Using routine clinical features to classify adult-onset diabetes at diagnosis: the StartRight prospective observational study
Authors:
- Julieanne KnuppJKJulieanne Knupp
- Anita V HillAHAnita V Hill
- Nicholas Jm ThomasNTNicholas Jm Thomas
- Timothy J McDonaldTMTimothy J McDonald
- Katherine G YoungKYKatherine G Young
- Diane P FraserDFDiane P Fraser
- Andrew T HattersleyAHAndrew T Hattersley
- Trevelyan J McKinleyTMTrevelyan J McKinley
- Beverley M ShieldsBSBeverley M Shields
- Angus G JonesAJAngus G Jones
Preprint
Abstract
Background It is not known which clinical features optimally differentiate type 1 and 2 diabetes at diagnosis. We aimed to determine which clinical features differentiate these conditions at diagnosis and develop classification models combining clinical features with and without islet-autoantibodies. Methods In this prospective cohort study, we recruited 1,800 adults (aged ≥18 years) diagnosed with diabetes within the preceding 12 months, excluding secondary or monogenic diabetes. The primary outcome was diabetes subtype defined by a combination of insulin treatment and endogenous insulin production (measured by C-peptide) at ≥3 years post-diagnosis. Models were developed in participants aged 18 to 50 years and validated internally, alongside validation in an older cohort (aged >50 years) and using UK primary care data (n=188,232). Results Eleven clinical features and routinely measured biomarkers discriminated type 1 from type 2 diabetes independently of diagnosis age and BMI. Lower age of diagnosis, BMI and waist-hip ratio, unintentional weight-loss, and higher presentation HbA1c or glucose were the most discriminative, with other features only weakly discriminative. Models integrating routine features with and without islet-autoantibodies, developed in those age 18 to 50 years at diabetes diagnosis, had high performance in internal validation (clinical features only: Area Under the Receiver Operating Characteristic curve (AUCROC) (95% CI) 0.94 (0.93, 0.96), clinical features and islet-autoantibodies: AUCROC 0.97 (0.96, 0.98)), and maintained high discrimination in older adults (age >50 years AUCROC 0.93 (0.90, 0.96), and 0.97 (0.94, 0.99)). Simplifying the models to a points-based score resulted in similar performance. In primary care data models and score were strongly predictive of outcomes associated with type 1 diabetes, including in those initially treated as type 2. Interpretation Lower age-at-diagnosis, BMI, and wait-hip ratio, unintentional weight loss and high presentation glycaemia are the most discriminative features for diagnosis of type 1 diabetes in adults. Models combining routine clinical features, with or without islet-autoantibodies, have high accuracy and could assist clinical classification and prioritisation of classification biomarker testing. Funding UK National Institute of Health and Care Research (NIHR) and Diabetes UK.Keywords
This publication has 38 references indexed in Scilit:
- Performance of European prediction models for classification of type 1 and type 2 diabetes in IndiansDiabetes & Metabolic Syndrome: Clinical Research & Reviews, 2024
- The INNODIA Type 1 Diabetes Natural History Study: a European cohort of newly diagnosed children, adolescents and adultsDiabetologia, 2024
- The challenges of identifying and studying type 1 diabetes in adultsDiabetologia, 2023
- Stability of clinical prediction models developed using statistical or machine learning methodsBiometrical Journal, 2023
- Identifying type 1 and 2 diabetes in research datasets where classification biomarkers are unavailable: assessing the accuracy of published approachesJournal of Clinical Epidemiology, 2022
- Type 1 diabetes defined by severe insulin deficiency occurs after 30 years of age and is commonly treated as type 2 diabetesDiabetologia, 2019
- Minimum sample size for developing a multivariable prediction model: PART II ‐ binary and time‐to‐event outcomesStatistics in Medicine, 2018
- Prediction models need appropriate internal, internal–external, and external validationJournal of Clinical Epidemiology, 2016
- Regression Modeling Strategies2015
- Index for rating diagnostic testsCancer, 1950