A time-to-event analysis to conceptualize and predict delay in type 2 diabetes diagnosis in primary care

Abstract

Delayed diagnosis of type 2 diabetes (T2D) increases the risk of diabetes-related health complications. Although signs of T2D are commonly identified in primary care, delays in diagnosis remain a significant challenge. The relationship between patient-level factors (e.g., demographics and healthcare utilization patterns), clinic-specific factors (e.g., primary care location), and the time to T2D diagnosis represents a critical, yet understudied, research area. We conducted a retrospective observational cohort study of 594 adults who received care from two primary care clinics within an integrated healthcare system in the mid-Atlantic region of the United States (2017–2023). Kaplan–Meier survival analysis and time-varying Cox proportional hazards models quantified the time from the first diabetes-range hemoglobin A1c (HgA1c \(\ge\) 6.5% [48 mmol/mol]) to first documented T2D diagnosis. Further, we developed a discrete-time proportional hazards model to estimate the conditional probability of diagnosis at each visit after the first diabetes-range HgA1c to explore time-to-event patterns. Patient-level features, primary care location, continuity of care, and visit regularity were evaluated in multivariate models as potential contributors to the time interval from first diabetes-range HgA1c to diagnosis. A Markov cohort state-transition model characterized clinical trajectories from undiagnosed state to T2D diagnosis over one year to explore the proportion of study population that remained undiagnosed. Median time to T2D diagnosis varied significantly between the two primary care locations (1.6 months vs. 2 months, respectively) which highlighted practice-level variation in diagnosis timelines. In adjusted multivariate analysis, primary care location, continuity of care, and visit regularity were not statistically significantly associated with time to diagnosis. The conditional probability of diagnosis given previous elevated lab observation was concentrated in the earlier follow-up visits after an elevated HgA1c. Markov cohort model revealed that 65.18% of individuals remained undiagnosed one year after the initial abnormal HgA1c. Our study highlights the role of clinic-specific factors in determining the time interval from first elevated lab signal to a documented diagnosis. Consistency in visiting the same primary care location and regularity of interactions accelerate the time to diagnosis. These findings underscore the need for data-driven approaches tailored to primary care clinics serving unique patient populations to facilitate timely diagnosis processes.