Two perinatal behaviour trials missed their endpoints, then found something after

Behaviour-change trials fail more often than drug trials, and they fail in a characteristic way: the primary endpoint does not separate, and a secondary or per-protocol analysis does. Two 2026 perinatal trials landed in exactly that position, and reading them well means being clear about which numbers were specified in advance.

Relapse after pregnancy

Many people who stop smoking in pregnancy start again after birth. BabyBreathe was designed to prevent that relapse.

Between 4 September 2021 and 3 August 2023, 887 participants were randomised — 442 to BabyBreathe and 445 to usual care — with one post-randomisation exclusion leaving 886 in the intention-to-treat analysis [s1].

At 12 months post partum, abstinence was 242 of 441 (54.9%) with BabyBreathe against 222 of 445 (49.9%) with usual care: an odds ratio of 1.23 (95% CI, 0.94 to 1.60, P = 0.13), a risk difference of 4.8% (−1.7% to 11.2%), and a number needed to treat of 21 [s1].

That is a null result on the primary endpoint. The confidence interval on the risk difference spans zero [s1].

A post hoc per protocol analysis, including only those who received interpersonal support as intended, favoured the intervention: 200 of 347 (57.6%) against 222 of 445 (49.9%), an adjusted odds ratio of 1.36 (95% CI, 1.02 to 1.81, P = 0.04), a risk difference of 7.2% (0.3% to 14.1%), and a number needed to treat of 13.9 [s1].

The authors describe this analysis as post hoc, which is the essential qualifier [s1]. It was not specified before the data were seen.

The deeper problem with per-protocol analysis is not multiplicity but selection. Restricting to participants who received the support as intended breaks randomisation, because receiving an intervention as intended is not random. People who engage with a support programme differ from those who do not — in stability, motivation and circumstances — and those same differences independently predict staying off cigarettes. Some of that 7.2% is the intervention. Some is the kind of person who completes it. The design cannot separate them.

Note also that the per-protocol comparison sets 347 intervention participants against the full 445 usual-care group [s1]. The intervention arm has been filtered for engagement and the control arm has not.

None of which makes the analysis worthless. The authors' conclusion — that a focus on fidelity and further targeting to groups with lower socioeconomic status may be warranted — is a reasonable reading of a trial whose intervention may have been diluted by incomplete delivery [s1]. It is a hypothesis for the next trial, not a finding from this one. No trial-related adverse incidents occurred [s1].

Safe sleep by text message

SMARTER tested video messages promoting infant safe-sleep practices, delivered by text to low-income families, in prenatal and postnatal periods.

Of 1,790 participants (86%) randomised, 1,383 (77%) completed the 60-plus-day survey and were included in the analysis [s2]. Of these, 818 (59.1%) were aged 24 to 34 years; 323 (23.4%) were Black, 325 (23.5%) Hispanic and 598 (43.2%) White [s2]. Participants were assigned to prenatal and postnatal intervention (n = 337), prenatal alone (n = 344), postnatal alone (n = 334), or control (n = 348), and 12% received the intervention in Spanish [s2].

Across groups, 87.6% to 92.2% of mothers reported usual supine position, 83.8% to 89.5% reported usual room sharing without bed-sharing, 75.6% to 84.5% reported no soft bedding use, and 72.5% to 76.0% reported pacifier use [s2]. There were no significant differences among groups for the primary outcomes [s2].

Among seven secondary outcomes, reported exclusive supine positioning at 60 days was higher with the prenatal intervention than control: 240 of 343 (70.6%) against 212 of 348 (60.9%), an adjusted risk difference of 9.7% (95% CI, 3.8% to 15.0%) [s2].

One secondary outcome out of seven separated. With seven comparisons, one crossing a conventional threshold is close to what chance alone delivers, and the trial's own conclusion is that the intervention did not have significant effects on the primary outcomes [s2].

There is a more interesting explanation available than either chance or effect, and it sits in the primary outcome numbers. Between 87.6% and 92.2% of mothers already reported usual supine positioning [s2]. When nine in ten people already do the thing, an intervention promoting it has almost nowhere to go. The trial may have been asking a population that had largely absorbed the message to absorb it again.

That the one positive signal concerns exclusive supine positioning is consistent with this [s2]. "Usual" was near-saturated at 87.6% to 92.2%; "exclusive" started lower at 60.9% in controls, leaving room to move.

A second limitation constrains everything: these are reported practices, from a survey [s2]. Safe-sleep behaviour is socially desirable and self-report is generous. And 1,383 of 1,790 randomised participants completed the survey, so 23% are missing [s2].

What these trials share

Both are well-conducted trials of plausible interventions that did not do what they were designed to do, and both are more useful published than not.

The common lesson is about denominators. BabyBreathe's usual-care arm achieved 49.9% abstinence at 12 months [s1]. SMARTER's control group already reported 87.6% to 92.2% usual supine positioning [s2]. Interventions are hardest to demonstrate where current practice is already good, and both faced that problem.

The differing treatment of their secondary findings is instructive. BabyBreathe's authors label their favourable analysis post hoc and recommend it inform future targeting [s1]. SMARTER's authors lead with the null primary result despite having a positive secondary [s2]. Both are honest reporting.

What to watch

Whether a BabyBreathe trial designed around fidelity — ensuring the interpersonal support is actually delivered — reproduces the per-protocol difference under randomisation [s1]. And whether safe-sleep interventions are redirected toward populations where baseline practice is poorer, rather than tested where nine in ten families already comply [s2].

This article describes published trial results. It is not medical advice.

Sources

• [s1] Effectiveness of the BabyBreathe intervention for maintaining postpartum smoking cessation: multicentre randomised controlled trial, The BMJ, 2026;394:e100242.
• [s2] Safe Sleep Video Intervention via Text Messaging to Low-Income Families: The SMARTER Randomized Clinical Trial, JAMA, 2026;336(9):774–782.