while 18.5–23.9 kg/m2 served as the reference group. This approach allows clear assessment of underweight and obesity as potential risk factors.
Data Collection
After an overnight fast, venous blood samples were collected in the early morning for all participants. Laboratory measurements included serum prolactin, renal function parameters, hemoglobin, red blood cell count, white blood cell count, platelet count, serum albumin, serum calcium, serum phosphorus, and parathyroid hormone (iPTH). All biochemical tests were performed in the clinical laboratories of the three participating tertiary hospitals according to their respective standardized protocols.
Predictor variables and serum PRL measurements were obtained from the same baseline clinical assessment. Therefore, the model was designed to estimate the probability of concurrently present, laboratory-defined HPRL rather than the future incidence of HPRL.
Statistical Analyses
Data processing and statistical analyses were conducted using R software version 4.4.1. Continuous variables were assessed for normality. Normally distributed variables were presented as mean ± standard deviation (SD) and compared using independent-samples t-tests, whereas non-normally distributed variables were presented as median (Q1, Q3) and compared using Wilcoxon rank-sum tests. Categorical variables were summarized as counts and percentages and compared using the chi-square test or Fisher’s exact test, as appropriate.
Although multiple comparisons were performed for baseline characteristics in Table 1, no formal adjustment was applied because these analyses were primarily descriptive and intended to summarize baseline differences between groups rather than to test prespecified hypotheses. Accordingly, the corresponding P values should be interpreted as exploratory. Before imputation, the number and proportion of missing observations for each candidate variable were summarized and are reported in Supplementary Table S1. Missing values were imputed using random forest-based imputation implemented with the “missRanger” package in R. The predictors retained in the final model (age, BMI, eGFR, comorbid diabetes, and dialysis duration) had no missing values before imputation. For patients not receiving hemodialysis, dialysis duration was categorized as “no dialysis” rather than treated as missing.
Table 1.
Comparison of Demographic and Baseline Characteristics Between Patients with and without HPRL
| Variables | Overall (N = 346) |
Normal-PRL (N = 202) |
HPRL (N = 144) |
p value |
|---|---|---|---|---|
| Sex (%) | 0.128 | |||
| Female | 135 (39) | 72 (35.6) | 63 (43.8) | |
| Male | 211 (61) | 130 (64.4) | 81 (56.2) | |
| Age (Years old, Mean ± SD) | 59.35±7.03 | 57±6.57 | 60.64±7.42 | < 0.001 |
| BMI Kg/m2(%)a | < 0.001 | |||
| <18.5 | 38 (11) | 19 (9.4) | 19 (13.2) | |
| 18.5~23.9 | 175 (50.6) | 118 (58.4) | 57 (39.6) | |
| 24~27.9 | 67 (19.4) | 49 (24.3) | 18 (12.5) | |
| ≥28 | 66 (19.1) | 16 (7.9) | 50 (34.7) | |
| Disease Course, Median (Years, Q1, Q3) |
3.00 (2.00, 6.00) | 2.00 (1.00, 5.00) | 5.00 (2.00, 7.00) | < 0.001 |
| eGFR, Median (Q1, Q3) | 7.87 (4.17, 29.46) | 19.99 (5.73, 52.59) | 4.60 (3.87, 6.97) | < 0.001 |
| CKD Etiology, n (%) | ||||
| Diabetic nephropathy | 111 (32.1) | 60 (29.7) | 51 (35.4) | 0.262 |
| IgA nephropathy | 18 (5.2) | 8 (4) | 10 (6.9) | 0.218 |
| Lupus nephritis | 26 (7.5) | 8 (4) | 18 (12.5) | 0.003 |
| Other Primary Glomerular Diseases | 93 (26.9) | 51 (25.2) | 42 (29.2) | 0.418 |
| Renal arteriosclerosis | 40 (11.6) | 31 (15.3) | 9 (6.2) | 0.009 |
| Comorbid Diabetes, n (%) | 148 (42.8) | 62 (30.7) | 86 (59.7) | < 0.001 |
| Com-Hypertension, n (%) | 277 (80.1) | 153 (75.7) | 124 (86.1) | 0.017 |
| Com-Coronary Heart Disease, n (%) | 49 (14.2) | 24 (11.9) | 25 (17.4) | 0.150 |
