Do we actually know anything about youth trans suicide rates?
(Author’s note: I’m not sure how much of a work in progress this article is. Usually, when I post something like this, people will hit me up with things I might have missed, or errors I’ve made/they think I’ve made. I think this is an important piece to be right on, so the piece itself might change a bit as time goes on.)
I’m going to present a bunch of scenarios. While reading them, pretend you are neither particularly well-informed, particularly religious, or particularly partisan:
One day, your 14-year-old child comes to you and asks for a moment of your time. They explain that they have discovered that they are trans, and want a series of expensive treatments, irreversible surgeries, and a great deal of support from you in order to transition to their non-birth-gender.
You are a voter, and politicians are doing things that have to do with parents’ obligations to their children re: issues of gender. Maybe they want to punish parents who won’t let their kids transition or to force them to let them. Maybe they want to punish parents who do let their kids transition or to keep them from doing so.
You are a taxpayer, and one that generally is at least minimally concerned with how tax dollars are spent. It has been proposed that gender transition be subsidized to some extent, similar to how other types of medical care are often subsidized under Medicaid/Medicare.
In all three cases, you probably care quite a bit no matter what your views on transgender issues are. And you should. Things like “surgery” and “being unhappy” matter a lot; again, this is true no matter what personal beliefs you hold on this issue.
But if we remember, in this set of hypothetical scenarios you are supposed to be pretending you don’t hold much of a view at all - you are a fairly blank slate. So you go looking for answers. You find that for at least the past couple of years there’s a fairly vocal group promoting the view that withholding your support for your child’s transition is emotionally abusive and phobic. Even worse, they say, your reluctance puts their life at risk.
You are told that trans people are unusually suicidal compared to the general population because they are being forced to live in a state that feels inherently wrong to them at an existential level. The longer this goes on the more the risk compounds, they say; are you so close-minded and dedicated to your biases that you’d be willing to see them die?
And they aren’t entirely without sources. Even a superficial check turns up data points like this:
The numbers are stark, worrisome and should set off alarm bells: 52% of all transgender and nonbinary young people in the U.S. seriously contemplated killing themselves in 2020. More than half thought it would be better to be dead, rather than trying to live with rejection, isolation, loneliness, bullying and being targeted by politicians and activists pushing anti-trans legislation.
Or powerful graphics showing a halving of risk if you allow your child to transition, like this:
(Note: The last graphic is from a survey of adults, not children.)
If it seems like I’m oversimplifying the discourse space on the issue, that’s true. There’s plenty of voices on the other side (and some recent state government actions) telling you the opposite is true and that allowing your children to transition should be considered child abuse instead.
You aren’t a kid; you know there’s basically no chance that big-news-source articles or activists on either side are going to give you clean, unbiased data; there’s too much culture at stake for there to be a realistic chance of that. While [primary sources] aren’t immune to bias, they are still a hell of a lot better than random writer’s motivated interpretation of a university PR department’s exaggerated press release of a study’s findings.
That’s where I’m at; I want to know exactly what and how confidently we know the risk of suicide among trans youth. I think it’s an important enough thing that you should, as well.
The first and most important thing to note is that there’s two ways the suicide risk of trans youth is discussed. The first and much more popular way is to report on suicidal ideation (how much an individual reports they think about/considers suicide) and other measurables that don’t directly measure how many people committed suicide, while the second (and much less mentioned) measure has to do with successful suicide attempts and actual deaths.
Before jumping to conclusions, it’s important to note that suicidal ideation and related metrics are much, much easier to study than actual suicide rates in some groups. If you want to know how many people in a particular group think about suicide, you can just call and ask them. The difficulties that exist are mostly related to identifying large amounts of people in those groups, but that’s more of a “lots of work” problem than it is a “this data might not be possible to get” problem.
If you want to know how many gum-chewers committed suicide you need to trust that a government can first identify any particular death as a suicide consistently, and then also be confident that they are accurately noting whether a particular victim was a gum-chewer or not.
With some characteristics (race, gender as listed on driver’s licenses, etc.) you can be fairly confident the government is getting the latter right. In the case of a variable like “is trans”, the government would have to ask a lot of questions it just isn’t consistently asking right now.
Adding in the normal problems related to trying to make sense of dozens of different agencies at a state and federal level reporting information in different ways means population-level suicide data for the most part just doesn’t exist. The closest we can generally come to this is data extrapolated from clinical populations, but there’s not much of it.
When the bulk of your information about a risk is data about a related risk factor, it’s reasonable to ask how reliably that risk factor actually predicts the damage you are worried about. So before we get into the actual suicide rates, we need to take a look at both ideation and reported suicide attempts separately and see what each tells us about the other.
