Table 2.

Long-term effects of CBT compared with pharmacotherapy: ORs of response†

	N	OR	95% CI	I2‡	95% CI	NNT	95% CI	p§
CBT vs continued pharmacotherapy
All studies	5	1.62	0.97 to 2.72 *	0	0 to 79	10	¶
One possible outlier excluded††	4	1.77	1.04 to 3.01	0	0 to 85	8	4 to 71
CBT vs discontinued pharmacotherapy
All studies	8	2.61	1.58 to 4.31****	0	0 to 68	5	4 to 11
Three possible outliers excluded‡‡	5	2.47	1.45 to 4.22****	0	0 to 79	6	4 to 15
Subgroups (long-term effects)
Pharmacotherapy§§
SSRI	2	3.02	1.29 to 7.04**	0	¶¶	5		0.82
TCA	5	2.66	1.40 to 5.04***	0	0 to 79	6	4 to 15
Included sample
All	2	1.97	0.91 to 4.27 *	0	¶¶	9	¶	0.14
Responders	6	3.20	1.65 to 6.19***	0	0 to 75	4	3 to 8
Quality
All 4 criteria	5	2.31	1.28 to 4.16***	0	0 to 79	6	2 to 11	0.25
≤3 criteria	3	3.58	1.39 to 9.22***	0	0 to 90	4	2 to 10
Short-term effects
All the studies	9	1.15	0.74 to 1.79	53	0 to 78	20	¶
One possible outlier excluded†††	8	0.96	0.72 to 1.30	0	0 to 68		¶
Drop out from intervention‡‡‡	8	0.59	0.34 to 0.99**	48	0 to 77	9	5 to 143

*p<0.1.

**p<0.05.

***p<0.01.

****p<0.001.

†According to the random effects model.

‡In this column, I² is reported; we also tested whether the Q value was significant. This was the case in two comparisons (indicated with an asterisk*).

§The p value indicates whether the subgroups differ from each other.

¶The 95% CI includes zero and would result in a negative NNT; therefore, we do not report the 95% of the NNT here the 95% CI included zero; because this would result in a negative NNT, we do not report this here.

††Jarrett et al²⁶.

‡‡Blackburn et al, 1981; Jarrett et al²⁶; Evans et al²⁴.

§§One study examined phenelzine (Jarrett et al²⁶); this was not included in the analyses.

¶¶95% CI could not be calculated when degrees of freedom is lower than two.

†††Kovacs et al²⁷.

‡‡‡One study did not report data on drop out (Blackburn et al ²¹).

CBT, cognitive behaviour therapy; NNT, numbers-needed-to-be-treated; SSRI, selective serotonin reuptake inhibitor; TCA, tricyclic antidepressant.