429 - Systematic Implicit Bias in Clinical Decision Instrument Development: A Quantitative Analysis of Clinically Available Instruments on MDCalc

Poster Number: 429
Publication Number: 429.3310

Background: Clinical decision instruments (CDIs) are widely used across medical specialties to improve patient care through data-driven standardization. CDIs are tools that use clinical data (e.g. signs, symptoms, laboratory values) and develop a reproducible strategy to aid in clinical decision-making. While CDIs enhance clinical care in a wide range of settings, implicit bias during the development of CDIs may inadvertently perpetuate inequality within healthcare systems. MDCalc is a globally recognized repository of clinically available CDIs.

Objective: In this study, we aim to assess the potential implicit bias inherent in developing clinically deployed CDIs. To create equitable CDIs for all populations, bias within CDIs must be recognized to determine the limitations from which they can be used to evaluate patients.

Design/Methods: We performed a secondary analysis of the primary literature using CDIs found in a well-recognized clinical repository, MDCalc, which has a global reach across 200 countries and garners over 1 million daily engagements. We included any CDI to retrieve the primary literature on PubMed, Embase, or Medline. We excluded CDIs in which the primary development literature was not in English or was exclusively a validation study. We conducted four quantitative analyses of implicit bias using patient demographics, investigator teams, CDI predictor variables, and CDI outcome definitions.

Results: We included 597 of 690 primary CDI development studies. For patient demographics, we found that most patients included in the development cohort were White, n (72.5%), and male, n (55.1%). For investigator teams, authors are concentrated in a few global regions, North America, n (52%), Europe, n (31%), and mostly identified as male, n (71%). For CDI predictor variables, we found 13 CDIs explicitly using Race/Ethnicity, 10 using Family History, and 79 using Sex and Gender which may introduce bias.

Conclusion(s): Overall, CDIs developed and shared on MDCalc may be at risk for implicit bias. We found that participant demographics, investigator demographics, predictor variables, and outcomes definitions are all sources of potential implicit bias. While CDIs are capable of streamlining decision-making for clinicians, we conclude that careful consideration of bias must be taken into account for the current usage and future development of CDIs. Future research should focus on identifying and warning clinicians of potential areas of implicit bias for CDI users.