Published on 19 September 2026, updated on 20 September 2026
Summer heat and youth suicide: what can a US county-level ecological study actually tell us?
The American Journal of Psychiatry · 2026 · online ahead of print, no volume or pagination · Jayaraman et al.
DOI 10.1176/appi.ajp.20250096
PMID 42337413
Scientific 62
Editorial 76
The essentials
Across 418,728 county-month observations covering the 48 contiguous states and Washington DC from 1980 to 2004, each additional degree Celsius of mean monthly temperature is associated, among 5 to 24 year-olds, with a suicide rate 0.75% higher across the full year and 2.68% higher in summer. Stratifying by ten-year age bands isolates a single significant band across the entire lifespan, 15 to 24 year-olds, at 2.97% per degree in summer, 95% CI 1.30 to 4.65. The other seasons show no significant association, which does not prove that no effect exists there. One caution governs the entire reading of this work: the unit of observation is the county-month, not the person. What the study shows is a variation in rates between hotter and colder months within the same county. No result of this kind licenses the claim that a given individual’s risk rises when it is hot. The period analyzed ends twenty-two years before publication.
The context
The question comes up in consultation through two doors. First through patients and their families, who spontaneously link spells of intense heat with poorer sleep and psychological agitation. Second through the organization of care: in many countries, notable heatwaves have turned extreme heat into a recognized public health concern, with vigilance plans that involve psychiatry without a clearly established clinical content for that vigilance. In France, for instance, the heatwaves of 2003, 2019 and 2022 played that role, and readers elsewhere will recognize an equivalent episode in their own national history.
Work linking temperature to suicide is mostly ecological: it compares territories and periods, not people. This study belongs to that family. It reuses the dataset already analyzed at the general-population level by Burke and colleagues in 2018, and applies to it a stratification by season and by age that the earlier work did not carry out. The authors themselves restrict their claim to priority, and it is worth citing as written: this study is, to their knowledge, the first to examine the relationship between ambient temperature and suicide in young people using a large national sample, at the county-month scale, over twenty-five years. The journal is a leading one, the data volume considerable, the estimate expressed per degree. None of these three qualities erases the design’s underlying limitation; if anything, they make it essential to name that limitation before the results rather than after.
Mechanism: what the study measured, and what it did not
Several mechanisms are commonly proposed to explain a link between heat and suicidal behavior, and the authors discuss them at length, drawing on other publications. None of them is measured here. The study measured a mean monthly temperature, a mean monthly precipitation, and a monthly number of suicide deaths, aggregated by county. It measured no biological variable, no sleep variable, no behavioral variable, and documented no intermediate step between temperature and death. The hypotheses below are therefore hypotheses, and the table presents them as such.
| Hypothesis raised in the literature | Status in this study |
|---|---|
| Serotonergic transmission | Not measured |
| FindingNo sample, no biological marker, no individual-level data enters the analysis. The pathway is cited in the discussion; it is not tested. | |
| Sleep debt from hot nights | Not measured |
| FindingSleep does not appear among the variables analyzed. This pathway’s plausibility comes not from this work but from the references it cites. | |
| Functional brain connectivity | Not measured |
| FindingNo imaging, no clinical or psychometric assessment is linked to the deaths counted. This lead comes from an earlier study cited by the authors. | |
| Reduced heat acclimatization in young people | Not measured |
| FindingThis is the hypothesis the authors invoke to explain the summer-specific signal, drawing on experimental data published elsewhere. Nothing in this dataset verifies it. | |
| Social exposure and changes in summer activity | Not measured |
| FindingCounty-month fixed effects absorb average local seasonality, including school schedules, but no data on activity, schooling or isolation is recorded. | |
In other words, the study documents an association between a weather variable and a suicide mortality rate. It documents no causal chain, and the authors explicitly write that prospective studies would be needed to establish causality. A reader who came away from this with a pathophysiological explanation would have added to the text something the text does not contain.