| Com-stroke, n (%) | 30 (8.7) | 18 (8.9) | 12 (8.3) | 0.851 |
| Hemodialysis duration, months | < 0.001 | |||
| No | 182 (52.6) | 142 (70.3) | 40 (27.8) | |
| ≤3 | 78 (22.5) | 39 (19.3) | 39 (27.1) | |
| >3 | 86 (24.9) | 21 (10.4) | 65 (45.1) | |
| Peritoneal dialysis duration, months | 0.173 | |||
| No | 317 (91.6) | 189 (93.6) | 128 (88.9) | |
| ≤3 | 17 (4.9) | 9 (4.5) | 8 (5.6) | |
| >3 | 12 (3.5) | 4 (2) | 8 (5.6) | |
| U-Alb (%) | 0.012 | |||
| ≤150 | 42 (12.1) | 32 (15.8) | 10 (6.9) | |
| >150 | 304 (87.9) | 170 (84.2) | 134 (93.1) | |
| U-24hTP, mg/24h (%) | 0.018 | |||
| ≤140 | 35 (10.1) | 27 (13.4) | 8 (5.6) | |
| >140 | 311 (89.9) | 175 (86.6) | 136 (94.4) | |
| PTH-ULN, ng/mL | 1.12 (0.56, 3.61) | 0.80 (0.42, 1.84) | 2.20 (0.88, 4.87) | < 0.001 |
| FPG-ULN, mmol/L | 1.08 (0.81, 1.39) | 0.93 (0.76, 1.27) | 1.29 (0.88, 1.51) | < 0.001 |
| U-ACR-ULN, mg/g | 18.48 (1.64, 45.64) | 12.50 (1.00, 40.12) | 22.70 (3.17, 66.22) | 0.014 |
Notes: aBMI was classified according to the Chinese adult BMI classification as follows: underweight (<18.5 kg/m2), normal weight (18.5–23.9 kg/m2), overweight (24.0–27.9 kg/m2), and obesity (≥28.0 kg/m2).
To construct the predictive model for HPRL, univariable logistic regression was first used as a broad pre-screening step (P < 0.10). This relatively permissive threshold was used to avoid prematurely excluding potentially relevant predictors while reducing the initial candidate set before LASSO selection. LASSO logistic regression was then performed using the “glmnet” package in R, with ten-fold cross-validation used to select the optimal penalty parameter (λ_min). Ten-fold cross-validation was used only for LASSO tuning and not for internal validation of the final model. Variables with non-zero LASSO coefficients were subsequently entered into multivariable logistic regression to estimate adjusted coefficients and derive a parsimonious, interpretable model; no variables were forced into the final model. Variables remaining statistically significant (P < 0.05) were retained in the final nomogram.
Multicollinearity among predictors entered into the multivariable model was assessed using variance inflation factors (VIFs). Dialysis status and dialysis duration were not simultaneously included because dialysis duration was categorized as no dialysis, ≤3 months, and >3 months, thereby incorporating dialysis status.
A nomogram was constructed using the “rms” package. Internal calibration was assessed using 1000 bootstrap resamples, calibration curves, and the Hosmer–Lemeshow goodness-of-fit test. Discrimination was evaluated by ROC analysis using the “pROC” package; the development-cohort AUC represented apparent performance and was not optimism-corrected. Sensitivity and specificity were calculated at the cutoffs generated by the ROC analysis pipeline. As an additional fixed-threshold analysis, the Youden cutoff derived from the development cohort was applied unchanged to the external validation cohort. Clinical utility was assessed by decision curve analysis using the “ggDCA” package. A sensitivity analysis was performed in the development cohort by including study center as an additional covariate in the final multivariable model. All statistical tests were two-sided, with P < 0.05 considered statistically significant.
Results
Baseline Characteristics
The key demographic and baseline characteristics of the study population are summarized in Table 1, with additional baseline characteristics provided in Supplementary Table S6. Unadjusted between-group comparisons showed differences in age, BMI distribution, disease course, eGFR, lupus nephritis, renal arteriosclerosis, comorbid diabetes, hypertension, hemodialysis duration, U-Alb, 24-hour urinary total protein, PTH-ULN, FPG-ULN, and U-ACR-ULN (all nominal P < 0.05).
