A relatively small number of patients account for a disproportionate number of psychiatric emergency room visits and thus represent a substantial burden upon this service [1–7]. The psychiatric emergency service (PES) is an important gateway into the mental health system and the early detection of these patients could be of clinical and administrative importance to overall mental health care delivery. Several studies have shown that heavy and intermediate PES users possess distinct diagnostic profiles [5–7]. These profiles become more heavily weighted with chronic and severe psychopathology as the number of visits increases [6]. Although distinct, these profiles are by themselves not sufficiently unique to be used as potential markers for early multiple visit patient detection. Correlating them with distinct patterns of PES use, such as temporal or geographical patterns, may provide useful additional information.

With regards temporal patterns of use the majority of return visits to the PES occur within 3 months of the first, a finding that appears independent of whether patients ultimately make a high or an intermediate number of return visits [6, 8]. Pasic et al. [7], found that using various definitions of what constitutes a high frequency user (partly on the basis of temporal patterns of use) resulted in different patient profiles. Differences however were mainly observed with regards variables such as enrollment in a mental health plan, homelessness and history of incarceration, variables that might not possess the same degree of pertinence from one country to another or, from one metal health care delivery system to another. As for geographical variables Oyewumi et al. [4] reported that up to 23% of their multiple visit patients (in Saskatoon, Canada) visited several, rather than the same PES. Whether this subgroup of multiple visit patients differed diagnostically from those making all of their visits at the same PES was not assessed.

There were several aims to the present study. First, to assess whether the diagnostic profiles of different multiple visit patient groups could be matched to distinct temporal or geographical patterns of PES use. If so, could this additional information be of use in their early detection.

Shorter time intervals between visits in patients with higher visit counts suggest that these patients might also differ with regards visit clustering. A single cluster was defined as 3 visits within a 3-month period (type 1 cluster) or, 3 visits within a 6-month period (type 2 cluster). Only 38% (N = 536) and 55% (N = 769) of the intermediate users versus 82% (N = 306) and 94% (N = 352) of those in the heavy user group had at least one type 1 or one type 2 cluster during their many visits, respectively. Larger cluster sizes of 6 (type 1a) or 7 (type 1b) visits per 3-months were also examined. This required isolating a cohort of 258 patients with 8 or more visits from the intermediate group as an adequate comparison group to the 373 patients with heavy PES use. Odds ratios showed that those in the heavy use group (independent variable) were six and fourteen times more likely to have a type 1a or a type 1b cluster (dependent variable), respectively (P < 0.005).

We examined the possibility that the higher number of patients with a cluster in the heavy use group might primarily reflect the overall greater number of visits made by these patients. In this case, the presence of a cluster, cluster size and cluster frequency should be related to the visit count. Linear regression of the number of patients with a type 1 cluster and the visit count for patients with 5 or more visits (N = 1277) showed a very strong relationship (P < 0.005, squared multiple R value of 0.84). Interestingly, regressing cluster sizes ranging from 0 (none) to 7 visits per 3-months (dependent variable) and visit count for patients with 8 or more visits (N = 631) showed a very weak relationship (squared multiple R value of 0.2). The frequency with which a 4 visits per 3-months cluster repeated itself in patients making 8 or more visits also showed a significant relationship to the visit count (P < 0.005, squared multiple R value of 0.7). As expected, the average number of visits per patient (21 visits/patient) for those with repeating clusters (N = 289 patients) was much higher than for those without (12 visits/patient, N = 342 patients). Repeating clusters of 7 visits per 6-months were also examined in 247 patients with 14 or more visits extracted from the heavy use group. Again, cluster frequency showed a significant (P < 0.005, squared multiple R value of 0.7) relationship with visit count and those (N = 89) with repeating clusters had a greater average number of visits per patient than those without (35 versus 20 visits/patient, respectively).

That some diagnoses are more prevalent in patients who exhibit a clustering pattern of PES use would be more clinically pertinent if the cluster occurred sufficiently early within the total time span over which a patient visited the PES. We therefore assessed the time at which the first cluster occurred (of any size or frequency, starting with a minimum type 1 cluster) in relation to the total time span for completing all visits for a selected group of multiple visit patients. Three groups were chosen. Patients with 4 visits (N = 766, 118 of which had clusters), 10 visits (N = 73, 37 of which had clusters) and those with 40 or more visits (N = 29, all had clusters). The average time span for each patient group to complete all of their visits was 720, 2504 and 5125 days, respectively whereas the average time to complete the first cluster was 357, 1050 and 654 days, respectively.

We next examined whether the patterns of PES use by multiple visit patients included admissions at other services. Data was acquired at services that were 11 km, 39 km and 256 km away from the main site. A subgroup of multiple visit patients at the main site was isolated using the criteria that their last visit fell within the time period of data acquisition for the multi center trial (N = 268 for the intermediate group and N = 132 for the heavy use group). Overall, 16% (N = 42) of the intermediate group patients and 10% (N = 13) of those with heavy PES use frequented at least one other service (between group differences were not significant).

