Statistical analysis

Patients were the unit of analysis, so patients in the same team shared the same team characteristics. They were also allocated the same staff characteristics, following the team approach of assertive outreach whereby patients are cared for by the whole team and not by one individual staff member.

Data were analysed using STATA 7.0 for Windows (StataCorp, 1999). Ten patient variables had up to 9% missing values, and 24% of patients had missing values on at least one variable. To avoid loss of precision, we imputed the missing baseline values using multiple imputation (Reference Clark and AltmanClark & Altman, 2003), so that all analyses were based on all subjects with the outcome observed. Because patients in the same team may not be independent, standard statistical techniques would produce incorrect standard errors. We therefore computed all standard errors by the robust method, allowing for clustering within teams (Reference RogersRogers, 1993). All analyses allowed for the sampling fraction (i.e. 0.37 for established patients and 1 for new patients; Reference Priebe, Fakhoury and WattsPriebe et al, 2003) by weighting by its inverse (Reference Horvitz and ThompsonHorvitz & Thompson, 1952). This tended to increase the standard errors by about 15%.

To predict the two dichotomous outcome variables, both univariate and multiple logistic regression was used. Univariate analyses related each outcome via logistic regression to each predictor. Quantitative variables were entered as such, and ordered categorical variables were entered as continuous. The multivariate model was selected from the team, staff and patient variables, starting with all variables that were univariately significant and using stepwise selection to include all variables that were significant independent predictors of either of the two outcomes, controlling for the effects of the other variables in the model. For variable selection, we used a liberal significance level of P<0.15. However, the statistical significance of associations was taken as P<0.05. Results were expressed as odds ratios for the presence v. absence of a characteristic, for a 10% increase in the percentage of contacts in the community, for a 10-year increase in age, for a one standard deviation increase in scores of staff burn-out and satisfaction and for a one unit increase in other variables.

What predicts outcome?

This naturalistic prospective study investigated assertive outreach practice under routine conditions in London. Other studies have suggested that community-focused services can be effective and reduce the number of days that patients spend in hospital (Reference Merson, Tyrer and OnyettMerson et al, 1992; Tyrer et al, Reference Tyrer, Merson and Onyett1994, Reference Tyrer, Evans and Gandhi1998, Reference Tyrer, Manley and Van Horn2000; Reference Tyrer and SimmondsTyrer & Simmonds, 2003) and contacts with the police (Reference Tyrer, Evans and GandhiTyrer et al, 1998; Reference Gandhi, Tyrer and EvansGandhi et al, 2001), but this study does not address the overall effectiveness of assertive outreach teams. Rather, it utilises the existing variation between teams, staff and patients to assess whether (and, if so, in what way) such characteristics predict outcome. Only two simple outcome criteria were used: whether patients were admitted and whether they were admitted compulsorily within a 9-month follow-up period. Other outcome criteria, such as patients’ psychopathology, quality of life and treatment satisfaction, may be equally or even more important targets of assertive outreach than preventing admissions. In some cases, voluntary hospitalisation might even be regarded as a positive outcome if it indicates a degree of engagement with services, although compulsory admission is an adverse outcome that services try to avoid. The advantage of these outcome criteria is that they are clearly operationalised and can be established on the basis of records alone, thus avoiding selection bias through non-response to research interviews.

The most important result is that certain characteristics of teams, staff and patients were all found to be predictive of outcome. This held true in multivariate analyses when the influence of all other variables was controlled for. Before concluding that these effects are causal, we must contemplate the idea that they may be due to confounding by unmeasured variables.

Team characteristics

With respect to team characteristics, weekend working was a strong predictor both of more voluntary admissions in general and of compulsory admissions in particular. The positive association between weekend working and admissions may reflect a greater illness severity of patients referred to teams with weekend working that has not been captured fully by the measured variables. However, there are also other possible explanations: teams that do not work at weekends, by definition, cannot admit any patients on two out of seven days of the week; and staff covering weekends will have to take time off during normal office hours. This might be disruptive to relationships with fellow staff and patients and have an adverse effect on patient outcome. Furthermore, a policy of weekend working may reflect a team philosophy with a stronger focus on risk containment than in teams that do not provide care on weekends. Such emphasis on risk containment may affect clinical decisions to admit patients voluntarily or involuntarily (Reference Tyrer, Morgan and Van HornTyrer et al, 1995). Similar explanations may apply to out-of-hours working, which in univariate analyses, also predicted higher admission and compulsory admission rates. When the influence of all other predictors, including weekend working, is controlled for, however, the effect was reversed (i.e. in addition to the impact of all other variables, extended working hours predicted lower compulsory admission rates), which reflects that the predictive values of some of the tested variables still overlap.

