Comparing All-Payer Claims Database Derived Diabetes Surveillance Estimates to Behavioral Risk Factor Surveillance System Estimates

BACKGROUND: All-Payer Claims Databases (APCD) are state-level compilations of insurance claims data. Minnesota’s APCD includes ambulatory, hospitalization and pharmacy claims from Medicaid, Medicare and Commercial payers and represents ~84% of the population.  APCDs may offer more reliable estimates of surveillance measures. We compare current estimates of diabetes prevalence to APCD-derived estimates. 

METHODS: APCD analyses were conducted with de-identified 2012 data (Onpoint Health Data). We used output from the Johns Hopkins ACG® System Version 10 to identify patients with diagnosed diabetes.  2012 Behavioral Risk Factor Surveillance System (BRFSS) data was analyzed using SAS 9.3 survey procedures. 2012 APCD data were analyzed using Linux SAS.  Counts and percentages were estimated, with percentages compared using Chi Square or Rao-Scott Chi Square tests.  

RESULTS:  

The 2012 APCD estimated the adult diabetes prevalence as 8.4%, compared to 7.3% (95% CI: 6.7-7.9) from 2012 BRFSS.  BRFSS data showed adults with diabetes were more likely to have health insurance than those without diabetes, but are more likely to not see a doctor within the last 12 months due to cost (13.7% vs. 10.4%, p=0.03). Eleven percent of adults with diabetes did not see a doctor for a diabetes-related visit in the last year.  While 84% of Minnesota adults are represented in the APCD, ~95% of adults 65+ years are included.  Modeled APCD estimates accounting for underutilization of health care (BRFSS) and underrepresentation of 18-64 year-olds show higher prevalence estimates than BRFSS, ranging between 8.1 and 9.0%. 

We also compared comorbidities among adults with diabetes. For depression, the APCD estimated 33.5% of adults with diabetes had depression vs. 20.6% (95% CI:17.4-23.8) from BRFSS. For COPD, estimates were 5.1% vs. 9.4% (95%CI:7.1-11.8) and for kidney disease, 9.3% vs. 7.7% (95% CI:5.5-9.8).

CONCLUSIONS:  APCD is a valuable dataset with promise for diabetes surveillance and strategic use of BRFSS data can enhance its utility.  The higher estimate of diabetes prevalence from the claims-based APCD is consistent with reports showing only moderate sensitivity (66-73%) of the self-reported diabetes data in BRFSS. Trends in healthcare utilization found in BRFSS can be used to adjust APCD estimates and to guide data analysis/interpretation.  Adjusting for healthcare utilization had a small effect on overall adult diabetes prevalence.  Finally, APCD-based and BRFSS estimates varied for some conditions, consistent with likely sources of error or bias in each dataset, demonstrating the need to assess the validity of estimates for each condition.  The APCD can provide a strong platform for diabetes surveillance.