Data Source and Variables

We abstracted demographic, clinical, and utilization data from the UCDCH EHR. It is important to note that during the years 2009 to 2017, no other hospitals in the communities included in our analyses had pediatric inpatient wards with the clinical capacity to care for children with neurologic conditions. In addition, we limited our analyses to patients in northern California, where there were no practicing pediatric neurologists using telemedicine outside of UCDCH. Therefore, it is a reasonable assumption that if pediatric patients cared for by UCDCH pediatric neurologists in the telemedicine or in-person clinics required hospital services, they would be transferred or admitted directly to UCDCH.

Sex, insurance status, and patient addresses were assumed to stay constant throughout the study period, and their values were designated as those recorded in the EHR at the time of data extraction. Insurance status was dichotomized into private (commercial employer-based) and nonprivate, which included public insurance (eg, Medicaid or managed Medicaid), self-pay, and no insurance. Addresses were geocoded and mapped to US Census tracts. Aggregate Census tract information was used to assign patients’ neighborhood median household income and education level (defined as the proportion of residents with a bachelor’s degree or higher) using the 2016 American Community Survey’s 5-year estimates.¹⁹ Geocoded addresses were used to estimate patients’ travel times to their outpatient neurology clinic (ie, time needed to travel from the patient’s home to the remote telemedicine clinic for the telemedicine cohort or to UCDCH for the in-person cohort), as well as patients’ travel time to UCDCH. Travel times were estimated using a proprietary geolocation application programming interface to compute travel distance and travel time between 2 points defined by their geographic coordinates, assuming motor vehicle speeds under standard traffic conditions.²⁰

International Classification of Diseases, Ninth Revision and International Classification of Diseases, Tenth Revision codes for up to 5 encounter diagnoses were used to determine whether the hospital encounter was related to a neurologic condition using manual review of codes and applying previously published criteria.^{21,22,23,24,25,26,27} Neurologic conditions were grouped into clinically relevant categories for comparison between the cohorts. Because most patients included in the study did not have any hospital encounters, we also compared patients’ neurology clinic diagnoses between the cohorts. In addition, International Classification of Diseases, Ninth Revision and International Classification of Diseases, Tenth Revision diagnosis codes recorded during hospital and/or clinic encounters were used to determine whether the patient had a complex chronic condition using a previously validated algorithm.²⁸