Mined diagnostic sequences reveal ordered illness paths tied to suicide death

Most suicide risk assessment tools treat each clinical diagnosis as an independent factor, weighing depression or anxiety on its own merits. A new study argues that the order in which conditions accumulate may itself carry prognostic information. Using temporal sequence mining on linked insurance claims and mortality records, researchers identified specific ordered diagnostic trajectories associated with suicide death, including paths that pass through physical illnesses before returning to psychiatric crises.

The cohort drew on the Maryland Suicide Data Warehouse and included 3,647,059 insured Maryland residents aged 10 years or older with claims records from January 1, 2016, to December 31, 2020. Through linkage with medical examiner data, the team ascertained 768 suicide deaths within that cohort. Diagnoses, coded in ICD-10-CM and grouped into Clinical Classifications Software Refined categories, were mined for sequential patterns, yielding 89,221 candidate sequences. When these were tested in time-varying Cox models, 1,816 sequences remained significantly associated with suicide death, with adjusted hazard ratios ranging from 2.4 to 134.1.

The structure of the significant trajectories was striking. Half ended in a physical condition rather than a psychiatric one, and 93.7 percent of those physical-ending sequences began with a psychiatric diagnosis. Among the suicide decedents, 62 percent had been exposed to at least one significant sequence, with a median of 16 such sequences per case. The median duration of a significant sequence was 18.7 months, and the median time from completing a sequence to death was 13.1 months, suggesting a window in which the accumulating pattern might be visible in routine claims data.

Landmark analyses illustrated how ordering changes risk interpretation. Among 26,356 patients with depression who later developed suicidal ideation, the path that ran through anxiety and then anemia was associated with a higher hazard of subsequent suicide death than other paths from depression to suicidal ideation, with an adjusted hazard ratio of 4.6 (95 percent confidence interval, 2.2 to 9.5). The anxiety-only path, by contrast, was not significantly associated, with an adjusted hazard ratio of 1.3 (95 percent confidence interval, 0.8 to 2.1). In a second landmark analysis of 149,215 patients with anxiety who later developed hypertension, the path that included a history of self-harm carried an adjusted hazard ratio of 32.0 (95 percent confidence interval, 16.6 to 61.6) for subsequent suicide death compared with other paths from anxiety to hypertension.

The authors conclude that the temporal ordering of clinical conditions may carry prognostic information for suicide death, and that trajectories incorporating physical illness within psychiatric sequences identified higher-risk groups. Because many significant sequences ended in physical conditions, the findings suggest that opportunities for risk detection may extend beyond psychiatric settings: a patient presenting with anemia or hypertension may be partway along a trajectory that began with a psychiatric diagnosis. The researchers also note that suicide risk signals may be fragmented across care settings and not apparent within any isolated encounter, which is precisely the kind of pattern that sequence mining across longitudinal claims is designed to surface.

The study, published open access in BioData Mining, is a retrospective observational analysis of insured Maryland residents, so its findings describe associations in that population and claims-based diagnostic records rather than established causal mechanisms or validated clinical screening criteria. Hazard ratios from the Cox models reflect relative hazard within the studied cohort, and the authors’ shared version is subject to further editorial processing before the final version of record. Even so, the scale of the cohort, the mortality linkage through medical examiner records, and the consistency of the ordered-path signals make a case that when a patient’s diagnoses are read as a sequence rather than a list, some suicide risk that is invisible encounter by encounter becomes visible across years of care.

Subject of Research: Temporal sequence mining of linked claims and mortality data to identify ordered clinical diagnostic trajectories associated with suicide death

Article Title: Identifying clinical diagnostic trajectories associated with suicide death using temporal sequence mining of linked claims and mortality data

Article References: Belouali, A., Kitchen, C., Haroz, E., Lehmann, H., Nestadt, P. S., Wilcox, H. C., & Kharrazi, H. (2026). Identifying clinical diagnostic trajectories associated with suicide death using temporal sequence mining of linked claims and mortality data. BioData Mining. https://doi.org/10.1186/s13040-026-00608-3

Image Credits: AI Generated

DOI: 10.1186/s13040-026-00608-3

Keywords: suicide risk, temporal sequence mining, claims data, diagnostic trajectories, Cox models, psychiatric diagnosis, physical illness, Maryland, ICD-10-CM, observational study, risk detection, BioData Mining