Design: Prospective cohort with group-based trajectory modelling and multinomial logistic regression
Population: 713 youth aged 6-18 with type 1 diabetes starting automated insulin delivery
Primary outcome: Trajectory group membership for time in range and HbA1c over 18 months
Effect: 4 time-in-range and 3 HbA1c trajectories; 12.7% reached time in range of 70% or more and 43.3% an HbA1c below 7%; lowest time-in-range group predicted by fewer boluses (odds ratio 2.46, 95% CI 1.78-3.39), higher insulin dose/kg (1.54, 1.25-1.91) and fewer appointments (2.69, 1.21-6.00)

Seven hundred and thirteen youth aged 6 to 18 were followed for 18 months after starting automated insulin delivery (AID), with time in range and HbA1c measured at baseline and at 3, 6, 12 and 18 months, and trajectory modelling used to find distinct patterns rather than an average.

Four time-in-range trajectories and three HbA1c trajectories emerged, and they shared a shape: fast improvement to three months, slower gains to six, then waning or merely sustained afterwards. Only 12.7% of the cohort reached time in range of 70% or more, although 43.3% reached an HbA1c below 7%. What separated the worst trajectory from the best was not the device. After adjustment, membership of the lowest time-in-range group went with fewer user-initiated boluses (odds ratio 2.46, 95% CI 1.78-3.39), higher insulin dose per kilogram (1.54, 1.25-1.91) and fewer clinic appointments (2.69, 1.21-6.00).

That changes what the three-month review is for. The improvement seen at three months is near-universal and tells you very little; the bolus count and the appointment record tell you where the child will be at eighteen months. A child who rarely boluses is not being carried by the algorithm — they are on the track that plateaus early, and the time to intervene is while the numbers still look good.

Record baseline bolus frequency and insulin dose per kg at AID initiation
At the three-month review, read the bolus count rather than the improvement
Treat a low user-initiated bolus rate as an indication for structured re-education, not reassurance
Flag missed appointments as a glycaemic risk factor in their own right
Set expectations with the family that gains slow after six months, so a plateau is not read as device failure

Why it matters

The assumption that the algorithm compensates for a child who rarely boluses does not survive eighteen months of data.

The statistics, in plain English

These odds ratios compare membership of the worst time-in-range group with the best, so an odds ratio of 2.46 for fewer boluses means those children were about two and a half times as likely to be in the poor group rather than the good one — it is not a risk of any single outcome. The appointment finding has a wide interval (1.21-6.00), which means the direction is reliable but the size is not. Trajectory groups are derived from the data, so they describe this cohort's patterns rather than categories that exist independently of it.