The study at a glance
| Element | Content |
|---|---|
| Population | Young people aged 5 to 24, 48 contiguous states and Washington DC, 1980 to 2004: 97,401 suicide deaths, 83.3% of them among boys and men. |
| DetailThe statistical unit is the county-month, not the person. No subject is followed individually. From 1989 onward, mortality data report deaths only for counties with more than 100,000 residents, which the authors state covers close to three-quarters of the US population. | |
| Exposure | Mean monthly ambient temperature, in degrees Celsius, from the PRISM interpolated climate data at 4 km² resolution. |
| DetailExposure is a territorial average, not an individual exposure measured for the people who died. Precipitation is adjusted for as a proxy for cloud cover. | |
| Comparator | The same county’s historical average for the same calendar month, with county-month and state-year fixed effects. |
| DetailCounty-month fixed effects neutralize what varies locally from one season to another; state-year fixed effects neutralize what varies at the state level from one year to the next, such as unemployment or prevention legislation. This is the study’s methodological strength. | |
| Outcome | Monthly county suicide death rate, standardized to the 2000 reference population. |
| DetailThe outcome is an aggregated rate drawn from mortality data, with no clinical information on the people who died. The baseline rate for 5 to 24 year-olds over the period is 0.496 per 100,000. | |
| Design and volume | Ecological county-panel study, 418,728 county-month observations, 104,682 per season. |
| DetailThe volume refers to the number of aggregated observations, not to a number of enrolled participants. | |
| Secondary analyses | By season, then, for the significant season alone, by ten-year age band and by sociodemographic and geographic subgroup. |
| DetailAge bands range from 5-14 to 65 and older. The publication explicitly describes this stratification as a post hoc analysis. Subgroups cover legal sex, income, the non-white population share, educational attainment, geographic division, and rural or urban status. A Bonferroni correction is applied to all reported p-values. | |
Quality control
| Domain | Judgment |
|---|---|
| Design | Ecological study |
| FindingThe results hold at the county level. No individual-level conclusion can be drawn from them, whatever the size of the database. The authors state this themselves in their limitations. | |
| County-month and state-year fixed effects | Strength |
| FindingThis design neutralizes baseline seasonality and stable local characteristics, ruling out some of the most obvious alternative explanations. Standard errors are clustered at the county level, and residuals are weighted by the county’s annual population. | |
| Period covered | Older data |
| FindingThe analysis ends in 2004, for a publication in June 2026. It covers neither the warming of the past two decades nor recent changes in youth suicide rates. The authors justify these boundaries by data availability, not by an analytical choice. | |
| External validity | US context |
| FindingClimate, urban form, access to care, and the determinants of suicide mortality differ between US counties of this period and any other country today. Even within the United States, the estimate varies by a factor of ten across geographic divisions. Transposing this result elsewhere cannot be assumed. | |
| Statistical modeling | Robustness checked |
| FindingThe primary model is a population-weighted linear fixed-effects regression, not a count model. The authors check the estimate’s stability across ten alternative specifications, linear and non-linear, and confirm in the supplementary material that Poisson and negative binomial models yield estimates of the same order, 0.85% and 0.86% per degree in young people versus 0.75% for the primary model. | |
The results
| Result | Estimate |
|---|---|
| 5 to 24 year-olds, all seasons | +0.75% per °C, 95% CI 0.34 to 1.16, p 0.00036 |
| ReadingAcross the full year, the association is weak. It is the stratification by season that brings out the signal. | |
| General population, all seasons | +0.73% per °C, 95% CI 0.54 to 0.93 |
| ReadingAcross the full year, young people are not distinguished from the general population. This study’s distinctive finding is therefore not an overall vulnerability, but its concentration in one season and one age band. | |
| Summer, 5 to 24 year-olds | +2.68% per °C, 95% CI 1.42 to 3.94, adjusted p 0.00012 |
| ReadingThe summer estimate is more than three and a half times the full-year figure. The variation between seasons is itself significant, Wald test p 0.018, and summer is statistically distinct from each of the other three seasons taken individually. | |
| Winter and spring, 5 to 24 year-olds | +0.59% and +0.12% per °C, not significant |
| ReadingConfidence intervals run from -0.07 to 1.25 in winter and from -0.88 to 1.12 in spring. Autumn is not significant either. A non-significant estimate does not prove the absence of an effect in these seasons: it indicates that, in this data, no association was detected there. The result also runs counter to the authors’ original hypothesis, which expected a peak in spring as much as in summer. | |
| Summer, 15 to 24 year-olds | +2.97% per °C, 95% CI 1.30 to 4.65 |
| ReadingThis is the only significant age band across the entire stratification, from age 5 to over 65, after correction for seven comparisons. It is the result underlying the study’s editorial interest. It remains a county-level estimate. | |
| Summer, 5 to 14 year-olds | +2.70% per °C, 95% CI -5.03 to 10.42, not significant |
| ReadingThe point estimate is of the same order as for 15 to 24 year-olds, but the interval is so wide that it permits no conclusion. The authors themselves present this point as a similarity in effect size, not as a result. | |
| Summer, 25 to 34 year-olds | +1.38% per °C, 95% CI 0.040 to 2.73, adjusted p 0.31 |
| ReadingNone of the age bands above 24 reaches significance after correction. The publication reports no direct comparison test between age bands: each band is tested separately. | |
| Translation into number of deaths | About 12 additional deaths per month per 1°C, the authors’ extrapolation |
