Diversity in people's reluctance to use medical artificial intelligence: Identifying subgroups through latent profile analysis

Haixia Wang, Qiaoqiao Sun, Li Gu, Kaisheng Lai, Lingnan HeFrontiers Artif. Intell 2022 · DOI

TL;DR

This paper investigates why people are hesitant to adopt medical AI, moving beyond general explanations to identify distinct subgroups with varying knowledge, attitudes, and behavioral intentions. By using latent profile analysis, the study reveals three key profiles, offering a more nuanced understanding of user reluctance and paving the way for targeted interventions.

What Problem Does This Paper Address?

The increasing integration of Artificial Intelligence (AI) into healthcare promises significant advancements in diagnosis and treatment. However, widespread adoption of medical AI faces resistance from the public. Previous research has often examined this reluctance through a "variable-centered" approach, looking at individual factors like knowledge, attitudes, and behaviors in isolation. This overlooks the possibility that specific combinations of these factors might define distinct groups of people with unique reasons for their hesitancy. This paper addresses this gap by employing a "person-centered" approach to identify and characterize these subgroups, aiming for a more comprehensive understanding of medical AI reluctance.

Methods

The core methodology employed is Latent Profile Analysis (LPA). LPA is a statistical technique used to identify unobserved subgroups (latent profiles) within a population based on patterns of observed variables. In this study, LPA was used to group participants based on their scores across four key dimensions: objective knowledge of medical AI, subjective knowledge of medical AI, negative attitudes toward medical AI, and behavioral intentions regarding medical AI use.

The researchers conducted two studies:

  • Study 1: An inductive approach to establish initial KAB (Knowledge, Attitude, Behavior) profiles of medical AI.
  • Study 2: Aimed to replicate the profiles found in Study 1 and further explore potential antecedents (predictors) of these profile memberships, such as trust and risk perception.

Model fit was assessed using various indices like AIC, BIC, and LMR, with a focus on finding a theoretically meaningful and statistically robust solution.

Datasets

The study utilized data collected from two online surveys:

  • Study 1: 328 participants.
  • Study 2: 388 participants.

The participants were recruited via convenience sampling. The paper does not introduce new benchmark datasets; rather, it uses survey data to analyze existing attitudes and knowledge regarding medical AI.

Key Insights and Ideas

  • Heterogeneity in Reluctance: The most significant insight is that reluctance to use medical AI is not monolithic. Instead, it is characterized by distinct subgroups with different combinations of knowledge, attitudes, and behavioral intentions.
  • Beyond Simple KAB: While the Knowledge-Attitude-Behavior (KAB) model is a useful framework, the study suggests its limitations in fully explaining complex phenomena like medical AI adoption. The identified profiles sometimes deviate from simple KAB predictions, indicating that the relationship between these factors is more intricate in the context of AI.
  • The "Sweet Spot" Profile: Profile 2, characterized by moderate knowledge, low negative attitudes, and high behavioral intention, emerged as the largest group in both studies. This suggests a segment of the population that is open to medical AI, provided certain conditions are met.
  • The Paradoxical Profile: Profile 3 presents an interesting paradox: high objective and subjective knowledge of medical AI, yet also the highest negative attitudes. This highlights that knowledge alone does not guarantee acceptance and that other factors, like perceived risk or trust, play a crucial role.
  • Predictive Power of Trust and Risk: Study 2 indicated that trust perception and risk perception are significant predictors of an individual's KAB profile membership, underscoring their importance in shaping attitudes towards medical AI.

Key Results

The study consistently identified three distinct profiles of medical AI reluctance across both studies:

  • Profile 1: Characterized by the lowest objective and subjective knowledge of medical AI, a middle level of negative attitudes, and the lowest behavioral intention. This group represents those who are least informed and least inclined to use medical AI.
  • Profile 2: Exhibited moderate objective and subjective knowledge, the lowest negative attitudes, and the highest behavioral intention. This is the largest profile, representing individuals who are generally open to medical AI.
  • Profile 3: Showed high objective and subjective knowledge, but also the highest negative attitudes and a middle level of behavioral intention. This group is knowledgeable but wary.

In Study 2, analysis of antecedents revealed that trust perception of medical AI significantly predicted being in Profile 2 (high openness) compared to Profile 1 (low openness). Conversely, risk perception of medical AI predicted being in Profile 1 or 3 (less open) compared to Profile 2.

Limitations and Future Work

  • Generalizability: The study acknowledges that the findings are based on convenience samples and may not be fully representative of all populations, especially across different cultural contexts. Future research should explore these profiles in diverse cultural settings.
  • Dynamic Nature of Attitudes: The KAB profiles might change over time. The authors suggest using latent transition analysis to track shifts in these profiles.
  • Professionals vs. General Public: The study primarily focused on the general public. The authors suggest that for medical professionals, the interplay between objective and subjective knowledge, and their impact on attitudes, might differ and warrants further investigation.
  • Intervention Design: Future work could focus on developing targeted interventions for each profile to address their specific barriers and promote the adoption of medical AI.

Summary updated

BibTeX
@article{wang2022diversity,
  author = {Haixia Wang and Qiaoqiao Sun and Li Gu and Kaisheng Lai and Lingnan He},
  title = {{Diversity in people's reluctance to use medical artificial intelligence: Identifying subgroups through latent profile analysis}},
  journal = {Frontiers Artif. Intell.},
  year = {2022},
  doi = {10.3389/frai.2022.1006173},
}
Figure