Compared with patients in the normal-PRL group, patients with HPRL were older and had a longer disease course and lower eGFR. The prevalence of comorbid diabetes and hypertension was also higher in the HPRL group. Regarding BMI distribution, obesity was more common in the HPRL group (34.7% vs 7.9%), whereas normal weight (39.6% vs 58.4%) and overweight (12.5% vs 24.3%) were less common; the proportion of underweight patients was 13.2% in the HPRL group and 9.4% in the normal-PRL group.
HPRL prevalence did not differ significantly between the development and external validation cohorts (43.2% vs 37.5%, P = 0.336), while differences in BMI distribution, eGFR, hemodialysis status, and dialysis duration were observed between the cohorts (Supplementary Table S5).
Variable Selection Using LASSO Regression and Logistic Regression
Univariable logistic regression analysis identified age, BMI, eGFR, lupus nephritis, renal arteriosclerosis, comorbid diabetes, dialysis, dialysis duration, U-Alb, U-24hTP, PTH-ULN, FPG-ULN, WBC-ULN, and U-ACR-ULN as variables meeting the prespecified screening criterion of P < 0.10 and therefore eligible for subsequent LASSO regression (Table 2). Results for the remaining candidate variables are provided in Supplementary Table S7.
Table 2.
Variables Meeting the Prespecified Screening Criterion in Univariable Logistic Regression Analysis for HPRL
| Variables | β | SE | z | OR (95% CI) | P value |
|---|---|---|---|---|---|
| Age | 0.057 | 0.011 | 5.205 | 1.059 (1.036–1.082) | < 0.001 |
| BMI, Kg/m2(%) | |||||
| <18.5 | 1.535 | 0.562 | 2.730 | 4.643 (1.542–13.982) | 0.006 |
| 18.5~23.9 | Ref. | ||||
| 24~27.9 | −0.631 | 0.386 | −1.635 | 0.532 (0.250–1.133) | 0.102 |
| ≥28 | 2.244 | 0.429 | 5.237 | 9.432 (4.073–21.845) | < 0.001 |
| EGFR, mL/min/1.73 m2 | −0.084 | 0.016 | −5.129 | 0.920 (0.891–0.950) | < 0.001 |
| Cause – LN (%) | |||||
| No | Ref. | ||||
| Yes | 1.041 | 0.517 | 2.013 | 2.833 (1.028–7.812) | 0.044 |
| Cause-renal arteriosclerosis (%) | |||||
| No | Ref. | ||||
| Yes | −0.982 | 0.428 | −2.293 | 0.374 (0.162–0.867) | 0.022 |
| Comorbid diabetes (%) | |||||
| No | Ref. | ||||
| Yes | 1.886 | 0.286 | 6.593 | 6.596 (3.765–11.557) | < 0.001 |
| Dialysis (%) | |||||
| No | Ref. | ||||
| Yes | 1.897 | 0.304 | 6.243 | 6.664 (3.674–12.087) | < 0.001 |
| Dialysis duration, months | |||||
| No | Ref. | ||||
| ≤3 months | 1.366 | 0.344 | 3.965 | 3.919 (1.995–7.698) | < 0.001 |
| >3 months | 2.008 | 0.351 | 5.729 | 7.449 (3.748–14.807) | < 0.001 |
| U-24hTP, mg/24h (%) | |||||
| ≤140 | Ref. | ||||
| >140 | 1.173 | 0.659 | 1.780 | 3.231 (0.888–11.749) | 0.075 |
| U-Alb, mg/L (%) | |||||
| ≤150 | Ref. | ||||
| >150 | 0.933 | 0.430 | 2.169 | 2.542 (1.094–5.909) | 0.030 |
| PTH-ULN | 0.126 | 0.041 | 3.066 | 1.134 (1.046–1.229) | 0.002 |
| WBC-ULN | 1.166 | 0.619 | 1.882 | 3.208 (0.953–10.799) | 0.060 |
| FPG-ULN | 1.806 | 0.352 | 5.135 | 6.084 (3.054–12.119) | < 0.001 |
| U-ACR-ULN | 0.006 | 0.003 | 2.212 | 1.006 (1.001–1.011) | 0.027 |
Notes: Variables with P < 0.10 in univariable logistic regression analysis were considered for subsequent LASSO regression. Results for the remaining candidate variables are provided in Supplementary Table S7.
Abbreviation: SE, standard error.