Suicidal Ideation and Risk
Suicidal ideation is essentially considering or actually wanting to kill oneself. This means the world of studying suicidal ideation is for the most part one of surveys; because both of these conditions involve thinking and feeling (and because mind-reading isn’t an option) attempts to quantify them essentially always start and end with asking people whether they’ve experienced it.
It’s pretty much universally accepted that rates of suicidal ideation in trans youth are unbelievably high. The Trevor Project recruits survey-takers through targeted ads on social media and reports that in 2021, 52% and 20% of transgender youth seriously considered or attempted suicide respectively. Here’s another survey (albeit one of adults) showing similar numbers:
The 2015 U.S. Transgender Survey (USTS), which is the largest survey of transgender people in the U.S. to date, found that 81.7 percent of respondents reported ever seriously thinking about suicide in their lifetimes, while 48.3 percent had done so in the past year. In regard to suicide attempts, 40.4 percent reported attempting suicide at some point in their lifetimes, and 7.3 percent reported attempting suicide in the past year.
This remains pretty consistent throughout all the data I’ve seen, and there’s no particular reason to doubt it unless you think they made serious errors in their survey method or misrepresented their data. They are not wrong to note that this is much, much higher than average. As noted here by the CDC, it’s much more typical to see significantly lower last-12-month ideation:
This is where most discussion of the matter ends, full-stop; these are massive differences in numbers, and the disparity is often used, by itself, to justify the argument for puberty blockers or transition in children. But since we are looking at trans youth as a group, it’s reasonable to ask how strongly self-reported suicidal ideation and suicide rates are related to suicide itself - i.e. how often thoughts become a grim reality.
While trans suicide rates themselves are not very well tracked, suicide rates having to do with more universal characteristics (age, sex, etc.) are, and the CDC offer these charts showing the prevalence of suicidal ideation and suicide attempts by age, again as determined by survey:
Judging by what they say, you would expect that you’d see the highest suicide rates in women between the ages of 18-25, with suicide rates dropping for both genders as they age. From the same page, here’s the actual observed suicide rate:
And that doesn’t track at all. Women between 18-25 commit suicide less than all but <14-year-old male and female children and > 75-year-old women. Men pretty consistently commit suicide at rates 3-5 times as high as women, but report that they think about it and plan it less.
Here’s the breakdown of suicide rates by race:
And we find that Hispanics, Blacks, and Asian/Pacific Islanders commit suicide at nearly identical rates, despite reporting 4.2%, 3.4%, and 2.3-2.8% respectively. Whites and Hispanics report they attempt suicide about the same amount, but Whites are successful at nearly 3x the rate.
I’m not doing deep, deep data analysis here, but it’s pretty clear even to casual analysis that there’s not a consistent association between suicidal ideation or suicide attempts as reported by groups and actual observed suicide rates in the same groups. You can pick the reason you think most likely as to why. It’s possible some groups are more reluctant to talk about their experiences with suicidality or relate them differently. It’s possible that “considered suicide” means different things to different cultures.
But given that some groups (say, those of different sexes) actually show a huge negative correlation between suicidal ideation/reported suicide attempts and actually successfully committing suicide, the responsible move would seem to be to note that suicidal ideation as expressed by a group doesn’t tell us a lot about what to expect from that group re: suicide rates.
That’s not the case here. Suicidal ideation as expressed by trans youth as a group is given as evidence of high suicide risk, full stop. Nothing we know that I can find suggests that’s the right move. As per XKCD, people are complex; the relationships between suicidal ideation, suicide attempt, and groups shows that.
No group of any kind that I was able to find in a week of looking that portrayed suicidal ideation and suicide attempts as a risk noted this messy relationship between those factors and actual suicide rates.
There’s no way someone working as closely with suicide as most people looking at this issue are could have missed the CDC data, But they ignore it so damn confidently that I’m sort of terrified I’m missing something big here and someone’s going to tell me so in the comments. Even so, every single thing I can see regarding the association of group suicidal ideation, suicide attempts, and suicide rates seems to indicate they made a giant leap of faith here.
Changes in individual behaviors and differences between groups are not the same thing. Please for the love of all that’s good note that I’m talking about GROUPS. I do not know the exact numbers related to how the risk of suicide changes between an individual who does not express thoughts of suicide or who reports suicide attempts changes compared to someone who doesn’t, but please take this kind of stuff very seriously.
