Healthcare Utilization, Chronic Condition Management, and Workplace Functioning Among Users of a Purpose-Built Mental Health AI (Ash): Cross-Sectional Study

• Comorbid mental and physical conditions are associated with worse disease management, higher healthcare utilization, and increased costs relative to either concern alone1, motivating behavioral-health support for people with chronic conditions9.Source: Introduction · p. 3
• Evidence on digital health cost savings is mixed, and downstream physical-health benefits from tools designed specifically for mental health remain untested.Source: Introduction · p. 4
• General-purpose models have documented safety limitations and can respond inadequately, particularly for users in crisis or at risk for delusional thinking17–19, whereas purpose-built mental-health models incorporate clinical appropriateness standards and safety guardrails20,21 and show preliminary efficacy for depression.Source: Introduction · p. 4
• The study compared active Ash users with non-users who had chronic conditions across five health domains and estimated associated cost savings.The domains were healthcare utilization, physical health and behaviors, chronic condition management, mental health, and workplace functioning.Source: Introduction · p. 4
• The authors frame the exploratory study and modest non-user comparison group as an initial effort to estimate association magnitudes and inform future adequately powered research.Source: Introduction · p. 4

Participants and Procedures

• The cross-sectional study recruited Ash registrants who enrolled six months before the survey and consented to research data sharing; active users had at least 6 sessions and non-users 1 or fewer over six months.Source: Participants and Procedures · p. 5
• In March 2026, participants were invited by email to complete a Tally questionnaire about healthcare utilization, physical health and health management, mental health, and workplace functioning.Source: Participants and Procedures · p. 5
• 5,879 active users were emailed; 461 (7.8%) opened the survey link and 267 (4.5% of those emailed) completed it.Source: Participants and Procedures · p. 5
• 15,270 non-users were emailed; 265 (1.7%) opened the survey link and 104 (0.7% of those emailed) completed it.Source: Participants and Procedures · p. 5
• Of survey participants, 65.2% (242 out of 371 total) reported at least one diagnosed chronic health condition, forming the final analytic sample of 169 active users and 73 non-users.Source: Participants and Procedures · p. 5
• The BRANY IRB deemed the study exempt under category 4(ii); respondents entered a $500 gift-card drawing, and non-users received an additional $10 completion incentive in reminder emails after low initial response.Source: Participants and Procedures · p. 5

Demographics

• Age was collected in six brackets (18–24 to 65+), alongside gender identity, health insurance type, and lifetime mental-health diagnosis.Source: Demographics · p. 6

Healthcare Utilization Measures

• Past-six-month primary-care, specialist, and psychiatrist visits used 0–4 scales; therapist/counselor visits used 0–5, and routine and emergency visits were dichotomized as any versus none.Source: Healthcare Utilization Measures · p. 6
• Urgent-care frequency was coded as repeat visits (two or more vs fewer than two) for main analyses; measures were self-reported, and repeat use was considered potentially indicative of an inadequately managed condition.Source: Healthcare Utilization Measures · p. 6

Health Behaviors and Physical Health Measures

• Five single-item measures covered sleep quality (fair or worse vs good or better) and weekly days of 30+-minute exercise (0–7).Source: Health Behaviors and Physical Health… · p. 6
• Among participants taking chronic-condition prescription medication, adherence was rated from rarely to always and coded as low (sometimes or less) or high (most of the time or more).Source: Health Behaviors and Physical Health… · p. 6
• Skipped or delayed care due to mental or emotional health was counted across five chronic-condition activities, including prescriptions, appointments, tests, exercise, and diet.Source: Health Behaviors and Physical Health… · p. 6
• Physical-health change versus six months earlier was rated from much worse to much better and coded as improved (somewhat better or better) versus not improved (about the same or worse).Source: Health Behaviors and Physical Health… · p. 7

Mental Health Measures

• Depression and anxiety were assessed with the PHQ-2 (23) and GAD-2 (24), respectively, using symptom frequency over the past 2 weeks and scores from 0–6.Higher scores indicated greater symptoms.Source: Mental Health Measures · p. 7
• Mental health improvement was defined as reporting “somewhat better” or more than 6 months earlier; daily activity impact was coded as moderate or greater versus a little or less.Both measures used five-point ordinal scales before dichotomization.Source: Mental Health Measures · p. 7

Workplace and Daily Functioning Measures

• Presenteeism due to mental or emotional health was measured over 6 months and dichotomized as monthly or more often versus a few days total or less.The item asked about significant reductions in work or school productivity.Source: Workplace and Daily Functioning Measures · p. 7
• Absenteeism was measured as missed work or school days due to mental or emotional health over 6 months and coded as monthly or more versus less than monthly.Source: Workplace and Daily Functioning Measures · p. 7

