“I’m fine, it’s whatever.”
That is the kind of reply a fifteen-year-old sends two days after leaving an emergency department. It is also, word for word, the sample reply on our Vasl Bridge page, because it is the sentence that best explains why we built the product. Read it as data and it says fine. Read it as someone who knows what happened on Saturday night and it says something closer to this: I am not going to tell you, and I want to see whether you notice.
The week after a mental health emergency visit is one of the quietest stretches in youth behavioral health care, and one of the most exposed. The crisis has been assessed. The young person has been judged safe enough to go home. A referral has been made, a phone number has been handed over, and an appointment, if one exists yet, sits somewhere past the weekend. Between the discharge desk and that first visit, in most systems, nobody is assigned to the young person at all.
This post is about that week. What the published numbers say about it, why the gap is so hard to see from inside a health plan or a hospital, and what we think a daily text, read properly, can and cannot do about it. It is also the first time I have written at length about Vasl Bridge, so I want to be precise about what it is.
Bridge is a text from Vasl every day, from the day a young person leaves the emergency department to the day of their first appointment. VLAP, our clinical language model, reads the replies. When a reply hides distress, VLAP flags it and surfaces it to the partner’s licensed clinician for their review. That is the whole product. Every part of it is deliberate.
What the numbers say about the week after
The clearest picture of the gap comes from a study Hoffmann and colleagues published in Pediatrics in 2023.1 They followed 28,551 Medicaid-enrolled children and adolescents, ages 6 to 17, who were discharged from an emergency department after a mental health visit between January 2018 and June 2019. Three quarters of them were between 12 and 17. Depressive disorders were the most common diagnosis, in 39.1% of the cohort.
Only 31.2% saw an outpatient mental health clinician within seven days of discharge. Within 30 days, 55.8% had. Turn that around and it reads worse. More than two out of three children left the emergency department and went a full week without a follow-up visit. Close to half went a month.
The children who missed follow-up were not spread evenly. Non-Hispanic Black children were less likely than non-Hispanic White children to have a follow-up visit within 7 days, and less likely again within 30. Children in fee-for-service Medicaid were less likely to get timely follow-up than children in capitated managed care. The gap is widest for the young people who already have the least room for it.
Then there is what happens next. In the same cohort, 6.5% of children had a return acute mental health encounter within 7 days. By 30 days it was 12.8%. By six months it was 26.5%. More than one in four came back.
The authors also asked whether follow-up made a difference. In their analysis, a follow-up visit within 7 days was associated with a 27% lower risk of a return visit for acute mental health care. I want to be careful with that number. It is an association in claims data, not a trial. Families who make the first appointment may differ in many ways from families who do not, and a claims study cannot separate all of that out. But the direction is the one the whole idea of follow-up rests on: a young person who gets connected is less likely to be back.
That logic is now written into how states and plans are measured. In 2022, follow-up after an emergency department visit for mental illness, at 7 and 30 days, was added to the CMS Child Core Set of quality measures. State reporting on the Child Core Set became mandatory starting in 2024. For a Medicaid plan, the share of young people seen within seven days is no longer a background statistic. It is a number the plan reports and is judged on.
So the gap is documented, it falls hardest on specific young people, it is associated with return visits, and it is measured. None of that is new. What has been missing is anything that actually happens during the seven days.
Why the week is so quiet
If the gap is that well documented, why does it persist? Some of it is capacity. There are not enough outpatient clinicians, and the ones who exist are booked out. That part is real, and nobody can text their way out of it. But a lot of the gap is quieter than a shortage. It lives inside the seven days themselves.
Think about what discharge looks like for a family on Medicaid. The visit may have run overnight. The parent or guardian may have missed a shift to be there, and may have to make it up. They leave with paperwork, a referral, maybe a list of numbers to call. The young person goes home, often back to school within a day or two, to the same phone, the same group chats and the same circumstances that brought them in. The first appointment has to be scheduled by someone, usually during working hours, often after a call or two that does not get returned. Every day in that week, the plan depends on a family that just went through a crisis doing administrative work on its own.
Nobody on the clinical side holds the young person during that stretch. The emergency department has discharged them. The outpatient clinician has not met them. The health plan sees a claim days or weeks later. From inside any one of those organizations, the week is a blank, and a blank is easy to miss in a report.
The industry has noticed. There are vendors whose job is to fill the blank with outreach: automated texts, calls and surveys after discharge. That is a reasonable instinct and I do not want to dismiss it. The trouble is what happens to the replies. Most check-in systems collect them. A reply gets logged, maybe scored on a simple scale, maybe checked against a list of alarming keywords. A young person who writes “I’m fine, it’s whatever” gets recorded as fine, and the dashboard stays green.
That is where language becomes the problem, and it is the problem we have spent years on. The young people with the lowest follow-up rates in the Hoffmann cohort overlap heavily with the young people standard language tools read worst. Black youth writing in African American Language (AAL). LGBTQ+ youth using vocabulary their communities built for themselves. Rural youth whose way of talking is thin in most training data. And teenagers in general, who learned a long time ago that saying the plain thing out loud gets an adult involved in ways they did not choose.
So distress shows up sideways. It shows up as minimization: “it’s not that deep but.” It shows up as a vernacular frame around hopelessness: “can’t keep doing this no more fr.” It shows up in coded terms youth communities developed to get past content filters, like “unaliving.” And it shows up as the flat, closed reply, the “fine,” the “whatever,” sent by someone two days out of an emergency department who is testing whether anyone is reading.
