Everything you were too afraid to ask

Suicidal Thoughts

How common these thoughts actually are and what separates thinking from acting, the world figures, why nobody can predict this including the people whose job it is, whether asking someone plants the idea, and what really stops people from saying anything, along with what genuinely works and why it is almost never persuasion, and where to go for help, wherever you are

Written by Ivana, psychologist and psychotherapist

32 minute read  ·  7,154 words

Maybe it’s a thought that turns up late at night: that it would be easier not to be here. Maybe it’s something a friend said weeks ago that you still can’t put down. If either sounds familiar, you’re in much bigger company than it feels like. About one adult in eleven has thought about suicide at some point in their life, and almost nobody mentions it.

Most advice about suicide is written for the person standing nearby: watch for the warning signs, know the risk factors. It’s well meant, and a surprising amount of it has either never been tested or was tested and didn’t hold up.

This article is about what the evidence can actually carry, and it’s written for you whether the thoughts are your own or you’re worried about someone else. Some of it is uncomfortable, especially how poorly anyone can predict this, including the people whose job it is. Some of it is more hopeful than the public conversation suggests, especially what happens to people who survive. And the most effective thing anyone has ever done about suicide, measured across whole countries, turns out not to have been persuasion at all.

More common than the silence around them suggests

In the World Health Organization’s World Mental Health Surveys, which asked the same questions in seventeen countries, about one adult in eleven had thought about suicide at some point in their life. Far fewer had made a plan, and fewer still had made an attempt.

A companion survey across twenty-one countries asked only about the past year, and found that about one adult in fifty had had these thoughts in that time. The most interesting part of that study is a comparison it almost throws away: the figures for richer and poorer countries were nearly identical. Whatever this is, it is not a problem of affluence and it is not a problem of poverty.

Those averages hide a big spread. A survey across ten sites on four continents found lifetime rates of suicidal thinking ranging from under 3% at one site to about a quarter at another. The link between thoughts, plans and attempts also differed from place to place, which is a good reason not to assume the whole thing works the same way everywhere.

Surveys of university students find higher figures, more than one in five over a lifetime. That gap shouldn’t be read simply as a fact about young people. The studies disagreed with each other a great deal, and the highest estimates tended to come from surveys with few responses and broad definitions. Part of the difference is age, and part of it is that the two sets of studies ask the question differently. Either way, having these thoughts is far more common than the silence around them suggests.

Wanting to be dead and planning to die are not the same thing

You might recognize a quieter version of this. Clinicians separate passive suicidal thoughts, meaning a wish to be dead or not to wake up, from active ones, meaning thoughts of taking action. A pooled analysis of 86 studies found passive thoughts in roughly one person in ten over a lifetime, though most of those studies were in English, so it shouldn’t be treated as a world figure.

What passive thoughts predict isn’t known yet. The study usually cited to show that they carry the same risk as active thoughts measured everything at a single moment, which can show that two things go together but never which came first. The best review of the evidence calls its own findings on this point preliminary. Meanwhile the most widely used clinical scale finds that the more severe the thinking, the higher the odds of a later attempt, which suggests that less active thoughts carry less risk.

So what can be said is narrower than some would like. Thoughts of not wanting to be here are common, and the research does not support treating them as a lesser category that can be waved away. Whether they predict later action as strongly as more active thoughts do is still unsettled. If that quieter version is what you’ve been living with, it doesn’t need to get worse before it’s worth talking about.

Most people who think about it never act on it

If you’re frightened by your own thoughts, or by something someone told you, this is the finding to hold on to. It holds in every dataset that has looked: most people with these thoughts never act on them.

How large that majority is depends on who is asked and how the question is worded. Across different groups, the share of people with these thoughts who go on to make an attempt ranges from about one in eight to about one in three. There is no single worldwide number.

