Predicting High-Risk CKD in Japanese People: Model Development and Validation Authors: Takaaki Kosugi, Masahiro Eriguchi, Hisako Yoshida, Hikari Tasaki, Masaru Matsui, Kunitoshi Iseki, Shouichi Fujimoto, Tsuneo Konta, Toshiki Moriyama, Kunihiro Yamagata, Ichiei Narita, Masato Kasahara, Yugo Shibagaki, Masahide Kondo, Koichi Asahi, Tsuyoshi Watanabe, Kazuhiko Tsuruya Published: 2026 ABSTRACT IntroductionValidated risk equations are recommended to identify patients at high risk of chronic kidney disease (CKD) progression; however, few models are applicable for early identification in primary care. This study developed and validated a predictive model to estimate the risk of progression to high-risk CKD according to the Kidney Disease: Improving Global Outcomes (KDIGO) risk classification.MethodsThis study used a nationwide Japanese health checkup cohort (2008–2014). The primary outcome was meeting the KDIGO high-risk CKD criteria at 5 years. The variables were selected using the least absolute shrinkage and selection operator method. The model was developed using Cox regression analysis in the Western Japan cohort and externally validated in the Eastern Japan cohort. A nomogram was constructed on the basis of this model to facilitate its clinical application. Net benefit was assessed using decision curve analysis.ResultsThe development and validation cohorts included 295,083 (13,453 events) and 119,225 (4593 events) participants, respectively. Age, sex, body mass index (BMI), systolic blood pressure (SBP), hemoglobin A1c (HbA1c), triglycerides (TG), uric acid (UA), estimated glomerular filtration rate, proteinuria, use of medication for hypertension and diabetes, smoking, and a history of stroke were selected as predictors. Discrimination and calibration were good in both the development cohort (C-index, 0.816; calibration slope, 0.998) and the validation cohort (C-index = 0.818; calibration slope = 1.042). Decision curve analysis confirmed the net benefit of the model.ConclusionsA prediction model was developed and validated to facilitate the identification of individuals at high risk of CKD progression in primary care. CKD is a major global public health concern, affecting approximately 10% of the worldwide population.1 Lower estimated glomerular filtration rate (eGFR) levels and albuminuria are associated with several serious outcomes, including mortality, cardiovascular disease, and kidney failure with replacement therapy.2 Importantly, the risk of adverse outcomes increases from relatively early stages of CKD2; therefore, early management of CKD is important for improving clinical outcomes. However, low patient awareness of CKD has been reported,3,4 and evidence from primary care-based data bases suggests that CKD may be underdiagnosed and under-referred in routine clinical practice.5,6 These challenges may be partly explained by the disease course of CKD, characterized by its asymptomatic nature and long-term progression in the early stages. As the number of patients with CKD continues to increase,1 early identification of CKD and the identification of patients at high risk for CKD progression at the primary care level are crucial.7 In the KDIGO 2024 Clinical Practice Guideline for the Evaluation and Management of CKD, the use of validated risk equations to estimate the absolute risk of kidney failure is recommended,7 and several risk equations are highlighted.8, 9, 10 However, because they include the urinary albumin-creatinine ratio (UACR) or cystatin C, these equations may be challenging to implement in primary care settings in Japan, primarily because of the limited health insurance coverage. In addition, these equations may not be valid for patients at high risk of CKD or with early-stage CKD.7 Several other prediction models are expected to be applicable to these patients,11,12 but they focus on a ≥ 40% decline in eGFR as the outcome, which may be unfamiliar to general physicians. In Japan, criteria for referral of patients with CKD to nephrologists have been established by the Japanese Society of Nephrology and are presented in the clinical practice guidelines.13 These referral criteria generally correspond to the “high risk” (orange) and “very high risk” (red) categories in the KDIGO CKD severity classification.7 This classification was reported to be clearly associated with adverse outcomes in Japan.14 Thus, the “high risk” and “very high