Mental health · September 2026
"If I stop answering anyone for five days, ask me why," Dana wrote in a good week, almost a year after her mother died. In November, she stopped answering.
It is 4:40 on a June morning and Dana is awake again, and this time the heat has nothing to do with it. She lies there and doesn't run through tomorrow's meetings, because she can't make herself care about them. That part is new.
Eight months ago her mother, Ruth, died at home under hospice, in her own bed, the way Ruth had planned it. The afternoon before, Ruth had been awake enough to talk about the house and Dana's brother and nothing much. Everyone told Dana she had done it right. She had slept in her old room at her mother's for most of those last three weeks and answered Ruth's app cards for her, and somewhere in there she let her own refill lapse.
She tells people she's tired. Nearly one woman in ten her age had major depression in 2024, as many as women ten years younger 1. Nobody has said that word to Dana, and she hasn't said it to herself.
Her mental health happens between the visits, at 4:40 in the morning and in the weeks she stops answering her phone. It is hers: her record, what she shares and when, and the words that come back to her on a bad day.
At Gale we are building care for women from 40 to 70, in the same app Dana has used since the hot flashes, made of ordinary visits that the insurance, HSA or savings she already has can pay for. Mental health sits beside the sleep log and the hot-flash count.
Health happens between the visits
I
In June Dana carries three layers of midlife, and researchers count each one on its own.
June, 243 days since Ruth died. Dana goes to work, does school pickup, answers her email, and feels nothing about any of it. She has started saying she's burned out, a word that makes it sound like the job.
The hot flashes came at 51, and Julie, her nurse practitioner, treated them. Ruth's diagnosis came in late August, when Dana was 53, and Ruth died in October.
One square per yearly interview with a woman in a long study of midlife. Before the menopause transition, 5.8 interviews in 100 found a major depression; during it, 9.1 2.
221 women at one US site. After the final period, 9.8 in this study 2; pooled studies find the raised risk clearest during the transition 3.
Three layers, each measured its own way. For Dana they landed inside three years.
The rise doesn't wait for the death. In Swedish national records of more than half a million women who lost a parent, diagnosed depression began climbing in the year before the death, during the dying 6.
A good death goes with better odds, and it is no guarantee. After a home hospice death like Ruth's, about 5 in 100 caregivers of people with cancer had grief that didn't ease, against about 22 in 100 after a hospital death 7.
We looked for a US study that counts the women carrying all three layers at once, and found none 8. Across those three years Dana got about the time with a doctor the average American gets, around an hour of exam time a year 910. Each clinician who saw Dana saw one layer.
II
The October Ruth died held three hard things, and the more land in one month, the more often a depression begins.
The October Ruth died was also budget month at work and her youngest's first term of high school. Dana went back to the ten o'clock meeting on the fourth day.
Piling up hard things over the years may not work the same way. A study that followed Australian women in midlife for decades found that however many stressful events they had been through, their depressive symptoms did not climb after the last one, and the women with the most events improved fastest 12.
Of every 100 Americans in their fifties, 44 have lost their mother, against 19 in 100 in their forties 13. Dana was one of the 44.
III
Dana says she's fine. Of adults over 50 with a mental illness who got no care, few thought they needed any.
Her brother calls on Sunday and asks how she's holding up. "Fine."
At her annual physical, on the waiting-room tablet, she marks "not at all" beside feeling down, because she doesn't want it in her chart. At 2 a.m. she types "I can't sleep and I don't feel anything" into a chatbot, reads a kind answer, and closes it. The mood app she downloaded in May hasn't been opened since the first week.
"I handled Mom," she tells herself. "I can handle this."
One square per American over 50 with a mental illness. 52 got some mental-health treatment that year 14.
Among adults who knew they needed care and went without, the top reason, given by 71.0%, was that they thought they should have been able to handle it on their own 15. The World Health Organization defines burnout, Dana's word, as a feature of work, and says it isn't a medical condition 16.
The typical mental-health app is still opened by about 3 in 100 of its users a month after download 17. Of adults who asked an AI for mental-health advice, 42 in 100 followed up with a professional 18. Dana didn't.
IV
In August Julie asks, and the people who could notice often don't.
On August 6 Dana had her routine menopause follow-up with Julie, by video, from the parked car. The hot flashes were quiet, but she kept waking at four. "I thought the patch was wearing off," she said, and asked to go up a step.
Julie said the patch was doing its job, and this sounded like something else. She asked the nine standard depression questions out loud and waited through the pauses. Dana's answers added up to 14, out of a possible 27, enough to take seriously. On the last question, about being better off dead, Dana said "not at all." Julie told her what they would have done if she'd answered otherwise: sat down together and written a safety plan, that same day.
Then Julie said the word nobody had said: this is depression, sitting on top of grief, and it is treatable.
Julie asked how much she drank. "A glass with dinner." It had been two since the winter. Julie asked her to keep it to one while she started a medicine.