Statistical Analyses

• Because hypertension prevalence differed between active and non-users, primary analyses adjusted for hypertension; demographic and chronic-condition differences were tested with chi-square or two-proportion z-tests.Source: Statistical Analyses · p. 8
• Binary outcomes were modeled as adjusted risk differences with HC3-robust standard errors, while count and symptom-score outcomes used linear regression and adjusted mean differences.Source: Statistical Analyses · p. 8
• Ordinal logistic regression on full response scales tested whether conclusions depended on primary-model dichotomization, collapsing categories when needed to retain at least 10 responses each.Source: Statistical Analyses · p. 8

Study Participants and Descriptive Data

• 242 participants with at least one chronic condition included 169 active users (69.8%) and 73 non-users (30.2%).Source: Study Participants and Descriptive Data · p. 8
• 73.1% of participants were female, 29.8% were ages 35-44, 44.6% had Medicaid, Medicare, or ACA Marketplace insurance, and 68.6% had a lifetime mental health diagnosis.Source: Study Participants and Descriptive Data · p. 8
• Active and non-users did not differ significantly in age, gender, insurance type, or lifetime mental health diagnosis (all ps > .05).Source: Study Participants and Descriptive Data · pp. 8, 10
• 57.4% reported overweight or obesity (BMI ≥ 25), 40.9% chronic pain, 33.9% high blood pressure, and 24.0% type 2 diabetes or pre-diabetes.Source: Study Participants and Descriptive Data · p. 8
• High blood pressure/hypertension prevalence differed between groups (z = -2.45, p = .014), while no other chronic condition differed significantly (all ps > .24).Source: Study Participants and Descriptive Data · pp. 8, 10
• 38.0% reported one chronic condition, 25.2% reported two, and 36.8% reported three or more.Conditions were assessed with a multi-select item; participants reported an average of 2.32 conditions (range = 1-7).Source: Study Participants and Descriptive Data · pp. 9–10

Outcome Comparisons by Domain

• Past-6-month visits to primary care providers, specialists, therapists or counselors, and psychiatrists, as well as any emergency department visit, did not differ significantly between groups.For primary care, specialist, therapist or counselor, and psychiatrist visits, ps > .44.Source: Outcome Comparisons by Domain · p. 11
• 9.5% of active users versus 23.3% of non-users reported repeat urgent care visits (RD = -0.15).Comparisons were adjusted for hypertension status.Source: Outcome Comparisons by Domain · p. 11

Health Behaviors and Physical Health

• 36.9% vs. 23.3% reported improved physical health (RD = 0.14), and 91.7% vs. 76.4% reported high medication adherence (RD = 0.15).Sleep quality and weekly exercise frequency did not differ significantly between groups.Source: Health Behaviors and Physical Health · p. 11
• 1.11 vs. 1.55 chronic-condition care activities were skipped or delayed (b = -0.44), while 45.0% vs. 28.8% reported no affected activities (RD = 0.17).The difference was primarily driven by prescription refills: 26.0% of non-users vs. 12.4% of active users skipped or delayed them (RD = -0.14).Source: Health Behaviors and Physical Health · p. 11
• 2.01 vs. 2.76 was the mean PHQ-2 score among active users and non-users, respectively (b = -0.73).Active users reported significantly lower depressive symptoms over the past two weeks.Source: Health Behaviors and Physical Health · p. 12
• Compared with non-users, active users were more likely to report improved mental health over the past 6 months (61.4% vs. 34.3%; RD = 0.27, p < .001).The likelihood that mental or emotional health moderately or severely affected daily activities did not differ significantly between groups (p = .337).Source: Health Behaviors and Physical Health · p. 12
• 24.2% vs. 45.2% reported monthly-or-more absenteeism due to mental or emotional health (RD = -0.20), while presenteeism trended lower (54.3% vs. 65.8%; RD = -0.13).The absenteeism difference was significant; the presenteeism difference was not (p = .055).Source: Health Behaviors and Physical Health · p. 12

RD [95% CI] p

• Respondents could select multiple activities; adjusted risk differences compared active users with non-users, controlling for hypertension, and negative values indicate lower probabilities among active users.The measure asked whether participants had skipped or delayed activities due to mental or emotional health in the past six months.Source: RD [95% CI] p · p. 13

