Twelve Trials Say Yes, Eleven Say No: CANMAT Puts Probiotics Third in Line

Patients arrive holding a bottle of multi-strain capsules and a headline about the gut-brain axis, and the honest answer has been "the data are mixed" – a phrase that sounds like evasion because it carries no number. CANMAT has now quantified exactly how mixed, and the answer is close to a coin toss. This is the first major guideline body to formally place microbiome-targeted interventions in a treatment line, which means the question in the consulting room has changed from "is there anything to this?" to "what, precisely, am I allowed to say?"

The box score nobody puts in the press release

Across the 23 included RCTs, 12 (n=781) reported statistically significant symptom improvement and 11 (n=795) did not. The sample sizes on either side of the ledger are nearly identical, which is what makes the split so informative: this is not a case of small negative trials being drowned out by large positive ones. The pooled figures cluster small and consistent. Pan et al. (2025) found an SMD of -0.26 across 34 RCTs of microbiome-targeted treatments; Goh et al. (2019) found -0.24 across 19 RCTs (n=1901); Chao et al. (2020) found -0.47. Amirani et al. (2020) reported a weighted mean difference of -9.60 on the HAM-D alongside reductions in CRP and IL-10. CANMAT's own report notes that an SMD of roughly 0.2 to 0.3 has been described as modest in prior CANMAT documents.

The trajectory of the estimates is more telling than any single one. Nikolova et al. published an SMD of 1.371 in 2019 – a figure that would rival established antidepressants – and the 2021 update did not so much revise that number as take it apart: 0.83 remained for adjunctive use, while probiotics taken on their own came out at -0.02. Ng et al. (2018) found no significant main effect, recovering a benefit only in a mild-to-moderate subgroup. Pan et al. found the effect more pronounced in studies from Asia and in participants without gastrointestinal comorbidity, and null in perinatal depression and in trials running longer than 12 weeks. Geographic clustering of positive results is a familiar signal in nutraceutical literature, and it is a reason for caution rather than a mechanism.

The most informative methodological fact in the report is a negative one: CANMAT declined to run a meta-analysis at all. Heterogeneity across strains, doses, durations (4 to 12 weeks), populations and outcome measures was judged to preclude formal pooling. On quality, the task force explicitly did not apply Cochrane RoB 2.0, AMSTAR-2 or GRADE, developed its protocol internally without formal registration, and acknowledges in its own limitations that the synthesis may be subject to selection and reporting bias. Several of the eight meta-analyses it reviewed include overlapping trials. Read plainly: a national guideline body looked at this literature and concluded it was too incoherent to average.

What to say when the bottle appears on your desk

Third-line adjunctive is a narrow permission, and it is worth reading how narrow. CANMAT's own rating table would place Level 3 evidence at second-line; the task force actively downgraded probiotics below that, citing lack of consistent replication across specific strains, heterogeneity in formulation and dose, and practical problems of product standardisation and regulatory oversight. The strongest signal sits in adjunctive use for partial responders to antidepressants, and even there it is thin. Nikolova et al. (2023, UK, n=49, 8 weeks) was a pilot explicitly powered for acceptability rather than efficacy; its HAM-D advantage held at week 4 only, and GAD-7 was null. Against that, Lin et al. (2024, Taiwan, n=32) found no between-group difference in symptoms (p=0.203) – and no significant shift in gut microbiota either, neither before versus after nor between groups. A trial that moves neither the microbiome nor the patient is a fair measure of how far the mechanism still sits from the clinic. Clinicians working in Russian-speaking settings should note Arifdjanova et al. (2021, n=119, multi-strain probiotic added to an SSRI over 6 weeks), which found no significant change on the HAMD-17. Outside the adjunctive question, CANMAT's monotherapy section carries Strodl et al. (2024, Australia, n=120), where improvement at 4 weeks had lost significance by weeks 8 and 16 – and where the probiotic came bundled with magnesium orotate and coenzyme Q10, so its own contribution cannot be isolated.

What not to promise: not monotherapy – the standalone evidence is limited and inconsistent, and CANMAT declines to recommend it. Not a named strain or dose, because the heterogeneity that blocked pooling also blocks any strain-specific advice. Not prebiotics, synbiotics or faecal microbiota transplantation: prebiotics and FMT were judged insufficient, synbiotics rested on a single small RCT, and the only FMT trial (Green et al., 2023, n=15) was underpowered for efficacy. Not benefit in perinatal depression, and not benefit sustained past 12 weeks. And not "it can't hurt": tolerability was good, with mild gas, bloating and transient discomfort dominating, but isolated bacteraemia has been reported in critically ill and immunocompromised patients, and CANMAT is blunt that adverse-event reporting is inconsistent, so the absence of documented harm is not evidence of safety.

The practical position is therefore small but real. If you have an adult on an adequate antidepressant trial with a partial response, no immunocompromise and no medical complexity, who is already inclined toward the intervention and can afford it, a multi-strain probiotic over 4 to 8 weeks is a defensible third-line add-on with a modest expected effect and a low harm ceiling – provided you name the effect size out loud, agree in advance on a stop date and an outcome measure, and drop it if the score has not moved. What the evidence does not support is initiating it in an antidepressant-naive patient, extending it indefinitely because "it's just a supplement", or letting it occupy the slot where a second-line augmentation strategy belongs. The gut-brain axis is a real biological system; the marketing around it has run roughly a decade ahead of the trial data, and this guideline is the first authoritative document to say so with numbers attached.