Development of a risk prediction nomogram for sarcopenia in hemodialysis patients

All patients undergoing hemodialysis at the First Affiliated Hospital College of Medicine Zhejiang University from March to June 2021 were enrolled in our study, and their medical records were retrospectively reviewed. Patients aged ≥ 18 years with a history of hemodialysis three times or more per week for at least 3 months were eligible to participate in this study. The exclusion criteria included implanted pacemaker or amputation surgeries; miss data of skeletal muscle mass index. Finally, 615 patients were enrolled in our study. They were randomly allocated to either the development cohort (n = 369) or the validation cohort (n = 246) in a 6:4 ratio. This study was performed in accordance with Declaration of Helsinki and was approved by the Clinical Research Ethics Committee of the First Affiliated Hospital of Zhejiang University School of Medicine (IIT20210808A), and all patients signed the informed consent form.

Based on the relevant literature [3, 6, 8, 12, 24] and consulting experts, a total of 27 predictors for sarcopenia were identified. The following data were collected from each patient: ① baseline demographics: age, gender, primary disease, duration of dialysis, etc.; ② self-rating anxiety scale (SAS) and self-rating depression scale (SDS) psychological status assessment indicators; ③ anthropometric indicators: height, post-dialysis weight, skeletal muscle mass index (SMI), handgrip strength, gait speed, mid-upper arm circumference, triceps skinfold thickness, body mass index (BMI) and mid-upper arm muscle circumference (MAMC),which was calculated as mid-upper arm circumference − 3.14 × triceps skinfold thickness [18]. ④ laboratory examination parameters measured before a mid-week predialysis session following enrolment: serum creatinine, serum uric acid, serum urea nitrogen, hemoglobin, serum albumin, serum prealbumin, C-reactive protein, serum phosphorus, serum calcium, blood lipid, serum potassium, parathyroid hormone and urea clearance index.

Handgrip strength test was measured before hemodialysis with an electronic grip strength meter (EH101; CAMRY). The patient was required to stand the feet and arms naturally positioned. Three measurements with an interval of 5 s were taken, and the maximum value was recorded. The gait speed test was also evaluated before hemodialysis, without the assistance of any tools, the patient walked 6 m twice at a normal gait speed, each gait speed was observed, and the mean value of two consecutive measurements was documented. Triceps skinfold thickness and upper arm circumference were measured at the end of hemodialysis. In addition, BIA (InbodyS10, BiospaceCo,Korea) was performed 15–20 min after hemodialysis to measure appendicular skeletal muscle mass (ASM), which was later used to calculate SMI as follows: SMI = ASM/height² [15].

The Zung self-rating depression scale (SDS) and Zung self-rating anxiety scale (SAS) were employed to assess patients' depression and anxiety levels; SAS and SDS have been found to have robust internal consistency with a Cronbach s alpha of 0.82 and 0.68, respectively. Both the SAS and SDS are measured on a scale containing 20 items, with each item scored on a four-point Likert scale, with “1” indicating no or little time, “2” representing a small amount of time, “3” referring to a lot of time, and “4” representing most or all of the time. The raw scores range from 20 to 80, which are subsequently converted into standard scores by dividing the sum of the raw scores by 80 and multiplying by 100 for further evaluation. the thresholds for identifying depression and anxiety were 50 points [25, 26].

Demographic characteristics and laboratory examination parameters were collected from medical records. The anthropometric indicators and SAS and SDS scales were determined by experienced nursing researchers; SAS and SDS were explained to patients by researchers using standardized guiding terms and were then filled in by the patients.

In this study, the diagnosis of sarcopenia was based on the AWGS 2019 criteria [27]. ①low appendicular skeletal muscle mass (ASM): BIA < 7.0 kg/m² in males and < 5.7 kg/m² in females ②low muscle strength: handgrip strength < 28 kg in males and < 18 kg in females; ③low physical performance: gait speed (6-m walk): < 1.0 m/s. The diagnosis of sarcopenia was made in patients with low ASM + low muscle strength OR low physical performance.

SPSS 26 software (IBM Corporation, Armonk, NY, USA) and R software (version 4.2.1, R Foundation for Statistical Computing, Vienna, Austria) were used to analyze the collected data. Among the 27 predictor variables, prealbumin, urea clearance index, parathyroid hormone, upper arm circumference, and skinfold thickness were identified with less than 5% missing data, which were replaced by the series mean of the continuous variables before the model was developed. To ensure more convenient clinical use, continuous variables such as age, serum albumin, serum creatinine, C-reactive protein, serum phosphorus, serum calcium, parathyroid hormone, triglycerides, cholesterol, high-density lipoprotein, low-density lipoprotein, and very-low-density lipoprotein, BMI were converted into categorical variables according to clinical references [27‐30]. Other continuous variables were converted to dichotomous variables according to the receiver operating characteristic curve using the optimal cut-off points of the analyzed variables with the Youden index criterion. Frequency or constituent ratio was used for statistical description, and the chi-square test was used for comparison between groups. Through univariate logistic regression analysis, variables with univariate analysis results of p < 0.05 were included in the multivariate logistic regression analysis. The backward stepwise regression method was then applied to select statistically significant factors used to construct the clinical risk prediction nomogram. The predictive nomogram was assessed in three aspects: discrimination, calibration, and clinical net benefit. The area under the receiver operating characteristic curve (ROC) was utilized to assess discriminatory ability. The calibration curve and Hosmer–Lemeshow goodness of fit test were used to assess calibration ability. The decision curve analysis (DCA) was used to assess clinical effectiveness. The model was internally validated using the Bootstrap method(resampling = 1000). A p-value < 0.05 was considered statistically significant.

