Abstract
Background and hypothesis Sarcopenia is common in patients undergoing maintenance haemodialysis (MHD); however, the current diagnostic support tools for sarcopenia are difficult to implement in dialysis clinics. This study aimed to develop a clinically friendly screening tool to predict sarcopenia using ubiquitous clinical data.
Methods This cross-sectional multicentre study enrolled 373 and 129 patients undergoing MHD in the derivation and external validation cohorts, respectively. The Asian Working Group for Sarcopenia diagnostic criteria were used as a sarcopenia reference standard. Candidate predictors, such as age, sex, body mass index, routine blood tests, and the one-item Clinical Frailty Scale (CFS) version 2.0, were used to develop an original web-based model and a paper-based point score system using backward elimination selection. The two tools were completed using optimism-corrected regression coefficients for each variable, derived by bootstrapping. Their performance was evaluated by examining the discrimination and calibration in the two cohorts.
Results In total, 98 (26.3%) and 44 (34.1%) patients in the derivation and validation cohorts were diagnosed with sarcopenia, respectively. For internal validation, the area under the receiver operating characteristic curve (AUROC) for the original model and the point score system were 0.97 (95% CI: 0.96– 0.98) and 0.95 (95% CI: 0.93–0.97), respectively. Calibration plots for the original model showed excellent agreement between the predicted and observed probabilities. In contrast, the point-score-based model underestimated sarcopenia in the moderate-risk range. For external validation, the original model achieved an AUROC of 0.97 (95% CI: 0.95–1.00), while the point score system achieved an AUROC of 0.91 (95% CI: 0.87–0.96). The calibration plots for both models showed similar performances to those of the internal validation.
Conclusion In patients undergoing MHD, our practical diagnostic support tool ‘the ABC2-Screener’ has good discrimination and calibration abilities and can be easily used at any medical facility.
Introduction
Patients undergoing maintenance haemodialysis (MHD) exhibit a high prevalence of sarcopenia owing to their exposure to chronic inflammatory conditions (1, 2), metabolic acidosis (3), malnutrition (4, 5), and inactivity (3, 6). In addition, the increased risk of fractures (7), cardiovascular events (8), and death (5, 9) associated with concomitant sarcopenia requires the need for accurate identification of sarcopenia among patients undergoing haemodialysis (6). Sarcopenia is diagnosed on the basis of a decrease in skeletal muscle mass, muscle strength, and/or physical performance (10, 11). Presently, special equipment such as dual-energy X-ray absorptiometry (DXA) and bioelectrical impedance analysis (BIA) are required to determine skeletal muscle mass. However, these devices are expensive and not available in all haemodialysis facilities. Furthermore, DXA has disadvantages such as radiation exposure. Consequently, there is a growing need for simple screening methods for sarcopenia that can be easily implemented in dialysis facilities regularly attended by patients. However, evidence to address this issue is insufficient.
Several studies on clinical prediction rules have been conducted to screen for sarcopenia in patients undergoing dialysis. However, many of these are nomogram-based models (12–14) and may not be preferred in clinical practice because of the complexity at first sight and inaccuracies associated with translation to final outcomes (15, 16). Additionally, in previously reported prediction rules, predictor variables included grip strength, a component of the criteria for diagnosing sarcopenia (14, 17), and irisin and fat-free mass index, which are not measured in the usual course of care (17). Therefore, there is a need for a reliable method to screen for sarcopenia in dialysis care settings based on data available in routine dialysis care, without performing examinations required for diagnosing sarcopenia.
To address this issue, we conducted a multicentre cross-sectional study to develop and externally validate the ABC2-Screener, a simple and reliable diagnostic support tool for predicting sarcopenia in patients undergoing maintenance haemodialysis. To ensure applicability in dialysis facilities, we focused on variables obtained in usual care processes (age, sex, body mass index [BMI], and routinely performed blood laboratory tests) and a 1-item Clinical Frailty Scale (CFS) (18,19), which rates frailty status based on clinician judgment. In addition, we aimed to present two versions of the ABC2-Screener: a paper-based point-scoring system and a user-friendly web-based programme.