Fourteen candidate variables were included in the least absolute shrinkage and selection operator (LASSO) regression analysis. At the selected λ_1se, 7 non-zero coefficients corresponding to seven predictor constructs were retained: age, BMI category, eGFR, comorbid diabetes, dialysis duration, FPG-ULN, and WBC-ULN Figure 2. These seven predictor constructs were subsequently entered into the multivariable logistic regression model. VIF values ranged from 1.02 to 1.21, indicating no evidence of problematic multicollinearity. Five predictors were retained in the final model: age, BMI category, eGFR, comorbid diabetes, and dialysis duration (Table 3). Center adjustment did not materially change the retained predictor estimates or improve model fit (likelihood-ratio P = 0.639; Supplementary Table S4).
Figure 2.
Variable selection using LASSO logistic regression. (A) LASSO coefficient profiles of the 14 candidate variables. (B) Selection of the tuning parameter (λ) by 10-fold cross-validation. The left and right vertical dotted lines indicate λ_min and λ_1se, respectively. λ_1se = 0.0191, corresponding to the minimum cross-validated binomial deviance, was selected; at this value, 10 non-zero coefficients corresponding to seven predictor constructs were retained, and these seven predictor constructs were subsequently entered into the multivariable logistic regression model.
Table 3.
LASSO-Derived Multivariable Logistic Regression Model to Predict HPRL in the Development Cohort
| Variables | β | SE | z | OR (95% CI) | P value |
|---|---|---|---|---|---|
| Age | 0.062 | 0.015 | 4.102 | 1.064 (1.033–1.096) | < 0.001 |
| aBMI kg/m2(%) | |||||
| <18.5 | 1.433 | 0.663 | 2.162 | 4.191 (1.143–15.360) | 0.031 |
| 18.5~23.9 | Ref. | ||||
| 24~27.9 | −0.925 | 0.528 | −1.752 | 0.396 (0.141–1.116) | 0.080 |
| ≥28 | 2.484 | 0.627 | 3.959 | 11.989 (3.505–41.011) | < 0.001 |
| EGFR, mL/min/1.73 m2 | −0.068 | 0.020 | −3.388 | 0.935 (0.899–0.972) | < 0.001 |
| Comorbid diabetes (%) | |||||
| 0 | Ref. | ||||
| 1 | 1.614 | 0.461 | 3.499 | 5.024 (2.034–12.411) | < 0.001 |
| Dialysis duration, months (%) | |||||
| No | Ref. | ||||
| ≤3 | 1.119 | 0.507 | 2.205 | 3.060 (1.133–8.270) | 0.027 |
| >3 | 1.312 | 0.502 | 2.615 | 3.715 (1.389–9.933) | 0.009 |
| FPG-ULN | 0.509 | 0.507 | 1.005 | 1.664 (0.616–4.495) | 0.315 |
| WBC-ULN | 1.376 | 0.954 | 1.442 | 3.958 (0.610–25.676) | 0.149 |
Notes: aBMI <18.5 and ≥28.0 kg/m2 were treated as risk categories, with 18.5–23.9 kg/m2 as the reference category.
Abbreviation: ULN, upper limit of normal.
Construction of a Nomogram Predicting HPRL Risk
Based on the five statistically significant predictors retained in the multivariable logistic regression model, a final parsimonious nomogram was constructed using age, BMI category, eGFR, comorbid diabetes, and dialysis duration (Figure 3). Each variable was assigned a score according to its regression coefficient, and the cumulative score was used to estimate the risk of HPRL in patients with CKD.
Figure 3.
Nomogram for estimating the probability of hyperprolactinemia (HPRL) in patients with chronic kidney disease (CKD).
To illustrate the application of this nomogram, consider a CKD patient with the following baseline characteristics: age 70 years, body mass index (BMI) of 22.5 kg/m2, eGFR of 10mL/min/1.73m2, comorbid diabetes mellitus, and a dialysis duration of more than 3 months. Using the nomogram, each parameter is assigned a corresponding score, and the total score is the sum of these individual scores. Based on the total score, the nomogram predicts a 90% probability of HPRL in this patient, indicating a high risk (Figure 3). This illustrative application demonstrates the potential use of the nomogram for individualized HPRL risk estimation, risk stratification, and subsequent clinical evaluation in patients with CKD.
