Blind Spots: Spatial analytics can identify nonrandom geographic variation in first episode psychosis program enrollments

Aim: To apply spatial analytics to an underway first episode psychosis program to identify areas of significant variation in the geographical distribution of program enrollees from an underlying at-risk population. Methods: Adaptive bandwidth kernel smoothing was used to estimate spatial density functions from program enrollee home addresses and a control population computed from US Census data. A relative risk surface derived from the ratio of these functions was used to discover under-represented areas, or areas from which fewer enrollees where produced than suggested by the underlying population density at the P < .05 level of statistical significance. As a test application of this analysis, a comprehensive list of primary care providers in the program catchment was extracted from the National Plan and Provider Enumeration System and spatially compared to the under-represented areas. Results: This approach identified under-represented areas containing 27.5% of the total program catchment area and 16% of the control population, yet had yielded zero program participants. These under-represented areas contained 179 primary care providers of the 2,337 in the total catchment area. Conclusions: Findings of nonrandom spatial variation in program enrollment is valuable data for those evaluating the impact of and implementing improvements for recruitment to specialty clinics serving geographically-defined catchments. Positive findings from this preliminary study warrant further development of the predictive model as well as measurement of the impact on enrollment from recruitment interventions driven by these findings.