Materials and Methods
Setting and Subjects
This multicentre cross-sectional study was conducted at four medical facilities providing outpatient haemodialysis services (Kameda Medical Center, Awa Regional Medical Center, Chikuseikai Munakata Clinic, and Munakata Clinic). The subjects were adult patients aged ≥ 18 years who underwent maintenance haemodialysis (three times a week) for ≥ 3 months. The exclusion criteria were as follows: (1) a history of pacemaker implantation or arthroplasty that rendered BIA measurement invalid or (2) a history of lower limb amputation, hemiplegia, or numbness due to carpal tunnel syndrome, which makes BIA measurement difficult.
A total of 502 patients were included in this study. The derivation set (n=373) included patients from two medical facilities (Kameda Medical Center and Munakata Clinic). The external validation set (n=129) included the remaining two medical facilities (Awa Regional Medical Center and Chikuseikai Munakata Clinic) (Figure 1).
This study was conducted in accordance with the Declaration of Helsinki and approved by the Ethical Review Board of Fukushima Medical University (number: ippan2021-292). All the patients signed an informed consent form. Data were collected between April 2022 and February 2023.
Reference standard: diagnosis of sarcopenia
The diagnostic criteria recommended by the 2019 Asian Sarcopenia Working Group (AWGS), which includes loss of muscle mass and decrease in muscle strength or physical performance, was used (11). To assess muscle strength, handgrip strength was measured before the haemodialysis session using an electronic handgrip meter (TKK 5101 Grip-D; Takei Rika Kikai Co., Ltd., Tokyo, Japan). The measurement was performed by providing maximal isometric contraction with the patient in a sitting position with the shoulder joint adducted, the elbow joint flexed at 90°, and the forearm and wrist in a neutral position. The measurements were repeated twice on each side and the average of the highest values on each side was recorded. The cut-off values for low muscle strength were < 28.0 kg for men and < 18.0 kg for women (11). Height-adjusted appendicular skeletal muscle mass (ASM) was measured 30 min after the haemodialysis session using multifrequency BIA (MLT-550N; Toray Medical, Inc., Tokyo, Japan) (20, 21). Based on the same analytical principle as the MLT-550N used in this study, fat-free mass (FFM) by BIA has been well validated with FFM by the dual-energy x-ray absorptiometry method (DPX-L; Lunar) (Pearson correlation coefficient = 0.88, standard error of estimate = 1.61, FFM [DEXA] = 0.99 x FFM [BIA] + 0.26 kg) (22). All patients were placed in the supine position with their arms away from the torso and legs not touching each other. The participants were asked to remain silent and stationary during the measurements. Reduced skeletal muscle mass was defined as an ASM < 7.0 kg/m2 in men and ASM < 5.7 kg/m2 in women (11).
Choice and measurement of candidate predictors
Age, sex, dialysis duration, BMI, primary renal disease, comorbidities, and laboratory blood data were collected from the medical records. The following blood laboratory tests were performed before dialysis on Monday or Tuesday, two days after the last dialysis: serum creatinine (Cr), blood urea nitrogen, potassium, phosphorus, calcium, albumin, C-reactive protein (CRP), and haemoglobin. Measurements were restricted to these blood tests because they are recommended to be assessed at least monthly in Japan and are measured worldwide (23). In all facilities, Cr was measured using an enzymatic assay, and albumin was measured using a modified bromocresol purple assay. To determine BMI, body weight was measured to the nearest 0.1 kg after the first haemodialysis session at the beginning of the week. Clinical Frailty Scale (CFS) version 2.0 is a global rating scale for frailty based on the investigator’s clinical judgment using available clinical information. The scale was rated from 1–9 (1 = very healthy, 2 = good, 3 = well-controlled, 4 = frail, 5 = mild frailty, 6 = moderate frailty, 7 = severe frailty, 8 = very severe frailty, and 9 = terminal illness). For classification purposes, a CFS score of 1–3 was considered ‘non-frail’, 4 was considered ‘pre-frail’, and > 5 was considered ‘frail’ (18, 19).
Statistical analyses
To describe candidate predictors, continuous variables were summarised as median and interquartile range (IQR), and categorical variables as number of patients and percentages.