Model Performance
In the development cohort, the apparent AUC was 0.914 (95% CI, 0.879–0.949), compared with 0.896 (95% CI, 0.834–0.958) in the independent external validation cohort (n = 96). At the analysis-pipeline cutoffs of 0.346 and 0.305, sensitivity and specificity were 88.9% and 79.6% in the development cohort and 77.8% and 83.3% in the external validation cohort, respectively. Applying the development-derived Youden cutoff of 0.392 unchanged to the external cohort yielded a sensitivity of 72.2% and specificity of 86.7% (Supplementary Table S2). The Brier scores were 0.114 and 0.129, the calibration intercepts were 0.000 and 0.356, and the calibration slopes were 1.000 and 0.764 in the development and external validation cohorts, respectively. Hosmer–Lemeshow P values were 0.747 and 0.142, respectively. DCA showed that the nomogram generally provided a higher net benefit than the treat-all and treat-none strategies across threshold probabilities of approximately 0.10–0.90 in both cohorts (Figure 4). This finding suggests potential decision-analytic benefit rather than established clinical utility, as the effects of risk-guided prolactin testing on clinical management or patient outcomes were not directly evaluated.
Figure 4.
Model performance in the development and external validation cohorts. (A and D) Receiver operating characteristic (ROC) curves; (B and E) calibration curves; and (C and F) decision curve analyses (DCA) in the development cohort (n = 250) and external validation cohort (n = 96), respectively.
Abbreviation: AUC, area under the ROC curve.
The full nomogram was also compared with models based on eGFR alone and eGFR plus dialysis duration (Supplementary Table S3). In the external validation cohort, AUCs were 0.896, 0.881, and 0.933, respectively; DeLong tests showed no significant differences (P = 0.623 and P = 0.087). Thus, the full model did not show a statistically significant improvement in discrimination over the simpler models.
Discussion
This multicenter retrospective investigation developed and externally corroborated a nomogram for hyperprolactinemia risk in patients with chronic kidney disease. The finalized algorithm integrated five routinely accessible clinical metrics: age, BMI, eGFR, comorbid diabetes, and dialysis duration. The model demonstrated good discrimination in both the development and independent external validation cohorts, although some degree of miscalibration was observed in the external cohort.
Previous studies of prolactin abnormalities in CKD have primarily focused on their prevalence, pathophysiology, and associations with kidney function, dialysis, and related clinical abnormalities.4,5,7 In contrast, the present study focused on individualized risk prediction by integrating five routinely available clinical variables into a nomogram and independently validating the model in a third center. Thus, the novelty of this study lies not in establishing the association between CKD and HPRL, but in translating existing clinical associations into an externally validated tool for individualized HPRL risk estimation.
Compared with simpler clinical models, the full nomogram showed greater discrimination in the development cohort, but did not demonstrate a statistically significant AUC advantage over eGFR alone or eGFR plus dialysis duration in the external validation cohort. These findings suggest that eGFR and dialysis exposure account for a substantial proportion of the predictive information for HPRL. The incremental value of the additional predictors therefore requires confirmation in larger external cohorts.
Our findings are consistent with previous evidence showing that prolactin levels are frequently elevated in patients with advanced CKD and dialysis-dependent kidney failure, mainly because of impaired renal clearance and altered dopaminergic regulation of prolactin secretion.18–20 In the present study, lower eGFR and longer dialysis duration were independently associated with HPRL, supporting the close relationship between declining renal function, dialysis exposure, and prolactin accumulation. These findings indicate that prolactin assessment may be particularly relevant in patients with advanced kidney dysfunction or prolonged dialysis exposure.
Metabolic status also appeared to be important, as previous studies have suggested an association between prolactin and glucose–lipid metabolism.21 A recent observational study in patients with CKD reported associations between prolactin levels, inflammatory biomarkers such as IL-1β and TNF-α, and indices of glycemic dysregulation.5 In our study, comorbid diabetes and extreme BMI categories were retained in the final model, suggesting that metabolic factors may contribute to HPRL risk in patients with CKD.22
Proteinuria-related variables and parathyroid hormone (PTH) differed between patients with and without HPRL in baseline comparisons and univariable analyses but were not retained after multivariable adjustment.23 Elevated PTH and phosphate levels are well-recognized features of CKD–mineral and bone disorder (CKD-MBD)24 and may therefore reflect the severity of underlying kidney dysfunction rather than provide independent predictive information. Their associations were attenuated after adjustment for eGFR and dialysis duration, supporting this interpretation. In addition, substantial missingness in proteinuria-related variables may have further limited their contribution to the final nomogram.
