Multimodal Risk Factor Graph Mapping, Metabolic Subtyping, and Digital Intelligent Chronic Disease Management for Prediabetes: An Integrated Cohort and Case-Control Clinical Study (DM-PIONEER)

This is a multicenter observational study integrating prospective and retrospective cohort data with a case-control component to investigate the development and progression of prediabetes in Chinese adults. The study is designed to characterize the heterogeneity of prediabetes by combining clinical, metabolic, lifestyle, environmental, genetic, and longitudinal health data. No pharmacological treatment is assigned by the study, and participation does not replace or modify routine clinical care.

The study has three major components. First, data from existing cohorts, including the previous cohort, participating hospitals, and health examination programs will be harmonized to identify clinical, behavioral, environmental, and genetic factors associated with prediabetes and its long-term progression. Multivariable statistical methods and machine-learning approaches will be used to identify important risk factors and explore interactions among genetic background, environmental exposure, lifestyle, and metabolic characteristics.

Second, a prospective observational cohort will undergo standardized baseline and follow-up assessments. Data collection may include questionnaires on health status and lifestyle, anthropometric measurements, oral glucose tolerance testing, biochemical measurements, body composition assessment, liver transient elastography, pulse wave velocity, urine testing, and continuous glucose monitoring. Blood and other biological samples will be collected for selected research analyses, including genetic testing and metabolomic profiling. Participants will be followed longitudinally to characterize changes in glucose metabolism and related metabolic phenotypes.

Third, a case-control component will compare participants with prediabetes with individuals who have normal glucose regulation but other metabolic risk factors. Genetic, metabolic, behavioral, and environmental characteristics will be compared between groups to identify factors associated with abnormal glucose regulation. Genetic information from genome-wide genotyping will also be used to construct polygenic risk scores, which will be integrated with clinical and metabolic data.

Longitudinal data will be used to identify metabolic subtypes of prediabetes and to develop prediction models for future changes in glucose regulation. Unsupervised clustering and other machine-learning methods will be used to characterize distinct metabolic phenotypes. Time-series data will also be used to develop and validate prediction models, including Transformer-based models, for estimating the risk of future progression of glucose abnormalities.

The planned observational cohort will include at least 30,000 participants. This sample size was selected to provide an adequate number of longitudinal outcome events for multivariable risk modeling and high-dimensional machine-learning analyses. A separate case-control component will include approximately 3,000 participants, with equal numbers of prediabetes cases and metabolically at-risk controls.

Standardized operating procedures will be used across participating centers for participant recruitment, informed consent, data collection, laboratory testing, sample handling, follow-up, data management, and quality control. Study personnel and clinical research coordinators will receive standardized training before participating in study procedures.

Each participant will be assigned a unique study identifier. Study identifiers, barcodes, scheduled assessment times, sample information, and completion status will be used to maintain traceability between participants, clinical assessments, biological samples, and study data.

Data will be obtained from multiple sources, including hospital information systems, structured questionnaires, case report forms, laboratory systems, study examinations, wearable devices, and continuous glucose monitoring systems. Where applicable, registry data will be compared with source medical records, laboratory records, case report forms, and device-generated data to evaluate data accuracy and completeness.

Automated and manual data checks will be used to identify values outside predefined ranges, inconsistent information across variables, duplicate records, missing data, and other potential data-quality problems. Suspected errors will be reviewed against available source records before correction. Key study procedures and protocol deviations will be documented and remain traceable.

A standardized data dictionary and coding framework will be used across participating centers. The data dictionary will define study variables, data sources, formats, coding rules, allowable values, and relevant units. Data will be stored in a centralized, de-identified research database with access controls and data security procedures.

Quality monitoring will include routine centralized review and planned site-level monitoring. Monitoring will focus on informed consent, participant identification, completeness of key study assessments, timing of oral glucose tolerance testing, sample tracking, device data, protocol deviations, and consistency between source information and the research database.

Missing or incomplete data will be characterized before analysis. Depending on the extent and pattern of missingness, statistical approaches such as multiple imputation may be used. Sensitivity analyses will be performed to evaluate the robustness of major findings to missing data and different analytical assumptions.

Statistical analyses will include descriptive analyses, multivariable regression, longitudinal risk modeling, subgroup and interaction analyses, and machine-learning approaches. Logistic regression and time-to-event models will be used where appropriate to evaluate associations between baseline or longitudinal factors and changes in glucose regulation. Machine-learning and clustering methods will be used to identify metabolic subtypes and develop prediction models. Model performance will be evaluated using internal and, where available, external validation datasets.