Original model building for web-based programme
To create a clinical prediction model for use in a web-based programme, variable selection was performed using a logistic regression model without transforming the predictor variables. For variable selection, a backward elimination method was employed, with a significance level of 0.157 (24). As a result, age, sex, BMI, CFS score, and Cr level were retained. These variables were then incorporated into a logistic regression model. The internal validity of the model was evaluated using the 200-fold bootstrap method (25). Furthermore, an optimism-corrected (i.e. heuristic shrinkage) model for the regression coefficients of each variable was adopted as the original model.
Point-score-based system
A clinical prediction model based on a point-score system was developed for a paper-based, convenient operation in a clinical setting (16, 26). First, low values were used to examine the relationship between continuous variables (age, BMI, CFS score, and Cr level) and the presence of sarcopenia. Second, based on the shapes of their plots and our clinical perspectives, the variables were classified into three or four groups (i.e. age: < 70, 70 to < 80, ≥ 80 years; BMI: < 20, 20 to < 23, ≥ 23 kg/m2; CFS score: 1–2, 3–4, 5–6, 7–9; Cr: < 7, 7 to < 10, 10 to < 12, ≥ 12 mg/dL). The midpoint of each category was then determined and regression units were calculated by multiplying the difference from the midpoint of the reference category by the optimistic-corrected regression coefficient (16, 27). Finally, the regression units were divided by the smallest coefficient and rounded to assign integer scores to each category.
Assessing the performance of the two models
The performances of both the original and point-score-based models were evaluated via discrimination and calibration. Discrimination was assessed by calculating the area under the receiver operating characteristic curve (AUROC). Calibration curves were generated to compare the observed and predicted probabilities of sarcopenia. These analyses were performed through internal validation using the derivation cohort and external validation using the validation cohort.
All statistical analyses were performed using Stata, version 18.0 (StataCorp LP, College Station, TX, USA). The sample size was not predetermined, and all available data were used to ensure maximum statistical power. Patients with missing outcome data were excluded from the analysis. Statistical significance was set at p < 0.05.
Results
Patient Characteristics
Of the 396 patients enrolled in the derivation cohort from the two hospitals, 373 were included in this study. The patient characteristics are presented in Table 1. The median patient age was 71 years (interquartile range [IQR] 61–79). Of the 373 patients, 357 (95.7%) underwent weekly haemodialysis. Among the 129 patients enrolled in the external validation cohort from the remaining two hospitals, the median age was 73 years (IQR 65–81). Of the 129 patients, 127 (98.8%) underwent weekly haemodialysis. Notably, more patients in the validation cohort had sarcopenia and higher CFS scores than those in the derivation cohort (Table 1).
Model Developments
Table 2 shows the regression coefficients for the selected variables (age, sex, BMI, CFS score, and Cr level) and those after optimism correction. Furthermore, using these corrected coefficients, continuous variables, such as BMI, CFS score, and Cr level, were used to create a model for the point score system. Table 3 presents the scores assigned to each category. The total risk score is the sum of the scores for each selected risk factor, ranging from -14–25 points. Higher scores indicate a higher likelihood of sarcopenia. Table 4 lists the formulas for the original model and the point-score-based model. A user-friendly web-based programme derived from the original model was released (https://noriaki-kurita.jp/en/resources/abc2screener/).
Predictive performance in the derivation cohort (internal validation)
For discrimination, the AUROCs of the original and point score-based models were 0.97 (95% CI: 0.96– 0.98) and 0.95 (95% CI: 0.93–0.97), respectively (Figures 2a and 2c). For the calibration, the predicted probability of sarcopenia in the original model was almost identical to the observed probability. In contrast, in the point-score-based model, the risks were underestimated within the moderately predicted risk range (Figures 2b and 2d). Table 5 shows the proportion of sarcopenia along with predicted risks by point scores and indicates that risks are underestimated in a subset of patients with scores ranging from 1– 14.