Therefore, these associations may reflect overall disease burden rather than independent predictive effects. Recent evidence also suggests that prolactin may be involved in cardiometabolic regulation in addition to its established reproductive functions.25 Further mechanistic studies are needed to determine whether metabolic abnormalities directly contribute to HPRL or whether both mainly reflect progressive kidney dysfunction.
The main strength of this study is the development of a clinically interpretable nomogram based on easily obtainable variables. Rather than recommending universal prolactin testing for all CKD patients, the model may support a risk-based strategy by identifying individuals who are more likely to have HPRL. This approach may improve diagnostic efficiency, reduce unnecessary testing in low-risk patients, and prompt earlier endocrine evaluation in high-risk patients.26–29 However, the nomogram should be viewed as a decision-support tool rather than a substitute for clinical judgment or biochemical confirmation.
Limitations
This study has several limitations. First, its retrospective design limits causal inference and may introduce selection bias. In addition, the relatively high AUC observed in the development and external validation cohorts should be interpreted cautiously, as the retrospective design, data-driven predictor selection, and relatively limited sample size may have contributed to optimistic performance estimates and potential overfitting. Second, although PRL values were normalized using assay-specific upper limits of normal, inter-center differences in laboratory platforms and dialysis practices may still have influenced the results. Third, data on macroprolactin testing, pituitary imaging, and detailed exposure to medications that may affect prolactin secretion were limited. Finally, long-term renal, cardiovascular, reproductive, and endocrine outcomes were not assessed; therefore, the prognostic significance of HPRL predicted by this model remains to be determined. In addition, the higher exclusion proportion at Center 3 may have reduced the representativeness of the external validation cohort and should be considered when interpreting the transportability of the validation results.
Conclusions
In conclusion, this multicenter retrospective study developed and externally validated a nomogram for estimating the probability of laboratory-defined HPRL in patients with CKD. The model integrates renal, metabolic, and dialysis-related factors and demonstrated good performance in an independent external cohort, suggesting potential value for prioritizing prolactin assessment. Further prospective multicenter validation and clinical-impact assessment are required before routine clinical implementation.
Acknowledgments
The authors thank all the participants for their contributions.
Funding Statement
This work was supported by grants from the Medical Research Project of the Jiangsu Commission of Health (No. ZDB2020023), and the Cross-disciplinary Scientific Research Project of Sir Run Run Hospital, Nanjing Medical University (No. YFHX-2023008).
Abbreviations
HPRL, Hyperprolactinemia; CKD, chronic kidney disease; LASSO, least absolute shrinkage and selection operator; AUC, Area Under the Curve; DCA, Decision curve analysis; ULN, upper limit of normal; BMI, Body mass index; eGFR, Estimated glomerular filtration rate; LN, Lupus nephritis; U-Alb, urinary microalbumin; 24-UTP, 24-hour urinary total protein; PTH, parathyroid hormone; FPG, fasting plasma glucose; U-ACR, urinary albumin-to-creatinine ratio; DN, Diabetic nephropathy; IgAN, IgA nephropathy; COM, Comorbidity; HD, Hemodialysis; PD, Peritoneal dialysis; Ca, calcium; P, phosphorus; ALB, Albumin; WBC, White blood cell; Hb, Hemoglobin; RBC, Red blood cell; PLT, Platelet.
Data Sharing Statement
The data underlying this article will be shared on reasonable request to the corresponding author.
Ethical Approval and Informed Consent
This study was conducted in accordance with the principles of the Declaration of Helsinki and relevant national regulations. The study protocol was reviewed and approved by the Institutional Review Boards of the three participating medical centers: Sir Run Run Hospital of Nanjing Medical University (Approval No.2024-SR-045), Jinling Hospital Affiliated to the Medical School of Nanjing University (Approval No.DZQH-KYLL-25-35), and Yijishan Hospital (The First Affiliated Hospital of Wannan Medical College) (Approval No.2025-148). Written informed consent was obtained from all participants before enrollment. All data were anonymized and handled confidentially to protect the privacy of the study subjects.
Consent for Publication
Consent for publication was deemed not applicable.
Author Contributions
All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.