A keyword list reads none of that well. We measured this directly this fall. Scored on suicide-attempt severity on a held-out test split, VLAP reached 0.83 ROC-AUC. The Suicide Risk Lexicon released this year by MIT’s McGovern Institute, the Child Mind Institute and Crisis Text Line scored 0.65 on the same texts. A 71-term clinical keyword rule we wrote ourselves, included so the comparison had a floor we controlled, scored 0.56. That is one benchmark, and it is not clinical validation. The method and its limits are on our Clinical Outcomes page, and I wrote about the comparison in detail last month. For this week, the point is simple. If the tool reading the replies is closer to a word list than to a reader, the replies that matter most are the ones it will wave through.
What a text a day can do, and what it cannot
Vasl Bridge is built around that week and nothing else. It starts at discharge. The social worker enrolls the young person with one thing, a phone number, and the first text arrives that day. There is no app to download, nothing to install and no account to set up before support starts. If the phone can send a text, it works.
Then Vasl texts every day. The messages are short on purpose. “Hey, it’s Vasl. How’s today?” “It’s Vasl. How are you holding up?” The young person answers by replying, in their own words, whatever is true. On the day of the first appointment the message changes: “First visit today. How are you feeling about it?”
We made three design calls early. Each one came from looking at how the week usually goes wrong.
The first was that Vasl texts first. The burden of reaching out should not sit with the person who just had the worst night of their year. A follow-up plan that depends on the young person or the family starting the contact is a plan that fails quietly on exactly the days when starting anything feels impossible.
The second was that the reply is free text. We do not ask young people to rate their mood from one to five or pick from a menu. A number is easy to collect and easy to fake. Words are harder to hide in, and they carry the signal VLAP is built to read.
The third, and the one that makes the product, was that every reply gets read. VLAP reads it against a taxonomy of 47 clinical signals across six categories, from hopelessness and social withdrawal to coded self-harm language and the minimization patterns that often come before a disclosure. When a reply carries one of those signals, VLAP flags it and surfaces it to the partner’s licensed clinician, with the young person’s exact words and the reasoning attached. The clinician reviews it and decides what happens next. “I’m fine, it’s whatever” reaches a clinician instead of a spreadsheet.
When the young person wants more than a daily text, more is there. Their mood map and their peer community open in the phone’s browser, still with nothing to install, and a culturally matched coach can reach out. None of that is required to get the daily check-in. The floor of the product is one text a day, and the floor has to work for someone who wants nothing else.
I also want to be specific about what Bridge does not do, because in this part of health care the boundaries matter as much as the features.
VLAP does not diagnose. It does not decide. It does not send alerts to parents, principals or anyone else on its own, and none of its output is wired to an automated action. It flags language for a clinician’s independent review, and the clinician is the one who acts. Crisis support is human-initiated, always, with 988 integration and licensed clinical supervisor review before any action. Vasl does not deliver therapy. The partner’s licensed clinicians do, and the purpose of the week is to get the young person to them.
Bridge also does not replace the first appointment, and it does not shorten the wait for one. What it does is make the days before that appointment less empty. It gives the young person a reason to keep answering, and it gives the clinician who will eventually see them some sense of how the week went, in the young person’s own words. The day-seven text asks how they feel about the first visit because the visit is the goal. Everything before it is there to make sure they arrive.
Who it is for, and who carries the cost
We built Bridge for the young people in the Hoffmann numbers: Black, Brown, LGBTQ+, rural and Medicaid youth, the ones least likely to get a follow-up visit and the ones standard tools hear least clearly. Those two facts are related. When the access gap and the language gap fall on the same young people, a check-in system that cannot read them cannot close either one.
We also built it for the organizations that carry the cost of the week going badly. I think it is worth being plain about that, because good intentions do not fund programs.
Medicaid plans are measured on 7-day follow-up after a mental health emergency visit. The measure sits in the Child Core Set, states report it, and the baseline in the largest study we have is 31.2%. A plan that wants to move that number needs something that happens during the seven days. A report about them afterwards does not move it.
Health systems carry the returns. In the Hoffmann cohort, 26.5% of children were back in acute mental health care within six months. For the discharging hospital, every one of those is another emergency visit, often longer and often harder to place, for a young person whose first visit already showed the system where the risk was. Bridge gives the discharging hospital something to hand the young person on the way out that keeps going after they leave the building.
There is an equity argument here that does not need dressing up. The young people with the lowest follow-up rates are disproportionately Black and disproportionately in fee-for-service Medicaid. A program that raises follow-up overall while leaving that gap where it was has not done the job. That is why anything we report from Bridge will be broken out by the populations the gap falls on, and why the language layer sits at the center of the product.
What we have not done yet is publish outcome results for Bridge, and I am not going to imply otherwise. What we have is a published benchmark for the language model that reads the replies, with its method and limits stated, and a product built around a gap that is measured, reported and paid for. The next step is to run it with partners who own that gap and measure what changes: 7-day follow-up, 30-day follow-up and return visits, each broken out by race, ethnicity and coverage type. If the numbers do not move, we will say so.
If you run a Medicaid plan or a children’s emergency department, that is the conversation we want to have. Bring your current 7-day rate and your discharge workflow. We will show you what the first text looks like, what VLAP reads in a reply, and where the flag lands for your clinicians.
The week after discharge has been quiet for a long time. A daily text will not fix the shortage of clinicians or the wait for a first appointment. What it can do is make sure that when a young person sends “I’m fine, it’s whatever” on day two, somebody qualified to hear it actually does.
1, Hoffmann JA, Krass P, Rodean J, et al., “Follow-up After Pediatric Mental Health Emergency Visits,” Pediatrics 151(3), 2023, doi:10.1542/peds.2022-057383. VLAP benchmark figures are from a held-out test split and are not clinical validation; see Clinical Outcomes for method and limits.