Fuller, Candace2015-12-082015-12-082013-09https://hdl.handle.net/11299/175699University of Minnesota Ph.D. dissertation. September 2013. Major: Epidemiology. Advisors: Kamakshi Lakshminarayan, James Pankow. 1 computer file (PDF); vi, 114 pages.Medicare claims (CMS) are a source of nationwide data on various disease conditions including stroke. Development of acute stroke hospitalization identification algorithms in CMS data could allow these data to be used more widely. Our aims were: 1) Link a validated, population-based, acute stroke hospitalization database to CMS data; 2) Validate the CMS Chronic Conditions Warehouse (CCW) stroke definition; 3) Develop and test algorithms to identify acute stroke hospitalizations in CMS data. Aim 1: We linked 90% of year 2000 Minnesota Stroke Survey (MSS) hospitalizations to CMS enrollment data and 74% to CMS hospitalization claims. No CMS claim was located for 16% of MSS hospitalizations linked to CMS enrollment data; 84% of these patients were enrolled in an HMO plan. Inclusion of the working aged and Medicare ineligible patients in MSS may account for inabilities to link some hospitalizations to CMS claims. Aim 2: When the CCW stroke definition was compared to acute stroke hospitalizations in MSS, sensitivity [SEN] was 97% and specificity [SPE] was 99%. However, we observed many false positives (positive predictive value [PPV] 78%). False positives increased when both CMS hospitalization and physician claims identified stroke cases. Aim 3: We used the Classification and Regression Tree (CART) modeling framework to develop and test algorithms to identify stroke in CMS data. The algorithm with best discriminative performance identifies cases with an acute stroke hospitalization discharge code in their hospital record in CMS hospitalization claims (Test data [TD]: SEN=90%, SPE=96%, area under the receiver operating curve [AUC]=0.94). Discriminative performance was also high for the algorithm identifying CMS hospitalizations for cases meeting the World Health Organization stroke definition (TD: SEN=91%, SPE=95%, AUC=0.93). Our CART algorithms are available for validation with other data sets. Potential utility of developed algorithms has broader implications for stroke epidemiology and health services research.enEpidemiologyStrokeThesis or Dissertation