Ming Ji: Conceptualization, funding acquisition, investigation, methodology, project administration, writing – original draft, and writing – review and editing.
Dongrui Liu: Data curation, formal analysis, methodology, software, visualization, writing – original draft, and writing – review and editing.
Xiaoren Peng: Data acquisition from Center 2, investigation, methodology, resources, and writing – review and editing.
Hongjing Zhao: Data acquisition from Center 3, data curation, investigation, methodology, and writing – review and editing.
Zhaowei Wang: Formal analysis, methodology, software, validation, visualization, and writing – review and editing.
Weijuan Deng: Data curation, formal analysis, validation, and writing – review and editing.
Hao Zhang: Formal analysis, methodology, supervision, validation, and writing – review and editing.
Hanbin Zhang: Data curation, investigation, validation, and writing – review and editing.
Changchun Cao: Conceptualization, funding acquisition, investigation, project administration, supervision, resources, writing – original draft, and writing – review and editing.
Disclosure
The authors declare no conflicts of interest in this work.
References
- 1.Romagnani P, Agarwal R, Chan JCN, et al. Chronic kidney disease. Nat Rev Dis Primers. 2025;11(1):8. doi: 10.1038/s41572-024-00589-9 [DOI] [PubMed] [Google Scholar]
- 2.Webster AC, Nagler EV, Morton RL, Masson P. Chronic kidney disease. Lancet. 2017;389(10075):1238–14. doi: 10.1016/S0140-6736(16)32064-5 [DOI] [PubMed] [Google Scholar]
- 3.Zhang Q, Lu J, Lei L, et al. Nomogram to predict rapid kidney function decline in population at risk of cardiovascular disease. BMC Nephrol. 2022;23(1):62. doi: 10.1186/s12882-022-02696-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Huang W, Molitch ME. Prolactin and other pituitary disorders in kidney disease. Semin Nephrol. 2021;41(2):156–167. doi: 10.1016/j.semnephrol.2021.03.010 [DOI] [PubMed] [Google Scholar]
- 5.Rusu CC, Moldovan D, Potra A, et al. The interplay of prolactin with inflammatory nutritional markers and NT-proBNP in chronic kidney disease. Int J Mol Sci. 2025;26(13):6347. doi: 10.3390/ijms26136347 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Lo JC, Beck GJ, Kaysen GA, et al. Hyperprolactinemia in end-stage renal disease and effects of frequent hemodialysis. Hemodial Int. 2017;21(2):190–196. doi: 10.1111/hdi.12489 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Petersenn S, Fleseriu M, Casanueva FF, et al. Diagnosis and management of prolactin-secreting pituitary adenomas: a Pituitary Society international consensus statement. Nat Rev Endocrinol. 2023;19(12):722–740. doi: 10.1038/s41574-023-00916-2 [DOI] [PubMed] [Google Scholar]
- 8.Dourado M, Cavalcanti F, Vilar L, Cantilino A. Relationship between prolactin, chronic kidney disease, and cardiovascular risk. Int J Endocrinol. 2020;2020:9524839. doi: 10.1155/2020/9524839 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Ghosh SK, Khandoker AH. A machine learning driven nomogram for predicting chronic kidney disease stages 3–5. Sci Rep. 2023;13(1):21613. doi: 10.1038/s41598-023-48815-w [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Yu P, Kan R, Meng X, et al. A nomogram for predicting the risk of CKD based on cardiometabolic risk factors. Int J Gen Med. 2023;16:4143–4154. doi: 10.2147/IJGM.S425122 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Li G, Jiang L, Li J, et al. Development and validation of a nomogram for predicting the 6-months survival rate of patients undergoing incident hemodialysis in China. BMC Nephrol. 2022;23(1):234. doi: 10.1186/s12882-022-02864-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Li N, Wang Z, Yang X, et al. Development and validation of a cardiovascular disease risk prediction model for patients with non-dialysis-dependent chronic kidney diseases based on the nomogram. Kidney Blood Press Res. 2023;48(1):7–17. doi: 10.1159/000527856 [DOI] [PubMed] [Google Scholar]
