Supplementary Material for: Validation of a Risk Prediction Equation for Incident Chronic Kidney Disease in a Hypertensive Non-Diabetes Cohort in Singapore Primary Care Patients

<b><i>Background:</i></b> Accurate identification of individuals at risk of developing chronic kidney disease (CKD) may improve clinical care. Nelson et al. developed prediction equations to estimate the risk of incident eGFR of less than 60 mL/min/1.73 m² in diabetic and non-diabetes patients using data from 34 multinational cohorts. We aim to validate the non-diabetes equation in our local multi-ethnic cohort and develop further prediction models. <b><i>Methods:</i></b> Demographics, clinical and laboratory data of hypertensive non-diabetes patients with baseline eGFR ≥60 mL/min/1.73 m² on follow-up with primary care clinics between 2010 and 2015 were collected. Follow-up was 5 years from entry to study. We validated Nelson’s equation and developed our own model which we subsequently validated. The developmental cohort included patients between 2010 and 2014 while the validation cohort included patients in 2015. Variables included age, sex, eGFR, history of cardiovascular disease, ever smoker, body mass index, albuminuria, cholesterol, and treatment. Primary outcome was incident eGFR &lt;60/min/1.73 m² within 5 years. Model performance was evaluated by C-statistics and calibration was assessed. <b><i>Results:</i></b> In the developmental cohort of 27,800 patients, 2823 (10.2%) developed the outcome during a mean follow-up of 4.4 years while 638 (12.8%) patients developed the outcome in the validation cohort of 4,994 patients. Applicability of Nelson’s equation was limited by missing albuminuria, absence of black race, and exclusion of non-hypertensive patients in our cohort. Nonetheless, the modified Nelson’s model demonstrated C-statistic of 0.85 (95% CI: 0.84–0.86). The C-statistic of our bespoke model was 0.85 (0.85–0.86) and 0.87 (0.85–0.88) for the developmental cohort and validation cohort, respectively. Calibration was suboptimal as the predicted risk exceeded the observed risk. <b><i>Conclusions:</i></b> The modified Nelson’s equation and our locally derived novel model demonstrated high discrimination. Both models may potentially be used in predicting risk of CKD in hypertensive patients who are managed in primary care, allowing for early interventions in high-risk population.

<b><i>背景：</i></b> 精准识别慢性肾脏病（chronic kidney disease, CKD）高危个体，可改善临床诊疗质量。Nelson等人基于34项跨国队列研究的数据，开发了预测方程，用于评估糖尿病及非糖尿病患者发生估算肾小球滤过率（estimated glomerular filtration rate, eGFR）<60 mL/min/1.73 m²的风险。本研究旨在本地区多民族队列中验证该非糖尿病患者预测方程，并开发更优化的预测模型。
<b><i>方法：</i></b> 本研究收集了2010至2015年间于基层医疗机构随访的基线eGFR≥60 mL/min/1.73 m²的高血压非糖尿病患者的人口学资料、临床及实验室数据。研究随访周期为入组后5年。本研究先验证了Nelson等人开发的预测方程，随后自研预测模型并开展后续验证。建模队列纳入2010至2014年的入组患者，验证队列纳入2015年的入组患者。纳入分析的变量包括年龄、性别、eGFR、心血管疾病病史、吸烟史、体质量指数、白蛋白尿（albuminuria）、胆固醇水平及治疗方案。主要结局为入组5年内发生eGFR<60 mL/min/1.73 m²的事件。采用C统计量（C-statistics）评估模型区分度，并通过校准分析评价模型拟合度。
<b><i>结果：</i></b> 建模队列共纳入27800例患者，平均随访4.4年后，共有2823例（10.2%）发生主要结局事件；验证队列共纳入4994例患者，其中638例（12.8%）发生主要结局事件。由于本队列缺失白蛋白尿数据、未纳入黑人种族人群，且未纳入非高血压患者，Nelson原方程的适用性受到限制。尽管如此，修正后的Nelson模型的C统计量为0.85（95%置信区间：0.84~0.86）。自研模型在建模队列和验证队列中的C统计量分别为0.85（0.85~0.86）和0.87（0.85~0.88）。模型校准度欠佳，预测风险高于实际观察到的发病风险。
<b><i>结论：</i></b> 修正后的Nelson方程与本地区自研的新型预测模型均表现出优异的区分度。两款模型均可用于基层医疗机构管理的高血压患者的CKD风险预测，从而为高危人群提供早期干预的机会。