In a recent study published in the journal Dr NPJ Digital MedicineResearchers conducted a scoping review to analyze artificial intelligence (AI)-based models for predicting type 2 diabetes mellitus (T2DM).
Revaluation: A Scoping Review of Artificial Intelligence-Based Methods for Diabetes Risk Prediction. Image credit: Created with assistance from DALL·E 3
The growing importance of AI in healthcare
AI is helping to develop predictive models for diabetes, a condition that is on the rise worldwide. Based on the risk profile, this strategy attempts to assess an individual’s risk of developing T2DM and related complications. AI allows the identification of high-risk patients and the development of individualized preventive approaches and focused therapies.
Key steps in creating an AI model
AI models must be developed step by step, including model creation, evaluation, and translation into clinical decision support. Internal or external validation can be used, with external validation being recommended for a more complete review of the model’s generalizability. AI-driven models have emerged as an effective method for developing prediction models for T2DM, allowing for individualized disease-prevention strategies.
In the current comprehensive review, researchers explored the applications of artificial intelligence-based predictive methods in diabetes risk prediction.
Data were systematically searched in the Scopus, PubMed, Google Scholar and IEEE-Xplore databases for relevant longitudinal studies using artificial intelligence-based models for human subjects and published between 1 January 2000 and 19 September 2022. In addition, references to included studies were screened to identify additional records.
Only peer-reviewed studies, original research and conference proceedings using medical information including electronic health records (EHRs), imaging and multiomics were included. The group excluded reviews, commentaries, editorials, letters, preprints, cross-sectional studies, not using AI, conducted on non-human subjects, included people with type 1 and gestational diabetes and with diabetes-related complications.
Data extracted included title, year of publication, name of first author, type of publication, country, study objective, sample population, study design, participant population, method used to diagnose T2DM, follow-up period and type of data source. In addition, the team recorded the number and type of methods used, AI type, specific algorithms and validation methods. One reviewer performed a data search, and two independently selected studies and data were extracted, and discrepancies were resolved by discussion or consultation with a third reviewer. A descriptive synthesis approach was used for analysis.
What the data reveals
Initially, 1105 records were identified, of which 853 underwent title- and abstract screening, 64 underwent full-text screening, and 40 were considered for the final analysis. Most of the research was published in the previous four years. Sample sizes ranged from 244 to 1,893,901 individuals, with diverse populations including China, Finland, California, and Kuwait.
Most common data types and algorithms
Most studies were of the retrospective cohort type, analyzing data from large personal datasets and publicly accessible databases such as the Canadian Primary Care Sentinel Surveillance Network (CPCSSN), the San Antonio Heart Study (SAHS), and the Tehran Lipid and Glucose Study (TLGS). ). Fasting blood glucose levels of 126 mg/dL or greater and glycated hemoglobin (HbA1c) of 6.5% or greater are the most commonly used criteria to confirm a diagnosis of T2D.
Most studies used unimodal (n=30) artificial intelligence models, with only 10 using multimodal approaches. Unimodal models showed an area under the curve (AUC) value of 0.8, multimodal models were higher (AUC, 0.9). Classical machine learning (ML) models were used in most (n=10) studies, with EHRs (including financial demographic data, family history of diabetes, lifestyle factors, anthropometric measurements, glycemic characteristics, serum lipids, cholesterol and TG levels) being the most common. Data modality used.
In addition, multi-omics (such as single nucleotide polymorphisms (SNPs), metabolomic measurements and microbiota data) were predominant, while medical imaging was the least used. Classical machine learning models use decision trees (DT) with their variations, such as fast, unbiased, efficient statistical trees (QUEST) and classification and regression trees (CART). Linear regression modeling, random forest (RF) classifier, support vector machine (SVM), Naïve Bays (NB) classifier, extreme gradient boosting (XGBoost), and KNN were used in 10 studies, nine studies, eight studies, four studies. study, and four studies, respectively.
Building AI predictive models for T2DM has faced limitations at various stages: associated with underlying data, model building and evaluation, and clinical translation.
Validity and interpretability: A critical aspect
Thirty-nine studies conducted internal validation, while only five conducted external validation. Most studies have used area under the curve (AUC) values for discriminative measures. Note that only five studies provided model calibration. Fifty percent of studies used interpretable methods to identify risk predictors, and most models generally reported known ones. Fasting blood glucose, body mass index (BMI), age and serum triglyceride (TG) were the most frequently documented T2D risk predictors. Metabolic markers include α-tocopherol, mannose, glucose, mestranol, iboflavin, hydroxysphingomyelin C14:1, and phosphatidylcholine acyl-alkyl C40:5. Imaging-based biomarkers for diabetes-related retinal disease include vascular tortuosity, retinal hemorrhage, cotton wool staining, and venous dilatation.
Future directions and challenges
Based on the results, AI models have shown promise in predicting T2DM development, but hurdles must be overcome before realizing their full potential. To assess the potential benefits of AI models, extensive validation and evaluation through clinical trials and prospective studies is required. The effectiveness of AI in medicine is not autonomous but a collaborative effort between AI models and human cognition.
Despite limitations and obstacles, researchers must use AI technology to speed discoveries and their translation into clinical practice for patients and healthcare professionals.