What the 988 study found, and what it could not prove

In April 2026 a research letter in the Journal of the American Medical Association reported something unusual for this field: a large fall in suicide deaths that lined up with one specific change in how help is delivered. The change was the launch of 988, the three-digit number that replaced the ten-digit American crisis line in July 2022.

Over the two and a half years that followed, the authors estimated, suicides among people aged 15 to 34 in the United States were around 11% lower than they would otherwise have been. That is 4,372 deaths fewer than the model predicted. The finding was reported widely, largely as good news and largely stripped of the qualifications the authors attached to it.

It deserves to be understood properly, for two reasons. It is the strongest evidence yet produced that a crisis line can reach far enough into a population to move mortality, which is not the same thing as proof that it did. And it has since been formally challenged in the same journal, in an exchange that says a good deal about how much a headline figure can carry.

What the study did

The design is easier to follow than the statistics suggest. The researchers took every recorded suicide in the United States from 1999 to 2024, from national death certificate data, and arranged the numbers by quarter. They then used the period from the start of 1999 to June 2022, the month before 988 launched, to build a model of how the trend was already moving. Finally they asked what that model predicted for the thirty months after the launch, and compared it with what actually happened.

The gap between the two is the finding. It is a before-and-after comparison against a projected baseline, not a randomised trial, and the authors describe it in those terms.

What it found

• 35,529 suicide deaths were recorded among people aged 15 to 34 between July 2022 and December 2024, against 39,901 expected from the pre-2022 trend.
• That is 4,372 fewer deaths, or 11.0% below expectation, with an uncertainty interval running from roughly 8.7% to 13.1%.
• The fall was larger where the line was used more. In the ten states with the biggest increases in 988 contact volumes, deaths came in 18.2% below expectation, against 10.6% in the ten states with the smallest increases.
• Among people over 65, who contact the line far less often, the fall was smaller.
• Running the same model against England, where no equivalent three-digit national lifeline exists, did not produce a comparable drop. The researchers also checked deaths from cancer over the same period as a negative control, and found nothing similar.

For context, the launch came with more than 1.5 billion dollars of federal investment in crisis centre capacity and staffing, and contacts to the line more than doubled over the following three years, with use disproportionately high among younger adults.

Why the state comparison is the important part

A national fall in suicides that happens to coincide with a national launch is weak evidence on its own. The country was also emerging from the pandemic, other mental health services were expanding, and suicide rates in the United States had already been moving.

The dose-response relationship is what makes the finding harder to dismiss. If 988 were irrelevant and something else were driving the improvement, that something else would have to be distributed across American states in almost exactly the same pattern as 988 call volume. That is possible, but it requires more explaining than the straightforward reading. It is also why the study is taken more seriously than most single-country before-and-after analyses.

The argument it started

On 17 August 2026 the journal published a letter challenging the result. Its central point is not about statistics but about what counts as a suicide death.

The study's outcome measure excluded deaths coded as unintentional poisoning and as poisoning of undetermined intent. Dr Andrew Herring's letter argues that this is a substantial omission, because a share of deaths coded that way are in fact suicides. He cites one study in which more than 20% of deaths recorded as unintentional or undetermined were judged to have been suicides. He also notes that the analysis does not speak to people with opioid use disorder, among whom suicidal thinking is common, or to the period after a non-fatal overdose, when risk rises sharply.

The authors published a reply in the same issue. The rights and wrongs of that exchange are for the journal's readers to settle. What it establishes for the rest of us is narrower. The 11% figure is not a measurement of suicides. It is a measurement of deaths recorded under a particular definition, compared against what that same definition would have produced without 988. Reasonable researchers disagree about whether that definition is the right one, and the disagreement is not a scandal. It is how evidence of this kind gets tested.

What the study cannot show

It is observational. The authors say so, and no careful reading of it can produce the sentence “988 prevented 4,372 suicides”. Three limits are worth holding on to.

• Causation is not established. The analysis shows a pattern consistent with the line having an effect. It does not isolate that effect from everything else changing at the same time. A stronger design would compare places where the line was rolled out at different times for reasons unrelated to suicide rates, and that is not what this was.
• The comparison groups are suggestive rather than decisive. The over-65 result and the England result both point the same way, which is reassuring, but neither is a controlled experiment.
• One study, however large, is one study. An independent researcher quoted in the coverage welcomed the finding and asked for replication. That is the correct response to almost everything published in this field.

What it means outside the United States

The study is American and so is the number, but the design question is not. The intervention tested was not a new therapy or a new medicine. It was three digits, a budget for the people who answer the calls, and enough public awareness that a person in distress knows what to dial.

Every country that already runs a crisis line has made some version of that bet. What this study suggests is that the parts which look administrative, such as how memorable the number is, whether the line is staffed well enough to answer, and whether anyone knows it exists, may matter more than the parts that are usually debated. The authors' own emphasis is blunter: sustained funding is the variable that matters.

For a reader outside the United States, the more immediately useful point is that the equivalent numbers in your own country are worth knowing before you need them. Ours are listed on the Get support page, and the link to it is in the footer below. The same argument applies to the number reserved at European Union level for emotional support helplines, which is explained on that page.