France could not establish the benefit of a digital insomnia therapy, which is not the same as finding none
- France's device commission CNEDiMTS, part of the Haute Autorité de Santé, judged on 15 July 2025 that SOMNIO v1.1, a digital CBT-I app used without a therapist, lacks sufficient expected benefit to be listed for reimbursement in adult chronic insomnia. The ground given is a file with a low level of evidence, and the same opinion notes that both studies it kept point to fewer insomnia symptoms.
- In the main German trial (n=238, open label, waiting-list control), the Insomnia Severity Index at 8 weeks differed by -7.60 points (95% CI -8.67 to -6.54), a self-reported sleep measure. Depressive symptoms (ADS-K, -5.48) and trait anxiety (STAI-T, -5.73) moved as well, and HAS counts both as exploratory secondary results.
- The main objection is missing information on hypnotic and psychotropic use during follow-up, which HAS terms a major confounding factor; a control arm without in-person CBT-I and roughly 30% missing data at one year come next. The first vote, on 20 May 2025, was unanimous; after ResMed's hearing on 15 July the opinion was kept by 12 votes, with 2 abstentions.
- A separate randomised trial in 140 adults with depression and insomnia, with the PHQ-9 as primary outcome, came out in Depression and Anxiety on 26 May 2025, six days after the first vote: PHQ-9 at 12 weeks was 3.34 points lower than with a waiting list (95% CI -4.70 to -1.98). The opinion's list of supplied publications does not include it, and the partly redacted transcripts of both sessions do not mention it.
The benefit of SOMNIO "ne peut être établi" – cannot be established – in the indication ResMed claimed. Earlier, the opinion's summary of the data says both retained studies suggest an effect in reducing insomnia symptoms. The two sentences belong to CNEDiMTS, the device commission of France's Haute Autorité de Santé, which on 15 July 2025 found the expected benefit (Service Attendu) of SOMNIO v1.1, a self-guided digital CBT-I programme, insufficient for reimbursement in adult chronic insomnia. What lies between them is attribution.
Insomnia Severity Index
8.56 against 15.95 was the mean score at 8 weeks in the pivotal trial (Schuffelen 2023, Sleep), which randomised 238 German adults with chronic insomnia to SOMNIO or to a waiting list, both on top of usual care. The adjusted difference of -7.60 points (95% CI -8.67 to -6.54) is self-reported insomnia severity, scored at baseline and at 8 weeks in an open-label trial; 40.7% against 1.7% reached remission, an ISI below 8.
Depressive symptoms and trait anxiety were secondary endpoints: -5.48 on the ADS-K (95% CI -6.96 to -4.02), -5.73 on the STAI-T (-7.37 to -4.16). HAS treats every secondary endpoint as exploratory, since multiplicity of tests was not taken into account, and it set aside a post-hoc analysis of participants with marked depressive or anxiety symptoms. The file therefore tests an intervention in sleep against a sleep outcome, and psychiatric change enters it only as exploratory data.
Comorbid depression
Six days after the commission's first vote on 20 May, and eight days before its draft opinion of 3 June, Depression and Anxiety published a trial of the same app whose primary outcome was psychiatric (Schuffelen 2025). It enrolled 140 adults with a diagnosed depressive disorder and insomnia disorder, excluding bipolar disorder, psychosis and suicidality. Against a waiting list, 12 weeks of SOMNIO plus care as usual gave an adjusted PHQ-9 difference of -3.34 points (95% CI -4.70 to -1.98), and -2.83 at 24 weeks (-4.23 to -1.43). Remission, 5.7% against 1.4%, did not differ significantly.
This trial is not among the four publications ResMed supplied, and neither side mentions it in the published transcripts of the two sessions, which are redacted in places. Whether it reached HAS in writing, those documents do not say.
Hypnotic use after randomisation
Jean-Louis Pépin, a Grenoble sleep physiologist speaking for ResMed at the 15 July hearing, argued that people in CBT-I tend to cut their hypnotics, so any drift would have narrowed the gap between arms. He agreed it would have been better had the trial reported medication in detail at each visit. Participants had been asked to keep their sleeping pills and psychotropics unchanged, and the study offers no data on what they took over time. The opinion names this a major confounding factor that rules out concluding with certainty that the effect belongs to SOMNIO alone.
In-person CBT-I
About 200 trained professionals for 6.7 million people with chronic insomnia in France: ResMed cited that expert estimate at the hearing to argue that usual care, what patients actually receive, is the fair comparator. After ResMed left, one member said the argument skips a step, because measuring against usual care presumes that digital CBT-I works, and she said she found no proof of that. The vote kept the opinion, 12 in favour and 2 abstentions. Online recruitment, a self-reported primary outcome in an open trial and close to 30% of 12-month data missing complete the opinion's list.
HAS could not attribute the fall in insomnia scores to the app with certainty, and a benefit that cannot be established from a file is a different finding from a benefit shown to be absent.
This is a reimbursement opinion, not a study, and it covers one product, SOMNIO v1.1, in one claimed indication in France. It does not rule on digital CBT-I as a method: the same opinion restates that guidelines place CBT-I first line in person or in digital form, provided each digital tool is evaluated rigorously. The psychiatric results in the file are exploratory secondary endpoints of an open-label trial. The 2025 depression trial was also open, with a waiting-list control, and the authors' listed affiliations include Mementor DE GmbH, the manufacturer. The published transcripts are redacted versions and the applicant's slides are not among the published documents, so whether the 2025 trial was submitted in writing cannot be determined from them. SOMNIO's reimbursed status in Germany as a DiGA rests on the manufacturer's press release of 27 October 2020 and a remark by HAS staff at the 20 May session; the BfArM register itself was not read. The HAS table prints an interval for nightmare frequency that cannot be read as it stands, and it is not used here.