JMIR Publications

ABSTRACT

Background:

Tracheostomy is a frequently performed procedure in critical care settings, but procedures are often inconsistently coded in electronic health records (EHRs), with explicit designation as elective or emergency frequently absent. This coding ambiguity limits the ability to identify planned tracheostomy cohorts for observational research on outcomes and time toxicity. Common data models such as the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) enable large-scale federated research, but require validated computable phenotypes to ensure reliable cohort identification across heterogeneous data sources.

Objective:

To develop and validate a computable phenotype that identifies elective tracheostomy procedures from EHR data standardized to the OMOP CDM, enabling scalable and reproducible analysis of tracheostomy-related time toxicity in critically ill patients.

Methods:

We conducted a retrospective observational study using EHR data from the Johns Hopkins Health System from 2017 to 2024, comprising approximately 2.1 million patients with data mapped to the OMOP CDM. A series of cohort definitions were developed using standardized clinical code sets (International Classification of Diseases, 10th Revision [ICD-10] and Current Procedural Terminology [CPT]) from the Observational Health Data Sciences and Informatics (OHDSI) Standardized Vocabularies. To classify tracheostomy procedures lacking explicit urgency coding, we compared covariate prevalence and temporal relationships (e.g., intubation timing relative to tracheostomy) between explicitly coded elective and emergency cohorts. Six candidate computable phenotypes with stepwise inclusion and exclusion criteria were evaluated using PheValuator, a validated probabilistic phenotype evaluation tool.

Results:

Among 3552 patients with a tracheostomy procedure identified between 2017 and 2024, 2484 (69.9%) were explicitly coded as elective and 107 (3.0%) as emergency; the remaining 961 (27.1%) lacked explicit urgency classification. Covariate analysis revealed significant differences in intubation timing, drug exposures, and procedure codes between the explicitly coded groups. The best-performing computable phenotype (Cohort #202), which used inpatient visit-based attribution of planned and emergency codes, achieved a sensitivity of 0.88 (95% CI 0.84-0.91) and a positive predictive value (PPV) of 0.81 (95% CI 0.77-0.84), with an F1 score of 0.84.

Conclusions:

The proposed computable phenotype effectively distinguishes elective from emergency tracheostomy in structured EHR data. This approach enables large-scale, reproducible studies of tracheostomy-related time toxicity across heterogeneous OMOP-mapped data sources and provides a generalizable framework for phenotyping intent-ambiguous procedures across federated research networks.