A total of 615 patients undergoing hemodialysis were eventually included in this study and were randomly divided into the development cohort (n = 369) and the validation cohort (n = 246) in a 6:4 ratio, as illustrated in (Fig. 1). Among the enrolled patients, 381 (62%) were male and 234 (38%) were female. The mean patient age was 60.07 ± 14.34 years, the median duration of dialysis was 60 months (interquartile range 21–110). The most common reason for dialysis was chronic glomerulonephritis (430 patients), followed by diabetic nephropathy (101 patients), and other causes (hypertensive nephropathy, polycystic kidney disease, lupus nephritis, etc.). There were 102 patients (16.60%) diagnosed with sarcopenia and the demographic parameters of patients with or without sarcopenia are summarized in (Table 1).

Univariate logistic regression analysis was performed with the included 27 independent variables, and the results demonstrated that age, urea clearance index, serum creatinine, upper arm muscle circumference, as well as levels of blood uric acid, albumin, C-reactive protein, serum phosphorus, alkaline phosphatase, BMI, prealbumin, and triglycerides were significantly different between the two groups (P < 0.05), as presented in (Table 2). Next, significant independent variables obtained from the above univariate logistic regression analysis were included in multivariate logistic regression analysis following a backward stepwise regression method. The results exposed that age, C-reactive protein, and serum phosphorus levels, as well as BMI and MAMC, were independent risk factors for sarcopenia in HD patients, as outlined in Table 2.

The risk prediction model for sarcopenia in HD patients was then established based on the formula: P = 1/(1 + e^Y) where e stands for the base of natural logarithm, Y = -3.316 + 0.651 × age ≥ 60 years + 1.168 × C-reactive protein ≥ 3 mg/L + 2.769 × serum phosphorus < 1.13 mmol/L-1.011 × serum phosphorus 1.13 ~ 1.78 mmol/L + 1.428 × BMI < 18.5 kg/m² -2.122 × BMI 18.5 ~ 24.9 kg/m² + 1.422 × MAMC < 22.64 cm. A nomogram of the risk prediction model for sarcopenia in HD patients was also plotted, as depicted in (Fig. 2). The corresponding score for each variable was obtained by crossing to the nomogram, and the total score was used to predict the risk of developing sarcopenia in HD patients.

The validation of this prediction model was performed by area under the curve and the calibration curve. Our data indicated that the AUC for this model was 0.869 (95% CI (0.822 to 0.915) (Fig. 3), with a sensitivity, specificity, and Youden index of 77% and 83%, and 0.60, respectively, in the development cohort. Similarly, the AUC of the validation cohort was 0.832 (95% CI (0.765 to 0.900) (Fig. 4), with a sensitivity of 70%, a specificity of 88%, and a Youden index of 0.58. In addition, the Hosmer–Lemeshow goodness-of-fit test showed promising fit (development cohort \(\chi\) ² = 4.001, P = 0.911; validation cohort \(\chi\) ² = 13.941, P = 0.124). In the development cohort and validation cohort, the calibration curve showed agreement between the observation and prediction as displayed in (Figs. 5 and 6). The model was then internally validated using the Bootstrap method with an internal validation C-statistic of 0.783.

The clinical validity of this predictive model was assessed using decision curve analysis (DCA). The DCA for the development and validation cohorts are displayed in (Figs. 7 and 8), which signaled that patients could benefit from this novel predictive model when the threshold was set to 10 ~ 80% and 10 ~ 70% for the development and validation cohorts, respectively. The positive predictive value, negative predictive value, positive likelihood ratio, negative likelihood ratio for this nomogram were 45%,93%,4.34,0.418, Accuracy was 82%.

To make this predictive model more convenient for physicians to use in clinical practice, we modified the nomogram into a scoring system with integer points: age ≥ 60 years(18points), C-reactive protein ≥ 3 mg/L (30points), serum phosphorus < 1.13 mmol/L (100points), serum phosphorus 1.13 ~ 1.78 mmol/L (27 points) and BMI < 18.5 kg/m² (93 points), BMI 18.5 ~ 24.9 kg/m² (56 points), MAMC < 22.64 cm(37 points). Then, the risk score model of sarcopenia in hemodialysis patients was established. The total score was 0–350 points of the nomogram, and the higher of the total score, the higher the risk of sarcopenia. The weights for each feature are list in Table 3 for calculation without a nomogram. the AUC of the scoring system was 0.867 (95% CI (0.830 to 0.903), with a sensitivity, specificity, and Youden index of 88% and 73%, and 0.611. The optimal cutoff value of 121 was taken as the risk threshold, if the total score < 121 points was classified as the low-risk group of sarcopenia, and the total score > 121 points was classified as the high-risk group.