Sensitivity Analyses

• Absenteeism and medication adherence remained significant in sensitivity analyses.These analyses used ordinal logistic regression on the full response scale for variables that had been dichotomized.Source: Sensitivity Analyses · p. 14
• Routine-care variables remained non-significant in sensitivity analyses.The analyses used the full response scale for variables that had been dichotomized.Source: Sensitivity Analyses · p. 14
• The urgent-care finding attenuated to a strong trend in sensitivity analyses.The analyses used ordinal logistic regression on the full response scale for variables that had been dichotomized.Source: Sensitivity Analyses · p. 14

Discussion

• The study reports a first examination of healthcare utilization and chronic-condition management among users of a purpose-built mental health AI who were managing chronic conditions.Participants reported high chronic-condition prevalence (65.2%), but whether findings generalize to adults with chronic conditions who have not used Ash remains untested.Source: Discussion · pp. 14, 15, 18
• Over the past 6 months, 61.4% versus 34.3% reported improved mental health, while repeat urgent-care visits were 9.5% versus 23.3% for active users and non-users, respectively.Active and non-users did not differ detectably in routine healthcare use.Source: Discussion · p. 15
• 36.9% vs. 23.3% of active users and non-users, respectively, reported improved physical health; the study did not measure physiological indicators and cannot establish causality.The authors propose physiological and behavioral pathways, with improved chronic-condition management supporting the behavioral pathway; the physiological explanation remains speculative.Source: Discussion · pp. 15–16
• Active users reported fewer skipped or delayed chronic-condition care activities, were more than twice as likely to fill prescriptions on time, and had higher medication adherence.Weekly exercise frequency and sleep quality did not differ significantly, suggesting observed benefits may be more closely tied to chronic-condition management than broader lifestyle change.Source: Discussion · p. 16
• Active users reported lower absenteeism, corresponding to about 3.60 estimated missed days over 6 months and $1,017,798 in estimated annual savings per 1,000 employees; presenteeism differed only marginally.The authors note the savings are estimates and their interpretation of the absenteeism-presenteeism difference is speculative.Source: Discussion · pp. 16–17
• Self-selected, cross-sectional groups, low completion rates, self-reported six-month recall, possible healthy-adherer bias, and a small non-user subgroup limit causal and generalizable conclusions.The authors describe the study as estimating effect sizes to inform adequately powered future work and suggest prospective designs linked to claims data for stronger inference and validation.Source: Discussion · p. 17
• In a naturalistic sample, active users of the purpose-built mental health AI reported better outcomes across mental and physical health, acute-care utilization, chronic-condition management, and workplace functioning than registered non-active users.The comparison involved self-selected groups.Source: Discussion · p. 18
• The cross-sectional design and self-selected comparison groups prevent causal interpretation; the discussion identifies longitudinal, claims-based research as an important next step.Source: Discussion · p. 18
• Consistent differences across largely independent outcome domains suggest scalable, purpose-built tools such as Ash may support working-age adults managing co-occurring mental and physical health concerns.Source: Discussion · p. 18

Sensitivity analysis

• Table S1 reports a hypertension-adjusted ordinal-logistic sensitivity analysis using full ordinal response scales, with sparse upper categories collapsed to retain at least 10 observations each.Proportional odds ratios above 1 indicate higher odds of a more frequent or favorable response category among active users; primary adjusted risk-difference estimates appear in Table 3 of the main text.Source: Table S1 · p. 24, Supplemental Materials · p. 24

Estimated cost savings

• The annual savings estimates scaled regression differences to a hypothetical 1,000-employee workforce and should be interpreted as plausible ranges under stated assumptions, not causal effects.Assumptions included national rather than employer-specific benchmarks and mapping an ordinal absenteeism scale to estimated days; cross-sectional measurement, differential nonresponse, and absent baseline outcomes preclude causal interpretation.Source: Supplemental Materials · p. 25
• $56,572 in estimated annual urgent-care savings corresponded to 0.26 fewer visits per person over 6 months among active users than non-users.Active users averaged 0.50 visits versus 0.74 visits; the adjusted difference was 0.26 fewer visits (95% CI: 0.02–0.50), assuming an average visit cost of $164.Source: Supplemental Materials · p. 25
• $1,017,798 in estimated annual absenteeism savings corresponded to 3.16 fewer missed workdays per person over 6 months among adults aged 25 and older.The sensitivity analysis included 213 adults; the adjusted difference was 3.16 fewer days (95% CI: 0.96–5.36; p = .005), and the estimate applied a $247 median daily wage.Source: Supplemental Materials · p. 25