This prospective cohort study used unsupervised machine learning to identify clinical and metabolic subtypes of recent-onset adult type 1 diabetes (within 12 months of diagnosis, aged 18–69 years) in the German Diabetes Study (n=600, mean follow-up 6.1 years), validated in the DPV registry (n=483, mean follow-up 3.1 years).
Three subtypes were identified: classic type 1 diabetes (GADA-positive, lower BMI, higher insulin sensitivity, severe β-cell loss), mixed type 1 diabetes (GADA-positive, higher BMI, lower insulin sensitivity, better β-cell function — positioned ~6.1 years earlier on the β-cell decline trajectory), and a GADA-negative subtype with lower genetic type 1 diabetes risk. Mixed type 1 diabetes carried markedly higher retinopathy risk vs. classic type 1 diabetes (aHR 4.93, 95% CI 1.71–14.24) and differed significantly in hypoglycaemia (log-rank p=0.0019) and cardiovascular autonomic neuropathy (log-rank p=0.028) rates.
Machine learning clustering was based on five variables and may not capture all axes of heterogeneity; the GADA-negative subtype's lower genetic risk raises uncertainty about whether these individuals truly have autoimmune type 1 diabetes; follow-up duration varied widely (SD ~4.7–5.0 years), limiting long-term complication comparisons.
Clinicians should recognise that adults with recent-onset type 1 diabetes who have higher BMI and lower insulin sensitivity (mixed type 1 diabetes) are likely at an earlier stage of β-cell decline and face higher complication risk — this group may represent a priority window for disease-modifying interventions.
Explore related topics