Background:
Administrative databases seldom include detailed clinical variables or final vital status, limiting the scope of population-based studies. We demonstrate a comprehensive process for integrating 3 databases (statewide all-payor inpatient hospitalizations [hospitalizations], Paul Coverdell National Acute Stroke Program Registry [registry] and Registry of Vital Statistics and Records [vitals]) into a single statewide stroke database.
Methods:
The 3 MA databases spanned 2008-2017 among 49 hospitals covering over 80% of the state’s stroke volume. Our integration process was composed of 3 phases: 1) hospitalizations-registry linkage, 2) hospitalizations-vitals linkage, and 3) final integration of all 3 databases (Figure). Following the assessments of linkage feasibility based on data uniqueness levels, rule-based deterministic linkage on indirect identifiers were applied in the first two phases. We validated the linkages by comparing additional patient variables not used in the linkage process.
Results:
Using 47,113 stroke admissions in the hospitalizations database, and 43,487 admissions in the registry from 01/01/2008 to 09/30/2015, we were able to link 38,493 (80.7%) of encounters, 95% of which were validated. There were 391,176 deaths reported between 01/01/2010 and 03/06/2017 in the vitals database; 10,660 encounters (27.7%) in the hospitalizations were linked to deaths, which reflects the cumulative mortality over the 7 year period among all registry-linked ischemic stroke hospitalization records.
Conclusion:
We demonstrate that a high-quality integration of statewide hospitalizations, clinical registry and vital data is achievable based on a data linkage strategy that leverages indirect identifiers. Our data integration framework which takes advantage of rich clinical data in registries with long term outcomes in claims or vital records may allow for larger scale outcome studies at reasonable cost.
