A Comparative Effectiveness Trial of Strategies to Implement Firearm Safety Promotion as a Universal Suicide Prevention Strategy in Pediatric Primary Care

The primary goal of our parent study, the Adolescent Suicide Prevention in Routine clinical Encounters
(ASPIRE) trial, is to compare the effectiveness of an electronic health record (EHR) nudge alone to an EHR
nudge plus facilitation with respect to delivery of a secure firearm storage program during pediatric well-child
visits. We captured clinician-reported reach in the EHR for the entire patient population (~45,000 visits) and
obtained fidelity (i.e., parent-reported receipt of the program) from a subset of parents with whom we engaged
in an intensive sampling approach to collect a post-visit survey. Reach was documented by clinicians for a
large sample of patients and therefore does not suffer from the potential non-response bias that may affect
parent-reported fidelity. However, reach may suffer from measurement error if clinicians incorrectly reported
program delivery or exited out of the EHR prompt without responding to the questions. The proposed aims of
this supplement will develop statistical methods to combine information contained in each outcome (reach,
fidelity), effectively mitigating their respective limitations. The proposed methods will offer a powerful statistical
analysis tool for comparing the effectiveness of the implementation strategies using multiple sources of data
from a cluster randomized trial like ASPIRE. In Aim 1, we will derive a statistical estimator for comparing
interventions in a two-arm cluster randomized trial to addresses non-response bias in survey-based fidelity by
leveraging the large, complete sample provided by EHR-based reach, the latter of which may be subject to
measurement error. We will also propose methods to obtain p-values and confidence intervals for the
comparison of study arms. The methods will account for correlation due to inherent clustering of outcomes
(patients within providers within clinics) while controlling Type I error and maximizing power to detect a
difference between study arms. In Aim 2, we will use simulation studies to assess the statistical properties of
the proposed estimator and inferential methods (i.e., p-values, confidence intervals). We will study bias,
variability, Type I error, power, and coverage of 95% confidence intervals using simulations across a range of
design factors such as the level of parent non-response in fidelity, amount of measurement error in reach,
magnitude of the treatment effect, and intra-cluster correlation. In Aim 3, we will apply the proposed methods to
data from ASPIRE. We will apply our methodology to jointly analyze parent-reported fidelity and EHR-based
reach from the active implementation period of the ASPIRE trial. The methodology developed within our
supplemental work will provide increased accuracy and precision in analyses of important implementation
outcomes collected as part of the ASPIRE trial and will offer a new analytic framework for future cluster
randomized trials in implementation science.

Grant Number: 3R01MH123491-06S1
NIH Institute/Center: NIH
Principal Investigator: Rinad Beidas