Predictive performance in the validation cohort (external validation)
The original model had an AUROC of 0.97 (95% CI: 0.95–1.00), which was comparable to that of the internal validation cohort (Figure 3a). In contrast, the point score-based model showed a lower AUROC value than that of the derivation cohort (0.91 [95% CI: 0.87–0.96]), as depicted in Figure 3c. The calibration plot of the original model showed excellent agreement across the range of predicted probabilities (Figure 3b). The calibration plot of the score-based model underestimated the risks across the moderately predicted probability range for sarcopenia, whereas it revealed better agreement between the predicted and observed probabilities for both the low- and high-risk ranges (Figure 3d).
Discussion
We hypothesised that our newly developed diagnostic support tool, the ABC2-Screener, could predict sarcopenia among patients undergoing haemodialysis. Our results demonstrated good discrimination and calibration with external validation of the ABC2-Screener. The tool consists of five variables (A: age, B: BMI, C2: CFS and Cr, and S: sex) and has the advantage of easy implementation in everyday clinical practice, as either web- or paper-based, without incurring additional costs.
Despite the requirement of high sensitivity for screening sarcopenia (28), low sensitivities of the SARC-F questionnaire and its modified versions recommended for the general population have been noted in a meta-analysis (28.9%–55.3% and 45.9%–57.2%, respectively) (29). Thus, a simple and practical clinical prediction rule for sarcopenia, as in our study, is required because of the low sensitivity of these screening methods among patients on dialysis (30). Several studies on clinical prediction rules for sarcopenia among patients on dialysis have been reported; however, they were limited in terms of validity and transferability. Xie et al.’s nomogram for patients undergoing MHD, consisting of four predictors (age, weight, sex, and grip strength), demonstrated good calibration in external validation (14). However, it was impractical because the predictor variables include grip strength, which is a diagnostic criterion for sarcopenia. Notably, clinical prediction rules should essentially be constructed with simple predictors that do not involve the measurements required to diagnose the target disease. Du et al.’s nomogram for patients undergoing haemodialysis, consisting of four predictors (age, BMI, calf circumference (CC), and serum Cr), also demonstrated good calibration in external validation (13). However, the inclusion of CC as a predictive variable requires a tape measure that is not always available at dialysis facilities. In addition, the application of CC is limited by leg oedema due to fluid overload and lower-extremity amputation. Cai et al.’s nomogram for patients undergoing haemodialysis consisting of five predictors (age, CRP, phosphorus, BMI, and mid-upper arm circumference [MUAC]) is limited in its application because it has not been externally validated and requires measurements of CRP and MUAC using a tape measure (12). The MUAC measurements, which can be useful in predicting sarcopenia in the general population (31), are typically measured in the non-dominant arm. However, MUAC measurements in patients undergoing haemodialysis are generally performed on the side without vascular access which does not necessarily correspond to the dominant side; thus, MUAC measurements may differ between the dominant and non-dominant sides (13,32). Consequently, this may lead to misclassification. Additionally, in a study evaluating a nomogram for sarcopenia in patients undergoing haemodialysis, the side of the MUAC measurement was not clarified (14). Other clinical prediction rules for patients on dialysis, including irisin and fat-free mass index, also have limitations due to the unavailability of measurements in usual care and the lack of external validation (17).
Considering the above, our research focused on developing a simple and practical clinical tool to predict sarcopenia using a combination of data routinely measured in haemodialysis facilities and a free frailty assessment instrument. Consequently, we developed the ABC2-Screener in two forms: a paper-based point score system, which provides a quick risk estimate based on total points, and a web-based programme, which automatically provides a risk estimate based on numerical values entered without revealing any complex mathematical formulas to its users (16). Blood test predictors (Cr) are measured at least monthly in Japan (23). The CFS, a 9-point rating tool developed and revised by the Canadian Society of Health and Aging, can be reliably implemented because it has been shown to have a sufficient agreement between gerontology professionals and amateurs (33). The predictive validity of the CFS has been demonstrated by a report in which the CFS rating based on medical personnel impressions predicted mortality among patients on haemodialysis (34). Thus, the data from these items can be collected through routine dialysis practices worldwide and do not incur new economic costs.
