An investigation of factors associated with psychiatric hospital admission despite the presence of crisis resolution teams

The aim of this study was to explore which of a comprehensive set of explanatory variables were independently associated with hospital admission within 8 weeks of an initial crisis presentation in a large cohort of patients with access to a CRT. Secondary objectives were to explore variables associated with admission within 6 months, with total bed days at 8 weeks and 6 months and with compulsory hospital detention.

Three different CRTs were investigated in the three studies whose results have been pooled for this analysis. Two were located in the London Borough of Islington and one in the London Borough of North Southwark. These are all deprived inner London boroughs with ethnically very mixed populations. The Islington sectors were served by well established community mental health teams (CMHTs) and a liaison team comprising of psychiatric nurses, a junior doctor and access to a consultant psychiatrist. The liaison team assesses any patient who attends the casualty department and is deemed to need mental health input. They operate between the hours of 8.00 am – 10.00 pm in the casualty department of the local general hospital. Two crisis houses, one women-only, were available as alternatives to hospital admission and were operating prior to the introduction of CRTs. The North Southwark sector was also served by a CMHT and out of hours liaison services at two local general hospitals. In addition there was an early intervention service for psychosis for the whole borough operating between 9.00 am and 5.00 pm. All the CRTs were multidisciplinary. The two Islington teams had junior doctors within the teams and senior medical input from local CMHT sector psychiatrists. The North Southwark service had a dedicated CRT consultant psychiatrist and a junior doctor on rotation from the local psychiatric training scheme within the team. All three teams provided 24 hour cover with a gate-keeping role, so that any patient assessed as requiring hospital admission could only be admitted if the CRT had agreed that home treatment was not feasible. The teams monitored symptoms, administered medication where necessary, identified and offered advice and help with social stresses that might be contributing to the crisis, and made appropriate follow-up arrangements with other local services once the crisis was resolving and discharge approaching. Patients could be seen twice daily if needed and were able to contact the CRT at any time. The CRTs arranged hospital admission if their intervention proved to be unsuccessful or the patient was deemed unsuitable for home treatment. The model implemented by all the teams essentially conformed to guidelines laid out by the Policy Implementation Guide [2]. The operation of CRTs has been described in more detail by Johnson [24].

The first Islington study was designed as a natural experiment comparing outcomes of patients presenting in crisis before and after the introduction of the South Islington CRT. Our study only included the group from the second phase of the study, who presented in crisis once the CRT had become available and begun to assess all patients presenting in crisis. No patient who met the criteria for being in crisis, as judged by an expert panel, was excluded [7]. The key features of the definition of crisis were that the person had to show deterioration in mental health or social functioning, and that this deterioration had led to concerns about their safety or the safety of others which were sufficiently great for an immediate change in clinical management to be required. The cohort of 123 patients used included in the current study was recruited over a 9 month period following the introduction of CRTs. Assessments were carried out on those identified as being in crisis immediately after identification of crisis, 6 weeks and 6 months afterwards.

The second Islington study from which patients have been drawn for the current analysis was the North Islington study [8], which used a randomised controlled trial design. Patients were eligible for the trial if they were resident in the study area and presented in a crisis severe enough for clinicians to judge that hospital admission was warranted. Further requirements were that either they had decision making capacity at the time of the crisis and consented to randomisation or that they lacked capacity, but had previously received information about the study and had not chosen to opt out in advance (1160 service users were contacted in advance, of whom exactly 100 opted out) or that they lacked capacity and had not already been informed about the study, but had a carer who was willing to give initial assent to their inclusion. Two hundred and sixty people were randomised during the recruitment period, while 104 were admitted to hospital without entering the study. The sample included in the analyses for the current paper consisted of the 135 participants who were randomised to the experimental group and whose care thus involved assessment by the CRT and intensive home treatment whenever feasible. This methodology was replicated in North Southwark where 100 patients were recruited to the experimental group. All three studies received local research ethics committee approval. In both studies those patients who had access to CRTs had significantly reduced admission rates compared to those who did not have access. For the South Islington study [7] admission rates at 6 weeks were reduced from 70% to 48% with a p-value of 0.002. In the randomised controlled study [8] hospital admission at 8 weeks was again significantly reduced in the experimental group with access to CRT (OR 0.19 95% CI 0.11–0.32). Neither studies showed any impact of CRT access on patients being admitted to hospital compulsorily.