Women 50 to 64 who binge drink: 84 in 100 were asked about alcohol at a visit, in person or on a form 20.
Women want the question: 76 in 100 say they want their doctor to ask about mental health as well as physical health 21. Asking early seems to matter. In a study of 86 caregivers of people who died of cancer, one thing stood out in who got mental-health care after the death: whether a health professional had talked with them about their own worries while the patient was still ill 22.
Nobody had asked Dana that. When Ruth was dying, Julie's office called about a missed visit, and Julie said she was sorry and to come back when she could.
I thought the patch was wearing off.Dana, to Julie
V
What happens to depression that nobody treats.
Ten minutes later, in the same visit, Dana asked whether it wouldn't just pass on its own. Julie said it did, for about half of people, and that nobody could tell in advance which half they were in.
One square per adult with depression, found in primary care, who got no treatment. By three months, 23 in 100 were well again 23.
Dana's depression had run since January, about 30 weeks. One study of untreated episodes put the median time to recovery at 23 weeks, and likely longer 24.
Treatment changes the odds. At about two months, 41 in 100 people responded to talk therapy, against 17 in 100 in usual care and 16 in 100 on a waiting list. Fewer got worse, 5 in 100 against 12 or 13 25.
The patch was unlikely to reach this. A patch-and-pill regimen like Dana's helped keep depressive symptoms away early in the menopause transition, and showed no clear benefit late in it or past it 26. Julie started a low-dose antidepressant that afternoon and wrote a referral.
VI
Dana gets one name and a time, and many people pointed toward care never get there.
The Saturday before, Dana had spent a morning on her insurer's list of therapists: two disconnected numbers, a full voicemail box, a practice that had left the network.
Julie's referral was one name, Maren Holt, a psychiatric nurse practitioner who does talk therapy, with a time the following Wednesday and the price, $250, on the screen before she booked.
In a trial of older adults referred to a specialty clinic, with help scheduling, getting there and paying: 49 in 100 engaged 29.
Dana took that first visit in her parked car at lunch. First Maren told her what goes where. Her record was hers to read, and nothing would go to Julie without Dana's yes. Maren would act without asking only if Dana were in danger, or someone else was, or the law required it. Then Maren asked about Ruth, and didn't look at the clock. At the end she said one thing about the medicine: "We look at the number together at six weeks. Not before."
In the Boston clinic, a first appointment soon went with showing up; a personal introduction from the doctor did not 27. The first weeks are where care loses people. About 22 in 100 people in mental-health treatment quit before their clinician thought they were done, and more than 70% of them left after the first or second visit 30.
After the second session, on day 12 of the pill, queasy and no better, Dana opened the app to cancel the third. She remembered the six weeks and left it booked.
VII
In one trial a care manager tracked people's scores between visits; in another, messages were the care.
Every other session Maren asked Julie's nine questions, out loud, and waited for the answers. The three of them had agreed in writing what the number would do: if it hadn't fallen five points by week six, Julie would raise the dose. Dana said yes to Maren sending Julie that one number.
It went 15, 13, 12. At week six, on September 17, Dana was two points below the 14 she had started at.
One square per older adult with depression in primary care. After a year of usual care, 19 in 100 had their symptoms cut in half 31.
The five points the three of them had agreed on is the smallest change on this questionnaire that researchers say matters to one person 33. When therapists were routinely shown their clients' scores, symptoms improved no more than without them, across 12 trials 34. When depression scores went to a clinician who used them to change the dose, people reached remission in a median of 4 weeks instead of 8, though by six months the difference in how many got there was no longer significant 35.
The day after week six, Julie raised the dose.
Your health is yours
VIII
In a good week Dana tells Maren about the wine, and surveys count what people keep from a clinician.
By the first week of October Dana was at 9, and it was a good week. She slept six and a half hours three nights running, laughed at something at work, and noticed herself laughing.
At that week's session she told Maren the wine had been two glasses, sometimes three, since the winter. They talked about it.
Of all the times people held something back, about half came down, at least in part, to one reason: they didn't want it in their medical record 36. It was Dana's reason on the waiting-room tablet, and the worry is earned. Federal regulators said one online therapy company passed people's intake answers to advertisers; it paid $7.8 million to settle 38.
Because it was a good week, Maren said this was the week to plan for the bad ones. They wrote two sentences, in Dana's words. The first was "If I sleep under six hours for a week, check on me." The second she said out loud before she wrote it: "If I stop answering anyone for five days, ask me why." Each listens only to what she chose for it: the first to her sleep, the second to the one tap Dana makes in Gale when she has talked to someone, a habit she had kept most days since August.
For the second sentence she named Lena, the friend who once gave her Julie's name, wrote a line for Lena, and decided that if Dana ever waved the reminder off twice, Lena would get it. Then they planned the anniversary week, day by day.
IX
The year after Ruth died, one day at a time, with what was offered and when Dana's care began.
