Association Between Children With Life-Threatening Conditions and Their Parents’ and Siblings’ Mental and Physical Health

This cohort study examines whether family members of children with 1 of 4 types of life-threatening pediatric conditions have higher rates of health care encounters, diagnoses, and prescriptions compared with families of children without these conditions.

Family members were operationally defined as individuals covered via the policy holder.
The policy holder was the parent of either the case or the control child and lived at the same residential address. Family members of case children were identified as case parents (any age) and case siblings (0 to 19 years of age), and similarly for family members of the control children. For both case and control family members, their initial study day was inherited from the originating case child's initial study day. Among case families, if no family member had insurance coverage after the initial study day, then the case family and the matching control families were dropped. For control families, if no member had insurance coverage after the initial study day, then that control family was omitted and the case-to-control ratio adjusted.

Cohort observation interval
The cohort observation interval was from 1 July 2015 until 31 December 2017. Families had different lengths of observation time, due to timing of cohort entry and different durations of insurance coverage.

Specification of diagnoses, prescriptions, and healthcare encounters
Claims files during the cohort observation period included information regarding diagnoses, prescriptions, and healthcare encounters. All diagnoses were recorded as ICD-10-CM codes. ICD-10-CM codes in the range from F10 to F59, as well as F50 to F98 for siblings only, were classified as "mental health" diagnoses; codes in the range from S00 to T79 and V00 to Y38 were classified as physical "trauma" diagnoses; all other codes were classified as "physical health" diagnoses.
Prescription information included generic drug names. Compound generic drug names were separated into single generic drug component names. The list of all observed generic drug component names was then matched to the Anatomical Therapeutic Category (ATC) coding system via the RxMix API. 31 Drugs matched to ACT codes with prefixes of N05A, N05B (excluding N05BB), N05C, N06A, N06C (excluding N05CM), and N03AE were specified as "mental health" prescriptions, while the remainder were specified as "all other" prescriptions. Healthcare encounters in the claims data included categories for hospitalizations, emergency department (ED) visit, and urgent care visits.
For each cohort member during the observation period, we summed the total count of diagnoses, of prescriptions, and of healthcare encounters, within each of the categories described above.

Statistical Analysis
We specified four main overall hypotheses regarding whether mothers, fathers, sisters, and brothers of case patients, compared to control patients' family members, experience increased rates of a composite measure of healthcare use, diagnoses, and prescriptions, implementing separate models for each of the four types of family members. In planned sub-analyses, we also analyzed each of the three outcome types (healthcare use, diagnoses, and prescriptions) separately, and further sub-analyses within each of these outcome types. Finally, we examined differences in bereaved case parents compared to control parents. Multivariable negative binomial regression model with logarithm link function was used to estimate the incidence rate ratio (IRR) with 95% confidence interval (CI) of case individuals with respect to the control individuals, on the © 2021 Feudtner C et al. JAMA Network Open.
person-level count data for each of the four cohorts separately. All implementations of this model adjusted for individuals' duration of time observed, age, and the race/ethnicity category specified in the data source (which included a category of other or missing) and accounted for any within-family clustering of sibling observations. A Pvalue of 0.01 was designated as the threshold of statistical significance of the 4 overall hypotheses, and of 0.05 for the sub-analysis comparisons.
Of note, we viewed several possible occurrences as being part of the potential causal pathway between exposure to a child with a LTC and adverse health consequences for other family members. Specifically, if the child were to have died (most likely due to the LTC), the subsequent bereavement is one mechanism by which adverse health outcomes could occur. If a co-parent or one of the other children in the family developed a physical or mental health condition secondary to exposure to the child with the LTC, and this occurrence than had an additional adverse consequence on the co-parent or other siblings, this could also be part of the causal pathway. We therefore did not view these occurrences as confounding per se.
All analyses were performed with Stata version 16.1 (StataCorp LLC, College Station, TX).
This study followed the STROBE cohort checklist when writing our report. 51 The study was registered on ClinicalTrials.Org (NCT03971344) 32,33 ; discrepancies between what was registered and how the study was conducted are reported in Supplemental Table   B.