The ABC2-Screener showed useful discriminative and calibration performance in external validation, particularly for web-based programmes. The discriminative ability of the ABC2-Screener in external validation was excellent for both the point-scoring system and web-based programme (0.9149 and 0.9708, respectively) and equal to or superior to those of the aforementioned clinical prediction models (especially the web-based programme) (12–14, 17). The external validation calibration of our tool showed a good fit across low to high predicted risk for the web-based programme while the intermediate predicted risk was underestimated for the point score system. The reason for the lower discrimination and calibration in our point-score system compared with that of our web-based programme appears to be a loss of accuracy owing to the categorization and rounding of continuous variable predictors to integer values to prioritise paper-based, intuitive implementation.
The clinical importance of our ABC2-Screener is its implementation in all haemodialysis facilities using existing equipment. The web-based programme can automatically calculate the predicted risk of sarcopenia instantly by entering numerical values for the three ABCs (age, BMI, and Cr), selecting a CFS value from a pull-down menu, and selecting sex. The paper-based point-score system can classify the predicted risk of sarcopenia from a summed risk score (-14 to 25 points) by selecting categories for the five applicable variables. With either system, a high predicted risk indicates a high probability of sarcopenia, whereas a low predicted risk eliminates the need for confirmatory testing for sarcopenia, such as DXA or four-point electrode BIA.
Nonetheless, this study had several limitations. First, the sample size was limited, especially for the validation set, which may have resulted in an inaccurate calibration of the point-score system. While our developed logistic regression model meets the requirement of 5–9 events per variable to accept bias (i.e. 44 events per 5 variables = 8.8) (35), a study suggests that at least 100 target events are required to obtain an accurate calibration curve (36). Second, the assay for blood laboratory values was not standardised. However, given that non-standardised measurements are subject to misclassification compared to uniformly assayed measurements, the magnitude of the association between serum Cr level and sarcopenia, for example, would have been conservatively estimated. Third, it is unclear whether the same performance would be observed in other ethnic groups, given that our study was conducted in a single race. However, many previous clinical prediction rule studies on sarcopenia among patients undergoing dialysis were also conducted in a single race. Therefore, future studies should validate the ABC2-Screener with more multicentre, multiracial, and large-scale samples.
In conclusion, we developed a new diagnostic support tool, the ABC2-Screener, for sarcopenia among patients undergoing haemodialysis and demonstrated good discrimination and calibration with external validation. The tool can be implemented in any haemodialysis facility using only routine clinical data, especially in those without the equipment necessary for confirmatory testing of sarcopenia.
Data availability statement
The datasets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request.
Funding
This study was supported by JSPS KAKENHI (grant numbers: JP19KT0021).
Authors‘ contributions
Research idea and study design: MM, TS, and NK; data acquisition: MM, AK, YM, HK, and TS; data analysis/interpretation: TA, TK, and NK; statistical analysis: TA, TK, and NK; supervision or mentorship: TK and NK. Each author contributed important intellectual content during manuscript drafting or revision, agreed to be personally accountable for the individual’s own contributions, and ensured that questions pertaining to the accuracy or integrity of any portion of the work, even one in which the author was not directly involved, were appropriately investigated and resolved, including documentation in the literature, if appropriate.
Conflict of interest statement
The authors have no conflicts of interest to declare.
Acknowledgements
The authors greatly thank the following researchers, research assistants, and medical staff members for their assistance in collecting the clinical information used in this study: Tetsuo Ueki, MD, Akio Munakata, MD, Yoshihiko Watanabe, MD (Munakata Clinic, Mobara-City, Chiba); Teruaki Koshiba, PT, Ms. Yayoi Takanashi, Reiji Masaki, NP, Takumi Toishi, MD, Tomohiko Inoue, MD, Shinnosuke Sugihara, MD, Kanako Nagaoka, MD, Junko Fukuda, MD, Mamiko Ohara, MD, PhD (Kameda Medical Center, Kamogawa-City, Chiba); Kenji Yamaguchi, MD (Awa Regional Medical Center, Tateyama-City, Chiba); Ms. Miyuki Sato (Fukushima Medical University Hospital, Fukushima-City, Fukushima).
Footnotes
↵* MM and TA share the first authorship.