| ReadingThe authors apply the summer baseline rate for 15 to 24 year-olds, 0.955 per 100,000, to the 43 million Americans in that age band, and infer that a summer 3°C above regional averages would correspond to more than 100 additional deaths per year. These are projections built from the estimate, not counted deaths. | |
| Differences between subgroups | No significant difference |
| ReadingSex, income, educational attainment, non-white population share, rural or urban status: no within-category comparison reaches significance after correction, and the abstract concludes that no sociodemographic difference was identified. The effect-size gaps the authors discuss remain trends that were not successfully tested. | |
Critical appraisal
| Domain | Judgment |
|---|---|
| Ecological fallacy | Inherent limitation |
| FindingAn association observed between county averages does not carry over to the level of the person. Nothing in this work says that the people who died were the ones experiencing the most heat, or that a given patient becomes more at risk above some threshold. This limitation is not a flaw in execution: it is built into the design, and the authors flag it first among their limitations. | |
| Temporality | Respected |
| FindingReverse causation is not plausible here: a month’s temperature does not depend on that month’s mortality. This is an argument in favor of the direction of the association, not a demonstration of causality. | |
| Exposure measurement | Approximate |
| FindingA monthly average interpolated at county scale captures neither heat spikes within the month, nor time spent indoors in air conditioning, nor travel outside the county. The authors note that larger counties introduce more measurement error. | |
| Residual confounding | Present risk |
| FindingDeterminants that vary within a county from one year to the next for the same month, alcohol consumption, local unemployment, access to care, are not measured. The authors also note that precipitation is an imperfect proxy for sunshine, which independently affects mood. | |
| Multiple comparisons | Corrected |
| FindingP-values are Bonferroni-corrected, multiplied by 4 for seasons, by 6 for pairwise comparisons and for subgroups, by 7 for age bands. The summer signal and the 15-24 signal survive this correction, which is the study’s strongest point. | |
| Power for subgroup analyses | Insufficient |
| FindingThe authors acknowledge that the rarity of suicide among young people limits power to detect differences between subgroups. The effect-size gaps they discuss, notably between women and men or between rural and urban counties, should therefore not be read as established differences. | |
| Internal consistency of the tables | One discrepancy noted |
| FindingIn the pairwise comparison table, the effect-size difference between summer and spring is given as 0.47 points, whereas the published seasonal estimates, 2.68% and 0.12%, yield a gap of 2.56 points, and 0.47 is also the value given for the spring-versus-winter comparison. The coefficient difference itself is consistent. Separately, the Pacific division is given as 0.72% in the discussion and 0.73% in the figure. These discrepancies do not call the main result into question, but they mean the seasonal values should be cited from the seasons table. | |
| Gap between data and publication | Twenty-two years |
| FindingAn association established over 1980 to 2004 does not automatically carry over to a warmer climate and to generations whose suicide epidemiology has changed. The authors note that their period precedes the west coast’s rapid warming specifically, which could explain the absence of a signal in the Pacific division. The direction of this gap, amplification or attenuation, cannot be determined from this data. | |
| Match between claim and evidence | Measured |
| FindingThe authors do not assert a causal link, write that prospective studies would be needed to establish one, and explicitly restrict their priority claim to the combination of a large national sample, the county-month scale, and twenty-five years. This is to their credit, and a further reason not to harden their conclusion when citing it. | |
Level of evidence
In the Oxford Centre for Evidence-Based Medicine’s levels-of-evidence table, 2009 edition, ecological studies fall under level 2c, alongside outcomes research. This classification says the essential thing: the design describes a geographic and temporal correlation, it does not test a hypothesis at the individual level. No risk-of-bias tool designed for trials or individual cohorts applies usefully here; the appraisal that follows rests on the characteristics specific to the panel design and on the limitations the authors themselves declare.
Confidence is high on one point: in this data, the hottest summer months are associated, at the level of US counties from 1980 to 2004, with a higher suicide rate among 15 to 24 year-olds, with a statistical design that rules out baseline seasonality as a lazy explanation, a correction for multiple comparisons, and stability confirmed across ten model specifications. Confidence is low, or nil, on everything else: on the move from county to person, on mechanism, on differences between subgroups, on transposing the finding to today’s climate, and on its applicability to any country outside this dataset.
The colleague test
What an experienced colleague would say if shown this study in two minutes, between two consultations.
The number is nice, but I don’t treat counties, I treat patients one at a time. What this tells me is that something happens at the population level in summer, not that a young patient of mine is going to do worse because it’s thirty degrees out. And as for why, nobody knows, the study measures none of that. Honestly, it doesn’t change a prescription. It just confirms for me that easing up on follow-up over a summer isn’t a good idea.
Translation for practice: this result belongs on the side of care organization and epidemiological literacy, not on the side of individual clinical decisions. It provides no screening criterion, no temperature threshold, and no therapeutic indication.
What you can do with this
- What you now understand: in older, large-scale US data on 5 to 24 year-olds, the association between temperature and suicide rate appears only in summer, and is significant only among 15 to 24 year-olds, with a county-level estimate of 2.97% per degree Celsius.
