Setting and Participants
Our study is part of a collaborative effort of the Hospital Medicine Reengineering Network (HOMERuN), a national network of investigators at 12 academic medical centers (Online Supplementary Appendix).10
Patients were eligible for inclusion in our study if they were discharged from a general medicine service and had an unscheduled readmission to a general medicine service within 30 days of the index discharge. Eligible patients were readmitted between April 1, 2012, and March 31, 2013; were 18 years of age or older; and spoke English as their primary language or had a caregiver present who spoke English as their primary language. At each site, within the eligible sample, we applied a random-digit generation schema to prospectively select up to five patients per week for study participation. Patients were approached for consent in their hospital room during the readmission. If a patient declined participation, or was too sick to participate, unavailable, or otherwise unwilling after identification, the next randomly selected patient was approached for enrollment.
Following patient enrollment, we emailed or faxed a survey to each patient’s PCP (when identifiable through electronic medical records), the discharging physician from the index admission based on the hospital discharge summary, and the admitting physician from the readmission based on the admission note. Only attending physicians were surveyed.
Our study was approved by the institutional review boards at the University of California-San Francisco (UCSF; the data-coordinating center) and all participating HOMERuN sites.
Survey Development and Data Collection
Surveys were developed by the HOMERuN investigators and pretested by the investigators as well as physicians not associated with the study. Successive refinements were made based on repeated rounds of feedback. The surveys to PCPs and inpatient physicians contained identical questions designed to ascertain the respondent’s opinion on factors potentially contributing to the readmission as well as the probability that different types of interventions might have prevented the readmission. We asked, “In your opinion, which of the following factors may have contributed to the readmission? (CHECK ALL THAT APPLY).” Contributing factors were grouped into five major categories, based in part on a conceptual framework of the ideal transition in care,11 and respondents were asked to endorse all pre-specified individual factors that applied within each of these categories: 1) patient understanding and ability to self-manage (e.g. patient or caregiver inability to manage his/her symptoms); 2) continuity of care and provider communication (e.g. insufficient communication with post-acute care providers regarding the post-discharge plan); 3) social supports (e.g. inadequate support for non-clinical issues); 4) problems with index (initial) admission (e.g. misdiagnosis or failure to treat the condition adequately); and 5) problems with triage after index (initial) discharge (e.g. patient inappropriately went/was sent to emergency department or inappropriately readmitted from emergency department). In addition, they were asked, “How probable do you think each of [the following] potential types of interventions might have been in preventing this readmission?”. Response options under each pre-specified intervention included: no probability, slightly probable, slightly less than 50-50, slightly more than 50-50, strongly probable, and nearly certain. See the Online Supplementary Appendix for full text of the surveys.
Surveys were developed and administered using the research electronic data capture (REDCap) application, an NIH-sponsored, HIPAA compliant, free secure web-based application.12 A unique survey link was emailed or faxed to each physician every 3 days until either a response was received or 14 calendar days elapsed.
Additional patient-level and admission-level data were collected via structured medical record review performed by trained research assistants.
We report the frequency with which PCPs, discharging physicians, and readmitting physicians selected each of the pre-specified factors contributing to readmission, grouped by major category (representing selection of at least one individual factor within that category) as well as by each individual factor. For strategies to prevent readmission, we report the frequency with which the three physician groups reported anything other than “no probability” for each of the potential preventive strategies (i.e. slightly probable, slightly less than 50-50, slightly more than 50-50, strongly probable, nearly certain). We chose this threshold because we were interested in any degree of preventability, and in order to maximize sample size for comparisons. To assess whether using a higher threshold to define endorsement of potential preventive strategies would lead to higher concordance levels, we performed a sensitivity analysis in which we used “slightly more than 50-50,” “strongly probable,” or “nearly certain” to indicate endorsement of a possible way to have prevented the readmission.
To gauge agreement among physicians on the factors contributing to and potential strategies to have prevented each individual readmission, we calculated kappa statistics for each dyad (PCP–discharging physician, PCP–admitting physician, admitting–discharging physician) for each major category of contributing factor and each potential preventive factor. As a measure of overall agreement for provider dyads, we calculated the average kappa for each dyad across all contributing factors, and across all potential preventive strategies. All analyses were performed using SAS software version 9.2 (SAS Institute Inc., Cary, NC).
To maximize sample size and make use of all available information, our main analysis included readmission events with at least one physician response (PCP, discharging, or admitting physician), and for comparisons, had non-missing data for the dyad being compared. For example, a readmission event with a missing PCP survey could still be included in the admitting and discharging physician tabulations and comparisons. Since this approach results in different patient subgroups, thus potentially influencing the results, we performed a secondary analysis in the subgroup of readmissions for which complete survey data were available (PCP, discharging, and admitting physician surveys all completed).