risk” categories of CKD (hereafter collectively referred to as “high-risk CKD”) can be considered a clinically meaningful end points, particularly for primary care physicians who need to identify individuals at risk. Early identification of these individuals allows for timely re-evaluation of risk factor management and consideration of kidney-protective agents, such as sodium–glucose cotransporter 2 inhibitors. In this study, we aimed to develop and validate a model to predict progression to high-risk CKD and construct a nomogram to support clinical decision-making in primary care. Methods This study was conducted in accordance with the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) statement (Supplementary Table S1).15 Study Participants This retrospective observational study was a substudy of the Japan Specific Health Checkups Study, which was based on a nationwide health checkup and guidance system implemented in Japan. The study participants were Japanese citizens aged 40–74 years from 27 Japanese prefectures. They underwent annual health checkups at preassigned clinics or hospitals between 2008 and 2014. The baseline was defined as the date of the first health checkup. The details of this study have been described previously.16 Participant exclusion was conducted in 2 steps—at the prefecture level and at the individual level. First, prefectures with a substantially high proportion (i.e., ≥ 50%) of missing data for relevant baseline variables (eGFR, proteinuria, blood pressure (BP)–related variables, glycemic variables, lipid-related variables, and UA) were excluded. For example, several prefectures in which eGFR data were missing for all participants were excluded. Individual-level exclusion may introduce systematic bias, whereas prefecture-level exclusion is less likely to introduce such bias because a high number of missing values at the prefecture level generally reflects regional policies or health care system characteristics rather than individual-level omissions. Second, participants were excluded according to the following criteria: missing eGFR or proteinuria at baseline; meeting the high-risk CKD criteria at baseline; a history of chronic kidney failure or dialysis; extremely high eGFR (≥200 mL/min/1.73 m2) at baseline; and fewer than 2 follow-up measurements of eGFR and proteinuria. Participants who met the KDIGO criteria for “high-risk” or “very high-risk” CKD at baseline were excluded on the basis of a single measurement because they already require cautious follow-up. Consequently, the study population encompassed both individuals without CKD and those with early-stage CKD. This broader population was targeted because annual health check-ups can delay the timely diagnosis of CKD on the basis of chronicity, and some individuals without established CKD, such as those with poorly controlled BP or hyperglycemia, may still experience a rapid decline in kidney function and/or an increase in proteinuria. The participants were geographically divided into development and validation cohorts, using data from western Japan for development and eastern Japan for validation. Variable Measurements Each participant completed a self-administered questionnaire to document their medical history, current medications, smoking status, alcohol consumption, and lifestyle habits (e.g., physical activity and dietary habits; Supplementary Table S2). Height, body weight, and waist circumference were measured, and BMI was calculated as weight (kg) divided by height squared (m2). According to the health checkup program, BP was measured in the seated position. Blood and urine samples were collected after a 10-hour overnight fast to measure laboratory parameters, including glycemic variables, lipid profiles, UA, kidney and liver function markers, and anemia-related parameters (i.e., hemoglobin, hematocrit, and red blood cell count).16 The results of dipstick proteinuria included the following 5 categories: (−), (±), (1+), (2+), and (3+). The eGFR was calculated using the following formula provided by the Japanese Society of Nephrology: eGFR (ml/min per 1.73 m2) = 194 × serum creatinine (mg/dl)−1.094 × age (years)−0.287 (× 0.739 for female).13 Primary Outcome The primary outcome was progression to high-risk CKD, equivalent to the “high risk” (orange) and “very high risk” (red) categories in the KDIGO CKD severity classification,7 at the time of follow-up health checkups. This outcome corresponds closely to the criteria for referral of patients with CKD to nephrologists established by the Japanese Society of Nephrology.13 In the present study, the criteria for high-risk CKD comprised CKD stages G3b–G5, CKD stage G3a with proteinuria categories A2–A3, and CKD stages G1–G2 with proteinuria category A3. Based on conversion formulas from a previous meta-analysis,17 proteinuria categories A2 and A3 were defined as (1+) and (2+), respectively, on the dipstick test. Statistical Analysis Continuous variables are presented as means with standard deviations or medians with interquartile ranges, as appropriate, and categorical variables are presented as numbers and percentages. Baseline variables were used to develop the prediction model. In addition to selection on the basis of medical knowledge and data completeness (low rates of missingness), candidate variables were screened using the least absolute shrinkage and selection operator (LASSO) method,18 given the large number of variables, particularly lifestyle factors. Hematuria was not included in the model owing to potential false positive results in the dipstick test, differing CKD progression mechanisms in glomerulonephritis, and the relatively high proportion of missing data (28.2%). Similarly, fasting plasma glucose and anemia-related parameters were not included because of the high frequency of missing data, 23.1% and approximately 70%, respectively. Liver function markers were also not included owing to their limited direct link to CKD progression. Specifically, a LASSO–Cox regression model was developed in the development cohort from western Japan, which included the following candidate variables: age; sex (female or male); BMI; waist circumference; SBP; diastolic BP; HbA1c; TG; high-density lipoprotein cholesterol (HDL-C); low-density lipoprotein cholesterol; UA; eGFR; proteinuria, categorized as (−), (±), or (1+); smoking (nonsmoker or smoker); drinking (nondrinker, occasional drinker, or regular drinker); use of medication for hypertension, diabetes, or dyslipidemia; a history of heart disease or stroke; and other lifestyle factors (Supplementary Table S2). The penalty parameter λ was selected using 10-fold cross-validation according to the 1-standard-error rule. The variables selected by the LASSO were subsequently included in the Cox regression model using the development cohort. Continuous variables expected to have a nonlinear association with the outcome (i.e., age, BMI, SBP, log-transformed TG, and eGFR) were modeled using restricted cubic splines with 3 knots. Missing data were imputed using multiple imputation on the basis of predictive mean matching under the assumption of missing at random. The imputation model included all predictors used in the derivation model, end point indicators, and the Nelson–Aalen cumulative hazard estimate.19 Considering the fraction of missing information, 10 complete data sets were generated, and results were combined using Rubin’s rules.20 A nomogram for predicting the 5-year risk of progression to high-risk CKD was constructed using the development cohort. Internal validation of the model was performed using 200 bootstrap resamples to assess optimism in model performance. Model discrimination was assessed using the optimism-corrected Harrell C-index, and calibration at 5 years was evaluated using calibration plots. External geographic validation was performed in the eastern Japan cohort. To evaluate the performance of the developed prediction model in individuals with complete predictor information, we used a complete-case analysis to reflect actual clinical settings in which all predictors are needed, which was reported to be similar to a multiple imputation approach in terms of external validation performance.21 To support the validity of the complete-case analysis, key variables were compared between participants included in the complete-case data set and those with missing data. To evaluate model performance in the validation cohort, individual risk predictions were calculated by applying the regression coefficients derived from the development cohort, and model discrimination and calibration were assessed. Decision curve analysis was performed in the eastern Japan cohort to evaluate the clinical use of the model at 5 years using a range of threshold probabilities.22 Statistical significance was defined as a P < 