We next examined whether some patients primarily visit multiple, rather than a single PES. Of the 12,083 diagnosed patients frequenting all 4 services during the two-year multi center phase of this study173 had 50% or more of their visits at the other services. Locally (either of the four services) their count was low (2 ± 3 visits per patient) but it increased to 4 ± 3 visits per patient when the non-local visits were included. Their clinical profile differed from that of either of the two multiple visit groups or, that of the 11,910 patients visiting a single one of the four sites. There was a predominance of substance abuse patients (30% versus 16% for the single site patients, OR of 2.3, P < 0.005) and patients with schizophrenia (20% versus 13% for the single site patients, OR of 1.7, P < 0.05).

Using a prolonged observation period as well as an extensive division of psychiatric diagnoses this study adds further detail to the previously described relationship between visit frequency and diagnostic profile in the PES [1, 6, 7]. Patients with adjustment, anxiety and depressive disorders and patients with psychoses of unknown etiology primarily made a low (1 to 3) number of visits. Personality and substance abuse disorders spanned both the low and intermediate range of PES use whereas patients with schizophrenia were primarily intermediate and heavy PES users. This latter finding was observed despite the fact that paranoid psychoses were excluded from the schizophrenia category, slightly reducing its proportion in comparison to our previous report [6]. Although diagnostic certainty in the PES has been questioned and remains a potential weakness of this as well as of all PES studies, schizophrenia, compared to other psychiatric diagnoses, has been shown to possess the greatest stability in different clinical settings [11, 12]. The present study also highlights the relative contribution of bipolar disorders to intermediate and heavy PES use (Table 1), a finding in keeping with their extensive use of inpatient and outpatient services [13, 14].

Temporal patterns of PES use also differed between the multiple visit groups. Patients in the heavy PES user group had significantly shorter time intervals between their consecutive visits. They were also more likely to exhibit a single 'cluster' pattern of service use than patients in the intermediate group. This finding could, at least in part, be attributable to the fact that heavy users made (by definition) a much higher number of visits. The only diagnostic category more predominant in patients with primarily single clusters (compared to those without clusters) was personality disorders. This was so even in the intermediate PES user group. Cluster frequency, although not cluster size, also correlated with visit frequency. However, repeating clusters of even small sizes (4 visits per 3 months) require a high number of visits (8 or more) and by and large these patients would have already been tagged heavy PES users. Nevertheless, this parameter does appear to provide additional clinical information inasmuch as the diagnostic profile of heavy users with a high level of clustering differed from the overall heavy user group profile. As opposed to the general profile, heavy users that cluster frequently were almost equally likely to have a personality disorder than to suffer from schizophrenia. They also made a higher average number of visits per patient than heavy users without repeating clusters. Some of these patients may be detectable at the intermediate PES user level as suggested by the higher proportion of personality disorders in patients with predominantly single clusters. In addition, clusters can be observed at an early stage, long before a patient's total number of visits have been completed.

Geographical patterns of PES use also appeared to provide potentially useful clinical information. Noteworthy was the substantial number of visits made by patients in both the intermediate and heavy PES user groups to multiple services. Oyewumi et al. ([15]) reported that 23% of frequent users made visits to multiple PESs whereas the figure reported here was closer to 14%. However, the three participating services in this study were chosen because they were either structurally or functionally dissimilar to the main site, not because they were the closest. Several other services were located in the main site's vicinity. As such it is more than likely that the percents reported here for both multiple visit groups frequenting other services are underestimations.

In addition, a subgroup of multiple visit patients was characterized by having 50% or more of their visits at other services. This subgroup appeared to possess a diagnostic profile different from that of the other multiple visit groups. This particular diagnostic profile was marked by a preponderance of substance abuse disorders. In the present study patients with substance abuse were not typically found to be heavy users of a single PES. Combining all user groups (including patients with single visits) they nevertheless represented one of the largest patient categories. Their total number of visits equaled that made by patients with schizophrenia, the heaviest of PES users (Table 1). Patients with substance abuse have also been shown to contribute significantly to the frequent user phenomenon in the medical emergency department [16]. As such, the overall burden to both emergency services represented by these patients would appear to be considerable. That they tend to visit multiple services suggests that this broad diagnostic category might be substantially underestimated if assessed purely at the local level.

Patients making a very high number of PES visits are much easier to describe that to treat. Many in the present study were (or had been) under active out/inpatient multi disciplinary care. An underlying premise of this study is that long-term data acquisition can lead to a more precise clinical description of the various PES user groups (and subgroups). This could ultimately lead to the development of treatment strategies tailored towards the common elements within these groups, rather than strategies geared towards each individual patient.

Clinically, the probability of finding a cluster (and ultimately, repeating clusters) at the local level rises along with the visit count. This relationship parallels that between 'visit count and diagnostic profile', at least until extremely high levels of PES use are reached. Making use of both findings, targeting specific diagnostic groups (personality disorders, schizophrenia) when clusters are present in their patterns of use could potentially help detect future heavy PES users at a much earlier stage. Early detection could help in the acquisition of large volumes of data for the study of group characteristics. On the policy side the above findings illustrate the pertinence of planning for long-term data acquisition in the PES, both locally and regionally, as it is highly unlikely that such levels of detection could be achieved using a 'paper' format. The finding that some frequent user subgroups visit several services further underscores this need. Better regional integration of individual PESs and, a more efficient means of communication with certain community services, especially those servicing substance abusers, might be considered. Alternatively, consolidating several services to a smaller number might offer substantial advantages in terms of quality of care and costs containment.