Her brother flew in on October 12. On the 13th, looking for Ruth's death certificate, Dana found the hospice folder in the drawer with the insurance cards. On the first evening of hospice the social worker had told her the family could have support "for up to 13 months." Dana had never called. She put the number in her phone.
On the 14th she and her brother drove to the lake Ruth liked and ate Ruth's terrible egg-salad sandwiches in the car. Lena came on the 15th. Maren's slot was the 16th, booked back in the good week.
Dana was sad and she was fine, and she noticed that both could be true on the same day.
One square per day after Ruth died. The three lit squares are the paid leave most employers give for a parent's death 39.
Medicare requires a hospice to make bereavement support available to the family for up to a year after a death 40.
The marks are Dana's: June, the day Julie named it, her first visit with Maren, and the agreement. Her care began on day 296.
Dana's dates are invented; she is a composite.
The anniversary sits on the calendar a year ahead. The year of support Medicare requires ends the same week 40.
In Swedish national records, more women who had lost a parent died by suicide in the days around the anniversary, including women bereaved in their fifties and early sixties 42. The date is known ahead of time, so the week can be planned. If you are thinking of ending your life, call or text 988.
The bad week that sat on no calendar came three weeks later.
X
Plans made ahead and contact after a crisis helped in some studies, and in two a named clinician held the plan or made the calls.
At the start of November Dana stopped answering texts. She didn't decide to; she just didn't, for five days: the family group chat, Lena, her brother's Sunday call. With no one to mark, the tap in Gale went unmade too.
On November 6 her lock screen showed one line: A note from the day you sat down with Maren. Open Gale. In the hospice weeks she had answered Ruth's cards for her.
She opened it, and her own sentence filled the screen. She read it twice. She wasn't ready to say it to anyone, not even Maren. She tapped Not now. Nothing else happened, which is what Not now is for.
Each row is a study. Left of the line means fewer bad outcomes; a bar crossing it means the study couldn't tell. After an emergency visit, a safety plan plus follow-up calls went with about half the suicidal behavior 43; pooled safety-plan studies agreed 44; caring texts cut attempts 45.
The first was a comparison, not a trial; the texts missed their main goals, and attempts were a secondary result 4345.
A crisis plan written with the treating team while a person was well halved the number admitted to hospital against their will, in a small trial 46.
At scale it made no clear difference; its authors said the plans weren't fully put in place at every site, and the review meetings they were folded into didn't act on what patients had asked for 46.
These studies followed people after a crisis or with severe mental illness, not women like Dana 4346.
A message can bring someone back to care. In a small US trial, when a clinic reached out by message or phone to people newly diagnosed with depression who hadn't started treatment, 24 in 100 started within 60 days, against 6 in 100 with usual care 48.
Dana's card does that much and no more: Go books Maren. After a Not now it waits 30 days before it asks again 49. The lock screen never says why it came.
XI
In December Dana's line reaches Lena, and women say they would turn to a friend more often than to a professional.
By December 11, two weeks before Christmas, Dana had gone quiet again for five days, and the card came back, five weeks after the first. She tapped Not now a second time.
In October she had decided what a second Not now would mean. Lena's phone got one line, the one Dana had written for her, and nothing else: I am having one of the weeks. Come and sit with me.
Lena's line is the one time a Not now reaches anyone.
People go to the people they have. One of Dana's was her mother, and she is gone.
Lena came on Saturday with two coffees, sat at the kitchen table, and asked her why. Dana told her. It took most of the second coffee. With Lena sitting there, Dana booked Maren. Maren saw her on Monday.
The coverage you already have pays for it
XII
What one therapy session costs, by who pays.
On January 1 Dana's deductible reset, and she used her HSA. She had taken most Thursday afternoons off work.
She had seen every price before she booked, and for the first time she added it up: a $250 first visit and fifteen follow-ups at $150. That came to $2,500.
About a third of private-practice therapists take no insurance. Cash averages $143 a session, and $196 for a psychologist who takes none 52.
At 65, Medicare lists $167 as its 2026 fee for an hour of therapy; her share after the deductible comes to about $33 53.
Our plan: a first visit costs $250 and a follow-up $150, shown before you book. A visit the app offers costs $150, $75 or $25, by income.
A Gale visit costs about what therapy costs, and it isn't the cheapest option. If your practice offers collaborative care, where a primary-care team with a care manager checks in between visits, take it: patients paid a median of about $8 a month 54.
Since 2025 a plan that comes with an HSA can cover video therapy before the deductible 55. It's worth asking yours.
The family leave Dana took in Ruth's last weeks also covers a mental-health condition of her own that needs ongoing treatment, the Labor Department says, including time off for regular appointments 56. It kept her Thursday afternoons.
XIII
In March Dana's brother works down his insurer's list, and most callers testing such lists got no booking.
On a Sunday in March her brother didn't ask how she was holding up. He said he hadn't slept properly since Christmas. His doctor had told him to see someone and pointed him to the insurer's website. He had called six names.
