Supporting data for “Case Finding of Pre-Diabetes and Evaluation of the Association of Dietary Patterns with Glycaemic Levels in Chinese People with Pre-Diabetes in Primary Care"

This PhD study first aimed to validate two locally developed non-laboratory-based multivariable risk models, developed using logistic regression (LR) and machine learning (ML) methods, for case finding of pre-DM and DM in a Chinese primary care population. It then evaluated the associations of dietary intake and patterns on glycaemic levels among a cohort of Chinese overweight/obese adults with pre-DM who were participants of an RCT on lifestyle interventions on glycaemic levels. The validation study was a cross-sectional study on 919 Chinese adults aged 18-84 without a prior diagnosis of DM recruited from primary care clinics. Each participant completed a questionnaire to provide data on the risk models’ predictors and attended a blood test on HbA1c and OGTT between April 2021 and January 2022. The LR model was converted to an additive risk-scoring algorithm for easy clinical application. The sensitivities of the models were 0.69 (ML), 0.72 (LR) and 0.77 (LR-risk-scoring algorithm) in this (external) primary care population. All prediction models and the scoring algorithm had area under the receiver operating characteristic curves (AUROCs) >0.7, suggesting satisfactory external discriminatory ability. The discriminatory powers were highest among participants with a lower pre-test probability of DM, e.g. those aged 18-44 years.However, the risks of pre-DM and DM estimated by the models were lower than the observed incidence in the primary care study population. Thus, recalibration was explored on the data. Simple recalibration of the LR model’s regression constant significantly improved the model accuracy, while extensive updating recalibration methods did not improve the accuracy any further. All recalibrated models had similar AUROCs to those of the original. The cohort study included 287 overweight/obese Chinese adults aged 40-60 with pre-DM. Each participant completed a 24-hour diet history recall and attended a blood test on HbA1c and OGTT at baseline and 12-month follow-up between October 2021 and September 2023. The baseline data showed that total daily caloric intake was positively associated with HbA1c level. Late eating (>20% total daily calorie intake) was associated with higher HbA1c level that was partially mediated by total caloric intake. The 12-month longitudinal data of 222 participants with complete baseline and follow-up measures showed that a decrease in total daily calorie intake and rectifying late eating pattern were associated with a reduction in HbA1c level, independent of the changes in BMI, quantities of individual nutrient intake and the RCT arm allocation.This study confirmed the validity of two local non-laboratory-based risk models for case finding of pre-DM in primary care. Restriction of total calorie intake should be the principal dietary modification for people with pre-DM. Rectifying late eating patterns is a promising additional strategy to lower glycaemic level. The study findings provide evidence to support screening and simple dietary modifications for people with pre-DM, which may attenuate the rising prevalence of T2DM.

本研究系一项博士研究生项目，首先旨在验证两种本地区开发的不依赖实验室环境的多元风险模型，该模型利用逻辑回归（LR）和机器学习（ML）方法构建，旨在对中国初级保健人群中糖尿病前期（pre-DM）和糖尿病（DM）的病例发现进行验证。随后，研究评估了膳食摄入和模式与一群中国肥胖/超重成年人血糖水平之间的关系，这些成年人为参与有关生活方式干预对血糖水平影响的随机对照试验（RCT）的研究对象。验证研究是一项针对919名年龄在18至84岁之间、无糖尿病前期诊断史的中国成年人进行的横断面研究，这些成年人来自初级保健诊所。每位参与者完成了一份问卷，以提供风险模型预测因子的数据，并在2021年4月至2022年1月期间接受了HbA1c和OGTT的血液检测。LR模型被转换为加性风险评分算法，以便于临床应用。在这些（外部）初级保健人群中，模型敏感性分别为0.69（ML）、0.72（LR）和0.77（LR风险评分算法）。所有预测模型和评分算法的受试者工作特征曲线下面积（AUROCs）均大于0.7，表明其外部区分能力令人满意。在具有较低糖尿病前期检测概率的参与者中，如18至44岁的参与者，其区分能力最高。然而，模型估计的糖尿病前期和糖尿病风险低于初级保健研究人群中的观察发病率。因此，对数据进行重新校准的探索。简单重新校准LR模型的回归常数显著提高了模型精度，而广泛的更新重新校准方法并未进一步提高精度。所有重新校准的模型与原始模型的AUROCs相似。队列研究包括287名40至60岁的肥胖/超重中国成年人，他们患有糖尿病前期。每位参与者在2021年10月至2023年9月期间完成了24小时膳食回顾，并在基线和12个月随访时接受了HbA1c和OGTT的血液检测。基线数据显示，每日总热量摄入与HbA1c水平呈正相关。晚餐摄入（占总日热量摄入的20%以上）与较高的HbA1c水平相关，部分是通过总热量摄入介导的。222名具有完整基线和随访测量的参与者的12个月纵向数据显示，减少每日总热量摄入和纠正晚餐模式与降低HbA1c水平相关，这种相关性独立于BMI的变化、个体营养素的摄入量和随机对照试验（RCT）组的分配。本研究证实了两种本地非实验室风险模型在初级保健中用于糖尿病前期病例发现的有效性。限制总热量摄入应成为糖尿病前期人群的主要饮食调整。纠正晚餐模式是降低血糖水平的潜在附加策略。研究结果为支持对糖尿病前期人群进行筛查和简单饮食调整提供了证据，这可能会减轻2型糖尿病（T2DM）的发病率上升趋势。