The evidence behind Bloom
Bloom is designed from the perinatal mental-health research — and we hold ourselves to the same standard we'd want from anyone calling a new mother. This page sets out what the evidence says, how strong it is, and where the open questions sit, including for Bloom's own approach.
Nothing on this page is a claim that Bloom diagnoses, treats or prevents any condition. Bloom screens, supports and connects — clinical decisions rest with qualified people.
How we grade evidence
Randomised trials, meta-analyses, national clinical guidelines or registries.
Large cohorts, national surveys, economic analyses — solid but observational or dated.
Pilots, early studies, indirect evidence — promising, unproven. Includes Bloom's own modality.
Perinatal depression and/or anxiety affects up to one in five Australian mothers [1] and around one in ten fathers and partners [2]. We grade the Australian figure Moderate because the canonical national collection dates from 2010 — there is no newer national prevalence registry, which is itself part of the problem.
Perinatal depression and anxiety were estimated to cost Australia A$877 million in the first year after birth [3]. UK modelling puts the cost of perinatal mental illness at £8.1 billion per annual birth cohort — with roughly 72% of that falling on child, not maternal, outcomes [4].
A 2025 umbrella review confirms the dominant risk factors: poor social support (risk ratio ≈ 3.6), a prior history of depression or anxiety (odds ratio ≈ 3.1), significant sleep disruption, birth trauma, and intimate-partner violence [5]. Mothers of babies admitted to NICU experience postnatal depression at 40–50% — three to four times the rate after term births [6]. Isolation and support are exactly the axis a proactive check-in works on.
In CDC analysis of a large US cohort, 57% of women with depressive symptoms at 9–10 months postpartum had reported no symptoms at the usual 2–6-month screen [7]. Symptom prevalence persists at ~15–16% into the second year and ~19% beyond two years [8]. Late onset is the norm, not the exception — single-time-point screening structurally misses it.
The NHMRC-approved 2023 COPE guideline recommends EPDS screening repeated through pregnancy and the postnatal year, structured psychosocial assessment, and — new in 2023 — screening fathers and non-birthing partners and asking about psychological birth trauma [9]. International bodies agree: USPSTF recommends screening all perinatal women for depression and anxiety, and referring those at risk to preventive counselling [10].
The EPDS (Cox, Holden & Sagovsky, 1987) is the world's most-used perinatal screening questionnaire [11]. An individual-participant meta-analysis across 58 studies found that at the common cut-off it identifies roughly 85% of women experiencing major depression, with ~84% specificity [12]. It screens — it does not diagnose. Bloom's conversations are structured around its ten wellbeing themes:
| Signal tier | Indicative range | What happens |
|---|---|---|
| Low | 0–9 | Routine summary to the care team |
| Moderate | 10–12 | Flagged for follow-up within days |
| High | 13+ | Prioritised for prompt clinician review |
| Self-harm theme | any | Immediate human escalation, whatever the score |
In a multisite RCT of 701 higher-risk mothers, proactive telephone peer support roughly halved depression risk at twelve weeks — 14% vs 25%, relative risk 0.54; about one case averted for every nine mothers supported [13]. A Cochrane review of psychosocial and psychological interventions (28 trials, ~17,000 women) found meaningful preventive effect, strongest when support is postnatally timed and directed to women at elevated risk [14]. An Australian replication of telephone peer support (DAISY, N=1,060) is under way [15].
This is human-support evidence. Bloom's call cadence and conversational approach are modelled on these protocols — but Bloom is a support-and-connection tool informed by that evidence, not a tested replication of it. See "Where Bloom's own evidence stands" below.
The SUMMIT trial (N≈1,230, Nature Medicine 2024) found behavioural-activation therapy delivered by trained non-specialists — and by telemedicine — was non-inferior to specialist, in-person delivery for perinatal depression and anxiety [16]. Telehealth interventions meaningfully reduce EPDS scores in meta-analysis [17]. Scalable, remote, conversational support is a legitimate part of the care landscape — as support, alongside clinicians.
Australian perinatal screening coverage has risen dramatically — yet an estimated four in five perinatal parents needing mental-health support receive no professional care [18]. The national psychology workforce shortfall is put at 57% of demand, with roughly a third of psychologists closed to new clients and out-of-pocket gaps of $80–100+ per session [19]. Meanwhile universal health contact largely ends around eight weeks postpartum — just before the late-onset window opens [7]. Bloom doesn't add clinical capacity; it makes sure the capacity that exists is pointed at the right mothers at the right time, and that no one waits until a missed appointment to be noticed.
We think you should be suspicious of any company in this space that doesn't say that plainly. The human-delivered versions of what Bloom does — proactive calls, repeated screening, warm referral — carry strong evidence. Whether an AI voice achieves the same is an open research question, and we treat Bloom accordingly: as a screening-support and connection tool under ongoing evaluation, never as a proven treatment.
Two design consequences follow. First, every Bloom deployment is structured to measure, with partner services, whether it actually helps. Second — because independent testing has found that generic mental-health chatbots routinely fail people in crisis, with none of 29 tested bots responding adequately to suicidal messaging [20] — Bloom never relies on AI judgement in a crisis: any sign of serious distress is priority-flagged to a human on call, and the mother is guided to PANDA, Lifeline or 000 during the conversation itself. We also do not claim to "detect depression from voice" — voice-biomarker research is promising but unvalidated in perinatal populations [21].
In the UK's ALSPAC cohort, children of mothers with persistent, severe postnatal depression had markedly higher risks of behavioural problems at age 3½ and depression at 18 — while briefer, milder episodes showed no such effects [22]. The risk concentrates where depression goes unnoticed and untreated, which is the case for early identification. And at the sharpest end: suicide is among the leading causes of maternal death in the year after birth — clearest in UK national audits, and likely understated in Australian statistics, which mostly count only the first 42 days [23]. Maternal mental health is not a soft topic.
References
Statistics on this page describe population research, not Bloom's own outcomes. Bloom's effectiveness is under evaluation with partner services. Compiled June 2026; we review this page as new evidence lands.
If you or someone you know needs support now: PANDA 1300 726 306 · ForWhen 1300 242 322 · Lifeline 13 11 14 · In an emergency call 000.