Dana stayed on the phone while he tried two more.
Callers to 360 listed psychiatrists booked 93. Saying they had private insurance, had Medicare, or would pay cash made no difference 59.
Many names on the lists see none of the plan's patients: in Medicare Advantage, 55% of the mental-health clinicians listed saw no one from the plan all year 60. For a new psychiatrist, the median wait across five states ran 67 days in person and 43 by video 61.
What Dana had was a name from a clinician who knew her, a price on the page, and a visit within the week. We are building Gale to offer that every time.
She told her brother what had worked: ask your doctor for one name and a time, not a list. He called his doctor's office on Monday. They found him someone for Thursday.
XIV
In April Dana asks about stopping, and a UK trial followed people who felt well enough to stop.
By April Dana had slept past five most mornings for a month, and her score had stayed under 5 since late January.
She asked Julie about stopping the antidepressant. Julie showed her what happened to people her age, most of them women, who felt well enough to stop.
Pooled across trials, 18 in 100 who stayed on after responding to treatment relapsed 63.
Dana agreed with Julie to wait until she had been well six months and past the second anniversary, then taper slowly in November with Maren alongside.
People who stop an antidepressant on their own most often say they felt better, 46.6% 64. Stopping doesn't have to be done alone. One review of pooled trials found that talk therapy given during the taper went with no more relapse than staying on the drug 65. About half of first episodes never returned over 23 years in one US city; a third recurred, and about one in seven never had a clear year 66.
At her next session with Maren, Dana added one sentence to her agreement, borrowed from the plan Julie wrote for her three years before, when the trouble was hot flashes: If it comes back after we taper, tell me. It listens to her sleep.
In May, Lena's father went into hospice. Dana called her and asked how she was sleeping, and who was looking after her.
What Gale does
Between the visits: what you see
In a good week, in a visit with a mental-health clinician, you write an agreement about the weeks that won't be good. Each sentence is yours, in your words, and each one listens only to what you chose for it: your sleep, whether you've been in touch with people, your medicine, how often you open the app, and a few others. Being in touch means the connections you mark yourself, with one tap; the app never sees your texts or your calls.
When one of your sentences comes true, your lock screen shows one line: A note from the day you sat down with Maren. Open Gale. Open it, and your own sentence fills the screen:
GALE · NOW
"If I stop answering anyone for five days, ask me why."
This message is from Gale's software. The words are yours.
Go books your clinician at a price you see first, and we're building it so you're seen within seven days. The app sends at most two of these in any 30 days, at least ten days apart, never between 9 p.m. and 9 a.m., and after a Not now it waits 30 days 49.
Some things don't wait for a card. Gale is not a crisis line. If what you type matches one of the crisis phrases the app watches for, it stops before any AI sees it and puts 988 and 911 in front of you, and nothing is saved. A yes to the last question on the depression questionnaire brings up the 988 screen at once.
Yours: what stays yours
The app is yours. What you log stays on your phone unless you send it. What you type to it has your name and identifying details removed on your phone before it goes anywhere.
Not now means not now: no clinician sees it on any screen. The one exception is the one you write. You can name up to four people on a sentence and write each of them a line; if you choose, a second Not now sends that line and nothing else.
A questionnaire is scored the way it would be on paper, never by AI. Your answers stay on your phone unless you check two boxes to save them, and sharing a result with a clinician will take its own permission, one visit at a time. Our AI model reads your record only if you turn that on. The app never names a diagnosis, never calls a number abnormal, never suggests a drug, and never acts as your therapist.
Yours: what Maren and Julie see
A booking, with a short brief you approved item by item. No stream of your sleep, no count of what you left out, and not your Not nows. They go through the brief with you at the visit.
Already paid for: what it costs
There is no new kind of plan to buy. Visits are ordinary visits, which your insurance, your HSA or FSA, cash, and at 65 Medicare can pay for. Today you pay by card, HSA or FSA and get a receipt to send your insurer; we are building the billing that sends it for you. Paying yourself, a first visit costs $250 and a follow-up $150, shown before you book. A visit the app offers you costs $150, $75 or $25 by income. If we miss the seven-day window on a visit you pay for yourself, it's free. The app carries no fee.
How we'll know it works
Nobody has yet tested this as one package: an agreement the patient writes, a signal from her own phone, and one clinician who answers 67. We have to prove it. No woman has used this yet. When women do, we'll count the days from a first card to a first visit, and how many are still in care at twelve weeks, against usual care, and publish the result either way.
For clinicians
You know her. Midlife, a parent recently dead or dying, a job, children, and a short answer to the screening question. Julie heard "I thought the patch was wearing off" and asked the nine questions out loud anyway. She may not bring it to you: nearly three in ten adults over 50 who have been depressed have kept it from a clinician 36, and primary-care doctors recognize about half of the depression that walks in 19.