What is better established is stranger and more useful. The things that predict whether somebody has these thoughts are largely not the things that predict whether they act on them. In a survey across twenty-one countries, mental disorders strongly predicted the start of suicidal thoughts, and predicted the move from thoughts to an attempt weakly or not at all.

Depression, for example, roughly tripled the odds of having these thoughts in the first place. Among people who already had them, it raised the odds of a planned attempt only about one and a half times. An American survey had shown the same pattern years before researchers gave the difference between thinking and acting a name.

One more pair of findings belongs here, and each would mislead on its own. In the largest British follow-up, people who went to the hospital after harming themselves had a risk of dying by suicide in the following year many times higher than that of the general population, and it didn’t fade quickly. And across long-term studies, more than nine in ten people who survived a serious attempt did not go on to die by suicide. Part seven comes back to what both halves mean if this is you.

Where this actually happens

Most research on suicide comes from the United States and Europe. Most of the deaths don’t happen there. The World Health Organization estimates that close to three quarters of the world’s suicides happen in low- and middle-income countries. That changes how the whole subject should look, because that research is usually read as describing everyone.

Around the world, the rate has fallen by about a third since 2000. That’s real progress, but it hasn’t been even. Over the same years the rate in the Americas rose, and in the African region it has barely moved, though the records there are thin enough that a flat line may mean missing records rather than no change.

You may have read that suicide is the third leading cause of death among 15 to 29 year olds worldwide. It’s true, and it’s one of the most misread facts in this field. Suicide ranks high among young people largely because young people die of very little else. Rates are generally higher at older ages, yet in the over-70s suicide doesn’t make the top ten causes of death, because so many other causes outrank it. Rank and rate move in opposite directions across a life.

The ranking itself shows this. Using 2019 data, WHO put suicide fourth in that age group, and it now puts it third. The rank rose while the rate among young people fell, because deaths from tuberculosis fell faster.

A ranking of countries is partly a ranking of record-keeping

If you’ve seen a table comparing countries, it helps to know what’s behind it. National rates differ by about a hundredfold, but WHO states plainly that only about eighty of its 194 member states keep death records good enough to estimate suicide rates directly. Everywhere else the figure is modeled.

The uncertainty around those estimates gives the game away. Lesotho’s estimate could be anywhere from about 22 to about 66 per 100,000 people. The United Kingdom’s sits between 8.3 and 9.0. Those two numbers are not the same kind of object and shouldn’t be set beside each other as though they were.

A review of 31 studies of how reliably suicide is counted found that over half reported under-counting of more than 10%, and nearly all of those studies came from Europe and North America. So even the research on how badly suicide is counted comes almost entirely from the countries that count it best.

There’s a further complication. Suicide is still a criminal offense in some countries. Five of the seven countries with the lowest rates in the 2021 data are in one WHO region, the Eastern Mediterranean, and that pattern fits a genuinely low rate just as well as it fits deaths going unrecorded. The data can’t tell those apart. Put together, a country’s place in a ranking tells you about its record-keeping as well as about its people.

Four figures that do not survive checking

Fifty years of trying, and no improvement

If you’re worried about someone, you might hope a professional could tell you exactly how much danger they’re in. It’s a natural hope. One group of researchers pooled 365 studies covering fifty years of work, all of which measured people first and then waited to see what happened. Prediction was only slightly better than chance for every outcome they examined, no type of risk factor did much better than any other, and prediction hadn’t improved across those fifty years. Following people for longer didn’t help either.

The authors were careful about what this shows. Almost every study they pooled tested risk factors one at a time, so the finding is that single risk factors, tested one by one over long periods, barely predict. It isn’t proof that no combination could ever do better, and the same authors called for exactly that work to be done. And being unable to predict is not the same as being unable to help, which is where this part ends up.

The arithmetic that defeats the checklist

The deeper problem isn’t that the risk factors are bad. It’s that suicide is rare, and rare events break prediction in a way that’s easy to say and hard to feel.

