Longitudinal surveys may apply fixed glycaemic thresholds across waves despite changes in laboratory methods. We quantified the sensitivity of estimated four-year diabetes incidence to alternative assumptions about cross-wave comparability of HbA1c and glucose in the China Health and Retirement Longitudinal Study.

Among 10,041 eligible baseline participants without diabetes, 6493 entered the paired analytic cohort. We compared observed definitions with scenarios that mapped 2015 HbA1c, glucose, or both markers to the 2011 scale. The HbA1c-independent definition relied on diagnosis, medication use, and one unconfirmed glucose measurement. Primary estimates used 2011 blood-sample weights, and a response-weighted analysis addressed selection into repeat biomarker measurement. Confidence intervals used 2000 community-cluster bootstrap replicates with all relevant mapping parameters re-estimated per replicate.

The observed HbA1c-inclusive definition yielded 10.11% incidence (95% CI 8.41–12.28), compared with 4.26% (3.58–4.94) for the HbA1c-independent definition. Estimates were 4.45%-4.58% after HbA1c-only mapping, 4.27%-4.52% after overall glucose-only mapping, 5.66%-5.67% after fasting-stratified glucose-only mapping, and 5.79%-5.90% after joint mapping. Response-weighted estimates were similar. Among 7433 participants with both markers at both waves, 61.6% had higher HbA1c but lower glucose in 2015. This discordance was compatible with a cross-wave shift in either or both markers.

Estimated incidence varied materially across assumptions about cross-wave HbA1c and glucose comparability. Observed definitions and mapping scenarios yielded divergent estimates, while the true incidence and relative contributions of HbA1c and glucose remained uncertain. Longitudinal diabetes surveillance should assess each marker affected by a laboratory transition and report clearly labelled sensitivity analyses when direct calibration is unavailable.