Abstract
•Suicidality phenotypes are complex, leading to numerous research definitions.•Heterogeneity in suicidality definitions hampers consortia research efforts.•Expert consensus ICD definitions are presented for suicidal ideation and attempt.•Real world data shows broad variation in ICD performance for suicidality.•ICD codes have poor sensitivity but good specificity for suicidality.
Suicidality, including suicidal ideation (SI), attempt (SA), and death (SD), represents complex and partially overlapping phenotypes that are moderately heritable. Suicidality definition heterogeneity impedes data replication and consolidation efforts by research consortia needed to address the sample size requirements of genetic research. The standardization of suicidality definitions would improve comparability of data across groups but has been insufficiently addressed in existing literature. Here, the Suicide Workgroup of the Psychiatric Genomics Consortium (PGC) provides International Classification of Disease (ICD) definitions and validation in real-world data for SA and SI.
The PGC Suicide Workgroup used published definitions coupled with expert consensus to develop ICD lists to serve as suicidality phenotype definitions. One SI and two SA lists were produced and evaluated for performance, including via sex stratification, against patient screening responses in multiple independent cohorts (total N = 21,772) with differing ascertainment strategies.
ICD code lists for suicidality component definitions were produced. SA ICD lists versus patient responses showed sensitivity of 15.4 % to 71.1 %, specificity of 67.6 % to 96.3 %, and positive predictive values of 0.57–0.92. SI ICD code performance versus patient report also varied in sensitivity (29.4 %-86.1 %), specificity (64.2 % to 90.6 %), and positive predictive values (0.67 to 0.98).
Lists of applicable ICD codes for SI and SA were developed that complied with C-SSRS definitions. Real-world application of ICD codes can vary substantially, perhaps dependent on clinician training and on cohort characteristics. Consistent training in use of ICD codes between sites may improve comparability of data sets.