Data collection
As previously described [6, 9], clinical and demographic data were obtained from all adult patients visiting the PES of a university teaching hospital in Montreal (Canada) from June 15, 1985, to June 15, 2004. The database originated June 15 1985 as an 'in-house' register of seven variables (name, sex, service sector, referral source, disposition, date and time of entry into and, departure from, the PES) kept by the nursing staff. It was gradually expanded to include 63 additional clinical (including diagnostic) and demographic variables by July 1, 1996. At this time it was used for research purposes until December 31, 2000. After this date only the nursing register data was collected up until September 2002. The full database was used once again (until June 15, 2004) when this site, as well as three other PESs, participated in assessing the multiple PES patterns of use. Overall, 19,740 patients made 38,420 visits (2 ± 3.5 visits per patient) during the 19-year period. Multiple visit patients were divided into an intermediate group (4 to 10 visits, 1408 patients, 8025 visits, 5.7 visits per patient) and, a heavy user group (11 or more visits, 373 patients, 7789 visits, 21 visits per patient), in accordance with previous reports [6, 7]. The average number of visits of the intermediate group approximated twice that of the standard deviation for all patients during the collection period and was equivalent to the 'high utilizers by standard deviation' group described by Pasic et al. [7]. Those with 1 to 3 visits (17,959 patients, 22,606 visits, 1.3 visits per patient) were included as a reference group. Data entry for the research database was performed by designated members of the nursing staff and by the principal investigator.
The 70 variables in the research database above were listed in a paper format and used as a triage instrument, for a two-year period beginning September 2002, for all patients visiting three other services. All services had strict rules regarding their respective geographical cathment areas that included procedures for integrating patients that did not have a fixed address (see [10] for a detailed discussion on sectors). The completed forms were forwarded to the principal investigator for data entry. The mains site, as well as two of the additional sites, were located in general hospitals and by protocol patients were triaged in the medical emergency department by both the nursing and medical staff and, if warranted, a psychiatric evaluation was requested. The fourth site was within a traditional psychiatric institute and operated as a walk-in clinic.
Diagnostic profiles were produced by attributing a "most probable" primary diagnosis, which was the diagnosis most frequently given during a patient's multiple visits. As over 60% of PES visits have been shown to occur during the daytime hours [6] only services that were covered by experienced, regular daytime psychiatric staff were included in this study. This was done in order to minimize diagnostic uncertainty. To further reduce diagnostic uncertainty diagnoses were grouped into broad categories. The latter differ from our previous report [6] as bipolar disorders and paranoid psychoses are now separate from the 'affective disorders' and 'schizophrenia' categories, respectively. Lastly, multiple-visit patients became increasing familiar to the regular PES staff, who routinely referred them to multidisciplinary treating teams (psychiatrists, social workers, nurse clinicians, psychologists) responsible for both their outpatient and inpatient care. These teams were primarily based upon the notion of maintaining continuity of care. As previously reported, over 80% of heavy users at the main site were, at one point during the time they frequented this PES, under the care of such teams [6]. During the multi center phase of this study 77% of heavy users and 67% of intermediate users were, at least at one point in time, under multidisciplinary outpatient care. As such, any further diagnostic uncertainty could be clarified by discussions with the treating team.
Primary data analysis
Data were analyzed by using the statistical analysis program Systat (Version 11). Categorical variables (such as the overall diagnostic profiles) were analyzed using the "crosstabs" section for one- and two-way tables. Independence of cell frequencies was assessed by the Pearson chi square statistic. Odds ratios (OR), adjusted for sex and age using the logistic regression modules, were used to assess the association between a given diagnosis (independent variable) and either frequency of PES use (dependent variable) or cluster frequency (independent variable). The associated 95% confidence intervals of all ORs were assessed and did intersect 1.0 (only P values are presented). ORs were not adjusted for race as Caucasians represented over 90% of the sample. Analyses of variance were used in order to assess whether the means of numerical variables differed significantly. Simple linear regression (Least Squares method with resulting squared multiple R values) was used to test the hypothesis that a significant relation exists between numerical variables.
This study was approved by the institutional review board (IRB) scientific subcommittees at all sites and exempted from full review other than at one site, where full IRB approval was required.