One review of risk tools used in psychiatric hospitals reports two numbers about the same tools. The first is the area under the curve, a standard measure of how well a test sorts a whole group into higher and lower risk. It was 0.83, which sounds respectable. The second is the positive predictive value, meaning the share of people labeled high risk who went on to die by suicide. It was fewer than one in two hundred. Both numbers are correct, and the gap between them is the whole story of suicide prediction.

Other reviews find the same thing. Across the clinical scales in use, only about 5% of the people they flagged went on to die by suicide. A review of fifteen tools found that none met a minimum standard of accuracy, and that they fail in two opposite ways: some flag almost everybody, and some miss almost everybody. One scale that is still widely taught picks up only about one in seven later attempts.

Clinical judgment without a scale is no better. Clinicians working without one correctly flagged only about three in ten of the people who later harmed themselves.

The problem people quote most is about the people the tools catch. The more worrying one is about the people they clear. In one review, roughly half of the suicides happened among people who had been placed in the lower-risk group. The authors of another put it plainly: relying too much on risk factors “may provide false reassurance and is, therefore, potentially dangerous.”

Machine learning has been offered as the answer, and the two most recent reviews of it disagree. One found its accuracy still too low to be useful for deciding who gets help. The other was more hopeful and found five models that worked well on groups they weren’t built on, but also found design weaknesses in almost every study, and that fewer than one model in ten had ever been tested outside its own data.

The clearest example in that research is a worked one. A model of typical accuracy, at one fixed setting, would be right about roughly one in a thousand of the people it flags in a group where suicide is rare, about one in six in a group where it’s more common, and about two in three where it’s most common. Only the base rate differs, meaning how common the outcome is in the group being tested. Researchers disagree about what to do with this. They don’t disagree about the numbers.

What this does and does not mean

If you’ve read this far, it would be easy to decide that nothing helps, and that would be wrong.

What has been shown is that long-range prediction from a checklist performs poorly across populations. That says nothing against asking someone how they are, and it isn’t a claim about what happens in a conversation. A previous attempt is still among the strongest single predictors anyone has, even though, on its own, it predicts poorly.

There’s also a distinction here that the field itself now treats as the important one. Risk factors describe groups of people over years. Warning signs describe one person over hours or days. They answer different questions, and mixing them up is a large part of why risk assessment disappoints.

So the conclusion the field has reached is that help should be allocated on the basis of need rather than on the basis of a risk score. It sounds like a modest sentence, but it changes the question. The question is no longer who is most likely to die. It’s who is struggling, and what would help.

Asking does not put the idea there

If you’re worried about someone, you may be afraid that asking will make things worse. Asking someone about suicide does not put the idea in their head. That is the best-supported claim in this article, and the evidence behind it is better than the short editorial most often quoted for it.

In a randomized trial, more than two thousand high school students in New York State were assigned either to a survey that asked about suicide or to one that didn’t. Two days later the two groups looked the same, both on distress and on suicidal thinking. Among the students already at highest risk, the ones who had been asked were doing slightly better rather than worse: those with symptoms of depression were less distressed, and those with a previous attempt had less suicidal thinking, than similar students who weren’t asked.

Two independent meta-analyses, one pooling 18 studies and one pooling 13, reached the same conclusion. Neither found any sign of harm, and the first found that, if anything, suicidal thinking fell very slightly after people were asked. An American study that asked the same people about suicidal thoughts again and again, over thousands of surveys, found that being asked more often wasn’t linked to more severe thinking.

The strongest recent addition is an experiment with 700 American adolescents, about three quarters of whom had had suicidal thoughts or behavior at some point. It was designed as an equivalence test, which goes further than failing to find a difference: it sets out to show that there is no meaningful effect. The effects on the wish to self-injure, the wish to die and mood were all within a few hundredths of zero.

