Addressing Variability and Inequities via Systematic Evidence-based Recommendations (AdVISER) for Head and Neck Imaging - PROJECT SUMMARY/ABSTRACT Approximately 141 million medical imaging examinations are performed in the United States annually. Over 12% of these examinations result in a radiologist recommendation for additional imaging (RAI). However, RAI rates vary among peer subspecialists, radiologists often disagree with RAI generated by other radiologists, and RAI language is often ambiguous and not readily actionable. Such variation in decision patterns is present in other professions and other areas of medicine, hypothesized to be a by-product of our personal biases and unique approaches to judgement, and a major source of professional errors. Referring provider failure to pursue clinically necessary RAI or radiologist failure to make clinically necessary RAI (medical imaging underutilization) can lead to diagnostic error. Referring provider pursuit of clinically unnecessary RAI (medical imaging overutilization) diminishes diagnostic excellence by creating waste and can lead to diagnostic delays. Furthermore, sociodemographic disparities appear to exist in RAI creation and completion. Our goal is to reduce variation and sociodemographic disparities in RAI creation, referring provider agreement, and patient adherence. Our innovation is the use of a validated natural language processing algorithm to enable a large- scale, theory-informed, versatile audit and feedback process for clinical adherence to evidence-based and equity-informed best practices across a multi-site health system. We will focus on head and neck radiology, which has the largest inter-radiologist RAI rate variation we have observed. Aim 1: Identify sociodemographic disparities in head and neck imaging RAI. We will measure the extent to which RAI rates (potential radiologist bias), referring provider agreement rates (potential referring provider bias), and RAI examination completion rates (patient factors including patient preferences) differ by patient socioeconomic status, sex, race, ethnicity, insurance, and income. Aim 2: Describe RAI-related epidemiology and diagnostic error for three conditions: thyroid cancer, lung cancer, and canal dehiscence syndrome. We will establish the 5-year healthcare system incidence of these diseases; measure the association of radiologist RAI rate with detection rate, sensitivity, and specificity; and measure radiology-related diagnostic error (false negatives, misdiagnoses, and delays to diagnoses). Aim 3: Develop and locally disseminate best practices for the studied conditions. We will generate local clinical consensus best practices using published evidence as well as gathered evidence from Aims 1 and 2, with a Delphi approach that includes a multi-disciplinary team of experts and also patients. Aim 4: Reduce RAI variation through a peer learning audit and feedback intervention. We will measure change in RAI rates, RAI rate variation, and referring provider agreement. We will also assess change of disparities in RAI creation, agreement, and adherence, and measure change in RAI actionability. If effective, this study will serve as a model for the development and audited implementation of best practices that reduce variation, diagnostic error, and disparities in medical imaging care.