If you're her nurse practitioner or her primary-care doctor, the prescription is probably yours: about three quarters of antidepressant and anti-anxiety fills for people with employer coverage are written in primary care 68. "I'm already on hormones" doesn't cover her; in older women past menopause, an estradiol patch did no better than placebo for depression 69. Ask during the caregiving, not only after the death 22. Ask about the wine, and say something when the answer matters.
If you're taking the referral, offer a first appointment soon. In one rollout, fewer than half of 1,999 referred patients reached an assessment, and 307 finished treatment 70. Measure, and say in advance what the number will do.
Tell her early what would make you act without asking, the way Maren did. One study found that most therapy clients who hid thoughts of suicide feared practical consequences, such as being hospitalized, more than shame 71.
The agreement does not replace a safety plan. If she answers yes to the last question, or tells you she is in danger, she needs a standard safety plan with you, that day. What the agreement asks of you is smaller: sit with her in a good week and help her write, in her words, what should happen in a bad one. You'll get bookings with a brief she chose, not a stream at midnight; a primary-care doctor's inbox already takes about 49 messages a day 72. And when she comes back in a bad week, she comes back because of a sentence she wrote with you.
SAMHSA's 2024 National Survey on Drug Use and Health: past-year major depressive episode in 9.6% of women aged 50–54 and 10.2% of women aged 40–44; the rate drops after 55 (5.9% at 55–59) 1.
As in part one: 3.207 office visits per person per year in 2019 9 × 18.0 minutes per primary-care exam, from one vendor's records in 2017 10 = 57.7 minutes; direct observation gives 13.3 minutes a visit 73 and physicians' own reports 23.5 74, or 43 and 75 minutes a year. The visit count leaves out hospital clinics, emergency departments and independent nurse practitioners, so it is a floor, and an average across a skewed distribution, not the typical person's experience 9.
Bromberger and colleagues, SWAN Pittsburgh, 221 women with yearly structured interviews: a major depressive episode at 5.8% of premenopausal visits, 9.1% of perimenopausal and 9.8% of postmenopausal ones 2. Across cohorts and a 2024 meta-analysis, the raised risk runs through perimenopause, and the meta-analysis found no significant raise after menopause overall 3. CDC's Behavioral Risk Factor Surveillance System caregiver module, 2021–22, 35 states and Puerto Rico, age-adjusted: frequent mental distress in 20.5% of caregivers and 13.6% of non-caregivers; 59.2% of caregivers were women 4. Malhotra and colleagues, 2025, pooled 49 studies of bereaved family caregivers: depression 27%, prolonged grief disorder 15%. The two overlap, so each gets its own grid 5. Each pair comes from a different study with a different measure; the chart places them side by side and adds nothing up.
Kendler and colleagues, Swedish national registers, 557,970 women who lost a parent: diagnosed major depression rose in the year before the death and declined slowly after it 6. Wright and colleagues, 2010, caregivers of people with cancer: prolonged grief disorder in 5.2% after a home hospice death and 21.6% after a hospital death (8 of 37), adjusted odds ratio 8.83. Home hospice was the reference group, and the study does not show home hospice as the lowest-risk setting 7. The search for women carrying all three layers: PubMed and Europe PMC searches returned about a dozen hits and no US source that counts women who are caregiving during the menopause transition or bereaved soon after it. The closest, the Women Living Better online survey of 2,406 women, measured how much midlife symptoms bothered caregivers and reported no count of them 8.
Kendler and colleagues, 1998, US female twins, 24,648 person-months: the chance that a major depression began in a month was 0.9% with no stressful event, 3.4% with one, 6.8% with two and 23.8% with three. One square is one month, not one woman 11. The same group's 1999 paper estimated that about a third of the association is non-causal, reflecting people prone to depression landing in harder months, as they reported it 75. The lifetime-total null is from the Australian Longitudinal Study on Women's Health, as its authors reported it: no group showed a rise in depressive symptoms after its last stressful event 12. Census Bureau data for 2022, both sexes: 44.3% of people in their fifties had lost their mother, against 18.9% in their forties 13.
SAMHSA, 2024 NSDUH, adults 50 and over with any mental illness: 52.0% received any mental-health treatment; 8.0% of the untreated perceived an unmet need. 48 × 0.080 = 3.8, drawn as 4; 48 − 4 = 44. Our arithmetic 14. The 2025 tables use a new estimation model and are not mixed in 76. The 71.0% who said they should handle it on their own is from the same survey, adults of all ages with any mental illness and an unmet need; SAMHSA names cost second, at 65.2%, and notes that the reasons after those two are not a ranking 15. The WHO text on burn-out does not mention caregiving or grief; applying it to Dana is our reading 16. The 42 in 100 who followed up with a professional after asking an AI is from KFF's 2026 tracking poll 18. The median 30-day retention of 93 popular mental-health apps was 3.3% 17.