0.05. All statistical analyses were performed using R version 4.4.2 (R Foundation, Vienna, Austria). Ethics Approval All procedures involving human participants were performed in accordance with the ethical standards of the institutional and/or national research committee at which the study was conducted (Fukushima Medical University; IRB approval numbers #1485 and #2771) and the 1964 Helsinki Declaration and its subsequent amendments. This study was conducted in accordance with the Private Information Protection Law and Ethical Guidelines for Medical and Health Research Involving Human Subjects enacted by the Ministry of Health, Labor, and Welfare of Japan. Results Baseline Characteristics of Study Participants The original data set included 933,490 participants from 27 prefectures. After excluding prefectures with substantially high proportions of missing data for relevant variables, 14 prefectures remained, comprising 655,775 participants. Of these, 414,308 participants from 14 prefectures were included in the analyses, comprising 295,083 participants from western Japan (development cohort) and 119,225 from eastern Japan (validation cohort) (Figure 1 and Supplementary Figure S1). The baseline characteristics of both the cohorts are presented in Table 1. At baseline, in the total cohort, the mean age was 62.4 years, 40.5% of participants were men, the mean eGFR was 76.3 ml/min per 1.73 m2, and proteinuria of (±) and (1+) was present in 7.9% and 2.7% of participants, respectively. The waist circumference tended to be larger in the development cohort, whereas the use of medications for hypertension and dyslipidemia tended to be more prevalent in the validation cohort. Fasting blood glucose levels and urinary occult blood data were unavailable for approximately one-quarter of the participants.Figure 1Flow of study participants.Table 1Baseline characteristics of development and validation cohortsTotalDevelopment cohort (western Japan)Validation cohort (eastern Japan)N414 308295 083119 225Missing (%)Age, y62.4 (7.6)62.3 (7.7)62.8 (7.3)0Male, n (%)167,996 (40.5)119,223 (40.4)48,773 (40.9)0BMI, kg/m223.1 (3.2)23.2 (3.3)22.9 (3.1)0.8Waist circumference, cm82.5 (11.8)83.8 (9.0)79.2 (16.4)3.3SBP, mmHg128.5 (17.4)128.3 (17.5)129.0 (17.2)0Diastolic BP, mmHg76.2 (10.7)76.0 (10.7)76.8 (10.6)0Fasting blood glucose, mg/dl97.0 (18.5)97.0 (18.6)97.2 (18.1)23.1HbA1c, %5.3 (0.6)5.3 (0.6)5.3 (0.6)2.5TG, mg/dl100 [72, 142]100 [72, 142]100 [72, 142]0HDL-C, mg/dl62.5 (16.1)62.2 (16.0)63.1 (16.4)0LDL-C, mg/dl126.5 (30.5)127.0 (30.8)125.1 (29.9)0UA, mg/dl5.1 (1.4)5.2 (1.4)5.1 (1.3)4.0eGFR, mL/min per 1.73 m276.3 (15.1)75.5 (14.8)78.3 (15.8)0Proteinuria, n (%)0 (−)370,185 (89.4)261,372 (88.6)108,813 (91.3) (±)32,935 (7.9)25,330 (8.6)7605 (6.4) (1+)11,188 (2.7)8381 (2.8)2807 (2.4)Urinary occult blood, n (%)28.2 (−)206,804 (69.5)140,602 (68.9)66,202 (70.8) (±)46,386 (15.6)32,782 (16.1)13,604 (14.6) (1+)26,390 (8.9)18,153 (8.9)8237 (8.8) (2+)13,126 (4.4)9017 (4.4)4109 (4.4) (3+)4718 (1.6)3409 (1.7)1309 (1.4)Current smoker, n (%)57,359 (14.1)39,864 (13.7)17,495 (15.3)1.9Drinking, n (%)12.8 None192,323 (53.3)148 587 (54.1)43,736 (50.6) Occasional82,871 (22.9)62,736 (22.8)20,135 (23.3) Regular85,924 (23.8)63,337 (23.1)22,587 (26.1)Medication for hypertension, n (%)109,810 (26.9)75,018 (25.7)34,792 (30.2)1.6Medication for diabetes, n (%)17,766 (4.4)12,492 (4.3)5274 (4.6)1.7Medication for dyslipidemia, n (%)60,997 (15.0)39,427 (13.5)21,570 (18.7)1.6History of stroke, n (%)11,533 (3.1)7509 (2.8)4024 (4.0)10.4History of heart disease, n (%)19,745 (5.3)14,184 (5.2)5561 (5.6)10.3BMI, body mass index; BP, blood pressure; eGFR, estimated glomerular filtration rate; HbA1c, hemoglobin A1c; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; SBP, systolic blood pressure; TG, triglycerides; UA; uric acid.Continuous variables are presented as means (standard deviations) or medians [interquartile ranges]. Flow of study participants. Table 1Baseline characteristics of development and validation cohortsTotalDevelopment cohort (western Japan)Validation cohort (eastern Japan)N414 308295 083119 225Missing (%)Age, y62.4 (7.6)62.3 (7.7)62.8 (7.3)0Male, n (%)167,996 (40.5)119,223 (40.4)48,773 (40.9)0BMI, kg/m223.1 (3.2)23.2 (3.3)22.9 (3.1)0.8Waist circumference, cm82.5 (11.8)83.8 (9.0)79.2 (16.4)3.3SBP, mmHg128.5 (17.4)128.3 (17.5)129.0 (17.2)0Diastolic BP, mmHg76.2 (10.7)76.0 (10.7)76.8 (10.6)0Fasting blood