Through simple linkages of longitudinal claims data, the algorithm assigned individuals into one of four categories. We evaluated the algorithm's sensitivity and specificity by comparing the predicted classification against self-report. The validation results were used to refine the algorithm.
When we classified people into two groups of any degree of functional limitation or no limitation, the sensitivity was 91 % and the specificity was 26 % . When classified into two groups of those needing proactive care coordination and all others, sensitivity was 83 % and specificity was 30 % . Thus, overall correct classification ranges from good to fair.
The algorithm utilizes claims databases readily available to many health claims payers. Adding Healthcare Common Procedural Coding System claims and number of prescriptions improves correct classification rates. Even when the claims data are incomplete and imprecise, ARCSv2 (ARCS version 2) can be used as an initial screen to identify people who should be included in the calculation of quality measures and who should be surveyed for consumer reported quality measurement. When using four classification categories, 69 % of the people with the greatest risk and need for care coordination are correctly identified. ARCS can increase the correct identification of people with disabilities by 400 % over random digit dialing of a general population. However, the ARCS should be further refined and validated in a larger population that includes more men with disabilities, children, and people without disabilities before it is used to compute quality measures using administrative data. Correct classification might be improved by incorporating information on comorbidities and specific medication categories.