What the evidence covers, and what it does not

How to ask, and what to do with the answer

If you’re getting ready to ask someone, you probably want to know what to actually say. On the wording, what exists is expert agreement rather than trial evidence, and it’s unusually useful because it has been compared across cultures. Expert panels in English-speaking countries and across Asia, along with refugee and immigrant panels, agreed on a shared core of advice, and later studies in Latin America, Indonesia and China reached the same place.

Ask plainly, and use the word: “Are you thinking about suicide?” In a small British study of people who had talked to a doctor about this, one thing they said themselves was that they wished the doctor would just say it. Another was that a good conversation made the subject feel ordinary rather than alarming: “he made it normal, not embarrassing or weird.”

Try not to ask in a way that tells the person which answer you want. A question like “You’re not thinking of doing anything stupid, are you?” tells them both that the subject is shameful and that an honest answer wouldn’t be welcome.

Ask when there’s time to hear the answer. If you’re still worried, ask again another day, because a no on a Tuesday isn’t a permanent answer. Some people who say no are telling the truth about that moment, not about the month.

What you do with the answer is the part most often got wrong, and here there is evidence as well as agreement. Researchers interviewed people who had said no to a suicide question at a health care visit and then made an attempt within the next two months. What held them back wasn’t only stigma. It was fear of an overreaction, and of losing control over what happened next. The researchers concluded that listening without judgment, and showing you care without overreacting, may help people get past the fear of saying so.

So your role is to ask, to listen, to stay, and to help the person connect to something. It isn’t to assess risk, and it isn’t to take responsibility for the outcome.

One finding sharpens all of this. Across eight American emergency departments, screening every patient for suicide risk was tried on its own, and later screening was followed by an intervention. Screening alone changed nothing. Screening followed by something being offered reduced attempts, by about five people in every hundred. Being asked the question changes very little. Being asked and then offered something changes more.

The fear is not the one you would guess

If you’ve kept these thoughts to yourself, you may have put it down to shame. Shame is part of it for many people, but it usually isn’t what comes first.

In a study of psychotherapy clients who had hidden suicidal thoughts from their own therapist, seven in ten said they feared practical consequences outside the therapy room. The one they named most was being hospitalized against their will, which they believed could follow from sharing even mild thoughts. Nearly half said they would be more honest if that threat were somehow reduced. Shame and embarrassment mattered, but they came second.

A separate study, in which people described their reasons in their own words, reached the same place. The main reason for holding back was fear of what would happen, especially being hospitalized against their will, and the reasons people did speak up were wanting help and having a good relationship with their therapist.

And the people from part four, who had denied suicidal thoughts at a health care visit before an attempt, named three fears together: stigma, overreaction, and losing control over their own choices.

If this is what has kept you quiet, it makes sense. Two things follow. The first is that the fear is specific: it’s about what will happen next, not only about feelings. The second is that anyone urging people to speak up, including this article, is asking them to accept a risk, and it’s more useful to make that calculation accurate than to pretend it isn’t being made. The researchers who measured this fear concluded that people often overestimate what telling someone leads to, which isn’t the same as saying the fear is groundless.

Caring without overreacting

What helps people answer honestly isn’t complicated. In a study of American university students who had had suicidal thoughts, the biggest barriers to answering a direct question honestly were about stigma, and what helped was wanting emotional support and wanting the person asking to understand them. Shame and fear of consequences sit side by side. The same students said their most honest answers went to mental health professionals.

Stigma is real, and it also isn’t the biggest barrier to getting help. A large review of help-seeking for mental health in general, not suicide in particular, found that stigma was only modestly linked to seeking help less and ranked fourth among the barriers people named. The stigma worry people reported most often was about what happens once they tell someone.

One more finding comes from very small numbers, so treat it as a signal rather than an estimate. On screening questionnaires, people who skip the suicide question score more like the people who say yes than like the people who say no. The point to carry away needs no statistics: when you ask directly, silence or a changed subject is not a no.

