The study design was an open (unblinded) 4-period crossover cluster-randomized controlled trial of 23 family physician practices.
The study was performed in the emergency department of the Sir Mortimer B. Davis–Jewish General Hospital (SMBD–JGH) in Montreal, an adult university teaching hospital with 637 beds. The emergency department's annual census for 2000–2001 was about 60 000 visits.
Family physicians were selected for recruitment on the basis of the frequency with which their patients consulted the hospital's emergency department; a minimum of 100 annual visits per physician clientele was required. Physicians were recruited by telephone, letters of invitation and information sessions. Consenting physicians, rather than patients or visits, were then randomly assigned to the initial intervention or control group to prevent contamination of effect between the intervention and control phases (i.e., so that family physicians would not receive information through electronic and conventional means on different patients at the same time). To ensure balance between the 2 study arms, family physician practices underwent stratified randomization on the basis of the mean age (< 65 v. ≥ 65 years) and annual rates of emergency department visits (< 200 v. ≥ 200) of their clientele. Stratified randomization was achieved by a separate randomization procedure performed within each of the strata.10
A computer-generated random allocation sequence was concealed until all family physician practices were assigned to begin the study in either the intervention or control phase.
Patients whose family physician was participating in the study were approached for recruitment upon presentation to the emergency department between 18 June 2001 and 2 April 2002. For their visit to be eligible, patients had to be 18 years of age or older, have been seen by the participating family physician at least once within the previous 2 years, and be able to provide informed consent as evaluated by the Short Portable Mental Status Questionnaire.11
Recruitment occurred on weekdays between 0800 and 2200 except on statutory holidays. Participants agreed to have their clinical information extracted from their emergency department medical chart and sent to their family physician.
Each of the 4 crossover periods was 10 weeks long. During each period, the results of the patient's visit were communicated to the participating family physician by way of either the intervention strategy or the control (usual practice) strategy depending on the physician's allocation at the time of the visit. At the completion of the study, all family physicians had received results by both strategies.
When family physicians were in the intervention phase, they received detailed clinical information of their patients' visit to the emergency department through a secure Web-based standardized communication system (SCS). The SCS program automatically issued advisory e-mails once per day (at 0700) to all family physicians whose patient or patients had presented to the emergency department within the previous 24 hours. The e-mail also provided a link to a secure Web page where the family physician could view and print a medical report with details of the emergency department visit, including the patient's name, presenting symptoms, emergency department diagnosis, disposition (room assignment if admitted to hospital), specialty consultation reports, laboratory test and electrocardiography results, imaging reports, discharge planning information and suggested follow-up, as well as any new medications or modifications to existing medication regimens. Emergency physicians were aware of which patients were having SCS reports issued because the names of the family physicians receiving SCS reports were posted in the emergency department.
When family physicians were in the control phase, they received a carbon copy of the first page of the emergency physician's notes by regular mail within 1–2 weeks of the visit to the emergency department, the standard practice at the SMBD–JGH emergency department.
Resource utilization was defined in both settings as aspects of health care provision that consumed manpower and materials. The primary outcome or measure of resource utilization included repeat visits to the emergency department within 14 days of the initial visit; secondary outcomes in the emergency department included repeat visits within 28 days, admissions to hospital, mean length of stay in the emergency department, and requests for laboratory tests, imaging and specialty consultations. These data were collected through the hospital's database. Other secondary outcomes, which measured resource utilization in the family physician's office at the follow-up visit, included requests for laboratory tests, imaging and specialty consultations, and duplication of these requests with those made in the emergency department; these were obtained through a questionnaire sent by e-mail to the family physician 21 days after the patient's visit to the emergency department.
The post-visit questionnaire also asked physicians whether data regarding the patient's visit to the emergency department had been received, and received within 48 hours of that visit. Physicians were then asked to evaluate the legibility, comprehensiveness and usefulness of the data received as well as the knowledge gained regarding the patient's visit as a result of that data; these variables were evaluated using a 5-point Likert scale, with 1 representing the extreme negative (e.g., “strongly disagree” that the data were legible), 3 representing neutrality, and 5 representing the extreme positive (e.g., “strongly agree” that the data were legible).
On the basis of a rate of repeat visits within 14 days — deemed to be the most important component of resource utilization and the one most open to influence — of 14% (obtained from the hospital database), a power of 80% and a significance level of 5%, we calculated that 951 patient visits per study arm were necessary to detect a minimally important relative difference in repeat visits of 30% (i.e., 4.2% absolute difference). Adjustment for a cluster effect without the benefit of a precise estimate of intraclass correlation raised the estimated final sample to 2000 patient visits.
The crossover design was chosen as a means of enhancing the statistical power of the study. The study was budgeted to run over a 10-month period, and we chose crossover periods of 10 weeks each to ensure that each period was long enough for physicians to adapt to the communication strategy being used and yet short enough to allow 4 periods in total.
All analyses were based on the intention-to-treat principle (i.e., irrespective of family physician compliance with the SCS intervention). Data were collected on each patient's visit, which created clusters associated with each family physician. Correlation of measurements within each family physician cluster was taken into account in all analyses. Repeated-measures techniques were also used to account for multiple emergency department visits by patients. The Generalized Estimating Equations approach (for binary outcomes, counts)12
and Mixed Model Analysis (for continuous outcomes)13
were used to account for the intracluster correlation and repeated measurements. The working correlation structure was set to be compound symmetry (or exchangeable form). Because of the 4-period crossover study design, 2 lag effects (first and second crossovers from the control to the intervention phase) and 2 washout effects (first and second crossovers from the intervention to the control phase) were also included in the model as potential confounders. The differences in outcomes between intervention and control phases were assessed through mean differences or odds ratios with their corresponding 95% confidence intervals. The odds ratios are presented as the odds of an outcome associated with increased resource utilization (i.e., increased number of repeat visits, tests and consultations) during the intervention phase over the odds of this same outcome during the control phase.
For the post-visit questionnaire sent to family physicians, variables scored using a 5-point Likert scale are reported as dichotomized variables whereby a score of 4 or 5 was counted in the affirmative and 1–3 in the negative.
The research protocol of this study was approved by the SMBD–JGH Research and Ethics Board.