Toko User Care Program · Clinical outcomes
The science behind Toko
Measuring what matters.
A longitudinal analysis of mental-health measurement across 197 participants and seven validated instruments — and what it means for the organizations that fund care.
01
Executive summary
Toko's User Care Program (UCP) measures progress with the same validated questionnaires used in clinical research — and the clearest story in the data is about the people who arrived in genuine distress. Among participants who started clinically elevated, the program delivered measurable, clinically meaningful relief — and the effect grew the longer people stayed engaged.
The analysis draws on 12,085 responses from 197 participants, measured on seven validated clinical instruments — the same questionnaires used in research (GAD-7, PHQ-9, PCL-5, UCLA-6 and three well-being scales). Each participant serves as their own baseline in a within-person pre/post design, with instruments re-administered about every three months; the 191 participants with two or more timepoints enter the results below.
4 in 10
16% → 29%
−24%
−22%
−26%
−32%
02
Why this matters to organizations
For the employers, health plans and partner organizations that fund mental-health support, the economic stakes are well established. The World Health Organization estimates that depression and anxiety alone cost the global economy roughly US$1 trillion a year in lost productivity.1 In the United States, depression costs employers an estimated US$187.8 billion annually across healthcare, absenteeism and reduced performance.2
The burden is social as well as clinical. More than half of U.S. workers report regular loneliness, and preventable workplace loneliness is estimated to cost employers US$154 billion a year; lonely employees miss roughly five additional workdays annually and are about twice as likely to intend to leave.3 That research measured loneliness with the UCLA Loneliness Scale — the same instrument Toko administers, allowing direct comparison.
The recurring problem is not a shortage of need but a shortage of accessible, measurable support that reaches people before distress becomes crisis. Toko was built to close that gap — and, just as importantly, to measure whether it does.
The rest of this page is that measurement.
US$1T
US$187.8B
US$154B
03
The program and how we measure it
The UCP delivers continuous, personalized emotional support through AI-guided, peer-to-peer conversations. What distinguishes it analytically is that progress is tracked with instruments widely validated in clinical and academic settings, administered repeatedly so that each participant serves as their own baseline. This within-person design provides the most direct test of whether participants' symptoms changed over the course of the program.
Instruments
Seven validated questionnaires were administered, spanning symptoms, distress, and positive psychological resources:
| Instrument | Domain | Direction of improvement |
|---|---|---|
| GAD-7 | Anxiety | Lower score |
| PHQ-9 | Depression | Lower score |
| PCL-5 | Post-traumatic stress | Lower score |
| UCLA-6 | Loneliness | Lower score |
| WEMWBS | Mental well-being | Higher score |
| CD-RISC-10 | Resilience | Higher score |
| SCS-SF | Self-compassion | Higher score |
GAD-7, Generalized Anxiety Disorder 7-item scale; PHQ-9, Patient Health Questionnaire-9; PCL-5, PTSD Checklist for DSM-5; UCLA-6, UCLA Loneliness Scale (6-item); WEMWBS, Warwick-Edinburgh Mental Well-being Scale; CD-RISC-10, Connor-Davidson Resilience Scale (10-item); SCS-SF, Self-Compassion Scale–Short Form.
Two further instruments were collected but are not analysed here: the PC-PTSD-5, a brief screener that overlaps with the more comprehensive PCL-5 used to assess trauma; and the ProQOL, which was administered to too few participants with repeat measurements to support reliable inference. Both are noted for completeness and excluded on methodological grounds — redundancy and statistical power — rather than results.
Analytic approach
For every participant and instrument, we compared the baseline assessment (first response) with the most recent follow-up. Outcome instruments are re-administered at roughly three-month intervals, so successive measurements are spaced about a quarter apart. Only participants with at least two timepoints on a given instrument contribute to that instrument's result, yielding 735 paired observations from 191 participants — the subset of the 197 with repeat measurements whose two-or-more timepoints fall on an instrument analysed here (the remaining six repeated only on the two instruments excluded above). For each instrument we report the average change, the share who improved, the standardized effect size (Cohen's d), and a paired-samples significance test.4 Throughout, “improvement” means movement in the clinically favourable direction — down for symptom scales, up for well-being scales.
The cohort
197
191
7
12,085
The population skews young-adult and female (73% of the cohort) and spans several delivery contexts — community programs, partner organizations, and clinically-referred groups. On some symptom scales the average baseline score was modest — mean baseline depression (PHQ-9 ≈ 6), for example, sat in the mild range — which constrains the room for measured improvement across the full sample.
