A Swedish national register study followed 34,155 adults with type 1 diabetes and 150,026 matched comparators from 1998 to 2020. Every additional year of age at diagnosis was associated with a 1.7% lower probability of death from any cause (HR 0.98, 95% CI 0.98–0.99), which compounds to about 8.2% per five years. The association was steeper for cardiovascular death (HR 0.97 per year, 0.97–0.98) than for non-cardiovascular death (HR 0.99, 0.98–0.99).

The excess risk relative to the general population was concentrated where diagnosis occurred between birth and age 10: all-cause mortality HR 3.23 (2.92–3.58), cardiovascular mortality HR 5.73 (4.32–7.59). An adult diagnosed in their thirties carries a materially different long-term risk profile from one diagnosed at six, and the register is large enough to say so with precision.

What this does not do is give a mechanism. Duration of exposure to hyperglycaemia is the obvious candidate, but the cohorts span two decades of changing therapy and the study cannot separate age at onset from what treatment was available in that era. The practical use is in risk stratification, not in explanation: a 40-year-old diagnosed in childhood belongs in a different surveillance conversation from a 40-year-old diagnosed last year.

Ask for age at diagnosis, not just duration — for a young adult they are nearly the same number, and for a middle-aged one they are not.
In childhood-onset type 1 diabetes, start cardiovascular risk conversations earlier than the arithmetic of age alone would suggest.
Check lipids and blood pressure at every annual review in this group, whatever the age on the chart.
Record albuminuria annually; the kidney outcomes track the same gradient.
This is Swedish register data — Indian childhood-onset cohorts face different access to continuous glucose monitoring and insulin analogues, and the excess risk is unlikely to be smaller here.

Why it matters

Duration of diabetes and age at onset are not interchangeable, and risk scores that use only the first are losing information.

The statistics, in plain English

A hazard ratio of 0.98 per year looks trivial until it is compounded: twenty years of later onset is a substantially different risk. The confidence intervals are narrow because the cohort is large, but narrow intervals say nothing about confounding — they only say the estimate is precise within this dataset.