M. A. Zapf1, A. Kothari1, P. Kuo1, G. Gupta1, P. Wai1, J. Driver1, T. Markossian2 1Loyola University Chicago Stritch School Of Medicine,Surgery,Maywood, IL, USA 2Loyola University Chicago Stritch School Of Medicine,Public Health Sciences,Maywood, IL, USA
Introduction: Linkage of large administrative datasets containing information at the hospital and patient levels offers the opportunity to conduct timely research on the impact of organizational characteristics including health information technology (HIT) on surgical outcomes. We aimed to develop a methodology to create a tiered dataset to study the impact of HIT on surgical outcomes.
Methods: One patient-level dataset was linked with two hospital-level datasets for the years 2007-2009. These were: 1) the Healthcare Cost and Utilization Project – State Inpatient Database (HCUP-SID) from Florida containing patient-level hospitalizations including surgical outcomes, 2) the Dorenfest Institute for Health Information Technology (HIMSS) database that contains data about hospitals’ HIT adoption and use, and 3) the American Hospital Association (AHA) Annual Survey which contains detailed hospital information. The goal was to maximize the number of hospitalizations in HCUP-SID having hospital IT information. Hospital unique identifiers were used to link the HCUP-SID to the AHA database. The AHA database was linked to the HIMSS data via hospital Medicare number. After primary linkage, manual matching with ZIP code and hospital name increased the number of hospitalizations in HCUP by including observations that did not initially match with the HIMSS.
Results:Hospitalizations were generated from 247 (2007) and 246 (2009) hospitals in the HCUP-SID. Exactly 196 (79.3%, 2007) and 206 (83.7%, 2009) hospitals were directly matched. After manual matching, the numbers increased to 211 (85.4%, 2007) and 220 hospitals (89.4%, 2009). In the final dataset, hospital-level IT characteristics from 2,486,167 of 2,563,383 (97.0%) and 2,502,342 of 2,606,165 (96.0%) hospitalizations were identified. Of these, manual merging was responsible for linking 36,880 (1.5%, 2007) and 33,282 observations (1.3%, 2009). Manually merged hospitals had a smaller number of hospitalizations per hospital compared to directly matched (2,459 v 12,890 in 2007 & 2,377 v 11,985 in 2009, both p<0.05). In the final database, the number of common general surgery operations (appendectomy, hemorrhoidectomy, cholecystectomy, inguinal hernia, and thyroidectomy) tallied 64,213 and 65,942, while complex operations (colorectal resection, gastrectomy, esophagectomy, and kidney/liver transplant) were 25,812 and 25,917 in 2007 and 2009. In total, the linked databases contain over 100 patient-level variables and 1,288 possibly clinically associated hospital-level characteristics.
Conclusion:We demonstrated the feasibility of creating a tiered database using the HCUP-SID, HIMSS, and AHA Annual Survey datasets with a high match rate and minimal lost patient encounters. Manual merging was essential for capturing lower volume hospitals. Using the same approach, additional datasets at the hospital or area levels could be appended to our dataset with the goal of expanding our analytical scope.