Validating the Dasymetric Areal Interpolation Method to Inform Health Policy

Background: Hospitalization rates, particularly for preventable conditions such as diabetes-related limb amputations, are important measures of community health and can directly inform health policy. However, hospital data is not always aggregated in spatial units that are meaningful to policy makers. In Chicago, hospital discharge data are aggregated at the five-digit ZIP Code level, while most policy data are tabulated for census tracts aggregates which uniquely identify 77 neighborhoods in Chicago. Reconciling information tabulated at various geographic levels is a known challenge, and represents potential sources of error in data interpretation.

Methods: To overcome inconsistent spatial units between hospital discharge data and health policy needs in Chicago, we developed and evaluated a dasymetric areal interpolation method, which disaggregates and allocates incidence rates from ZIP Code to Chicago community area. Community level rates of diabetes-related hospitalizations were estimated by allocating the number of hospital discharges associated with diabetes (diagnostic ICD-9 codes 250.00-250.93) of a given ZIP Code to overlapping communities based on proportions within the population. Ancillary data census block data (i.e. age, gender, race) were used to generate population estimates. To evaluate the method’s accuracy, we obtained raw hospital discharges from a single academic medical center in Chicago and compared the actual discharge rates to the estimated discharge rates from the dasymetric areal interpolation method for 77 communities in the city. Statistical significance was measured using a one-tailed Chi-square test with a p-value of 0.05.

Results: A total of 6,534 diabetes-related hospitalizations occurred during the study period (1/1/09 – 12/31/11). The average patient age was 61 years, and 60.5% were female. An estimated 6,544 hospitalizations were calculated using the dasymetric method, for a difference of 10 persons. Variations between the actual and estimated hospital discharge rates by neighborhoods were not statistically significant, X²(76, N=6,534) = 54, p=0.97.

Conclusions: This study shows that the dasymetric areal interpolation method can be a valid means of converting spatial units (e.g. zip codes to neighborhoods) that can make existing health data meaningful to community-level policy makers.