If self-reported suicidal ideation as measured at a group level isn’t that informative by itself, then you have to look at actual suicide rates to determine the severity of the risk a particular group faces. This sounds terrible, but what you need to know at this point is how many people actually die; without it, none of the other group-level data means anything.
As mentioned, this isn’t easy to do because there just isn’t that much data yet. The data that does exist is hard to find because it’s hidden behind people claiming to have studied suicide rates who end up to be relying entirely or almost entirely on survey data, like this:
Denial of appropriate bathroom or housing access also elevates the risk for suicide. TGNC college students victimized in this manner are distressed and commit suicide at a relatively high rate.
Or news sources who have crafted their words so carefully to be technically true while still misleading a casual reader that it’s almost comically clear they know what they are doing. So you get stories like Comedian's death underscores high suicide rate among transgender people, that say stuff like this:
The U.S. suicide rate increased 33 percent between 1999 and 2017, despite falling in many other developed countries, including most of Western Europe, according to CDC data. And while this national trend is worrisome, the suicide rates for transgender and gender-nonconforming people are much higher than the national averages. According to the National Center for Transgender Equality’s 2015 U.S. Transgender Survey, 40 percent of adult respondents reported having attempted suicide in their lifetime — almost nine times the attempted suicide rate in the general U.S. population.
Once you’ve weeded through the people who have already determined the correct thing to allow you to know, you end up with very few studies that actually deal with actual deaths. All that I know of the deal entirely with data derived from clinical populations.
Very few studies I found have anything like a large sample, but a few stood out. The first is Wiepjes, et. al. 2020, which looked at some 8,263 and found their per-100000 people-years death rate was elevated:
Forty-nine people died by suicide: 41 trans women (0.8%) and 8 trans men (0.3%), which is 64 per 100,000 person years in trans women and 29 per 100,000 person years in trans men… The mean number of suicides in the years 2013-2017 was higher in the trans population (40 per 100,000 person years; 43 per 100,000 trans women and 34 per 100,000 trans men) compared with the Dutch population in this time frame (11 per 100,000 person years; 15 per 100,00 registered men and 7 per 100,000 registered women).
Note that this is an adult population, so it doesn’t map exactly onto the youth populations being discussed. Differences between suicide rates in different age ranges can vary pretty widely, so take it with a big grain of salt. It’s useful to include here as a level-setting device to give you an idea of what a reasonable ballpark looks like, but I wouldn’t include it at all if there were more and better pediatric studies to draw from.
There are only two studies I know of that do look at pediatric populations (both were mentioned on ACX’s most recent links post), and of the two by far the most striking results come from Cauwenberg, et. al. 2021, which got a lot of attention for them but relied on what seems to me to be a silly-small N=177:
Of all 177 participants, five (2.8%) adolescents are known to have committed suicide. According to the Agency of Care and Health in Flanders, in 2017 there were 0.9 suicides in the age group 10–14 years old and 7.2 suicides in the age group 15–19 years old per 100,000 inhabitants…
Biggs 2022 has by far the largest N (>15000 patients in their data) and finds a much less striking result (line breaks added for readability):
From 2010 to 2020, the four suicide deaths equate to 0.03% of the 15,032 patients. Taking the denominator as 30,080 patient-years, the annual suicide rate is calculated as 13 per 100,000 (95% confidence interval: 4 to 34 per 100,000). For comparison, the annual suicide rate in England and Wales between 2010 and 2020 for adolescents aged from 15 to 19 years averaged 4.7 (Office for National Statistics, 2021). This does not quite correspond to the age range of the GIDS patients, however.
At referral, the patients ranged in age from 3 to 17 years; only 7% were younger than 10. The mean was 14 years and the median 15. Most patients stay with the GIDS until transitioning to an adult service. Therefore, the average age of patients at any point in time will lie somewhere between 14 and 17. A better comparison is therefore the overall suicide rate for adolescents aged from 14 to 17 (available only for the entire United Kingdom for 2015–2017), which was 2.7 per 100,000 (Office for National Statistics, 2018; Rodway et. al.., 2020).
Comparison should also account for the difference between the sexes, because males are more likely to commit suicide than females. Of the GIDS patients, 69% were female. Adjusting for sex, the GIDS patients were 5.5 times more likely to commit suicide than the overall population of adolescents aged 14 to 17.
They also helpfully calculated a comparison to Cauwenberg’s results, since Cauwenberg was too lazy to calculate patient-years and per 100,000 rates:
Direct comparison can be made with the Belgian pediatric gender clinic (Van Cauwenberg et. al., 2021). Its annual suicide rate was about 70 times greater than the rate at the GIDS. This is especially puzzling because patients at the Belgian clinic scored better, on average, than those at the GIDS on tests of psychological functioning (de Graaf et. al., 2018).