The nine-question depression questionnaire (PHQ-9) scores each answer 0 to 3, for a total of 0 to 27. A cut-off of 10 gives 0.88 sensitivity and 0.85 specificity for major depression, and the smallest change that matters to one person is 5 points, as reported 33. Mitchell and colleagues, Lancet 2009, a meta-analysis of 41 studies and 50,371 patients: general practitioners recognized depression in 47.3% and recorded it in 33.6%. They are reported as separate figures, so they get separate grids. Accuracy was better when doctors reassessed over months 19. The alcohol figures are Gale's own weighted analysis of the NSDUH public-use files for 2015, 2017 and 2019, women 50–64 who binge drank and had a past-year visit: 84.0% asked (95% CI 81.7–86.2), 5.2% advised to cut down (3.7–6.8); men the same age 10.0% 20. Gallup and West Health, 2024: 76% of women want their doctor to ask about mental health as well as physical health 21. Lichtenthal and colleagues, 2011, 86 bereaved cancer caregivers, observational: talking with a professional about psychological concerns during the illness was the only significant predictor of later care after adjustment; several other variables were significant before it. It is a predictor, not a tested intervention 22.
Whiteford and colleagues, 2013: a model estimate, fitted to primary-care and wait-list studies, of remission in untreated adults recruited in primary care; wait-listed samples remitted less. A population figure, not anyone's personal odds 23. The median of at least 23 weeks is Posternak and colleagues, 2006, 130 untreated recurrences, with each episode counted only until treatment began; the authors call 23 weeks a lower limit. The 84 people who never got treatment recovered faster, a median of 13 weeks. "Untreated" here means no somatic therapy 24. Cuijpers and colleagues, 2021: response 41% with psychotherapy, 17% usual care, 16% wait-list; deterioration 5% against 12–13% 25. Gordon and colleagues, 2018: a year of estradiol patch plus progesterone prevented depressive symptoms early in the transition and showed no significant benefit late in it or after it 26. Morrison and colleagues, 2004, cited in the clinician section: in older postmenopausal women who were already depressed, depression scores fell 40% on an estradiol patch and 44% on placebo 69. Dana's "since January" and "about 30 weeks" are invented, and the study's 23 weeks describes other people's episodes, not hers.
Pace and colleagues, 2018, one Boston safety-net program, observational: 1,087 of 2,690 scheduled patients (40%) attended; warm handoff adjusted odds ratio 0.96 (0.79–1.18); a sooner appointment predicted attendance 27. Waitzfelder and colleagues, 2018, five health systems: 35.7% started treatment within 90 days, 53% with moderate or worse symptoms, half the odds at 60 and over 28. PRISM-E, 2004, randomized, older adults, mean age 73.5, 26% women: 71% engaged in integrated care against 49% with enhanced referral 29. Four populations, never chained into one funnel. Olfson and colleagues, 2009: 22% dropout, more than 70% of it after the first or second visit 30.
IMPACT, 2002, adults 60 and over: symptoms halved at 12 months in 45% with collaborative care and 19% with usual care, odds ratio 3.45 (2.71–4.38) 31. Simon and colleagues, JAMA 2022, 18,882 outpatients: self-harm over 18 months 3.27% with care management, 3.27% with usual care and 3.92% with online skills training; care management hazard ratio 1.07 (97.5% CI 0.84–1.37), skills training 1.29 (1.02–1.64); 31% ever engaged with care management 32. The two trials measured different outcomes and share a chart, not a comparison. Remission in 4 weeks against 8. At week 24, 73% against 63% had reached remission, a difference the trial could not separate from chance (P = .20) 35. No clear symptom benefit from routine score feedback to therapists: 12 trials, 3,696 participants, SMD -0.07 (-0.16 to 0.01), low-quality evidence; the two trials that added a decision tool did no better 34. The five-point minimum change on the PHQ-9 is as its researchers reported it 33. Dana's scores, 14, 15, 13, 12, 9 and under 5, are invented.
Levy and colleagues, JAMA Network Open 2019, a national online panel of adults 50 and over (one of two surveys in the study), among those who had been depressed: 29.0% had withheld it from a clinician; women were more likely to withhold at least one of the four experiences asked about (odds ratio 1.33) 36. The 52.7% who cited not wanting it in their medical record is a share of all nondisclosures, not of people 36. SAMHSA, 2024 NSDUH detailed tables, adults with an unmet need: privacy worry 18.3% for mental health and 31.7% for substance use; fear of losing a job, home or children 9.8% and 33.0% 37. The FTC charged that BetterHelp shared users' health-questionnaire answers with advertising platforms; the company settled for $7.8 million 38.
SHRM, 2016: three days of bereavement leave for a parent 39. 42 CFR 418.64(d)(1)(ii) requires a hospice to make bereavement services available "up to 1 year following the death." The social worker's "13 months" is what hospices commonly say; the rule says up to a year 40. Cherlin and colleagues, 2007, 161 family caregivers of hospice patients with cancer: about 30% used bereavement services in the year after the death, 47.6% of those with major depression 41.
Grotta and colleagues, JAMA Network Open 2023, Swedish registers 1990–2016, a case-crossover design. Women who lost a parent: odds ratio 1.67 (1.07–2.62) for suicide from the anniversary to two days after. Women who lost a mother: 2.29 (1.20–4.40). Women bereaved at 50–65: 2.53 (1.04–6.15), from the day before up to the anniversary. The absolute risk stays small. These figures stay out of the body on purpose 42. If you are thinking of ending your life, call or text 988.
Stanley and colleagues, nine VA emergency departments, a non-randomized comparison of 1,640 people, 88% men: odds ratio 0.56 (0.33–0.95) for suicidal behavior over six months 43. Nuij and colleagues, pooled safety-planning studies, relative risk 0.57 (0.41–0.80) 44. Comtois and colleagues, caring texts to service members: primary outcomes null; attempts, a secondary outcome, odds ratio 0.52 (0.29–0.92) 45. Henderson and colleagues, 2004, joint crisis plans, 160 people: compulsory admission 13% against 27%, relative risk 0.48 (0.24–0.95). CRIMSON, 2013, 569 people: 18% against 20%, odds ratio 0.90 (0.58–1.39); its authors said the plans were not fully implemented at every site and were folded into routine review meetings that did not act on patients' stated preferences 46. The marks are our own coding of who held the plan or made the calls, and no trial randomized who held it. CRIMSON is coded no, which is contestable, and nobody else has checked the coding. The pooled advance-statement estimate (de Jong 2016, relative risk 0.77, about a quarter fewer compulsory admissions) is not drawn, because its trials include both plotted crisis-plan trials; whether a plan must reach the decision to work remains a hypothesis 47. Simiola and colleagues, 2026, a pilot randomized trial of 309 Asian, Native Hawaiian or Pacific Islander, Black and Hispanic adults with a new depression diagnosis and no treatment by day 30: 24.2% started treatment within 60 days with outreach, against 5.9% 48. No study has tested the package 67. The card's 30-day wait is in Gale's code 49.
Pew Research Center, "Men, Women and Social Connections," January 2025, share of women extremely or very likely to turn to each for emotional support. The spouse or partner, mother and other-family shares count only women to whom the question applied, such as those with a partner or a living mother; the friend, professional and online shares count all women 50. The 14% and 24% are adults 50 and over and under 50, both sexes 50.
The private-coverage prices are Gale's own analysis of the 2023 Medical Expenditure Panel Survey microdata, adults 18–64 with private coverage: when the plan paid something (65.9% of visits), the patient's median share was $20; when it paid nothing (34.1%), the median payment was $130. These are per therapy visit, of any length. The categories are "plan paid something" and "plan paid nothing"; the data don't show network status or why the plan paid nothing 51. Zhu and colleagues, 2024: 35.1% of private-practice therapists take no insurance; average cash rate $143 a typical session, and $196 for psychologists who take none 52. CMS's 2026 national fee for a 60-minute psychotherapy session, CPT 90837, is $167.00 as listed; the $33 is 20% of it after the Part B deductible, our arithmetic 53. Cliff and colleagues, 2024, from Colorado all-payer claims for 2018–2019: median out-of-pocket for collaborative care $8.35 a month (commercial coverage $15, Medicare Advantage $0); among the 193 clinicians who billed for it, a mean of 12% of eligible patients received it 54.
The 2025 federal law permanently lets high-deductible plans cover telehealth, including remote therapy, before the deductible without costing the member HSA eligibility; it permits this and does not require it (IRS Notice 2026-5) 55. IRS Publication 502 lists psychiatric care and psychologists' fees as medical expenses, as reported 77. Hansen and Lambert, 2003, 4,761 patients: a 50% recovery rate was reached between 15 and 19 sessions 78. FMLA covers an employee's own serious mental-health condition, including intermittent leave for scheduled appointments, as the Labor Department's Fact Sheet 28O explains it; it does not cover bereavement itself 56. Gale's prices ($250, $150, and $150, $75 or $25 by income), the seven-day window and the free visit if we miss it are our plan's.
US Senate Finance Committee majority staff, 2023: 120 calls, 22 appointments, 39 wrong, dead or unreturned; 59 = 120 − 22 − 39, our subtraction 57. New York Attorney General, 2023: 396 listed clinicians across 13 plans, 56 appointments, 93 wrong, dead or unreturned; two-thirds of calls used the depressed adult relative's script and one-third the parent of a 14-year-old; 247 = 396 − 93 − 56, our subtraction 58. Malowney and colleagues, 2015: 93 appointments with 360 listed psychiatrists after two rounds, no difference by payer 59. HHS inspector general, 2025: on average across the 40 Medicare Advantage plans studied, 55% of listed behavioral-health clinicians saw no enrollees in 2023 60. Sun and colleagues, 2023, five states: median waits of 67 days in person and 43 by video 61.