- What you do not conclude: nothing about any particular patient’s risk. An ecological result does not translate into an individual assessment, and this study justifies no change in prescribing and no new screening practice.
- What you can say to a patient or family who asks you after a spell of extreme heat: yes, some studies observe a link between heat and suicide at the population level; no, the mechanism is unknown, and it says nothing about their particular situation.
- What you keep an eye on: continuity of follow-up over the summer, a period when appointments space out, teams rotate, and young patients lose their school-year structure. This vigilance was already warranted before this study; the work adds a further argument for it, as a lead, not a directive.
- What you are waiting for: more recent data, ideally from outside the United States, and individual-level analyses. The authors themselves point the way, ambulatory temperature monitoring, ecological momentary assessment of mood, actigraphy. Without them, transposing this finding to any other country remains a hypothesis.
Frequently asked questions
Does heat increase suicide risk?
This study cannot answer that at the individual level. It observes that hotter summer months are associated with higher suicide rates across US counties, between 1980 and 2004, among 15 to 24 year-olds. Moving from this observation to a statement about a person’s risk is an ecological fallacy, that is, attributing to an individual a relationship established across groups.
Why does a study published in 2026 stop at 2004?
The authors explain this by data availability. At the start, 1980 is the year the number of weather stations feeding the climate database used increased by half, which led the research community to recommend not going back further. At the end, after 2004, the federal public health agency stopped publishing suicide counts for counties whose monthly total falls below ten, which affects most counties. What remains is that the period analyzed precedes the recent acceleration of warming, which limits the finding’s present-day relevance, in a direction that cannot be determined in advance.
How many additional deaths does this represent?
The authors offer a projection, which should be read as such. Applying the summer baseline rate for 15 to 24 year-olds, 0.955 per 100,000, to the 43 million Americans in that age group, they estimate that one additional degree Celsius would correspond to about twelve additional deaths per month, and that a summer 3°C above regional averages would correspond to more than one hundred additional deaths per year. These numbers come from a calculation applied to the estimate; they are not counted deaths.
Should suicide-risk screening be stepped up during heatwaves?
This study does not support such a recommendation. It concerns a different country, a different period, and a different level of analysis. Heatwaves are a real public health concern in many places, and maintaining continuity of summer follow-up for at-risk young patients is already recognized good practice. This work offers an additional reason for vigilance, not a new screening criterion.
Is the mechanism known?
No, and the study measures none. The leads the authors raise in their discussion, brain connectivity, impaired sleep, reduced heat acclimatization in young people, changes in summer social activity, rest on other publications. This work does not test them.
Annotated bibliography
Source study. Jayaraman P, Wortzel JR, Wei G, Christ RR, Sharma Y, Nugent NR. Deadly Heat: The Association Between Ambient Temperature and Suicide in Young People in the United States. The American Journal of Psychiatry, 2026, online ahead of print June 24, 2026, no volume or pagination. Ecological county-panel study, 418,728 county-month observations, 48 contiguous states and Washington DC, 1980 to 2004. Contribution: a quantified estimate of the association between ambient temperature and youth suicide rate, stratified by season and age, with county-month and state-year fixed effects. Limitations: ecological design, no mechanism measured, period ending in 2004, no established difference between subgroups. DOI 10.1176/appi.ajp.20250096 · PMID 42337413
Study whose data and method this work reuses. Burke M, González F, Baylis P, et al. Higher temperatures increase suicide rates in the United States and Mexico. Nature Climate Change, 2018;8:723 to 729. Contribution: this is the work that established the general-population reference estimate, 0.68% per degree, and built the dataset reused here. Limitation: it stratified by neither season nor age, which is precisely this study’s contribution. Reference noted in the source publication’s bibliography; its identifiers were not independently verified against the primary source.
General framework for the heat-mental health link. Thompson R, Lawrance EL, Roberts LF, et al. Ambient temperature and mental health: a systematic review and meta-analysis. Lancet Planetary Health, 2023;7:e580 to e589. Contribution: the synthesis to which the authors anchor the whole field. Limitation: like the present study, it aggregates work that is mostly ecological. Reference noted in the source publication’s bibliography; its identifiers were not independently verified against the primary source.
The other references the authors cite, notably on brain connectivity in children exposed to heat and on the epidemiology of youth suicide, were not consulted in the primary source and are therefore not discussed here.
Editorial collections
Tags
Verified on August 12, 2026 against the full text of the publication and its supplementary material where available. This analysis underwent an independent double reading. The English version was checked for conformity on September 19, 2026, against the figures of the French version and against the source. How we verify what we publish
This analysis is intended for healthcare professionals. It does not constitute a prescribing recommendation and does not replace individual clinical judgment.
Analysis from Psychiatry Evidence Base, evidence-based psychiatry, explained with rigor.