What usually happens if you tell someone whose job it is

The plan itself is what has evidence behind it. A safety plan means working out in advance what your own warning signs are, what helps, who to contact and how to make your immediate surroundings safer. A meta-analysis of six studies tested it.

Safety planning reduced suicidal behavior by a little over 40%. It did not reduce suicidal thinking. That’s an unusual combination: a safety plan made people safer without reducing the thoughts themselves.

There’s a question that turns all of this from a fear into something you can do, and you can ask it before you share anything: “What happens if I tell you I have been having these thoughts?”

Changing what is within reach

You might expect the most effective prevention to be a conversation that changes someone’s mind. The strongest evidence points somewhere else. In Sri Lanka, the suicide rate rose roughly eightfold between 1950 and 1995. In 1995, and again in 1998, the country restricted the import and sale of a means that was in common use, and within about a decade the national rate had halved. The analysis estimated nearly 20,000 fewer deaths in the decade after than in the decade before, and found that unemployment, alcohol, divorce and the years of civil war didn’t track the fall the way the regulation did.

Bangladesh did something similar and measured it more carefully. Deaths by that route fell by about two thirds, and all unnatural deaths fell by a quarter, even though overall use of that means kept rising.

Eighteen studies of restricting access at places where deaths have happened repeatedly have been pooled, and together they found roughly a 90% drop at those places. Encouraging people to seek help at the same places was linked to about half as many deaths.

These are population studies rather than trials, because no one can randomize a country, and reviewers judged only five of twenty-seven national studies to have used the strongest methods available. They show what happened after a law changed, not proof that the law caused it. Even so, this is the most consistent body of evidence in the whole field, and almost none of it involves persuading anybody of anything.

Sri Lanka’s gains haven’t held. Rates among men there have been rising again since 2016. Restricting one route removed a route. It did not remove the distress.

Does it just move somewhere else?

You might wonder whether making one means harder to reach simply moves the deaths somewhere else. It’s a fair question.

Substitution is real, and it has been measured. In Bangladesh, deaths by another method rose by a fifth over the same period, and the authors reported it themselves. Set that beside the fall of a quarter in all unnatural deaths and the shape is clear: some people did turn to another means, not all of them did, and the total came down.

Sri Lanka shows the same shape. The rise in deaths by other means after its restrictions was small next to the fall they produced, and the researchers who tracked it describe the substitution as limited. So the answer sits between the most hopeful version and the most doubtful one: substitution happens, and where it has been measured it has been smaller than the fall that came with it.

Doing the same thing one household at a time did not work

One trial ties this whole part together, because it tested whether what works for a whole population also works one household at a time.

It gave households a practical way to do the same thing at home. It was a cluster-randomizedcluster-randomizedA trial that assigns whole groups, such as villages or schools, to each arm instead of individual people, usually because the intervention works at the level of the group. trial covering more than 223,000 people, which is enormous by the standards of this subject, and it found no difference. That trial tested one particular step, a locked box handed to every household whether or not anyone there was struggling. It doesn’t mean that, when someone is in crisis, putting distance between them and a means is pointless. Making the surroundings safer is still one of the steps of a safety plan.

Put that beside the screening result from part four, where asking the question on its own changed nothing and asking plus offering something did. A pattern emerges that runs against most people’s instinct. The strongest results come from changing what is within reach for a whole population at once. Leaving the task to each household didn’t work, and neither did a question with nothing offered afterwards.

The things with better reputations than evidence

Several common practices are widely treated as settled, which is exactly why their evidence is worth setting out.

Safety planning belongs here only in part. The meta-analysis in part five is real evidence, but its most-cited study compared sites in the United States veterans’ hospital system rather than randomizing people. It bundled the plan with follow-up phone calls, so the two can’t be separated, and most of the people in it were male veterans.