This reflects the distribution of symptom severity, not an absence of need: a substantial minority arrived clinically elevated — 30% above the GAD-7 anxiety threshold and 31% above the PHQ-9 depression threshold — and it is precisely among these participants that the gains are largest and most clinically meaningful.
| Characteristic | Participants, n (%) | Mean age, yrs (SD) |
|---|---|---|
| Female | 144 (73.1%) | 34.3 (12.0) |
| Male | 52 (26.4%) | 30.4 (8.6) |
| Gender not reported | 1 (0.5%) | — |
| All participants | 197 (100%) | 33.1 (11.2) |
Demographic profile of the analytic cohort (n = 197). Age was recorded for 104 of 197 participants (53%); group means use available ages (73 female, 31 male). Overall median age 28, range 21–68.
04
Headline outcome: anxiety
Anxiety is where the program's effect is clearest. Across 111 participants with repeated GAD-7 measurements, scores fell from a mean of 7.2 to 5.6 — a 23% reduction in the favourable direction. The change is statistically significant at p < 0.001 with a medium effect size (Cohen's d = 0.40), and a majority of participants (58%) improved. Among all seven instruments, this combines the largest sample with one of the largest, most reliable effects.
Whole-sample improvement by instrument
PCL-5
PTSD · d=0.64 · p<.05
32%
GAD-7
Anxiety · n=111 · p<.001
23%
WEMWBS
Well-being · p<.05
4%
UCLA-6
Loneliness · p=.48
3%
PHQ-9
Depression · p=.88
0%
Read alone, these whole-sample averages understate the program's impact, because most participants began with little symptom burden to relieve. The picture sharpens when the same instruments are restricted to participants who began clinically elevated: the improvements are several times larger and uniformly statistically significant (Figure 1B).
Improvement among participants who began above the clinical threshold
PCL-5
PTSD severe · n=6
41%
GAD-7
Anxiety ≥10 · n=26
24%
PHQ-9
Depression ≥10 · n=23
22%
UCLA-6
Loneliness · p=.004
11%
Several factors likely contribute to the prominence of the anxiety effect at the whole-sample level. The GAD-7 is sensitive to change over relatively short intervals, and its core symptoms — worry, restlessness, and difficulty relaxing — are among the first to respond to structured, consistent psychosocial support. This interpretation is consistent with the engagement and clinical-severity patterns reported in the following sections.
05
A dose–response relationship
A genuine treatment effect should follow a dose–response relationship, in which greater exposure is associated with greater benefit. The data display this pattern. When participants are grouped by the number of measurement points they completed — a proxy for sustained engagement — improvement increases monotonically with engagement. Because Toko measures outcomes about every three months, four measurements corresponds to roughly a year of sustained engagement.
Improvement by number of completed measurements
Anxiety improvement climbs from 16% at two measurements to 27% at three and 29% at four. Well-being, which barely moves after a single follow-up, shows clear gains once participants reach three or more measurements; resilience (CD-RISC-10) and self-compassion (SCS-SF) follow the same pattern, strengthening most clearly once participants pass three or more check-ins. This monotonic gradient is difficult to attribute to chance, regression to the mean, or selection effects alone, and is consistent with a genuine treatment effect.
It also has an operational implication: the measurable value of the program accrues primarily once participants progress beyond the first few check-ins.
06
The greatest impact for those who need it most
Because average baseline scores were mild, the most informative analysis restricts attention to participants who began with clinically elevated symptoms — those for whom support is most indicated and for whom measurable improvement is possible. Within these subgroups, the program produces its strongest effects in the entire dataset.
Mean scores at baseline vs. follow-up, clinical-range participants
GAD-7
Anxiety · scale 0–21
≥1014.611.1PHQ-9
Depression · scale 0–27
≥1014.711.5PCL-5
PTSD · scale 0–80
≥3148.228.5
| Subgroup (baseline severity) | n | Baseline | Follow-up | Improvement | Reliable improvement |
|---|---|---|---|---|---|
| Anxiety, GAD-7 ≥ 10 | 26 | 14.6 | 11.1 | 24% | 42% |
| Depression, PHQ-9 ≥ 10 | 23 | 14.7 | 11.5 | 22% | 43% |
| PTSD severity, PCL-5 ≥ 31 | 6 | 48.2 | 28.5 | 41% | 67% |
Table 1. Outcomes among clinically elevated participants.
