Individual treatment selection for patients with post-traumatic stress disorder: External validation of a personalised advantage index

Figures & data

Table I. Sample characteristics (after multiple imputation).

	Tf-CBT (N = 1155) Mean (SD) or %	Tf-CBT after PSM (N = 114) Mean (SD) or %	EMDR (N = 38) Mean (SD) or %
PHQ-9 pre	16.05 (6.18)	14.93 (6.62)	14.82 (6.43)
PHQ-9 post	11.43 (7.58)	10.37 (7.82)	9.05 (7.17)
GAD-7 pre	14.93 (4.81)	14.37 (5.27)	14.16 (5.00)
GAD-7 post	10.62 (6.52)	9.99 (7.09)	8.68 (6.40)
WSAS pre	21.11 (9.67)	17.13 (10.01)	16.87 (8.95)
WSAS post	15.74 (11.01)	13.63 (11.05)	11.16 (8.89)
Gender (female)	62.17%	59.65%	63.16%
Age	38.94 (12.79)	43.02 (14.87)	40.63 (12.50)
LTC	29.09%	30.70%	26.32%
Disability	8.83%	21.05%	15.79%
Employment pre
Employed	49.44%	47.37%	47.37%
Student	4.24%	0.00%	0.00%
Unemployed	3.55%	3.51%	5.26%
Long-term sick	17.40%	9.65%	7.89%
Other^a	25.37%	39.47%	39.47%
Medication^b pre
Prescribed	55.06%	50.00%	50.00%
Prescribed not taking	3.55%	4.39%	5.26%
Not prescribed	41.39%	45.61%	44.74%
Ethnicity^c
White	74.46%	77.19%	89.47%
Mixed/Multiple	3.64%	2.63%	5.26%
Asian/Asian British	7.62%	7.02%	2.63%
Black/Black British	10.22%	9.65%	2.63%
Other	4.07%	3.51%	0.00%
IAPT appointments attended	8.61 (5.15)	9.07 (5.26)	7.00 (4.51)
High intensity treatment sessions	7.50 (4.93)	7.88 (5.16)	5.68 (4.53)
Accessed low intensity interventions	65.63%	65.79%	44.74%

Note. EMDR = Eye-movement desensitisation and reprocessing; GAD-7 = Generalised Anxiety Disorder 7; LTC = Long-term medical condition; PHQ-9 = Patient Health Questionnaire 9; PSM = Propensity score matching; Tf-CBT = Trauma-focussed cognitive behavioural therapy; WSAS = Work and Social Adjustment Scale.

^a Employment Other = Voluntary work, homemaker, carer, or retired.

^b Medication = Antidepressant medication.

^c Ethnicity = Office for National Statistics ethnic group.

Table II. Treatment outcome in the total sample and matched Tf-CBT and EMDR groups.

Measure	Sample	d	95% Confidence interval
			Lower limit	Upper limit
PHQ-9	Total (N = 152)	0.74	0.58	0.90
	Tf-CBT (n = 114)	0.88	0.56	1.20
	EMDR (n = 38)	0.68	0.50	0.87
GAD-7	Total (N = 152)	0.89	0.71	1.07
	Tf-CBT (n = 114)	1.07	0.69	1.45
	EMDR (n = 38)	0.83	0.62	1.03
WSAS	Total (N = 152)	0.41	0.27	0.56
	Tf-CBT (n = 114)	0.62	0.32	0.93
	EMDR (n = 38)	0.35	0.18	0.52

Note. EMDR = Eye-movement desensitisation and reprocessing; GAD-7 = Generalised Anxiety Disorder 7; PHQ-9 = Patient Health Questionnaire 9; Tf-CBT = Trauma-focussed cognitive behavioural therapy; WSAS = Work and Social Adjustment Scale.

Effect sizes and confidence intervals calculated using the method described by Minami et al. (2008).

Table III. Model prediction accuracy (R²) and error (RMSE) of the Tf-CBT and EMDR prediction models in the development and validation samples.

Sample	Tf-CBT Model		EMDR Model
	R^2	RMSE	R^2	RMSE
Development sample (N = 225)	.28	6.45	.20	6.84
Development Tf-CBT (n = 150)	.38	5.97	.14	7.21
Development EMDR (n = 75)	.11	7.32	.35	6.05
Validation sample (N = 152)	.30	6.62	.45	5.85
Validation Tf-CBT (n = 114)	.28	6.92	.47	6.00
Validation EMDR (n = 38)	.38	5.64	.42	5.40

Note. EMDR = Eye-movement desensitisation and reprocessing; PHQ-9 = Patient Health Questionnaire 9; RMSE = Root Mean Squared Error; Tf-CBT = Trauma-focussed cognitive behavioural therapy.

R² was calculated by squaring the correlation (Pearson's r) between the predicted and observed post-treatment PHQ-9 scores and can be interpreted as the proportion of variance explained by the model.

RMSE was calculated by taking the square root of the mean of the squared differences between the predicted and observed post-treatment PHQ-9 scores.

Figure 1. Calibration plot comparing predicted and observed post-treatment PHQ-9 scores.

Table IV. Comparing the rate of reliable improvement in PHQ-9 score between cases who received their model-indicated optimal treatment (N = 87) and those who received their model-indicated suboptimal treatment (N = 65).

Treatment received	Reliable improvement
	Yes n (%)	No n (%)
Optimal	34 (39.08%)	53 (60.92%)
Suboptimal	27 (41.54%)	38 (58.46%)

Note. X² (1) = 0.02, p = 0.890.

Table V. Predicting post-treatment PHQ-9 score (square root transformed) from having received model-indicated optimal treatment, adjusted for pre-treatment PHQ-9 score and propensity score estimate (log transformed), among patients with a PAI > 1.92 (N = 93).

Effect	B	SE	β	t	95% CI		p
					Lower	Upper
Intercept	0.45	0.51		0.87	−0.57	1.46	.386
Optimal treatment	0.33	0.24	0.12	1.39	−0.14	0.81	.168
PHQ-9 pre	0.13	0.02	0.61	7.23	0.09	0.17	<.001
Propensity score (log)	−0.14	0.35	−0.03	−0.40	−0.84	0.56	.691

Note. F (3, 89) = 22.77, p < .001, R² = .43.

Minami, T., Serlin, R. C., Wampold, B. E., Kircher, J. C., & Brown, G. (2008). Using clinical trials to benchmark effects produced in clinical practice. Quality and Quantity, 42(4), 513–525. https://doi.org/10.1007/s11135-006-9057-z

 Web of Science ®Google Scholar

Data Availability Statement

In line with the requirements of the ethics review board for this study, requests for access to data are to be made in writing to the data controller, Michael Barkham, [email protected].