Researchers found that people who started using automated insulin delivery (AID) systems did not visit their diabetes specialists less. In a study of 1,608 adults with type 1 diabetes treated at four specialist clinics in Italy, 283 patients adopted AID. After careful comparison with similar non‑adopters, adoption was associated with larger routine outpatient engagement, not a reduction in specialist contact.

Automated insulin delivery combines a continuous glucose monitor, an insulin pump, and a control algorithm that adjusts insulin in real time. This “closed‑loop” system records glucose readings, insulin delivery, and the algorithm’s dosing decisions together. Patients still give mealtime insulin and handle alarms or technical problems, but the integrated data that AID produces can give clinicians new information that may require interpretation and follow‑up.

To estimate the effect of adoption, the authors matched adopters to comparable patients who had not yet adopted. They used a method called risk‑set coarsened exact matching to pair patients with similar histories and kept 181 adopters with suitable controls for analysis. Because patients started AID at different times, the team used a staggered event‑study approach and then adjusted for different trends in care before adoption by projecting the untreated group’s pre‑adoption path into the post period. These steps aim to reduce bias from clinical targeting and timing differences.

The adjusted results show sizable increases in specialist visits. The probability of a visit to a diabetologist rose by 16.3 percentage points one semester (six months) after adoption and by 39.3 percentage points after four semesters (two years). For context, the probability of a visit in the semester just before adoption was 55.2 percent. Measures such as HbA1c testing (a lab test that shows average blood sugar over months) and a broader process‑of‑care index moved in the same direction, but the evidence was weaker for those outcomes. The authors note that the visit estimates are sensitive to the assumption used to extrapolate pre‑adoption trends, so the findings should be read conditional on that assumption.