PS2-39: Key Challenges and Decisions in the Development of HMO Death Data

BackgroundThe mortality status of an HMO?s members, whether a person is alive or dead, is critically important to health care research. The uses of mortality status include determination of causes of death, death rates, and for selection of study cohorts. Unlike other clinical activity and health statuses, mortality status cannot be determined solely with a health plan?s internal data, even for well-defined populations, because health plan members often die outside of the plan?s care delivery and claims payment systems. For such members, determination of mortality status and date of death requires linkage of health plan data to external sources of mortality data, such as state death certificate or federal Social Security data. This person-level statistical linkage is a complex, multi-step process that involves many decisions, assumptions, and choosing of priorities. Even when death data from internal systems is available, internal data sources can disagree, likewise requiring complex decisions to determine a member?s ?true? mortality status.MethodsThis presentation highlights key challenges and decisions points in the development of death data logic at Kaiser Permanente Northern California (KPNC).ResultsKey decisions in KPNC?s death data logic include: handling internal membership data with conflicting personal identifiers, e.g., multiple Social Security Numbers associated with a single Medical Record Number; ranking, grading, and selecting from multiple matches between internal and external data; finding members with multiple MRNs; using utilization contact dates and membership activity dates to evaluate death records, judging reliability of internal health plan data, setting thresholds and criteria for confidence scoring, selecting between multiple possible dates of death, chart validation of final death data.ConclusionsThe development of death data logic at KPNC is a complex process that requires intricate decisions, careful analysis, and a higher degree of discretionary judgment than is typical of building other data areas. The quality of final results depends upon validation and iterative improvements to the process.