1. The adoption of medical artificial intelligence (AI) technologies has increased exponentially over the past few years, both in terms of total claims and the development of new current procedural terminology (CPT) medical AI procedures.
2. The use of medical AI technology is shown to be largely limited to urban areas and/or those with an academic hospital and is enriched in higher income areas.
Level of Evidence Rating: 2 (Good)
Summary of the study: Advances in medical AI technologies have increased rapidly, especially with the development of new large language models that provide generative capabilities. Prior to approval, device manufacturers are required to submit detailed safety and efficacy data to the Food and Drug Administration (FDA). Yet, after approval, there is scant evidence of how widely these technologies are used and what factors drive certain technologies to be adopted by specific communities.
This study aimed to characterize the adoption of medical AI devices by using a dataset of medical and pharmaceutical claims submitted to large commercial health insurance plans. The authors identified 16 billable AI medical procedures under a CPT code covering a wide range of technologies in various health specialties. Among all medical AI procedures, only a handful make up the majority of claims data.
In addition to examining adoption rates over time, the authors used zip codes associated with the claims data set to determine relationships between the use of medical AI products and demographics. Not surprisingly, there was a strong association between having an academic hospital in a zip code and the odds of using an AI medical procedure. Other factors that strongly predicted the use of medical AI included zip code median income and whether the zip code was classified as metropolitan. These analyzes suggest limitations to the adoption rate of AI in healthcare, perhaps primarily driven by structural and societal factors.
Click here to read the study in NEJM AI.
Related reading: How artificial intelligence medical devices are evaluated: limitations and recommendations from an analysis of FDA approvals
Thoroughly [cross-sectional study]: This study uses a large dataset of insurance claims to determine the variability over time and socioeconomic factors associated with the adoption of medical AI technologies. The authors examined CPT codes from the IQVIA PharMetrics Plus dataset for MedTech, consisting of approximately 11 billion claims from January 1, 2018, to June 1, 2023. The authors first examined 16 medical AI procedures based on 32 unique CPT codes and found overall exponentially increasing use of these codes over the dates they studied.
The authors then performed multivariate logistic regression to determine whether at least one occurrence of AI CPT code billing was associated with certain factors that could be extracted from the zip code. They categorized ZIP codes as having high median income (>$100,000 median annual household income) versus low, whether the ZIP code was metropolitan (by the USDA), and whether it had at least one academic hospital center (as defined by the American Medical Association colleges). Indeed, the authors found strong statistical significance (p < 0.001) for all three predictors, whereby those zip codes with an academic hospital were 17 times more likely to have had a medical AI procedure billed in the era they studied.
In conclusion, this study carefully assesses changes over time and geographic distribution of the use of medical AI technologies and highlights demographic barriers to wider adoption.
Image: PD
©2023 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without the express written consent of 2 Minute Medicine, Inc. Licensing inquiry here. No article should be construed as medical advice and is not intended as such by the authors or 2 Minute Medicine, Inc.
