Rosenthal, James Edward2014-11-032014-11-032014-08https://hdl.handle.net/11299/167432University of Minnesota Ph.D. dissertation. August 2014. Major: Health Informatics. Advisor: Stephan T. Parente. 1 computer file (PDF); xiv, 113 pages.The objective of this study is to evaluate Present on Admission (POA) indicators as a new data source for which to model hospital readmissions. POA indicators for have been in administrative claims data since 2008. POA indicators' primary purpose is to identify Hospital Acquired Conditions (HACs), which represent 0.14% of overall claims. The remaining non-HAC POA data then falls into a category called “other.” This study attempts to gage the secondary usefulness of POA indicators in aiding hospital readmission modeling. Methods This study used Medicare inpatient 5% Limited Data Sets (LDSs) for the years 2008 through 2011. Patient histories were assembled, index and readmission events were established, and datasets representing the primary diagnosis conditions of Acute Myocardial Infarction (AMI), Heart Failure (HF), and Pneumonia (PN) were extracted. CMS methodologies were followed consistent with the limitations of the source data. A base logistic regression model was created to approximate the CMS hospital readmission models. Three readmission periods were examined: 7 days, 15 days, and 30 days. To this base, three POA variables were developed to address the following research questions: P1) Does the presence of any POA=no indicator (condition occurred after admission to hospital) found on an administrative claim correlate to readmission? P2) Does the number of POA=no indicators found on administrative claim correlate to readmission? P3) Does the hospital-specific POA usage rate per year across all available claims correlate to readmission? These three POA variables were added to the three primary diagnosis datasets, and modeled across the three readmission periods, yielding a total of 27 individual statistical models. Results For variable P1, all three readmission periods for AMI were statistically significant at the 95% confidence level indicating an increased likelihood of readmission with odds ratios for 7-day: 1.276 (1.051, 1.547); 15-day: 1.269 (1.076, 1.494); 30-day: 1.316 (1.139, 1.520). HF 15-day odds ratio just exceeded statistical significance at 1.061 (1.009, 1.115). For variable P2, results were at the cusp of statistical significance, but probably not clinical significance at all readmission periods. For variable P3, HF and PN were significant, but showed a reduced likelihood of hospital readmission. The data for 2008 showed the widest errors, 2011 the narrowest, indicating an evolution toward more consistent POA use by providers. The odds ratio for 2011 30-day readmission in the HF dataset returned 0.604 (0.476, 0.765), and PN returned 0.730 (0.539, 0.987).Conclusions POA indicators are not a homogeneous form of data. POA indicators offer an added insight of patient complexity not previously available. POA has personalities based on the primary diagnostic condition. For AMI, there is a link between any POA=no condition during a patient stay and hospital readmission, but this is not true for HF nor PN. When aggregating POA data at the hospital level, HF and PN show a reduced likelihood of hospital readmission, but this does not hold for AMI. This effect could capture the provider's documentation maturity, which is linked to better discharge practices, which in turn reduces readmission.enHealth informaticsThesis or Dissertation