glucose, mg/dl97.0 (18.5)97.0 (18.6)97.2 (18.1)23.1HbA1c, %5.3 (0.6)5.3 (0.6)5.3 (0.6)2.5TG, mg/dl100 [72, 142]100 [72, 142]100 [72, 142]0HDL-C, mg/dl62.5 (16.1)62.2 (16.0)63.1 (16.4)0LDL-C, mg/dl126.5 (30.5)127.0 (30.8)125.1 (29.9)0UA, mg/dl5.1 (1.4)5.2 (1.4)5.1 (1.3)4.0eGFR, mL/min per 1.73 m276.3 (15.1)75.5 (14.8)78.3 (15.8)0Proteinuria, n (%)0 (−)370,185 (89.4)261,372 (88.6)108,813 (91.3) (±)32,935 (7.9)25,330 (8.6)7605 (6.4) (1+)11,188 (2.7)8381 (2.8)2807 (2.4)Urinary occult blood, n (%)28.2 (−)206,804 (69.5)140,602 (68.9)66,202 (70.8) (±)46,386 (15.6)32,782 (16.1)13,604 (14.6) (1+)26,390 (8.9)18,153 (8.9)8237 (8.8) (2+)13,126 (4.4)9017 (4.4)4109 (4.4) (3+)4718 (1.6)3409 (1.7)1309 (1.4)Current smoker, n (%)57,359 (14.1)39,864 (13.7)17,495 (15.3)1.9Drinking, n (%)12.8 None192,323 (53.3)148 587 (54.1)43,736 (50.6) Occasional82,871 (22.9)62,736 (22.8)20,135 (23.3) Regular85,924 (23.8)63,337 (23.1)22,587 (26.1)Medication for hypertension, n (%)109,810 (26.9)75,018 (25.7)34,792 (30.2)1.6Medication for diabetes, n (%)17,766 (4.4)12,492 (4.3)5274 (4.6)1.7Medication for dyslipidemia, n (%)60,997 (15.0)39,427 (13.5)21,570 (18.7)1.6History of stroke, n (%)11,533 (3.1)7509 (2.8)4024 (4.0)10.4History of heart disease, n (%)19,745 (5.3)14,184 (5.2)5561 (5.6)10.3BMI, body mass index; BP, blood pressure; eGFR, estimated glomerular filtration rate; HbA1c, hemoglobin A1c; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; SBP, systolic blood pressure; TG, triglycerides; UA; uric acid.Continuous variables are presented as means (standard deviations) or medians [interquartile ranges]. Baseline characteristics of development and validation cohorts BMI, body mass index; BP, blood pressure; eGFR, estimated glomerular filtration rate; HbA1c, hemoglobin A1c; HDL-C, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; SBP, systolic blood pressure; TG, triglycerides; UA; uric acid. Continuous variables are presented as means (standard deviations) or medians [interquartile ranges]. Meeting the Criteria for High-Risk CKD In the development cohort, 13,453 of 295,083 participants (4.6%) met the criteria for high-risk CKD during a median follow-up period of 44.7 months (IQR, 32.4–52.8). In the validation cohort, 4593 of 119,225 participants (3.9%) met the criteria during a median follow-up of 46.2 months (IQR, 33.8–51.6). Selection of Candidate Predictors Using LASSO Among the candidate variables included in the LASSO–Cox regression model, age, sex, BMI, SBP, HbA1c, log-transformed TG, UA, eGFR, proteinuria, medication for hypertension and diabetes, smoking, and history of stroke were selected as predictors with nonzero coefficients using the 1-standard error rule (Supplementary Figure S2). Model Development, Internal Validation, and Nomogram Using the Development Cohort Using the development cohort with multiple imputations, a multivariable Cox regression model was fitted on the basis of the predictors selected using LASSO. Internal validation using 200 bootstrap resamples showed good discrimination, with an optimism-corrected Harrell C-index of 0.816 (95% confidence interval [CI], 0.812–0.820). The calibration plot on the basis of optimism-corrected predictions showed good agreement between the 5-year predicted and observed risks (Figure 2a). Although a mild underestimation at higher predicted probabilities was observed in this plot, the calibration slope was 0.998 (95% CI, 0.983–1.014), indicating minimal deviation throughout the predicted risk range. The complete mathematical formula and parameters for individual risk calculation of high-risk CKD at 5 years are provided in Supplementary Figure S3.Figure 2Calibration plot of predicted and actual 5-year risk of reaching the criteria for high-risk CKD. (a) optimism-corrected calibration plot of the development cohort, (b) calibration plot of the validation cohort. The dashed line represents the ideal calibration. Calibration plot of predicted and actual 5-year risk of reaching the criteria for high-risk CKD. (a) optimism-corrected calibration plot of the development cohort, (b) calibration plot of the validation cohort. The dashed line represents the ideal calibration. A nomogram was constructed on the basis of the Cox model to estimate the 5-year probability of progression to high-risk CKD (Figure 3). To prioritize clinical usage, some variables with limited contribution to risk prediction (i.e., BMI, UA, and a history