Frequency and Predictors of Prescription-related Issues After Hospital Discharge

Sunil Kripalani, MD, MSc; Megan Price, MS; Victoria Vigil, MPH, CHES, CPHQ; Kenneth R. Epstein, MD, MBA


Journal of Hospital Medicine. 2008;3(1):12-19. 

In This Article


Information for the present analysis consisted of deidentified clinical, administrative, and survey data provided by a national hospitalist management group, IPC—The Hospitalist Company. At the time of the study, IPC employed more than 300 physicians working at 170 community hospitals in 18 regions across the United States. As part of their daily patient management, physicians entered clinical and administrative data into a proprietary Webbased program. At discharge, physicians completed discharge summaries in the same program. These summaries were faxed to the outpatient physicians scheduled to see the patients and were transmitted electronically to a call center. The call center attempted to contact all patients at home to assess their clinical status and satisfaction and to assist with any postdischarge needs. The call center staff made up to 2 attempts to reach each patient by telephone within 3 days of discharge. Patients who were reached were interviewed using a scripted survey. Any identified medical needs were addressed in a separate follow-up call by a nurse.

Patients were included in this analysis if they were at least 18 years old, were treated by an IPC hospitalist, had been discharged between January 1, 2005, and December 31, 2005, and were successfully surveyed by telephone 48-72 hours after hospital discharge. If patients had more than 1 discharge during the study period, only the first survey and its corresponding hospital stay were included. The analytic plan was approved by the Emory University Institutional Review Board.

Hospitalists recorded the age, sex, and insurance coverage of each hospitalized patient and noted the discharge diagnoses and medications on the discharge summary. Primary diagnosis and length of stay were determined from hospitalist billing data, which were entered daily into the Web-based program. Each patient's severity of illness was classified as minor, moderate, major, or extreme using a commercially available program (3M Health Information Systems) that considered patient age, primary diagnosis, diagnosis-related group (DRG), and non-operating room procedures.

A common patient identifier code linked these data with patient-reported information obtained from the call center. Patients indicated whether they picked up their prescribed medications and if they had any trouble understanding how to take their medications. For the present analysis, patients were considered to have a prescription-related issue if they had problems filling or taking medications prescribed at discharge, a composite variable defined as including not picking up discharge medications, not knowing whether discharge medications had been picked up, not taking discharge medications, or not understanding how to take discharge medications.

Initial analyses included construction of frequency tables to estimate the distribution of prescriptionrelated issues across patient demographic characteristics, insurance type, clinical diagnosis, and number of medications, as well as among users of certain high-risk classes of medication. Some continuous variables, such as age and number of medications, were categorized (age into clinically relevant categories, number of medications into tertiles). Separate variables were created for clinical diagnoses by mapping DRGs to 26 major diagnostic categories (MDCs) so that comparisons could be made based on the frequency of prescription-related issues for those with a primary diagnosis pertaining to a particular organ system versus those with a primary diagnosis outside that organ system. The 10 most common MDCs were circulatory, digestive, respiratory, nervous, skin-subcutaneousbreast, kidney-urinary, musculoskeletal-connective, hepatobiliary-pancreas, endocrine-nutrition-metabolic, and infectious. The categories of the severity of illness variable were reduced to 3 by combining "major" and "extreme" because there were so few in the extreme category.

Unadjusted odds ratios were calculated based on a logistic regression model relating any prescription-related issues to each possible covariate 1 at a time (ie, not adjusted for any of the other covariates). Adjusted odds ratios were obtained through stepwise building of a logistic regression model. Initially, all possible covariates were entered into the model, and the model was then reduced using Wald test results to assess the significance of dropped parameters. All analyses were conducted using SAS version 9.1 (Cary, NC) and a significance level of 0.05.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: