A systematic review of prognostic multivariable risk prediction models for type 2 diabetes in prediabetic populations

Individuals with prediabetes, defined by impaired glucose tolerance (IGT) and/or impaired fasting glucose (IFG), are at increased risk of developing type 2 diabetes (T2D). This systematic review evaluates the reliability and potential biases of existing multivariable risk prediction models that estimate the future risk of T2D in adults (aged ≥ 18 years) with prediabetes. We included models developed with multiple predictors and, where available, externally validated or updated. All models were intended for use in primary care following a confirmed diagnosis of prediabetes. We systematically searched PubMed, MEDLINE, EMBASE, and IEEE Xplore. Studies were excluded if they reported only population-level incidence rates, focused on diabetes treatment, used animal models, used only single predictors, or reported disease incidence without individual-level risk prediction. Data extraction followed the CHARMS checklist. Risk of bias and applicability were assessed using PROBAST. Findings were synthesized narratively. Certainty of evidence was evaluated using the GRADE framework. The review protocol was registered with PROSPERO (CRD42024212695). Twenty-one studies evaluating twenty-two prognostic risk models were included. Common final predictors included FPG, BMI, and age, in studies conducted in China, Denmark, Korea, the USA, and the UK, among others. Discrimination, assessed using the AUC, ranged from 0.64 to 0.81, indicating varied predictive accuracy. FPG was the most frequently reported final predictor, followed by BMI. Associations of age and gender with progression to T2D were inconsistent. Most studies had unclear risk of bias due to incomplete reporting of model development, lack of information on blinding procedures, and insufficient detail on handling of missing data. Based on GRADE, the overall certainty of evidence was rated as low. Existing multivariable risk prediction models for T2D in individuals with prediabetes show varied predictive performance and important methodological limitations. While common predictors such as FPG and BMI were consistently identified, the overall certainty of evidence was rated as low, primarily due to study limitations and the rarity of the outcome, despite strengths in indirectness and publication-related criteria. Adherence to established reporting standards, such as TRIPOD, is essential to support the development of more reliable, generalizable, and clinically applicable prediction models.