Table 3.
Primary outcome |
Secondary outcomes* |
||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Primary analysis: Cox proportional hazards for first event |
Sensitivity analysis: frailty model, clustering by therapist |
Sensitivity analysis: frailty model, clustering by NHS trust |
Sensitivity analysis: including unclassified hospital attendances† |
Cox proportional hazards for first event (within 12 months) |
Recurrent events |
||||||||
Hazard ratio (95% CI) | p value | Hazard ratio (95% CI) | p value | Hazard ratio (95% CI) | p value | Hazard ratio (95% CI) | p value | Hazard ratio (95% CI) | p value | Hazard ratio (95% CI) | p value | ||
Fixed effects | |||||||||||||
Treatment: family therapy (vs treatment as usual) | 1·14 (0·87–1·49) | 0·33 | 1·09 (0·81–1·46) | 0·50 | 1·14 (0·87–1·48) | 0·33 | 1·15 (0·89–1·49) | 0·27 | 1·09 (0·81–1·48) | 0·56 | 1·05 (0·76–1·44) | 0·78 | |
Sex: female (vs male) | 1·60 (0·98–2·61) | 0·059 | 1·61 (0·98–2·63) | 0·057 | 1·54 (0·94–2·50) | 0·084 | 1·65 (1·03–2·65) | 0·039 | 1·60 (0·92–2·79) | 0·094 | 1·27 (0·77–2·10) | 0·34 | |
Age group: 15–17 years (vs 11–14 years) | 0·70 (0·53–0·92) | 0·011 | 0·69 (0·52–0·91) | 0·0095 | 0·70 (0·53–0·93) | 0·012 | 0·75 (0·58–0·98) | 0·038 | 0·72 (0·53–0·99) | 0·043 | 0·67 (0·50–0·92) | 0·012 | |
Previous self-harm episodes: at least three (vs two) | 1·22 (0·78–1·92) | 0·39 | 1·21 (0·77–1·91) | 0·41 | 1·22 (0·78–1·92) | 0·38 | 1·20 (0·78–1·85) | 0·41 | 1·31 (0·77–2·22) | 0·32 | 1·52 (0·92–2·49) | 0·10 | |
Type of index episode | .. | 0·033 | .. | 0·035 | .. | 0·023 | .. | 0·020 | .. | 0·071 | .. | 0·064 | |
Combined (vs injury) | 1·83 (1·14–2·96) | .. | 1·83 (1·13–2·98) | .. | 1·85 (1·15–2·98) | .. | 1·90 (1·20–3·02) | .. | 1·80 (1·05–3·09) | .. | 1·20 (0·66–2·18) | .. | |
Poisoning (vs injury) | 1·03 (0·69–1·54) | .. | 1·02 (0·68–1·53) | .. | 1·00 (0·67–1·49) | .. | 1·09 (0·74–1·60) | .. | 1·00 (0·63–1·57) | .. | 0·72 (0·45–1·16) | .. | |
Referred via hospital: yes (vs community) | 1·31 (0·93–1·86) | 0·12 | 1·33 (0·93–1·88) | 0·11 | 1·39 (0·99–1·95) | 0·060 | 1·24 (0·88–1·74) | 0·21 | 1·27 (0·86–1·88) | 0·23 | 1·98 (1·18–3·32) | 0·0096 | |
NHS trust | .. | 0·094 | .. | 0·15 | .. | .. | .. | 0·077 | .. | 0·14 | .. | 0·049 | |
Random effects | |||||||||||||
Main therapist | .. | .. | .. | 0·37 | .. | .. | .. | .. | .. | .. | .. | .. | |
NHS trust | .. | .. | .. | .. | .. | 0·065 | .. | .. | .. | .. | .. | .. |
‡p values are for type of index episode (injury, poisoning, or combined).
Further secondary analysis using adjusted probit regression found similar estimates for the intention-to-treat, as treated, and complier average causal effect analysis. There was a similar effect of family therapy receipt in the complier average causal effect analysis (parameter estimate 0·12 [SE 0·13], p=0·34) compared with the standard intention-to-treat estimate of the allocation of family therapy (0·11 [SE 0·10], p=0·24), and the as treated estimate (0·10 [SE 0·10], p=0·31); with no significant differences detected between trial groups, or receipt of family therapy.
47 unclassified attendances in 41 participants were classed as being related to self-harm, thus contributing new primary outcome events for 18 participants, and earlier primary outcome events for nine participants.