Einstein/Montefiore Department of Medicine

Using Patient Data for Health Services Research

(Data)base to Bedside? Using Patient Data for Health Services Research

by Julia Hess


Image: Will Southern, MD, Chief of the Section of Hospital Medicine.

Hospital databases, long known for their value in illuminating clinical studies, can be a goldmine for research that tackles some of the nation’s most complex, challenging healthcare issues. However, sophisticated adjustment is vital to finding the “truth” in the data, according to Dr. William (Will) Southern, Associate Professor of Clinical Medicine, Associate Director of the Weiler Division Medical Service, and Chief of the recently established Einstein-Montefiore Section of Hospital Medicine.

Outcomes research has been conducted since the early 1900’s, when Dr. Ernest Codman, founder of the American College of Surgeons and an advocate for hospital reform, challenged hospitals to “analyze their results to find their strong and weak points, compare their results with those of others, and welcome publicity, not for their successes, but for their errors.” (Codman’s controversial approach caused him to lose staff privileges at Massachusetts General Hospital, though he eventually established his own End Result Hospital, where he rigorously pursued performance measurement and improvement objectives.) Health services research reached its “golden era” in the 1980’s, when investigators increasingly focused on how patient outcomes were affected by episodes of care, providers, and type and number of procedures performed.

Today, health services research informs policy making, leads to improvements in clinical practice, and helps shape future healthcare funding and delivery. It can also change physician practice behavior—nearly overnight, in some cases. One Canadian study merged outpatient prescription data with 1,400,000 hospital inpatient records to study the relationship between antibiotic prescriptions and admissions for hypoglycemia or hyperglycemia. The outcomes suggested that patients had 4.3 odds of developing hypoglycemia and 16.7 odds of developing hyperglycemia after taking Gatifloxacin. To demonstrate how quickly this finding changed physician behavior, a Montefiore study found that its inpatient service physicians wrote over 900 Gatifloxacin prescriptions in January 2006; that September, following the results of the study, one prescription was written.

The effect can be dramatic, and the data illuminating. However, the quality of the data needs to be carefully examined and researchers need to think carefully about the limitations of observation research when using the data. When making comparisons between dissimilar patient groups, factors such as age, case mix, Charlson Comorbidity Index score, accuracy of ICD-9 (disease) coding, and lab results need to be accounted for. Analyses need to adjust for the differences between patient populations. “Researchers need to think carefully about what questions can be answered using these data” said Southern. “They need to know what to measure and adjust for to bring their analyses closer to the truth.”

Traditional administrative datasets have expanded as hospital data has become more sophisticated. Datasets range from relatively small (e.g., patient data from one particular hospital or system) to quite large (regional and national). Smaller datasets, like Montefiore’s, tend to include richer data. Montefiore’s dataset is accessed primarily using Clinical Looking Glass (CLG), a quality improvement software tool. Accessible data include all inpatient and outpatient visits, ICD-9 code diagnoses, lab tests, prescriptions, imaging, and other orders for patients throughout the network. Used for decision support, outcomes measurement, performance improvement, and investigation, CLG allows investigators to query the system and receive relevant data relatively instantaneously. The data is accessible to all of Montefiore’s investigators, and currently has over 400 active users.

Department of Medicine faculty are using the data to investigate a number of areas, including an intervention to improve diabetes care for patients with HgbA1c over 8% (Erica Spatz), pre-operative factors associated with post-surgical C. difficile (Lawrence J. Brandt), and the association between income and prevalence of disease (Robert Ostfeld). A number of published studies have also used the data: mortality rates before and after computerized physician ordering was initiated in the pediatric and neonatal ICUs (Adam Keene), the benefit of IVC filters (Henny Billett and Laurie Jacobs), and the effect of hospitalists on inpatient care (Southern). Southern used the dataset to examine comparable patient groups assigned to hospitalist vs. non-hospitalist teaching teams for differences in length of stay, readmission, and mortality between the two patient groups. He found that hospitalists reduce length of stay on the inpatient service with no ill effect on readmission or mortality.

Occasionally, natural experiments can be exploited to compare similar groups.  In one example, patients are assigned randomly to teaching teams upon admission to the hospital.  Because of this random assignment, patients assigned to hospitalists and non-hospitalists are equivalent, allowing for a meaningful comparison of patient outcomes. Creative data analysis can explore new questions and even open new fields of study, and Montefiore has a wealth of easily accessible, high quality data for investigators to perform rigorous analyses and generate publishable results.

In 1917, Codman said of his advocacy, “I am called eccentric…[but] such opinions will not be eccentric a few years hence.” Nearly a century later, as hospitals implement more sophisticated information technology, and health systems learn to communicate with each other, larger and richer datasets will likely become available for research and increasingly influence the advancement of scientific knowledge. “It is an exciting time, with great potential,” said Southern, whose studies have been published in the Archives of Internal Medicine. “We have this data at our fingertips, and know that our findings can make an immediate difference for our patients.”

Calendar

Thursday, December 04, 2014

Traditional Data Sources for Public Health Surveillance and Rapid Detection of Emerging Infectious Disease
John Samuel Brownstein, PhD
8:00 AM : Forchheimer Medical Science Building 3rd Floor Lecture Hall

Grand Rounds

Einstein (East) & Moses (West) Campus

Archive

Contact Us

Department of Medicine
Albert Einstein College of Medicine
Jack and Pearl Resnick Campus
Belfer Building - Room 1008
1300 Morris Park Avenue
Bronx, NY 10461 (directions)

718.430.2591
Fax: 718.430.8563

 
Click here to log in