Table 2 shows the baseline variables used for this study. These were obtained from the best available data source: patients supplied socio-demographic and clinical information when feasible, but this was supplemented by staff reports and case notes where interview information could not be obtained. The main presenting symptoms were identified and rated by staff, as was cooperativeness at time of assessment. Ratings of the severity of clinical and social problems were collected using the Health of the Nation Outcomes Scales (HoNOS, [25]). Current risk of self harm, violence, self-neglect with serious lack of caution or vulnerability from exploitation from others was evaluated using staff ratings on the Threshold Assessment Grid (TAG) [26] for the North Islington and North Southwark studies. In the South Islington study a set of structured questions on severity of risk, eliciting ratings on a five point scale encompassing no risk, mild risk, moderate risk and severe risk was used. For the purposes of this analysis, a rating of moderate risk on the scale used in South Islington was treated as equivalent to a rating of moderate risk on the TAG.

The primary outcome measure on which analysis was based was whether the patient was admitted within 8 weeks of the initial crisis assessment. Secondary outcome measures were whether the patient was admitted within 6 months, whether the patient was admitted compulsorily under the 1983 Mental Health Act and bed days in hospital at both 8 weeks and 6 months.

The choice of variables was based on previous literature, as reviewed in the introduction, with 3 further variables (risk of unintentional harm to self, for example through self neglect or incautious behaviour, risk from others and which crisis team was providing the service) chosen on clinical grounds as potentially important but not investigated in previous studies. All the variables used are shown in Table 2. The TAG variables relating to severity were not normally distributed in their initial form and were converted into binary variables indicating whether or not a risk of at least moderate severity was present.

Univariate tests (chi squared tests and t tests) were first used to compare those admitted and those not admitted by 8 weeks on each explanatory variable. Logistic regression was then used to test which variables retained an independent effect after adjustment for other explanatory variables. All analysis used STATA statistical software, version 8 (StatCorp, 2003). Less than 10% of the data were missing, but exclusion of all cases with missing data would nonetheless have resulted in substantial loss of data. To avoid this, we used multiple imputation, which fills in the missing values based on values of other variables and a missing at random assumption [27]. Unlike other methods of imputation, multiple imputation acknowledges uncertainty about the missing values by creating several imputed datasets. Each imputed dataset is analysed separately and the results are combined in a way that correctly allows for uncertainty about the missing values [28]. We generated five imputed datasets using the ice command [29], including all variables in Table 2 in the imputation model. We analysed the imputed data using the micombine command [29]. This logistic regression procedure was repeated for admission by 6 months and compulsory detention, and linear regression was used to analyse bed use.

This study incorporated data collected from three different studies using methods that were not identical, although the core set of measures was largely the same. Recruitment methods differed between the South Islington study and the other two, probably resulting in fewer exclusions of severely ill patients in South Islington than elsewhere. Thus the comparison between teams should be treated with considerable caution, even though adjustment has been made for some of the baseline differences that may have resulted from different methods of recruitment. We tested a comprehensive set of variables informed by literature or based on clinical grounds. However, some important variables may have been omitted. Recognised standardised measures were used for ratings of the severity of symptoms and social problems and of risks, but these scales are very broad and global, and more differentiated and sensitive measures of domains such as psychotic symptom severity might have yielded significant associations. The imputation of some missing values (no more than 6% of values for any variable) should be noted. It should also be noted that all findings are associations only: causality cannot necessarily be assumed.

Finally it should be mentioned that this study was carried out in inner city London which represents an area of high deprivation with a transient and cosmopolitan population. The combination of a greater severity of illness and rootlessness means that this population is not typical of the UK as a whole [30]. The high local density of large general hospitals with very busy casualty departments and the high local threshold for admission are each factors which may influence local CRT practices.

Our findings identify factors that may limit CRTs' capacity to treat people at home. They also suggest that some factors expected in theory to limit considerably scope for home treatment may not in fact do so.

In future research regarding CRTs, greater investigation of the mechanisms underlying these differences in admission rate would be of interest. Analysis of the process of decision making in the early management of crises may enhance understanding of the variations described. Some of the risk factors for admission are potentially amenable to change: for example, avoiding initial assessments in the casualty department might prevent some admissions. A greater understanding is also desirable of patients' views and the factors that make some of them unwilling to cooperate with assessment. In our sample, only a small number of patients were admitted after a week or more of community management, so that there was relatively little scope for examining the characteristics of this group for whom a substantial period of home treatment appears to have failed to prevent admission: focusing on them may be of interest in future investigations. With regard to implications for in-patient wards, this study suggests that the group admitted despite the availability of home treatment is likely to be a relatively challenging one, with high representation of young patients who are unwilling to cooperate with care and pose risks of self neglect or harm to others. In-patient staff is thus likely to require substantial support and training, and good staffing levels in order to manage these groups. The effects on in-patient services of the introduction of intensive home treatment have yet to be examined. Finally, identifying the characteristics of patients who are not successfully managed by home treatment allows consideration of other alternatives to admission that may meet their needs better. For example, residential alternatives to admission, such as crisis houses, may be better placed than CRTs to manage the needs of those who neglect themselves.