Design: Systematic review and meta-analysis of randomised controlled trials
Population: 8 trials, 1,811 women with postpartum depression
Primary outcome: EPDS score and EPDS-defined clinical improvement
Effect: SMD -0.59 (95% CI -0.77 to -0.41); improvement RR 1.84 (95% CI 1.59 to 2.12)

A systematic review and meta-analysis pooled eight randomised trials of cognitive behavioural therapy (CBT) against usual care in 1,811 women with postpartum depression, searching databases to December 2025.

CBT lowered Edinburgh Postnatal Depression Scale (EPDS) scores by a standardised mean difference of -0.59 (95% CI -0.77 to -0.41), a moderate effect, and raised the EPDS-defined rate of clinical improvement (RR 1.84, 95% CI 1.59 to 2.12). There were possible gains in perceived social support and maternal anxiety, and no difference in a measure of infant temperament. The authors grade the certainty of evidence from very low to moderate.

For obstetric teams the useful message is that a structured talking therapy has a measurable effect and is worth referring for when the EPDS is raised, alongside the usual safety assessment. The authors point to non-specialist and digital delivery as the next step, which matters where perinatal psychiatry is scarce.

Screen with the EPDS at the postnatal visit and act on a raised score rather than reassuring.
Ask about thoughts of self-harm directly whatever the total score.
Offer or refer for structured CBT for mild to moderate postnatal depression.
Refer to psychiatry for severe symptoms, psychotic features or risk to mother or baby.
Revisit symptoms at later contacts: postnatal depression can start or worsen after the first weeks.

Why it matters

It puts a number on what a talking therapy offers a mother who may prefer not to take medication while breastfeeding.

The statistics, in plain English

A standardised mean difference of -0.59 is conventionally a moderate effect. A risk ratio of 1.84 means women given CBT were nearly twice as likely to meet the improvement threshold. Both intervals sit well clear of no effect, but eight trials with varied formats and certainty graded very low to moderate mean the true size could be smaller.