This article is more than 5 years old. Information may no longer be current.
Socioeconomic factors may alter risk prediction in type 2 diabetes models
Diabetes prediction models that do not account for social demographic information may overestimate or underestimate disease risk depending on the socioeconomic status of the patient, according to findings published in the American Journal of Preventive Medicine.
“With the growth of risk prediction modeling to inform preventive interventions, much research has focused upon improving prediction models,” Paul J. Christine, MPH, of the Center for Social Epidemiology and Population Health at the University of Michigan School of Public Health in Ann Arbor, and colleagues wrote. “Relatively little research has explored the added value of using social and environmental information (eg, individual- and area-level [socioeconomic status]) in diabetes prediction models, with the exception of several models outside the U.S. This is surprising, given pronounced disparities in [type 2 diabetes] by [socioeconomic status] and geographic area, and recognition of the importance of social and area-level factors in the development of diabetes. [Socioeconomic status] may capture aspects of diabetes risk, including psychosocial stress and health behaviors, which are not accounted for by traditional risk factors.”
Christine and colleagues analyzed data from 5,021 white, black, Hispanic and Chinese adults without diabetes or cardiovascular disease at baseline participating in the Multi-Ethnic Study of Atherosclerosis from 2002 to 2012. A baseline physical examination took place in 2002. Follow-up exams were completed an average of 1.6, 3.1, 4.8 and 9.5 years later, and incident type 2 diabetes was determined at each follow-up exam. Researchers measured family history of diabetes, systolic blood pressure, hypertension treatment, waist circumference, waist-to-hip ratio and smoking status, and created individual-level, area-level and composite socioeconomic indices for participants. Researchers used Cox proportional hazard models to estimate 10-year incident diabetes risk; one employing only non-laboratory variables (clinical model) and one also using laboratory variables (laboratory model).
During a median of 9.2 years of follow-up, 615 individuals developed diabetes.
Researchers found that several socioeconomic characteristics were predictive of type 2 diabetes, independent of traditional risk factors. However, no socioeconomic variable altered the overall ability of the models to discriminate between those who would and would not develop diabetes. Individual- and area-level socioeconomic variables also did not reclassify substantial numbers of individuals across risk categories, the researchers noted.
In the clinical model, the strongest socioeconomic predictors were individual-level socioeconomic status index (HR per standard deviation [SD] increase = 0.91; 95% CI, 0.82-1) and area-level percentage of adults with a bachelor’s degree (HR per SD increase = 0.91; 95% CI, 0.83-1.01). In the laboratory model, the strongest individual-level predictor was categorical household income (HR comparing highest with lowest category = 0.75; 95% CI, 0.58-0.95), although continuous household income and household income per capita also were predictive. At the area level, researchers wrote, the socioeconomic status index was more predictive (HR per SD increase = 0.91; 95% CI, 0.83-1.01).
Models without socioeconomic status predictors generally underestimated risk for low- socioeconomic status individuals or individuals residing in low-socioeconomic status areas (underestimates ranging from 0.31% to 1.07%) and overestimated risk for high-socioeconomic status individuals or individuals residing in high-socioeconomic status areas (overestimates ranging from 0.7% to 1.3%).
“For instance, those in neighborhoods with the lowest educational attainment had observed risks that were on average 1.06% (95% CI, 0.54-1.57) higher than the predicted risk, whereas those in neighborhoods with the highest educational attainment had observed risks 1.2% lower than predicted (95% CI, 1.61-0.78),” the researchers wrote. “The addition of area-level education improved model calibration across tertiles of area-level education, with mean differences between observed and predicted risks narrowing for each group.”
Most socioeconomic status predictors reclassified 2% to 3% of individuals in each risk category, according to researchers.
“Although differences between observed and predicted risks in the different socioeconomic status groups may not be very meaningful on the individual level, there are more sizable implications at a population or health system level, as even a small systematic underestimation of risk could result in a substantial number of individuals who would not receive the preventive measures they need,” the researchers wrote. “This may become more important with the increasing use of prediction models and the development of thresholds for potential interventions, which are not currently well defined for [type 2 diabetes].” – by Regina Schaffer
Disclosures: The researchers report no relevant financial disclosures.