Emergency Medicine Practitioners Overall Well-being Enhancement With Ambient AI Scribes

We will fit a linear model to describe the effect of Ambient tool introduction on our co-primary outcomes under the intention-to-treat (ITT) framework with a random effects structure to describe within provider variability. While we anticipate high or complete survey completion rates, in the event that not all surveys are completed and returned, we will perform a primary analysis on completed surveys only, and perform sensitivity analyses accounting for potentially systematic survey non-response bias using a response weighting strategy, using provider, scheduling, and patient encounter characteristics to create survey response weights (within each survey time period), then reweighting observations to account for non-response patterns. In analyses for both co-primary outcomes, we will perform Wald type hypothesis tests for inferences on the overall Ambient treatment effect and compare p-values to 0.05 / 2 to conservatively account for multiple comparisons using Bonferroni's method.

In exploratory analyses for all secondary outcomes, we will fit Bayesian generalized linear models with appropriate link functions and likelihood choices (ie, log and Poisson for count data, logit and Bernoulli for binary responses) to for each encounter, while accounting for serial correlation overall through time (indexed to the day) using an autoregressive modeling structure, within provider patterns using a provider level random intercept, weekday, weekend, and holiday effects using fixed effects terms, and specify a flexible interrupted time series treatment effect using a semi-parametric, thin plate spline to describe how the introduction of Ambient AI affects the relationship between providers and notes, and how that relationship evolves over time. In sensitivity analyses, we will additionally evaluate how these effects and their evolution may differ by treatment wave. All inferences will be based on describing the mean and 95% credible intervals for the treatment effect daily.