Among clinically anxious participants, anxiety fell 24% and 42% achieved a reliable improvement — a magnitude of change that is clinically meaningful, not merely statistical. Depression illustrates the same principle: flat across the full sample, it falls 22% once the analysis is restricted to participants who began clinically depressed. This is the central interpretive point of the paper: the program's effect is concentrated among participants with elevated baseline symptoms, and whole-sample averages therefore understate it.
07
Depression and loneliness: flat overall, strong where distress is real
Two domains warrant closer examination, because their whole-sample results are open to misinterpretation. Across the full cohort, depression (PHQ-9) and loneliness (UCLA-6) showed essentially no net change — depression moved 6.0 → 6.1 (p = .88) and loneliness 13.1 → 12.7 (p = .48). Considered in isolation, this could be read as an absence of effect.
It is more accurately characterized as a floor effect. Most participants entered the program with depression and loneliness scores already in the low, sub-clinical range, leaving little symptom burden to relieve and constraining any change in the mean. The relevant question is how these domains responded among the minority who did present with elevated symptoms.
Whole sample vs. clinically elevated at baseline
PHQ-9
Whole sample · p=.88
≥106.06.1PHQ-9
Clinical ≥10 · −22%
≥1014.711.5UCLA-6
Whole sample · p=.48
≥1213.112.7UCLA-6
Clinical · −11% · p=.004
≥1215.713.9
Among participants who began above the clinical thresholds, the program produced clear movement: loneliness fell 11% (15.7 → 13.9, p = .004) and depression fell 22% (14.7 → 11.5) among those clinically depressed at intake. The contrast with the flat whole-sample figures indicates that benefit is concentrated among participants with the greatest baseline need.
The most acute case: high depression and high loneliness together
We then isolated the participants facing both burdens at once — those who began clinically elevated on depression (PHQ-9 ≥ 10) and on loneliness (UCLA-6 ≥ 12) simultaneously. This co-occurring profile is common and clinically serious: 116 participants (about 28% of those measured on both) presented this way at baseline. Among the subset with repeat measurements, both domains improved substantially.
Participants who began high on both depression and loneliness
UCLA-6
Loneliness · −26% · p=.007
≥1216.712.4PHQ-9
Depression · −31% · p=.06
≥1014.410.0
| Measure (high on both at baseline) | n | Baseline | Follow-up | Improvement | p |
|---|---|---|---|---|---|
| Loneliness (UCLA-6) | 15 | 16.7 | 12.4 | −26% | .007 |
| Depression (PHQ-9) | 15 | 14.4 | 10.0 | −31% | .06 |
Table 2. The most acute subgroup — elevated on both depression and loneliness at baseline.
In this most acutely affected subgroup, loneliness fell 26% — a statistically significant change with a large effect size (p = .007, d = 0.82) — and depression fell 31%, moving the group mean from the “moderately severe” range (14.4) to the clinical threshold (10.0). The depression change did not reach conventional significance (p = .06), a result attributable to the small repeat-measure sample (n = 15) rather than to a weak effect.
The conclusion for these two domains is therefore twofold: among participants who began with low depression and loneliness, there was no significant change, consistent with minimal baseline symptom burden; among those who began in genuine distress, the improvement was substantial, and for loneliness it was statistically significant.
08
Trauma and post-traumatic stress
Among participants who completed the extended PCL-5, trauma severity fell from a mean of 32.2 to 22.0 — a 32% reduction with a large effect size (Cohen's d = 0.64, p < 0.05), and 71% improved. Within the most severe cases at baseline (PCL-5 ≥ 31), the average drop reached 41% and two-thirds achieved a reliable improvement.
These reductions are substantial and consistent with the broader pattern of larger gains among the most symptomatic participants. They are based on a small subgroup, however, and are therefore presented as a preliminary signal warranting focused case-study or expanded measurement rather than as a primary claim. The direction and magnitude of the effect support continued tracking as the trauma cohort grows.
Well-being, resilience and self-compassion
Beyond symptom reduction, the program aims to build positive psychological resources. Well-being on the WEMWBS rose significantly (47.6 → 49.4, p < 0.05), and resilience (CD-RISC-10) and self-compassion (SCS-SF) both trended upward. As Figure 2 shows, these resources strengthen most clearly with continued engagement — a reminder that building capacity, unlike relieving acute symptoms, tends to take more time.
Patterns across participants
Improvement was broad rather than confined to one group, but a few patterns are worth noting. Women, who make up the majority of the cohort, showed a particularly large anxiety reduction (29%, with 60% improving). Men showed comparatively larger gains in resilience. Participants supported through partner-organization contexts showed some of the strongest anxiety and well-being movements, though these subgroups are smaller.