Other team variables often regarded as relevant in the assertive outreach literature, such as multi-disciplinary working, high percentage of contacts in the community and integration of health and social care, do not predict outcome when the influence of other factors is controlled for. These factors therefore may be less important for the effectiveness of teams than has been suggested on the basis of reviews (Reference Mueser, Bond and DrakeMueser et al, 1998; Reference Catty, Burns and KnappCatty et al, 2002). The findings might encourage service providers to be more flexible over these aspects of assertive outreach, and not necessarily adhere to detailed prescriptions lacking research evidence.

Staff characteristics

Staff satisfaction and burn-out was averaged at a team level reflecting the team approach of assertive outreach. Although job satisfaction did not have an impact on outcome, staff burn-out did. It is interesting to note that in the multivariate model two components of burn-out – depersonalisation and high personal accomplishment – were associated with reduced hospitalisation and compulsory admission at 9-month follow-up. This meant that those with more negative views of their patients, and those who viewed themselves more positively regarding their work with their patients, were less likely to have these patients admitted to hospitals. This is surprising given that, univariately, high depersonalisation and low personal accomplishment were associated with admissions, and that high depersonalisation correlated significantly with low personal accomplishment. Thus, the results at the multivariate level could be due to the confounding masking effect of personal accomplishment on depersonalisation.

The impact of staff burn-out is independent of the way the team is organised and of the characteristics of the clients as far as both aspects have been captured by the variables used in this study. How to improve staff morale in assertive outreach teams and maintain it at a level that is as high as possible remains an open question and is an appropriate subject for further research. The findings also suggest that staff burn-out might affect the results of randomised controlled trials comparing assertive outreach with other forms of treatment, particularly when the experimental service is new and has a more charismatic leadership than the service in the control condition.

Patient characteristics

The patient characteristics identified in the univariate analyses as predictors of the two outcome criteria were very similar. This was expected because the two criteria are not independent: hospital admission included compulsory admissions. A higher total number of previous admissions, voluntary or compulsory admissions in the last 2 years, physical violence in the last 2 years and no contact with other services predicted poorer outcome on both criteria. In multivariate analyses, however, different and specific events in the past seem to be the best predictors of similar events in the follow-up period (i.e. hospital admissions in the past predict further admissions, and a history of compulsory hospital admissions is the best predictor of compulsory admissions in the future). One might conclude that where treatment has failed in the past it is more likely to fail in the future, and those patients for whose care the assertive outreach teams have been specifically set up (i.e. those with a history of voluntary and compulsory admissions), still have the poorest outcome. Assertive outreach teams face the same problems with these patients as generic community mental health teams, despite their superior resources and targeted approach. This implies that teams with a high percentage of this core group of patients managed by assertive outreach on their case-load inevitably tend to achieve a less favourable average outcome, and what teams can realistically accomplish will depend on the history of their patients.

Contact with other services emerged as a very powerful, independent predictor of favourable outcome. To some degree, patients’ contact with other services might simply reflect a higher level of engagement, a greater willingness to accept support and better skills to seek and receive it. Thus, patients’ attitudes and skills may explain the predictive association. Nevertheless, the fact that contact with other services alone reduces the risk for voluntary and compulsory admissions by around 50% may be seen as evidence for the importance for multi-agency working with this group.

Implications and future research

The findings of the study point at the complexities of predicting outcome under routine conditions. Aspects of how the team is organised, staff burn-out, patients’ history and their contact with other services have been identified as independent significant predictors and should be considered in research as well as clinical practice. In the UK, the decision on whether assertive outreach should be implemented has been taken, and assertive outreach teams will be part of established services for some time to come. The challenge now is to evaluate how the teams work and to improve their effectiveness. This study provides some indication about what factors may have to be targeted in the processes of clinical governance and service development.

The Pan-London Assertive Outreach Study Group

Tom Burns, Christine Wright, Peter James, Adele Greaves, Christine Benfell, Nan Greenwood (St George's Hospital Medical School, London); Paul Bebbington, Sonia Johnson, Joanne Billings (University College London and Camden and Islington Mental Health and Social Care Trust); Stefan Priebe, Walid Fakhoury, Joanna Watts (Unit for Social and Community Psychiatry, Barts’ and The London School of Medicine); Matt Muijen, Ian Ryrie, Rebecca Walwyn (Sainsbury Centre for Mental Health); Ian White (Medical Research Council Biostatistics Unit, Cambridge).

CLINICAL IMPLICATIONS

1. ▪ Aspects of how assertive outreach teams work, staff burn-out and patient characteristics are associated independently with outcome.

2. ▪ Weekend working predicts higher admission/compulsory admission rates.

3. ▪ Patients who are in contact with services other than assertive outreach have a much lower risk of voluntary and compulsory admission.

LIMITATIONS

1. ▪ Only voluntary and compulsory admissions within a 9-month follow-up period were taken as outcome criteria. Other criteria such as patients’ views and quality of life were not considered.

2. ▪ Only 25 predictor variables were tested.

3. ▪ This is a naturalistic observational study, and the associations found may reflect confounding rather than causal relationships.