of stroke) were omitted from the nomogram while remaining in the adjusted Cox model. The Harrell C-index and calibration slope on the basis of this nomogram-based prediction were 0.813 (95% CI, 0.809–0.817) and 1.026 (95% CI, 1.013–1.040), indicating that the discrimination and calibration did not change substantially.Figure 3Nomogram for estimating 5-year risk of reaching the criteria for high-risk CKD. Tick marks between ages 40 and 60 years were omitted because the contribution of age to risk is minimal. The association between eGFR and risk was U-shaped, with point values reversing at 85 ml/min per 1.73 m2. Scoring points for categorical variables (i.e., male sex, proteinuria, BP medication, diabetes mellitus medication, and smoking) are explicitly labeled. Note that patients with eGFR < 60 ml/min per 1.73 m2 and proteinuria (1+) already meet the criteria for high-risk CKD, regardless of the nomogram score. Nomogram for estimating 5-year risk of reaching the criteria for high-risk CKD. Tick marks between ages 40 and 60 years were omitted because the contribution of age to risk is minimal. The association between eGFR and risk was U-shaped, with point values reversing at 85 ml/min per 1.73 m2. Scoring points for categorical variables (i.e., male sex, proteinuria, BP medication, diabetes mellitus medication, and smoking) are explicitly labeled. Note that patients with eGFR < 60 ml/min per 1.73 m2 and proteinuria (1+) already meet the criteria for high-risk CKD, regardless of the nomogram score. Geographical External Validation Baseline characteristics were generally similar between the participants included in the complete-case external validation cohort and those excluded because of missing data, indicating that the exclusion of cases with missing values was unlikely to have introduced substantial systematic bias (Supplementary Table S3). The model showed good discrimination in the validation cohort, with a Harrell C-index of 0.818 (95% CI, 0.810–0.826). Calibration plots showed good agreement between 5-year predicted and observed risks, with a calibration slope of 1.042 (95% CI, 1.017–1.066) (Figure 2b). DCA DCA showed that the model provided a higher net benefit than treating all or none for threshold probabilities ranging from < 0.05 to approximately 0.3, indicating clinical usage across a broad range of risk levels (Figure 4).Figure 4Decision curve analysis comparing the net benefit of the model with treat-all and treat-none strategies. Decision curve analysis comparing the net benefit of the model with treat-all and treat-none strategies. Discussion In this study, we developed and validated a prediction model for high-risk CKD, corresponding to the criteria for referral of patients with CKD to nephrologists. The model showed good discrimination and calibration, and the DCA showed clinical usefulness. The model performance was consistently high in both the development and external validation cohorts, which indicated that the model is validated for and potentially applicable to the Japanese general population. In primary care, early recognition of patients at high risk of CKD progression is important and may be facilitated by this model. For high-risk patients, timely recognition of CKD may lead to optimization of BP and glycemic and lipid control, as well as reinforcement of smoking cessation counseling. Furthermore, this may provide an opportunity to initiate the use of kidney-protective agents, including sodium–glucose cotransporter 2 inhibitors, in primary care. Recent meta-analyses have shown that sodium–glucose cotransporter 2 inhibitors are effective even in the early stages of CKD,23 suggesting that initiating treatment with these agents in high-risk patients at the primary care level may help improve long-term outcomes. This prediction model does not recommend referral to a nephrologist before patients meet the CKD referral criteria. It is designed to identify patients at a high risk of CKD progression and enable primary care physicians to recognize risks and optimize management in a timely manner. For patients in the early stages of CKD, using progression to high-risk CKD criteria as an outcome may be more common in primary care. The clinically meaningful risk thresholds for implementation of this model warrants further consideration. For kidney failure risk prediction, specific risk thresholds have been proposed—for example, a 5-year