If we ignore that Biggs’ results might be 25% lower than stated (one of their presumed suicides has not been definitively declared to be), ignore that Wiepjes is dealing with a more suicide-prone older adult population, and trust Bigg’s estimate of seeing 1/70th the suicide rate of Cauwenberg we are left with numbers that range from 13, 46.5 and 910 per 100000 patient-years.
In the last quote, Biggs calls this discrepancy “puzzling”. Yeah, no shit, Biggs.
A difference in numbers that big means something weird, but it’s hard to say what. It could be the kind of strange results you see when N is small; 177 is not a lot, and it could be they got a particularly suicidal cohort. It could be that Biggs and Wiepjes are both wrong, despite both having much larger N. It’s even possible that whatever gender-affirming care the Belgian clinic offers is massively backfiring to the tune of order-of-magnitude worse outcomes, although I don’t expect that’s what’s happening.
I think it’s more likely that it has something to do with these two easy-to-miss comments from Cauwenberg:
The client database compromises all adolescents between 12 and 18 years, who were referred to our pediatric clinic between January 1st 2007 and December 31st 2016. These referrals are made by parents, other health care providers, general practitioners and sometimes by the adolescent themselves. Once referred, clients are put on a waiting list and contacted as soon as one of the psychologists of the pediatric team has room to start-up a new client…
In Belgium, the pediatric gender clinic in Ghent is the only pediatric clinic that offers gender affirming care. Consequently, waiting lists are unacceptable long.
If there’s about 900,000 Belgians in the relevant age range and we use the usual .3% US estimate of trans prevalence, we end up with ~2700 trans children in Belgium at any given time; the program saw 235 children total over 10 years. As the only clinic offering what they believe is lifesaving care in the entirety of the country, it would be shocking (and likely a massive ethical failure given their priors) if they weren’t triaging by severity in some way or another.
To put it another way, the difference in the numbers is more than explained if there’s even a small amount of triaging by perceived suicide risk at a clinic that admits only 1% of those affected by the condition it treats in a population it serves. And that could be what’s happening here, but note that this guess is nowhere near enough to justify completely ignoring the Belgian study. When you only have two studies looking at pediatric trans suicide rates, you should be pretty reluctant to throw one out.
So what does this mean in real terms? That depends an awful lot about which study you pay attention to, how triaged you believe each clinical population to be (or how representative of a “population average” trans person you think those clinical trans patients are), and how much you think gender-affirmative care helps.
To break this down, let’s look at this as if each of the two pediatric studies got it exactly right and produced findings that perfectly represent reality.
The Cauwenberg Universe
Let’s take a look at the Belgian clinic’s reality first. If triaging did not occur and the study population was fairly representative of western trans adolescents, then their data implies that your hypothetical trans child has about 25x the suicide risk of a >75-year-old man, the most suicide-prone group I’ve mentioned in this article. In absolute terms, this is close to a 1% chance of killing themselves every year of their adolescence.
Unless you think gender-affirmative care makes this worse, there’s no indication in the study that medical care/transition would help improve this rate. The clinic doesn’t have a bad reputation and every person the study counted was receiving care, so these are best-case numbers unless the kind of care they provide is, on net, harmful.
The Biggs Universe
The Biggs universe is a bit more complex. First, only about half of the people they count were receiving care, but the suicide rates were the same:
Although two out of the four suicides were of patients on the waiting list, and thus would not have obtained treatment, this is not disproportionate: the waiting list contributed nearly half of the total patient-years.
Biggs sets the probability of a particular trans child committing suicide in the biggsverse at something like .013% per year. But as opposed to the Cauwenberg timeline, it’s possible for gender-affirmative treatment to make those numbers better - for one, only about half the study population was seen by the clinic, which means our hypothetical 50% reduction in suicide rate would have saved a life. There’s also some people who say that GIDS doesn’t really treat people at all, which would mean two potential lives in this cohort, and a .0065% suicide rate.
Considering Turban et al
Turban et al, 2020 is one of the more popular studies analyzing data related to gender-affirming care and mental health outcomes, primarily because of the huge effects it claims:
And make no mistakes, those asterisked achieved-statistical-significance odds ratios are a big deal. If true, and if we get anything like “full value” from the reduction (i.e. the number of suicides drop proportionally) this would be a huge deal.