ANTLER, New England Journal of Medicine 2021, 478 patients in 150 UK practices, average age 54, 73% women: relapse within a year 39% staying on against 56% tapering, hazard ratio 2.06 62. To join, a patient needed at least two past episodes or two or more years on an antidepressant, which Dana does not have, so her risk is likely lower than this chart shows 62. Geddes and colleagues, 2003, pooled randomized trials: 18% against 41% 63. Kazdin and colleagues, 2024: of people who stopped without their prescriber's advice, 46.6% felt better; the 5.0% is "enabling factors" such as cost 64. Breedvelt and colleagues, 2021: psychological therapy during the taper, as reported 65. Eaton and colleagues, 2008, first episodes followed for 23 years in one US city: about half never returned, 35% recurred, about 15% never had a year free 66. Julie's sentence in part one read "If they come back after we taper, tell me"; Dana's version changes only "they" to "it."
The clinician section draws on 36, 19, 68 (Health Care Cost Institute: about three quarters of antidepressant and anxiolytic fills for employer-insured Americans come from primary-care providers), 69 (Morrison 2004, older postmenopausal women already depressed), 22, 70 (Carroll 2026: 1,999 referred, 986 assessed, 307 completed), 35, 34, 71 (Blanchard and Farber 2020, as reported: 70% of those who concealed cited fear of practical consequences outside therapy) and 72 (49.3 inbox messages a day per primary-care physician).
All of them are composites, not real people. Invented: every date in Dana's year and the day counts (243, 296), 4:40 a.m., six and a half hours, the questionnaire scores (14; 15, 13, 12; 9; under 5), "since January" and "about 30 weeks," "a glass" and "two, sometimes three," day 12 of the pill, the six days to the first visit, the five quiet days, the five weeks between cards, the visit count and bills (16 visits: $250 + 15 × $150 = $2,500), the Thursday afternoons, the two coffees, the brother's six names and two more calls, the lake and the sandwiches, "past five for a month," every line anyone speaks, and every other count in Dana's story. Dana's five-day sentence and Lena's line are the sample sentences in our app's own code, a test fixture no real woman wrote; the fixture's other rule is sleep under six hours for a week, and Dana's wording of it, "check on me," is ours 79. In the fixture the named person's sentence sits on the sleep clause; we moved it to the five-day clause for Dana's story. Dana's later sentence, "If it comes back after we taper, tell me," adapts Julie's line from part one. The one-tap habit of marking who she talked to is invented; the tap itself is the app's. Maren is a psychiatric nurse practitioner because our code bills agreement visits as office visits, which fits a nurse practitioner better than a psychologist 80.
Built, in Gale's code: the agreement's data sources, with a trigger on typed phrases rejected by design 81. The being-in-touch source counts the connections she marks herself in the app, with no contact names, and is checked on the phone; the app has no access to her texts or calls. That sentence rests on a direct read of the phone app's code on September 22, 2026: the source reads only the connections she logs, each a circle (family, a friend, community) and a mode (in person, a call, messages), keeps no contact names, and labels the brief "Someone logged" or "Nobody logged". The one lock-screen line 82; the send to a named person after a second Not now 83; the card limits and quiet hours 49; fixed-rule questionnaire scoring 84; the two save consents and the separate share permission, which today exists on the web 85; the 92-pattern crisis net that runs before any AI model 86; and the privacy switches changed on 2026-09-22, with AI data sharing now off by default and no withheld count in the clinician brief 87.
Not built, or not yet true. No sentence can fire on mood, and there is no mood tracker 88. A questionnaire taken on the phone can't reach the chart yet, so the share permission has no phone-to-chart path today, and in this story the clinicians ask the questions in the visit 89. Insurance billing is switched off: patients pay by card, HSA or FSA and can get a superbill 90. The therapist directory is a demonstration of 30 synthetic profiles, none in network 91. A crisis message sent to a clinician through the web thread reaches no one at the practice today 92. The crisis net lets through phrases like "I can't do this anymore" and "I've been drinking more since my mom died" 86. Not now events are stored where a clinician's database role could read them, though no screen shows them 93. The behavioral-health consent form is a draft not yet reviewed by counsel, and Gale has no store for psychotherapy notes 94. One clause type, a "detector" a clinician names, can fire on flags computed from a supervised chat, so the accurate claim is that each sentence listens to the sources she chose and no sentence matches her words directly 81. The card's footer line is our plan's wording. An in-practice referral with a guaranteed time, and the seven-day window, are our plan; today the directory is a demonstration.
These are being fixed. No real woman has used any of it: Gale runs on synthetic data, and the switch for contacting a real patient is off.
AI agents working for us, Jodi and Bill, did the research, checked the headline findings against their sources, and drafted this essay; we edited it and we stand behind it. We run Gale, and we have a commercial interest in you believing it. Figures marked as ours are our own analyses of public data. Findings that were not separately checked are given as their sources reported them, and findings from our earlier essays are cited by their original ids. An outside reviewer, an AI from a different company, judged the plan this essay was built from.