Introduction Electronic health records-based disease surveillance can be used to efficiently estimate incidence of health conditions, but limitations between using claims and membership-based information need to be evaluated. To examine the impact of membership-based and utilisation-based (ie, healthcare encounter-based) definitions on population composition and diabetes incidence estimates in 2021. Methods We included between 1.4 and 1.6 million members 18 to <45 years of age in Kaiser Permanente Southern California without a diabetes diagnosis prior to 2021, applying membership criteria (continuous membership; membership in 1 and 2 years prior) and, separately, utilisation criteria (≥1 healthcare encounter in 2021 and ≥1 healthcare encounter in the 2 years prior; ≥1 healthcare encounter in 2021 only) to determine at-risk populations for diabetes. Incidence was determined by any diabetes-related diagnosis code in 2021. Population characteristics were calculated for and compared across membership and utilisation definitions. Unadjusted and age, sex and race and ethnicity-adjusted incidence rates per 1000 persons were calculated overall and by diabetes type. Results Utilisation-based cohorts at risk for diabetes were older (36.8%–37.5% aged 35 to <45 years) and had a higher proportion female (54.6%–56.8%) compared with membership-based cohorts (35.8%–35.9% aged 35 to <45 years; 51.6%–51.9% female). After adjusting for demographics, incidence rates applying membership criteria were lower compared with applying utilisation criteria across all age, sex, and race and ethnicity groups, but followed similar patterns. The highest adjusted incidence rates were observed for adults 35 to <45 years (range: 7.65–10.61 per 1000), males (range: 3.78–5.91 per 1000), and Hispanic individuals (range: 5.03–7.09 per 1000) within all cohorts. Conclusions Requiring multiple encounters in utilisation-defined denominator populations resulted in estimates of diabetes incidences that approximated those identified using more restrictive membership definitions. These findings suggest that diabetes incidence estimates can be harmonised using utilisation definitions across diverse health systems with different levels of granularity in data capture.
Wei et al. (Thu,) studied this question.