The prevalence of type 2 diabetes mellitus (T2DM) is increasing throughout the world. Six out of the top 10 countries with the highest number of adults with diabetes in 2021 were in Asia. Reliable type 2 diabetes risk prediction models can identify individuals at risk of developing T2DM, which may provide a basis for decision-making in the prevention and intervention of T2DM.

To perform a systematic review of risk prediction models for T2DM, providing a reference for the prevention and treatment of T2DM.

In April 2021, we searched for studies on risk prediction models for T2DM in Asian adults in databases of PubMed, EmBase, and the Cochrane Library from inception to April 1, 2021. Two reviewers independently screened the literature, extracted data, and evaluated the risk of bias and applicability of included studies using the Prediction model Risk Of Bias Assessment Tool (PROBAST) . A descriptive analysis was used to summarise the basic characteristics of the models and the risk of bias and applicability of included studies.

A total of 31 studies were included, among which 17 are prospective cohort studies and other 14 are retrospective cohort studies. Logistic regression and Cox regression were widely used to construct the models. The models were externally validated in 5 studies, internally validated in 22 studies, and externally and internally validated in 4 studies. The number of predictors included in the models ranged from 3 to 24, with performance measured by the area under the curve of receiver operating characteristic curve lying between 0.62 and 0.92. There was a high risk of bias in the included studies, which may mainly due to inappropriate treatment of continuous variables and missing data, and ignoring the overfitting of the model.

The included prediction models may have proven to have good predictive performance, which could support medical workers in early identification of the population at high risk of T2DM. Recommendations for future studies developing risk prediction models for T2DM with good performance and low risk of bias are as follows: improving methods for data modeling and statistical analysis, and attaching great importance to external verification and recalibration of the models.

2型糖尿病（T2DM）的患病率不断上升。2021年成年糖尿病患者人数最多的10个国家中，有6个是亚洲国家。可靠的T2DM发病风险预测模型能够识别有患T2DM风险的个体，并可为开展针对性的预防干预工作提供决策依据。

系统分析、评价亚洲T2DM发病风险预测模型，以期为T2DM的防治提供参考。

于2021年4月，计算机检索PubMed、EmBase、the Cochrane Library获取有关亚洲T2DM发病风险预测模型的研究，检索时限均为建库至2021-04-01。由2名研究者独立筛选文献、提取资料后，应用预测模型研究偏倚风险评估工具（PROBAST）评价纳入文献的偏倚风险和适用性。采用描述性分析法对模型的基本特征及纳入研究的偏倚风险与适用性评价结果进行总结、分析。

共纳入31项亚洲T2DM发病风险预测模型研究，其中17项为前瞻性队列研究，14项为回顾性队列研究。纳入研究多采用Cox回归、Logistic回归构建模型；5项研究仅对模型进行了外部验证，22项研究仅对模型进行了内部验证，4项研究采用内部验证与外部验证相结合的方法对模型进行了验证。模型的受试者工作特征曲线下面积为0.62~0.92，包含预测因子数量为3~24个。纳入研究均存在较高的偏倚风险，主要原因为对连续变量的处理不合理、对缺失数据的处理不合理、忽略了模型的过度拟合问题等。

纳入的模型具有良好的预测效能，可帮助医务人员早期识别T2DM发病高风险人群。未来，应对数据建模及统计分析方法进行改进，开发性能优良、偏倚风险低的预测模型，注重对模型进行外部验证和重新校准。