Sample
Participants in the OCTET Follow-up Study were the 336 patients who were participating in the OCTET Trial. We applied no further inclusion or exclusion criteria.
Objectives
Our objectives were to investigate:
the association between compulsion and levels of disengagement
the effect of randomisation arm (CTO or non-CTO) on levels of disengagement and readmission rates
the association between CTO use and readmission to hospital for patients who experienced time on a CTO
the differential impact of baseline characteristics on the effect of duration of compulsion on discontinuity of care.
The first objective was to test the hypothesis that a longer time under compulsion would increase disengagement from mental health services or discontinuity of contact. These were defined as:
The second objective was to test the hypothesis that at 36-month follow-up, patients in the CTO arm of the trial, compared with those in the non-CTO arm, would:
be more likely to experience disengagement
experience more discontinuities
have a reduction in readmissions to hospital.
The third objective was to test the association between CTO use and (only for patients who experienced time on a CTO, regardless of the arm to which they were randomised):
hospitalisation rate
time to readmission
duration of admissions.
We also aimed to use subgroup analysis to test the hypothesis that the levels of disengagement (time to disengagement and discontinuity of treatment over time) would differ according to factors measured at baseline, and to test for a centre effect by comparing London patients to those in the other sites.
Data collection
We collected all data from medical records at 36 months after randomisation. This continued until 22 February 2014.
Outcomes
Disengagement outcomes
Two variables were used to measure the level of disengagement during the 36-month follow-up period from the first discharge (the date when the patient left the hospital) to the end of the study:
Time to disengagement Disengagement was defined as no service contact for 90 days (or longer) and no return to contact. Time to disengagement was therefore calculated as the number of days from the point of first discharge from hospital to the last clinical contact, if that last contact occurred at least 90 days (3 months) before the end of the study follow-up period. This was a continuous variable, expected to be skewed. Data were censored for patients who died, emigrated or were discharged or imprisoned, and the 90-day period no contact requirement was calculated to that date rather than 1095 days.
Discontinuity of treatment over time This was defined as the number of time periods of 60 days or more in community care without a contact with services. This was a continuous variable expected to be skewed.
The explanatory variable duration of compulsion was defined as follows:
Duration of compulsion This was defined for duration of total compulsion and for duration of community compulsion. Total compulsion was defined as the number of days under any legal compulsion (e.g. Sections 2, 3, 4, 136 or 37, CTO, and 40/48 of the MHA) during the 36-month follow-up period, including time under the initial Section 3 between randomisation and discharge to randomised status. This variable included inpatient and outpatient compulsion times.
Community compulsion This was defined as the number of days in the community on a CTO and excluded inpatient compulsion. Time under voluntary status was not included in this variable.
Readmission outcomes
We used the following readmission outcome variables:
Readmission to psychiatric hospital This was defined as a binary outcome (patient readmitted to hospital (voluntary and involuntary) during the 36-month follow-up period versus patient never readmitted).
Duration of admission Number of nights of psychiatric hospitalisation from first discharge from hospital to 1095 days. This included voluntary and involuntary hospitalisations; nights on recall were not included unless the recall ended in revocation.
Time to first readmission Number of nights from first discharge to first readmission. For patients (in both arms) who remained in hospital for the duration of the study, the time to first readmission was counted as zero. Nights on recall did not count as a readmission unless the recall ended in revocation.
Number of readmissions from first discharge to 1095 days This included voluntary and involuntary hospitalisations but not any recall that did not end in revocation of the CTO.
Community treatment order compulsion outcomes
The following CTO compulsion outcomes were used for patients who experienced a CTO at any point during the study (non-randomised group):
Data management
The study researchers entered most of the data directly into an Access® 2010 database (Microsoft Corporation). They subsequently cleaned these data using an Excel 2010 database. They recorded data for the first 66 patients in clinical research forms. For these 66 patient interviews, different researchers double-entered the data, compared them against each other, and discussed and corrected discrepancies under supervision. The statistician responsible for the analysis performed additional data quality evaluations. These included range checks and logical and consistency checks that might not be picked up by checks at the individual patient level by research staff. In the case of variables that were a function of other variables (e.g. length of a particular hospitalisation), these were checked by automatic calculation of its values. We froze the final cleaned data before starting the analysis.
Statistical methods
As with the OCTET Trial, the research team wrote and signed off a detailed statistical analytical plan before analysing any data (see Appendix 2). All analyses were done according to the statistical analysis plan, except for the analysis of non-randomised CTO group.