- 13.Wu J, Li X, Zhang H, et al. Development and validation of a prediction model for all-cause mortality in maintenance dialysis patients: a multicenter retrospective cohort study. Ren Fail. 2024;46(1):2322039. doi: 10.1080/0886022X.2024.2322039 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Wang F, Xu J, Wang F, et al. A dynamic online nomogram for predicting renal outcomes of idiopathic membranous nephropathy. BMC Med Inform Decis Mak. 2024;24(1):173. doi: 10.1186/s12911-024-02568-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Zhao X, Zheng C, Su Q, et al. Development and validation of a nomogram for predicting hyperphosphatemia in non-dialysis patients with chronic kidney disease. BMC Nephrol. 2025;26(1):512. doi: 10.1186/s12882-025-04445-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Xu W, Yang Q, Li L, Xiang Y, Yang Q. Construction and validation of a risk prediction model for mild cognitive impairment in non-dialysis chronic kidney disease patient. Kidney Blood Press Res. 2024;49(1):556–580. doi: 10.1159/000540025 [DOI] [PubMed] [Google Scholar]
- 17.Wang Z, Du W, Wei X, et al. Clinical obesity among Chinese adults: prevalence, multimorbidity burden, and associations with physical activity. Nutrients. 2026;18(6):983. doi: 10.3390/nu18060983 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Herrington WG, Judge PK, Grams ME, Wanner C. Chronic kidney disease. Lancet. 2026;407(10523):90–104. doi: 10.1016/S0140-6736(25)01942-7 [DOI] [PubMed] [Google Scholar]
- 19.Kiyani M, Chua MWJ. It keeps getting bigger: a patient with a hypothalamic tumor and hyperprolactinemia. Cureus. 2025;17(3):e80793. doi: 10.7759/cureus.80793 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Rojhani E, Rahmati M, Firouzi F, et al. Prolactin levels and chronic kidney disease and the subsequent risk of cardiovascular events: a long term population based cohort study. Sci Rep. 2025;15(1):7198. doi: 10.1038/s41598-025-87783-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Hofbauer S, Horka L, Seidenberg S, et al. Metabolic and inflammatory parameters in relation to baseline characterization and treatment outcome in patients with prolactinoma: insights from a retrospective cohort study. Front Endocrinol. 2024;15:1363939. doi: 10.3389/fendo.2024.1363939 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Ke X, Wang L, Zhao Y, et al. Serum prolactin levels were positively related to metabolic indexes and disorders in male obese patients. Endocrine. 2024;84(3):1097–1107. doi: 10.1007/s12020-024-03743-1 [DOI] [PubMed] [Google Scholar]
- 23.Di Filippo L, Doga M, Resmini E, Giustina A. Hyperprolactinemia and bone. Pituitary. 2020;23(3):314–321. doi: 10.1007/s11102-020-01041-3 [DOI] [PubMed] [Google Scholar]
- 24.Cannata-Andía JB, Martín-Carro B, Martín-Vírgala J, et al. Chronic kidney disease-mineral and bone disorders: pathogenesis and management. Calcif Tissue Int. 2021;108(4):410–422. doi: 10.1007/s00223-020-00777-1 [DOI] [PubMed] [Google Scholar]
- 25.Therkelsen KE, Abraham TM, Pedley A, et al. Association between prolactin and incidence of cardiovascular risk factors in the Framingham Heart Study. J Am Heart Assoc. 2016;5(2):e002640. doi: 10.1161/JAHA.115.002640 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Wang Z, Li J, Le Y, et al. Combinatorial classification model for predicting antipsychotic-induced hyperprolactinemia risk in female patients with schizophrenia. Am J Transl Res. 2025;17(11):9076–9083. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Cho YS, Lee JS, Byrne JV. A predictive algorithm for evaluating elevated serum prolactin in patients with a sellar mass. J Clin Neurosci. 2015;22(1):155–160. doi: 10.1016/j.jocn.2014.07.020 [DOI] [PubMed] [Google Scholar]
- 28.Lyu L, Yin S, Hu Y, et al. Hyperprolactinemia in clinical nonfunctional pituitary macroadenomas: a STROBE-compliant study. Medicine. 2020;99(41):e22673. doi: 10.1097/MD.0000000000022673 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Vilar L, Vilar CF, Lyra R L, Vilar CF, Lyra R, et al. Usefulness of prolactin levels in predicting the etiology of hyperprolactinemia in a cohort of 770 patients. Arch Endocrinol Metab. 2024:68. doi: 10.20945/2359-4292-2023-0391. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data underlying this article will be shared on reasonable request to the corresponding author.