Hospital Presenting Self-Harm and Risk of Fatal and Non-Fatal RepetitionSystematic Review and Meta-Analysis

For thirty years, clinicians have been refining how they treat people who arrive at emergency departments after self-harm. Better assessment tools, improved follow-up protocols, and growing awareness. The field has not been standing still. And yet when Carroll, Metcalfe, and Gunnell pulled together one hundred seventy-seven papers spanning those three decades, they found something that stops that progress narrative cold: the rate of repeat self-harm and suicide in this population hasn't moved. Not in the 1980s, not in the 1990s, not now. That stillness is the finding this lecture is built around. Self-harm presentations are among the most common reasons for emergency hospital admission, and they carry an elevated risk that persists long after the patient walks out the door. Carroll and colleagues frame it simply: a hospital presentation for self-harm is the single strongest known risk factor for subsequent suicide. That framing matters because it positions every one of these episodes as a prevention opportunity — a moment where the healthcare system has contact with someone at serious risk. The question the review was built to answer is what actually happens after that contact. The answer, drawn from more than a century's worth of accumulated cohort data, is both precise and troubling. The core numbers first. The pooled estimate for suicide within one year of a hospital self-harm presentation is one point six percent. By five years, that rises to three point nine percent. Carroll and colleagues translate that into a phrase worth holding onto: one in twenty-five patients who present to hospital for self-harm will die by suicide within five years. Non-fatal repetition runs higher and faster — about one in six patients, sixteen point three percent, repeats self-harm within the first year alone. These are averages across a large and heterogeneous evidence base, presented with confidence intervals. But even at the lower bound of those intervals, the risks are substantial. Before going further into what those numbers mean, it's worth pausing on something the review exposes about how they were generated — because it turns out the method of counting repetitions changes the count dramatically. When Carroll, Metcalfe, and Gunnell grouped studies by how they ascertained repeat episodes, the estimates split apart. Studies using hospital admissions data produced a one-year non-fatal repetition rate of thirteen point seven percent. Studies using patient self-report produced twenty-one point nine percent. That's not a rounding difference — it's a gap of nearly eight percentage points. The authors explain exactly why it exists: hospital admissions data miss everyone who presents to an emergency department and is discharged without being admitted. Admitted patients are estimated to represent only about half of the hospital-presenting self-harm population. So if you measure repetition using admissions records, you're looking at a fraction of the real burden. One trial cited in the review makes the gap concrete: the control arm showed an eleven percent repeat rate by hospital records and an eighteen percent rate by patient report, from the same patients. This is not a methodological footnote. If you're running a trial of an intervention designed to reduce repeat self-harm, the outcome you choose will determine whether you think your intervention worked. A treatment that genuinely helps people but only reduces repeat attendances at the same hospital — without changing whether people harm themselves at all — will look effective under one measure and meaningless under another. Carroll and colleagues are direct about this: researchers and service evaluators need to choose and report their outcome measures carefully because the measurement changes the finding. The regional variation in the data raises its own questions. European studies produced a pooled one-year non-fatal repetition rate of seventeen point one percent. Asian studies came in at ten percent. That difference held up statistically when the authors split studies by location. Study location combined with follow-up method explained twenty-two point seven percent of the between-study variation in repetition estimates — which is informative, but also means more than three-quarters of the variation remains unexplained. The heterogeneity across studies was very high, with an I-squared statistic exceeding ninety percent. Carroll and colleagues are honest about this: much of what drives variation between studies is still unclear. What the data do tell us about risk within the population is more precise. Fatal repetition — death by suicide — is considerably higher in males than females: two point seven percent versus one point two percent at one year. Older cohorts show roughly double the one-year suicide rate of younger cohorts, and age alone explained forty percent of between-study variation in fatal repetition. Age and sex together explained nearly seventy percent. Cohorts with above-median rates of prior self-harm — where more than forty-three point seven percent of patients had harmed before — showed a one-year non-fatal repetition rate of nineteen point six percent, compared to fifteen point two percent in cohorts below that threshold. The risk isn't uniform. Older patients, males, and those with a history of prior self-harm are carrying a disproportionate share of it. Now back to the finding that demands the most attention: thirty years of data, and the line hasn't moved. The pooled one-year non-fatal repetition rate did not differ between studies published before the year 2000 and those published after it. The paper states this plainly — despite over three decades of research, the incidence of non-fatal repeat self-harm has not changed. That is not a finding you can wave away with methodological caveats, though caveats do apply. The measurement problems described earlier mean that true change could be hidden — if more recent studies use less comprehensive ascertainment methods, they might systematically undercount repetition relative to older studies, masking a genuine improvement. Or genuine improvement might exist in some settings and not others, averaging out across a global evidence base. Carroll, Metcalfe, and Gunnell don't resolve this. What they insist is that we have to face the question rather than assuming progress is occurring beneath the noise. The implication is pointed. Incremental refinements to existing approaches have not, in aggregate, shifted outcomes. The review identifies two things that need to change. First, measurement. If patient self-report captures a substantially larger burden of self-harm than hospital records, then future trials and service evaluations should use patient-reported repetition as their primary outcome — not because it is more convenient, but because it is more complete. An intervention that doesn't move the hospital admissions number might still be changing lives; the reverse is also true. Second, and more fundamentally, genuinely novel interventions are needed. The unchanged rates over thirty years highlight the need for new approaches, particularly targeting the groups the data identify as highest risk — older patients, males, and those with prior self-harm histories. The human scale of the problem keeps pulling the analysis back to ground level. One in twenty-five. That is the number Carroll, Metcalfe, and Gunnell return to, and it's worth sitting with. These are people who came through a hospital door — who were, in some sense, reachable. The emergency department contact is a moment. What happens in the months and years after that moment, across a quarter century of changing clinical practice, has not materially improved. That is what this review establishes. It is also, in the most direct sense, a description of what remains to be done. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.