This was a prospective study conducted at the Ottawa Hospital, a multisite tertiary-care teaching hospital. The protocol was approved by the Ottawa Hospital Research Ethics Board.
Consecutive patients discharged to independent or residential living from the general internal medicine service at the Ottawa Hospital over a 14-week interval in 2002 were eligible for the study. The general internal medicine service is staffed by an attending physician and several house staff. Patients are admitted from the emergency department (ED), and most are previously unknown to the attending physician. If necessary, discharge to the community is facilitated by home-care coordinators and social workers. At discharge, patients receive a handwritten interim report, which they take to follow-up physician visits. In addition, an official discharge summary is mailed to these physicians. Follow-up care is performed by the general internist if this is thought to be necessary.
We reviewed each patient's medical chart to record demographic data and information on chronic illnesses and the hospitalization. We telephoned the patient approximately 30 days after discharge. If the patient consented, a registered nurse or physician administered a structured telephone interview to determine the posthospital course. We first determined, by means of a complete review of systems, whether the patient had any new or worsening symptoms. To assess the severity of any such symptom, we asked how the symptom affected physical functioning and what the patient had done to help resolve the symptom, including whether the cause had been determined. Timing of the symptom in relation to the hospitalization was documented. Finally, we recorded the date, location and reason for all physician visits, ED visits and hospital readmissions.
If the patient was unable to complete the telephone interview because of cognitive or language difficulty, we interviewed the caregiver or someone else living in the home.
If the patient visited the study hospital's ED or was readmitted to the study hospital, the medical record was reviewed to determine the reason for the encounter and its outcome. For an ED visit or admission to another hospital, we relied on the patient's description of the outcome, since the chart was unavailable. This occurred with 5 patients.
If we were unable to contact a patient after 3 months, we used a number of means to identify outcomes. First, we reviewed administrative databases to determine if ED visits or readmissions had occurred at the study hospital. Then, if we had not determined the patient's outcome, we checked with the registration departments of 2 local community hospitals to determine if the patient had visited their ED or had been admitted. Finally, we used the regional vital statistics registry to determine if the patient had died.
Patients were considered to have had an adverse outcome if, after discharge, they had new or worsening symptoms, a physician or health-facility visit that was unscheduled (i.e., not booked at the time of discharge), an ED visit or readmission to hospital, or if they had died. For such patients, 1 of us (A.J.F.) systematically summarized information from the chart review, telephone interview and records of any postdischarge ED visit or rehospitalization. The outcome summary included a detailed description of all outcomes, including time of onset, severity, health services used and resolution. When a patient had more than 1 adverse outcome, we prepared an outcome summary for each outcome.
All outcome summaries were independently reviewed by 2 certified general internists (H.D.C. and C.W.). Each summary was rated with standard techniques from ease; 4, the outcome was possibly due to treatment [50/50 chance] and was more likely due to treatment than to disease; 5, the outcome was probably due to treatment; and 6, the outcome was definitely due to treatment). If both reviewers judged that the adverse outcome was probably or definitely due to medical management (rating 5 or 6), it was classified as an AE. If both reviewers judged that the adverse outcome was not caused by medical management (rating 4 or less), it was not classified as an AE. If there was disagreement, a third certified internist (A.J.F.) rated the event independently.
The 2 reviewers independently rated the type, severity, preventability and ameliorability of all the AEs. As in previous studies,3,4,5,6,15
AE type was classified as adverse drug event, procedure- related injury, nosocomial infection, fall, therapeutic error, diagnostic error or other. AE severity was categorized as serious laboratory abnormality only, 1 day of symptoms, several days of symptoms, nonpermanent disability, permanent disability or death. Symptoms had to interfere with a patient's activities of daily living to be categorized as nonpermanent disability. AEs were considered preventable if the 2 reviewers agreed, using implicit judgement, on whether the outcome could have been prevented. AEs were considered ameliorable if the 2 reviewers agreed that the outcome severity could have been substantially reduced if alternative actions or procedures had occurred. If the 2 reviewers disagreed in their assessments of preventability or ameliorability, the third reviewer rated the outcome summary independently.
These reviews had moderate to high interrater reliability. For AE judgements the reviewers agreed 86% of the time on initial review (κ = 0.61). The third reviewer rated 22 of the 37 discrepant event summaries as indicating AEs. For preventability ratings there was 73% agreement (κ = 0.44), and for ameliorability ratings there was 90% agreement (κ = 0.77).
For the descriptive results, we described proportions and their 95% confidence intervals (CIs), as calculated by the Wilson score method. In addition, we used the Student's t, Wilcoxon rank-sum and χ2 tests to determine the univariate association between outcomes of interest and continuous, ordinal and categorical variables, respectively. We used multiple logistic regression to measure the independent association of patient characteristics with the likelihood of an AE, including variables that were significantly associated according to the univariate analysis (p < 0.10).