Talking therapies aimed directly at suicidal behavior have some trial evidence, though none of it is large. In one American trial with adults seen after an attempt, ten sessions of cognitive therapy roughly halved further attempts over eighteen months without changing how often people had the thoughts. In another, with women with borderline personality disorder, dialectical behavior therapy halved attempts compared with treatment by experienced therapists of other schools. In adolescents, its advantage was clear at the end of treatment and had faded six months later. None of these trials has shown fewer deaths.

Caring contacts, meaning brief, regular messages of care sent to someone after a crisis, have a weaker evidence base than their reputation. Pooled trials found no clear drop in the number of people who harmed themselves again, and trials in Australia and New Zealand found no difference either. A WHO trial across five countries did find fewer deaths, and in the same trial found no difference at all in repeat attempts.

Lithium’s reputation here rests on older pooled analyses. The trial designed specifically to test it, in United States veterans, was stopped early because it had become clear it wouldn’t find a benefit, although it kept lithium levels lower than usual and so may have tested a weaker version of the treatment.

Clozapine is the one medicine with a clear positive trial for suicidal behavior in schizophrenia, although that trial had far too few deaths to show a difference in deaths either way. Ketamine reduces suicidal thinking within a day, and after a single infusion the effect has been measured for about a month. No trial has been large enough to show whether it reduces suicidal behavior or deaths.

Crisis lines have a weaker evidence base than their visibility suggests. Most studies measure how callers feel by the end of a call, and they do find distress falling, but few follow people up afterwards, and the best-known studies had no comparison group of people who didn’t call. That is an argument for better research, not against picking up the phone, and this article ends by pointing you toward a directory of them.

No-suicide contracts, written promises not to act, were standard practice for decades, and they were never tested in a randomized trial at all. The case against them is an absence of evidence plus a serious ethical objection: they tended to reassure the clinician more than they protected the patient. They were quietly dropped rather than disproved.

If it is you

The evidence doesn’t support the idea that anybody, including a professional with a form, can sort you into safe or not safe. What it does support is that being asked and then offered something helps, and that a concrete plan made in advance reduces the chance of acting even when it doesn’t make the thinking stop.

If what’s stopping you from telling someone is fear of what happens next, that fear is one of the most commonly documented reasons people stay quiet, and it isn’t irrational. You can ask about it first.

And if you’ve already survived an attempt, both halves of the earlier finding are yours: the risk in the following year is higher, which is a reason to have people around you, and more than nine in ten people in your position do not go on to die by suicide.

If it is someone else

Don’t appoint yourself the person who predicts what will happen. The whole of part three is an argument that nobody can do that, and trying will make you frightened and them guarded. The job is smaller and more achievable than the one people imagine, and the smaller version is the one with evidence behind it.

If someone tells you and then asks you to keep it entirely to yourself, you’re allowed to say that you want to help and that you can’t promise silence. A promise you will break is worse than a boundary you keep.

Why there is no phone number in this article

Close to three quarters of the world’s suicides, 73%, happen in low- and middle-income countries, and a number for one country tells most readers nothing and suggests the article wasn’t written for them.

WHO’s own guidance points the same way for a different reason. It advises naming only a limited number of services, because a long list overwhelms people and leaves them unsure where to turn.

So this article gives one main route. Our helplines page lets you pick your country and shows the crisis lines and emergency numbers for it. If you prefer, befrienders.org lists centers in a smaller number of countries. Each line on our page shows when it was last checked, and befrienders.org was checked before this article was published. If you are in immediate danger, your local emergency number is the right call.

The most striking thing about this research is how much of the confident public advice about suicide rests on expert agreement rather than evidence, and how much of the well-evidenced material points somewhere nobody expects. Nobody can reliably tell who will act. Asking does not make things worse. And the approaches with the strongest numbers behind them changed what was within reach rather than changing anybody’s mind. If these thoughts are yours, or you’re worried about someone who has them, none of that asks you to be certain of anything. It asks you to say something, and to let someone answer.