The study has major, major problems, among them that there were very strong indicators that a large amount of his sample didn’t know what puberty blockers even were (seriously, read the comments on that article, it’s rare to see so brutal a bloodbath). Even if the study isn’t actually true, it’s useful here because it provides an “optimistic study” ceiling.
Attentive people might have caught that “suicide attempts resulting in inpatient care” went up, but I’m resisting the urge to jump on that. I’ve already registered that I think this study is kind of shitty and that I don’t really think surveys are particularly reliable ways to know the truth. I’m not going to waste a bunch of time figuring out how to contort enough to say this survey is right on this one point, especially when the p-value is so weak.
I’m approaching the limit of words substack allows in an email without clicking through, which means I have probably once again failed to be efficient in relaying data. What we want at this point is a visual comparison of the risk each study presents, and the effect of Turban’s reduction in suicidal ideation on observed suicide rates at different levels of imagined effectiveness.
It’s time to appeal to the incredibly ugly efficiency of Excel:
I’ve adjusted for Turban in four ways - assuming his past 12-month suicidal ideation and his ill-defined “lifeline” suicidal ideation reductions translate 1:1 into prevented suicides, as well as looking at if each person prevented from past 12-month ideation represented a quarter or a tenth of a prevented suicide. Note that there really shouldn’t be any adjustment for Cauwenberg at all, unless you think they were withholding puberty blockers from people who wanted them; again, every person in their study received treatment of some form or another.
Since Turban 2020 is so bad and because he has no idea how those suicidal ideation numbers translate to suicides, this is not *much* better than just randomly guessing at how much gender affirmative care might help, if it does at all.
And, finally, here’s the same chart, but with the 15-24 baseline death rate (for the US; I couldn’t find age-group death rates for Belgium. Their overall age-adjusted rate is about ~915/100000):
That last chart should give you pause. According to Cauwenberg, your suicide risk if you are a trans youth1 is 13 times higher than all other forms of mortality combined. It’s tens of times higher than the highest risk of suicide of any other group I’ve been able to identify. It’s much higher than the one other comparable study and much higher than studies dealing with older populations would lead you to expect even if you expected higher youth suicide rates, which you shouldn’t.
Cauwenberg saw all this, barely mentioned it, and did essentially no work to figure out why; they didn’t even ask what triaging methods were used to select patients. Something is up there; all Cauwenberg had to do to make their study 100 times more known was make it clear that they had discovered trans teenagers had suicide rates 25 times higher than the next most at-risk groups we know about, and not doing that stinks of knowing something we don’t.
But I’m some jackass on the internet. It’s entirely possible I’m doing the math wrong2 here and someone is going to stumble in, notice it, and let me know I should be very very embarrassed about my quant skills. Fair enough. Let’s say that happens and you decide that Cauwenberg is much less clearly bad than I think it is: what then?
The answer is that you should still have a very, very high level of uncertainty about what’s going on here. For all you’ve heard about trans children being at risk for suicide, did you know we didn’t have any idea about the actual suicide rates until a year or so ago? Did you know before reading this there are only two studies that disagree with each other to an absurd degree on what those suicide rates probably are?
Did you know before reading this that reported suicidality differs hugely between groups? Did you know that when comparing those groups to each other, the correlation between suicidality seems about as likely to be negatively correlated to actual suicide than not?
You probably didn’t; I didn’t, before writing this. And yet there’s people on both sides very confidently telling you they know what’s going on with a very high level of confidence. There are people on both sides who will tell you that transitioning or not transitioning a child constitutes abuse serious enough that the decision should be out of the parents’ hands. But without a clear way to interpret reported suicidality for a particular group into actual suicide risk, and with hard data on suicide risk restricted to a couple studies dealing entirely with clinical populations, they simply can’t do this in a data-driven way.
There are ways that a person can still honestly argue that one course of action or another is better, say by using logic or common sense. This is another way of saying that people will have different opinions about this stemming from their pre-existing beliefs, and that’s not something I generally disapprove of.
But if someone comes to you saying they’ve examined the data, that they’ve found the science is clear and that these findings imply a clear moral obligation to allow or disallow transition, they are telling you something that it’s not possible for them to have learned from the science, at least at this time. You should adjust your belief in them and your confidence in what they are saying accordingly.
I have mostly ignored differing suicide rates by gender here in favor of comparing both-gender rates where possible. What little I did see seems to indicate that suicide rates track what you’d expect from birth sex, i.e. a transwoman has a suicide rate much higher than a transman.
I am a Qual who is just barely smart enough to fake small amounts of quant. If I got something real super-duper wrong, please let me know; I’d much rather know about it than end up looking dumb forever.