Diagnostic Billing Codes Vs. MINI PAS-ADD Clinical Interview: Big Data Accuracy for Identifying Psychiatric Comorbidities in Adults with ASD

Thursday, May 11, 2017: 5:30 PM-7:00 PM
Golden Gate Ballroom (Marriott Marquis Hotel)
K. J. Cottle1, M. Newman1, A. V. Bakian2, H. Coon1, J. L. Davis1, A. J. Fischer1 and D. Bilder2, (1)University of Utah, Salt Lake City, UT, (2)Psychiatry, University of Utah, Salt Lake City, UT
Background:  Researchers and clinicians report that 11-80% of individuals with ASD suffer from co-occurring psychiatric disorders. Researchers have begun taking advantage of the ease and availability of electronic medical records to study ASD and its comorbidities. Although this method of data collection is widely used to report prevalence estimates for medical conditions, the diagnostic accuracy of this data source for psychiatric conditions is unclear. Psychiatric conditions are more difficult to assess and could be biased by clinician preference, hospital setting, or documentation practices; therefore; it is possible that diagnostic billing code data are not a reliable source of case status identification. However, medical billing records still provide one of the only feasible opportunities to estimate co-occurring psychiatric disorders in large population-based samples.

Objectives:  Evaluate the accuracy of using diagnostic billing codes to identify psychiatric comorbidities in adults with ASD.

Methods:  This analysis is part of a larger study investigating medical, psychiatric, and social outcomes of 582 adults with ASD. The Mini PAS-ADD Clinical Interview was administered to assess co-occurring psychiatric disorders. The Utah Population Database provided diagnostic billing codes on psychiatric conditions from the two largest health care systems in the catchment area. All analyses were conducted between Mini PAS-ADD case status and diagnostic billing code case status for depression, expansive mood (bipolar disorder), anxiety, obsessive compulsive, psychosis, and an aggregated variable representing any co-occurring disorder. Chi-square tests were used to test for association. Spearman’s rho was used to measure the strength of correlations. Lastly, sensitivity, specificity, positive predictive value, and negative predictive value estimates were calculated.

Results:  The study sample consisted of 214 participants (84.6% male; mean age 35.4 years, SD=9.8) who had a completed Mini PAS-ADD interview. Fifty-four percent of participants had normal IQ. Significant differences were found in the proportion of ASD cases identified with co-occurring depression [χ2(1)=22.5, p <0.001], expansive mood [χ2(1)=15.0, p <0.001], obsessive compulsive [χ2(1)=9.2, p <0.005], psychosis [χ2(1)=23.3, p <.001], and any co-occurring disorder [χ2(1)=4.0, p <0.047] between the Mini PAS-ADD and diagnostic billing code approaches. In contrast, no difference was identified in the proportion of ASD cases with co-occurring anxiety [χ2(1)=1.5, p=0.22] between the two approaches. There were significant correlations for depression (r=0.32, p< 0.001), expansive mood (r=0.27, p<0.001), obsessive compulsive (r=0.21, p<0.002), psychosis (r=0.33, p<0.001), and any co-occurring disorder (r=0.14, p<0.05). Sensitivity estimates were relatively low, ranging from 24.1-58.2, and specificity estimates were relatively high, ranging from 67.9-92.7.

Conclusions: This preliminary analysis indicates that proportions of co-occurring disorders between data from diagnostic billing codes and diagnoses from the Mini PAS-ADD were discrepant. In addition, the sensitivity estimates were relatively low. Given that semi-structured clinical interviews are a preferred measure of co-occurring psychiatric conditions and the Mini-PAS ADD is empirically validated, the results suggest that diagnostic billing code data may not be a sensitive method of reporting co-occurring psychiatric diagnoses in adults with ASD for many co-occurring conditions. These associations will be explored further using an expanded data set.