Distinguishing incident and prevalent diabetes in an electronic medical records database.

PURPOSE: To develop a method to identify incident diabetes mellitus (DM) using an electronic medical records (EMR) database and test this classification by comparing incident and prevalent DM with common outcomes related to DM duration.
METHODS: Incidence rates (IRs) of DM (defined as a first diagnosis or prescription) were measured in 3-month intervals through 36 months after registration in The Health Improvement Network, a primary care database, from 1994 to 2012. We used Joinpoint regression to identify the point where a statistically significant change in the trend of IRs occurred. Further analyses used this point to distinguish those likely to have incident (n = 50 315) versus prevalent (n = 28 337) DM. Incident and prevalent cohorts were compared using Cox regression for all-cause mortality, cardiovascular disease (CVD), diabetic retinopathy, diabetic nephropathy, and diabetic neuropathy. Analyses were adjusted for age, sex, smoking, obesity, hyperlipidemia, hypertension, and calendar year.
RESULTS: Trends in DM IRs plateaued 9 months after registration (p = 0.04). All cause-mortality was increased (hazard ratio (HR) 1.62, 95% CI 1.53-1.70) among patients diagnosed with DM prior to 9 months following registration (prevalent DM) compared to those diagnosed after 9 months (incident DM). Similarly, the risk of DM-related complications was higher in prevalent versus incident DM patients [CVD, HR 2.24 (2.08-2.40); diabetic retinopathy, HR 1.31 (1.24-1.38); diabetic nephropathy, HR 2.30 (1.95-2.72); diabetic neuropathy, HR 1.28 (1.16-1.41)].
CONCLUSION: Joinpoint regression can be used to identify patients with newly diagnosed diabetes within EMR data. Failure to exclude patients with prevalent DM can lead to exaggerated associations of DM-related outcomes.