kidney failure risk of 3%–5% for nephrology referral, a 2-year risk of > 10% for initiating multidisciplinary care, and a 2-year risk of > 40% for vascular access planning and modality education.7 However, no established thresholds currently exist for predicting progression to an earlier stage of high-risk CKD. In this study, our DCA showed a positive net benefit across threshold probabilities of approximately 5%–30%. Within this range, the model may support individualized clinical decision-making, whereby the intensity of management can be tailored according to the predicted risk and the expected impact of interventions. Identification of patients at higher predicted risk may prompt more intensive implementation of evidence-based interventions, including lifestyle modification, optimization of metabolic risk factor control, intensification of kidney-protective therapies, and consideration of nephrology referral when appropriate. Future studies are needed to determine risk thresholds associated with specific management strategies and clinical outcomes. The variables selected for our model are diverse but readily measurable in routine clinical practice and health-screening settings. As anticipated from existing prediction models,8 eGFR and proteinuria were major contributors to risk; however, the aggregate contribution of other clinical factors also incrementally increased risk. It should be noted that because prediction models differ from causal models, interventions targeting these selected factors may not necessarily result in improved outcomes. In particular, the use of medications for hypertension and diabetes should not be interpreted as causal risk factors but rather as indicators of a clinical state associated with a higher risk. Nevertheless, re-evaluating and optimizing the management of these factors remains essential in routine clinical practice. In our model, both lower and higher eGFR levels were associated with a higher risk, with the lowest risk observed at an eGFR of 80–90 ml/min per 1.73 m2, as illustrated in the nomogram. The increased risk observed at higher eGFR levels may reflect glomerular hyperfiltration or an apparently low serum creatinine level in patients with severe conditions.24 When eGFR is considered to be excessively high, closer attention to the presence of proteinuria may be warranted. In terms of proteinuria, trace (±) and (1+) levels were associated with an incremental increase in risk compared with negative findings. Although trace proteinuria may be associated with a higher likelihood of false-positive results,25 our findings are consistent with those of our previous study.26 According to the equation for the urine dipstick test to UACR (crude model), the predicted UACR is approximately 25 mg/g when the dipstick test shows trace proteinuria.17 Recent evidence indicates that higher levels of UACR within the normoalbuminuric range are associated with CKD progression,27 supporting our findings. The variables selected by LASSO aligned with our prior expectations on the basis of clinical knowledge, particularly regarding variables related to BP, lipid levels, and glycemic-related factors. Among the bp–related variables, SBP and the use of antihypertensive medications were selected, whereas diastolic BP was not. Including interaction terms with treatment or pulse pressure could potentially improve the model fit,28 but this would make the nomogram more complex. In contrast, we used restricted cubic splines to account for the nonlinearity between SBP and outcome; however, the resulting scores remained monotonic in the nomogram. Nonetheless, this does not imply that excessively lowering BP is preferable. Dyslipidemia, particularly high TG and low HDL-C, is an important comorbidity in CKD and is mutually associated with CKD progression.29, 30, 31 Although waist circumference and low HDL-C were not selected in the prediction model, markedly high TG levels contributed to increased risk points, consistent with their role as a component of metabolic syndrome. Regarding glycemic factors, fasting plasma glucose was not included in the model because of a high proportion of missing values. In addition, although an interaction between treatment and HbA1c could potentially improve the predictive performance,32 it again was not incorporated to avoid making the nomogram overly complex. For clinical implementation, we envision that the model