Disengagement
‘Time to disengagement’ is a time-to-event outcome. We therefore performed the analysis using a proportional hazards model adjusting for duration of compulsion and the stratification factors [gender (male/female), schizophrenia (yes/no) and duration of illness (< 2 years, > 2 years)]. We present the results as HRs with 95% CIs.
‘Discontinuity of treatment over time’ is a count outcome. We therefore analysed it using a negative-binomial regression model and adjusting for duration of compulsion and stratification factors. We present the results as IDRs, which are interpreted in the same way as relative risks.
We conducted this analysis for the whole sample (not splitting by trial arm).
Readmission
We conducted the analysis for the readmission outcomes using the intention-to-treat population. We analysed the readmission outcomes in the same way as the disengagement outcomes, using multiple regression models with adjustment for the stratification factors. The type of regression model depended on the data distribution. We assessed all model assumptions.
We compared the trial groups for time to disengagement with the non-parametric Wilcoxon rank-sum test because of the violation of the proportional hazards assumption of the proportional hazards model.
Similarly, the model used in the analysis for discontinuity of treatment was adjusted for trial arm, reporting its coefficient (and 95% CI and two-sided p-value) interpreted as an IDR.
We analysed psychiatric hospital readmission in the 36-month follow-up period, as a binary outcome, using log-binomial regression that was adjusted for the trial arm indicator and the stratification factors. We present the results as the relative risk of readmission in the CTO group compared with the non-CTO group, with appropriate 95% CIs and two-sided p-values.
The number of readmissions and duration of readmission are count outcomes. We analysed these using negative-binomial regression models and adjusting for trial arm indicator and the stratification factors. We present the results as IDRs.
The time to first readmission from first discharge to 1095 days is a time-to-event outcome. We therefore performed this analysis using a proportional hazards model, adjusting for the trial arm indicator and the stratification factors. We present the results as HRs with 95% CIs and also present Kaplan–Meier plots. We calculated the median readmission time with 95% CIs.
Community treatment order compulsion
Analyses of association of duration of CTO and readmission outcomes were conducted only on patients who had at least one CTO event so the sample size was restricted to 198 people. Owing to model assumptions violations, the explanatory variable CTO compulsion was split into quartiles.
We analysed the association between CTO compulsion groups and readmission rate using Poisson regression with robust error variances,199 as the log-binomial model was not possible because of model instability. The results are presented as relative risks with 95% CIs.
We performed the analysis for the time to readmission using the non-parametric Kruskal–Wallis test. The proportional hazards model was not used because of the violation of the proportional hazards assumption.
We analysed the duration of readmissions using a negative binomial regression model adjusting for the categorical CTO compulsion and stratification factors. We present the results as IDRs with 95% CIs.
Sensitivity analyses
We conducted a sensitivity analysis for the variables measuring the disengagement outcome, which consisted of repeating the above analyses without adjusting for the stratification factors.
Subgroup analyses
To identify patients’ baseline characteristics associated with a differential effect of duration of compulsion on discontinuity of care, we fitted the same model as for the disengagement outcome, with the inclusion of an additional interaction effect for the interaction between the duration of compulsion and the relevant subgroup variable. The p-value for the interaction test was the p-value of interest, as this was the test of the stated hypothesis. The significance of the compulsion variable was not considered of interest here.
We evaluated the following groups, defined a priori, the first three of which were stratification factors:
gender: male versus female
diagnosis: schizophrenia versus other
duration of illness: < 2 years versus ≥ 2 years
age: ≤ 40 years versus > 40 years
ethnicity: white versus others
born in UK: born in UK versus born in another country
marital status: (single + separated/divorced) versus married/cohabiting
accommodation: independent versus (supported + homeless)
living status: living alone versus living with others
educational level: ≤ 12 years versus > 12 years; tertiary education (yes/no)
BPRS: ≤ 33 versus > 33
GAF: ≤ 49 versus > 49.
A centre effect was evaluated through a similar subgroup analysis for both the disengagement and readmission outcomes. The subgroups were defined by the variable:
Changes to protocol from original proposal
With the exception of the change in the primary objective, detailed above, the analyses we conducted were all in line with the spirit of the analyses detailed in the original proposal. As with the OCTET Trial, we made some changes to the original proposal when writing the statistical analysis plan. In particular, we used more sophisticated adjusted regression models as the primary comparisons, using the simpler unadjusted tests for secondary sensitivity analyses. We performed the adjustment for the variables used in the stratified block design method of randomisation. We did not conduct any minimisation process.
We performed all analyses over the 36-month follow-up period from randomisation (i.e. from randomisation to 1095 days) and not over the 24 months after the OCTET Trial ended, as the original proposal indicated.
