MamaLift Plus Against Its Own Sham: What the Paper Reports and What the Registry Records

A New Jersey digital therapeutics company put its postpartum depression app through a properly sham-controlled pivotal trial and reported a wide separation on the primary endpoint. In the authors' own conflicts statement, Curio Digital Therapeutics developed MamaLift Plus and funded this investigation, and all four authors are its employees. That does not make the result wrong, and the comparator is better built than most in this field. It does mean the structural questions carry more weight than usual. On several of them the paper and its own registry record do not say the same thing.

How the trial was built

SuMMER was decentralised and conducted in the United States, with participants distributed across the country and the largest shares from Texas, California, New York, Florida and New Jersey. Recruitment ran entirely through paid advertising on Meta platforms between 18 April and 24 May 2023, with the advertisements describing a paid research study; women who completed all study activities received a US $100 gift card. Of 2,177 people assessed for eligibility, 141 were randomised 2 to 1: 95 to MamaLift Plus, 46 to the sham.

Entry was narrower than a one-line description suggests. A screening Edinburgh score of 13 to 19, that is mild to moderate symptoms. Then a meeting with a licensed mental health provider who administered a diagnostic clinical interview based on DSM-5 criteria together with a HAMD-17, with only scores in the 8 to 23 range eligible to proceed. Anyone previously diagnosed with serious mental illness was excluded, as was anyone who had taken part in an earlier Curio study.

The sham was not a waiting list. It reproduced the interface, colour scheme and feature set of the product, carried content of general interest to women in the postpartum period, and both arms were told they were using the SuMMER study app. What it omitted was the therapy content. On the primary endpoint, an improvement of at least 4 points on the Edinburgh scale at week 8, 86.3% (82/95) responded on the app against 23.9% (11/46) on the sham, P<0.0001. Mean improvement was 8.34 points against 1.89. The secondary endpoint, falling below 13 on the same scale, gave 83.2% against 32.6%.

The paper also carries a second secondary endpoint, prespecified and described as supportive: improvement on the clinician-rated HAMD-17 at end of treatment. It is mostly missing, and the authors say why. "The supportive EOT HAMD-17 was intended to be collected from all participants. However, most participants exited the study before EOT HAMD-17s could be initiated." It exists for 43 of the 141 randomised women, 33 on the app and 10 on the sham, and among them mean improvement was 12.1 points (SD 5.87) against 9.4 points (SD 9.52), P=0.28. Three things make this a weak read rather than a second look at the same finding: it is a different instrument from the primary, on a different scale range; the subsample is defined by having survived to end of treatment; and the HAMD-17 baseline was taken at the screening visit that gated eligibility, which builds regression to the mean into any within-arm change. The authors do not call the comparison underpowered. They write that the missing data "had a minimal influence on the results" and that "a trend can be discerned" in the product's favour. With 10 women in the sham arm, P=0.28 carries no information about whether the arms differ, and a difference that small on that sample is not a trend either.

What the registry records

NCT05958095 is worth opening beside the paper.

Registration is retrospective. The record was first submitted on 14 July 2023 and posted on 24 July 2023. The same record gives the trial's actual start as 1 April 2023 and its actual primary completion as 24 May 2023. The filing therefore happened after the trial had finished. Prospective registration exists so that endpoints are fixed in public before anyone has seen the data; a record entered afterwards cannot perform that function, whatever the reason for the delay.

The registry lists two primary outcomes: the proportion of women improving their Edinburgh score by at least 4 points, and the proportion improving to below 13. The paper reports the first as the primary endpoint and the second as a secondary one. Both were met here, so nothing in the conclusion turns on it, but a co-primary outcome moved to secondary is a change to the primary analysis and the paper does not flag it.

Masking in the registry is recorded as single, with participants the only masked party. The paper describes the trial as double-blind in its abstract and in its methods.

No HAMD outcome appears in the registry at all. The clinician-rated endpoint the paper prespecifies as its second secondary has no registry counterpart.

One further point of description. The registry's lead sponsor is Healthcare Innovation Technology Lab, with Curio Digital Therapeutics listed as a collaborator, and the responsible party is a principal investigator at that lab. Curio's role, stated in the paper, is that it developed MamaLift Plus and funded this investigation. Developer and funder is the exact description, and it is the one to use.

For your practice

Two structural facts belong beside the headline. The first is the authors' own admission in their limitations: "there was no test for blindness, without which the study investigators cannot be certain that control arm participants did not alter their behavior due to knowledge of their allocation." The sham was built to be indistinguishable; indistinguishability was designed for, not measured. The second is that all four authors work for the company that developed and funded the product and hold a financial stake in it. They describe mitigations: a predetermined analysis plan, deidentified data during analysis, randomisation codes held by a contract research organisation, and study conduct largely outside the author group.

What the trial does show is a large difference in self-reported response over eight weeks in mild to moderate symptoms, against a comparator that was seriously made. A skeptical reading should not flatten that into nothing. What it does not show is that the difference is visible to a clinician rating the same patients, that the blind held, or that anything remains after week 8. The authors name the missing follow-up as their main limitation and say 30-day and 90-day designs are planned.

So when a self-guided app arrives on your desk with a responder rate attached, four questions do most of the work. Who filled in the primary scale, the patient or a clinician? Was blinding tested, or only intended? Does the registry entry match the published account, and was it filed before the data existed? And how long after the last module did anyone measure anything?

On placement, the paper is explicit: MamaLift Plus "is intended to be used as an adjunct to clinician-managed outpatient care", and the registry describes both arms as receiving treatment as usual alongside the app. Concurrent care continued throughout; 16 of the 95 women in the product arm and 14 of the 46 in the sham arm reported taking antidepressants. Nothing here was tested as a stand-in for care a woman cannot get, so it should not be offered as one. Alongside the care she is already receiving, with the follow-up appointment booked, it has eight weeks of self-reported evidence behind it and nothing past that.