This blog was authored by Dr. Jeffrey S. Markowitz, who specializes in epidemiology and health data analytics.

Suicide is notoriously difficult to predict. We know many warning signs and risk factors, including mental illness, previous attempts, acute emotional distress, social isolation, and major life crises. Yet knowing which individual will die by suicide remains extraordinarily difficult. Some circumstances, however, can bring several risks together.

Recent reports described conditions in U.S. Immigration and Customs Enforcement (ICE) detention that seemed particularly relevant to suicide risk and prevention. Human Rights Watch and Physicians for Human Rights reported concerns involving detention conditions, medical care, staffing, and oversight. In addition, they documented a rise in deaths during the second Trump administration. An  Associated Press investigation described problems with mental-health screening, suicide prevention, isolation, and monitoring, while reporting at least 10 suicides since January 2025.

Those reports led me to a somewhat different question: Did suicide account for a disproportionately large share of deaths in ICE detention after taking into account the age and sex of those who died?

Finding the records

ICE publishes Detainee Death Reports, or DDRs, describing deaths in custody. I expected standardized government reports and assumed they would be anonymous. Instead, the reports varied enormously, and the people who died were identified by name.

Some DDRs offered little beyond basic information about the person and the death. Others contained detailed accounts of detention history, medical and mental-health care, interactions with staff, and deterioration over hours, days, or longer. That unevenness became one of the study’s central challenges.

I developed a detailed codebook defining the information to extract and reviewed all 113 DDRs one at a time with ChatGPT’s assistance, applying the same definitions to every case. I made the final coding decisions myself. Some deaths were clearly suicides; others were not. When the record did not support a confident determination, I classified the death as indeterminate rather than forcing it into the suicide or non-suicide group.

Once the reports had been analyzed, I thought the difficult part was over. I submitted the study to Injury Epidemiology. Then peer review changed the paper.

A question that changed the paper

The editor returned “Minor Revisions,” then asked a deceptively simple question: 22 of the 113 deaths were suicides. But was that actually more than we should have expected?

I conducted a proportionate mortality analysis comparing the observed suicides with what would be expected based on the age and sex of those who died. At first, the excess looked dramatic. But the suicide decedents were relatively young, and suicide accounts for a larger share of deaths at younger ages in the U.S. population. Adjusting for age and sex reduced the excess but did not eliminate it. Among ICE detainees, suicide still accounted for more than twice the share of deaths expected from the age- and sex-specific pattern of mortality in the U.S. population. That became the study’s central finding.

The reviewers also questioned how reliable information extracted from such uneven reports could be. I had used AI in the original abstraction, so I turned to it again as a test of reproducibility. A second AI system independently reviewed all 113 DDRs using the same definitions without seeing my original coding. Agreement was excellent for some variables; for others, it was not. Mental-health history, prior self-harm, and placement on suicide watch, for example, proved too inconsistent to retain. The final paper contained fewer variables than the one I submitted, but I trusted those variables more.

What I found

Of the 113 deaths, 22 (nearly one in five) were classified as suicides, while another 10 could not be classified confidently. All 22 suicide decedents were male. Their average age was about 37 years, compared with 48 among those who died from other causes. They also died earlier in detention: a median of 27 days after entering ICE custody compared with 52 days for non-suicide deaths. Twenty-one of the 22 suicides involved hanging or ligature.

What was less obvious was who might have been recognized beforehand as being at risk. Suicidal ideation was documented in only 2 of the 22 suicide deaths and a previous attempt in only 3. Acute emotional distress and psychiatric problems were documented more often among suicide than non-suicide deaths, but no warning sign identified most of those who ultimately died by suicide. If a risk factor was missing from a DDR, however, I could not assume it had been absent, only that it was not documented.

Uncertainty extended even to the deaths themselves. Only 18 of the 113 DDRs (15.9%) contained a final cause of death, and only nine (8.0%) included an official manner of death. Some narratives described much of what happened but stopped short of providing the final medical examiner or coroner determination. Uncertainty was not simply a limitation of the study; it was a feature of the public record itself.

Behind the numbers

One thing I had not anticipated was how personal these records would feel. The DDRs identify detainees by name, and some describe the final hours or days of a person’s life in considerable detail. Yet my job as an epidemiologist was to step back and look for patterns that could be measured and compared. Holding those two perspectives, the individual person and the population represented by the data, was one of the more difficult parts of the study.

Conclusion

The study began with a simple question about suicide in ICE detention. Providing an answer required more than counting deaths; it required finding usable data, deciding what could be learned from an uneven public record, checking whether the information could be reproduced reliably, and finding an appropriate population for comparison.

After adjustment for age and sex, suicide still accounted for more than twice the share of deaths expected. That excess could not simply be explained by the demographic composition of those who died. Suicide was unusually prominent among deaths in ICE detention.

I am also grateful to the editor and reviewers, whose thoughtful questions helped strengthen the study in important ways.

Read Dr. Jeff Markowitz's paper here, published in Injury Epidemiology.