Tracheostomy and Long-term Ventilator Dependence (LMTV) Computable Phenotype

Patients with LTMV are a small, heterogeneous population who are at significant risk of morbidity and mortality. They place an enormous stress on the healthcare system and caregivers, requiring lengthy hospital stays and frequent readmission. Research efforts to improve understanding of the population and improve patient phenotyping, better stratify risk, and personalize care based on these features is limited by the small population size and heterogeneity at a single center. Thus, there is a need to develop multisite collaboration to improve potential to learn from variation and test interventions in this population. 
 
PEDSnet is a network of large children’s hospitals which facilitates sharing of clinical data derived from the electronic health record for all patients seen in the outpatient, inpatient and emergency department settings. The network population is biased toward sicker, more complex patients and so children with LTMV dependence would be well-represented and receive the majority of their clinical care within the network institutions. The network has high quality observational data shared at regular intervals to describe much of the relevant care in the population including diagnoses, medications, encounters, procedures, and vital signs. However, identification and validation of the LTMV population has never been performed in PEDSnet. 
 
We hypothesize that children with LTMV dependence can be reliably identified within the PEDSnet data. Additionally, we anticipate that the population will show significant variation in healthcare utilization represented as “days away from home” both across the population and at the site level. We plan to test these hypotheses with the following aims: 
 
Aim 1: Develop a computable phenotype for LTMV dependence using diagnosis codes and validate that phenotype at 5 network sites (CCHMC, Colorado, Lurie, Seattle, Stanford). 
We will use a previously validated, sensitive computable phenotype for PHIS data and apply it to the PEDSnet data. Patients meeting criteria will be re-identified at the site level and a clinical collaborator at each site will verify that the child meets inclusion criteria. The computable phenotype will then be revised to achieve a goal sensitivity and specificity of >90%. 
 
Aim 2: Describe variation in “days away from home” across the network population and at the site level controlling for key covariates including age, sex, race/ethnicity and comorbidities.  
For the identified cohort, we will evaluate the total hospital days and days with outpatient encounters to develop a “days away from home” count where the patient does not require interaction with the healthcare system. We will then describe the variation in this metric stratified by site and by the covariates described. 
 
These aims will demonstrate the potential utility of PEDSnet to accurately identify and describe key outcomes in this high-risk population across a large number of sites. Future work will focus on augmenting PEDSnet data with additional key outcomes in the population and developing interventions to reduce variation and improve outcomes across sites.