could be applied through automated calculation using electronic health record systems or web-based risk calculators, as well as the accompanying nomogram. Because the model estimates individualized absolute risk, its application is expected to rely primarily on computerized calculation or nomogram rather than manual computation. In settings where automated implementation is available, model complexity is less likely to impose an additional burden on clinicians, allowing predictive performance to be prioritized over further simplification. Even parsimonious absolute risk prediction models, such as the 4-variable Kidney Failure Risk Equation,8 are typically implemented using electronic calculators. For clinical settings where such digital system is not yet available, the use of the nomogram remains practically important. Further investigation is needed to facilitate its clinical implementation. This study has several strengths which should be highlighted. We used a nationwide cohort with a large sample size covering a wide range of geographic regions. Although approximately half of the participants were excluded from the source cohort, we addressed missing data to minimize potential selection bias by applying prefecture-level exclusion and multiple imputation. In addition, the use of both internal and external geographic validations ensured the robustness and validity of the results. The variables used in our model are readily available in primary care and health-screening settings, and the model does not require repeated measurements over time, suggesting that it has wide applicability. In recent years, eGFR slope, 30%–40% decline in eGFR and change in UACR have been recognized as important surrogate markers by nephrologists.33, 34, 35 However, these measures may not be fully implemented in primary care settings. Recent data from Australian general practice indicate substantial gaps in follow-up testing for abnormal eGFR and UACR.36 Thus, in primary care settings, prediction tools on the basis of the data available at a single time point are useful. Despite these strengths, this study had certain limitations. First, the generalizability of our model to other countries remains unclear. Although the model included predictors that are easily available in many countries, including qualitative urine dipstick testing, the risk of CKD progression and the impact of each variable may vary by ethnicity and region. Indeed, the Kidney Failure Risk Equation, for example, required distinct regional calibration adjustments when applied outside North America.37 Therefore, whether our model can be directly applied to non-Japanese populations warrants careful evaluation. Nevertheless, we believe that our approach of using widely accessible screening markers to predict high-risk CKD provides a practical framework applicable to other international settings Second, the Japan Specific Health Checkups study data sets used in this study did not include information on mortality; therefore, death was not considered a competing risk in the model. However, given the characteristics of the study population, the effect of not accounting for competing risks may be limited. Finally, UACR data were unavailable in this study; therefore, some misclassifications within the KDIGO risk categories may have occurred during the outcome evaluation. Specifically, defining outcomes on the basis of the dipstick results may have led to an overestimation of high-risk CKD. A previous health checkup study in Japan reported that the prevalence of micro- and macroalbuminuria was 63.9% and 6.6%, respectively, among patients with dipstick proteinuria (1+) and 40.6% and 50.0%, respectively, among those with dipstick proteinuria (2+).38 However, from a screening perspective, avoiding underestimation is preferred to prevent missing patients with truly high-risk CKD. Given that the implementation of a quantitative proteinuria test remains insufficient,39,40 subsequent quantitative proteinuria measurements are required for high-risk patients. In conclusion, we developed and validated a model for predicting the risk of progression to high-risk CKD and confirmed its good discrimination and calibration. We believe that this model may facilitate the identification of individuals at a high risk of CKD progression warranting early management in primary care. Disclosure All the authors declared no competing interests.