We report these as directional observations; the cohort is not evenly balanced across groups, and the dataset is best read as evidence that benefit is widely distributed.
09
From outcomes to value
Clinical improvement is the goal; for a funding organization it is also a financial lever. Toko's observed outcomes map directly onto the well-documented economics of untreated anxiety. The model below is illustrative and intentionally conservative — a scenario, not a claims-based guarantee.
An illustrative model: 1,000 covered employees
| Step | Basis | Result |
|---|---|---|
| Covered population | Modeled cohort | 1,000 employees |
| Elevated anxiety at baseline | 30% above GAD-7 cutoff (observed) | ~297 employees |
| Expected responders | 46% move below cutoff (observed) | ~137 employees |
| Avoided cost per responder | Reduced absenteeism + presenteeism | ~US$4,785 |
| Estimated annual savings | Responders × avoided cost | ~US$0.65M |
Table 3. Illustrative annual productivity-cost savings for a 1,000-employee population.
The per-responder figure is anchored in published estimates that distressed and lonely employees lose roughly five additional workdays per year to absenteeism, with presenteeism costs typically larger still. Even before accounting for retention — where distressed employees are about twice as likely to intend to quit — the modeled return is substantial.
“Every US$1 invested in treating anxiety and depression returns an estimated US$4 in better health and productivity.”
Toko's observed responder rates fall squarely within the range underpinning that 4:1 return — with the added advantage of continuous measurement, so impact can be verified for each population rather than assumed.
10
Methodology notes and limitations
We apply the same rigor to the study's limitations as to its results, and note the following:
- 01
Single-arm design. This is a within-person pre/post analysis without a control group, so it cannot fully separate the program's effect from natural recovery or regression to the mean. The dose–response gradient and the concentration of effects among clinical cases both argue against those alternatives, but do not eliminate them.
- 02
Self-selection in follow-up. Results are based on the 191 participants (of the 197 with repeat measurements) whose repeats fall on an analysed instrument. People who stay engaged may differ from those who do not, which can bias estimates in either direction.
- 03
Floor effects. Low average baseline severity compresses whole-sample averages; subgroup analyses among clinically elevated participants give a truer picture of impact.
- 04
Small subgroups. The PCL-5 trauma findings and the "high-on-both" acute subgroup rest on few participants with repeat measures and should be treated as preliminary, directional signals.
- 05
Multiple comparisons. We report results across seven instruments and several severity and engagement subgroups without formal correction for multiple testing, so individual p-values close to 0.05 should be read with caution; the anxiety result is emphasised because it is large, dose-dependent and significant at p < 0.001.
- 06
Illustrative cost model. Savings figures are scenario estimates from published cost ranges, not a claims analysis of any specific population.
None of these caveats undercut the core finding. The most robust result — a significant, medium-sized, dose-dependent reduction in anxiety, amplified among those in clinical need — is exactly the pattern one would expect from an effective program, and does not depend on any single comparison.
11
Conclusion
Measured with the same validated instruments used in clinical research, Toko's User Care Program shows a clear and credible signal of benefit. Anxiety falls by roughly a quarter on average and by more among those who arrive anxious; the improvement grows with engagement; trauma severity drops sharply in those affected; and positive resources like well-being and resilience build over time. Where the whole-sample data appear flat — depression and loneliness — this is largely attributable to low baseline symptom levels, and even in these domains the analysis of participants in genuine clinical need reveals substantial, clinically meaningful change.
The evidence supports a clear and defensible conclusion: among individuals experiencing clinically meaningful psychological distress, the program is associated with measurable improvement that increases with continued engagement. As the cohort grows, more extensive measurement of the trauma and depression subgroups — and, ideally, a matched comparison group — will further refine these estimates.
Notes
- 1World Health Organization. Depression and anxiety disorders cost the global economy an estimated US$1 trillion per year in lost productivity.
- 2Greenberg PE, et al. Depression is estimated to cost U.S. employers roughly US$187.8 billion annually across healthcare, absenteeism, and lost productivity.
- 3Cigna / Ipsos Loneliness Index, measured with the UCLA Loneliness Scale. Preventable workplace loneliness is estimated at US$154 billion annually.
- 4Cohen's d for paired change (d≈0.2 small, 0.5 medium, 0.8 large). Significance from paired t-tests, corroborated with Wilcoxon signed-rank tests.