The Value of Artificial Intelligence in Laboratory Medicine

Ketan Paranjape, Michiel Schinkel, Richard Hammer, Bo Schouten, Rishi Panday, Paul Elbers, Mark H.h. Kramer, Prabath W.b. NanayakkaraAmerican 2020 · DOI

TL;DR

This paper surveys stakeholders in laboratory medicine to gauge their understanding and adoption of Artificial Intelligence (AI). While many see AI's potential value, there's a significant knowledge gap and a strong need for education and clearer implementation strategies to overcome adoption barriers.

What Problem Does This Paper Address?

The field of laboratory medicine is rapidly digitalizing and automating, leading to an increasing integration of Artificial Intelligence (AI). However, there's a perceived lack of understanding among clinical laboratorians regarding what AI is, its capabilities, limitations, and how to effectively implement it. This gap hinders the adoption of AI, despite its potential to improve diagnostics, streamline workflows, and reduce healthcare costs. The paper aims to assess the current perspectives on AI's value in laboratory medicine and identify the anticipated challenges and solutions for its implementation.

Methods

The study employed a mixed-methods approach, combining a web-based survey with qualitative analysis.

  • Survey Development: A 21-question survey was designed through an iterative process involving an online discussion board with key stakeholders and subsequent online group chats for refining questions and answer options.
  • Data Collection: The survey was distributed via email to 302 laboratory medicine practitioners who were part of Roche's Strategic Advisory Network (SAN). Data was collected using the Confirmit software platform.
  • Quantitative Analysis: Categorical data was analyzed using Pearson's chi-squared test. Perceived value of AI was compared across different age groups and experience levels with AI. Microsoft Excel was used for analysis, and multiple-choice data was presented as percentages.
  • Qualitative Analysis: An inductive approach of direct content analysis was used for open-ended questions. Two researchers independently screened answers, drafted a framework of themes, and resolved discrepancies through consensus.

Datasets

The study did not introduce new datasets or benchmarks. Instead, it collected data from a survey administered to 128 participants from Roche's Strategic Advisory Network (SAN).

Key Insights and Ideas

  • AI Hype vs. Reality: A significant portion of the medical community perceives AI as valuable, but this perception seems to be driven more by general hype than deep understanding. Many respondents are unsure about the specific reasons for AI's value or how to comfortably adopt it.
  • Education is Paramount: The survey strongly indicates a critical need for AI education across all levels of laboratory medicine professionals. This education should cover the technology itself, its practical applications, and how to integrate it into existing workflows.
  • Implementation Challenges are Real: High investment costs, a lack of proven clinical benefits, and the complexity of decision-making processes are significant barriers to AI adoption.
  • Patient-Centricity is Key: While the focus is on AI's impact on laboratory processes, the ultimate goal is patient benefit. The paper touches upon the need to consider the patient's perspective and how AI's role might be communicated.
  • Gradual Integration is Preferred: A promising strategy for AI implementation involves introducing new AI tools alongside existing ones, allowing practitioners to gain comfort and experience the added value before fully relying on them.

Key Results

  • Current AI Use: Only 15.6% of surveyed organizations currently use AI in laboratory medicine.
  • Future Adoption: A substantial majority (66.4%) believe they might use AI in the future.
  • Perceived Value: Approximately 81% of participants felt AI would be valuable in their organization within the next 5 years.
  • Barriers to Adoption: The most frequently cited barriers included high investment costs, lack of proven clinical benefits, the number of decision-makers involved, and privacy concerns.
  • Solutions for Adoption: Key solutions identified were education on AI's value, streamlined implementation and integration into workflows, and research to prove clinical utility.
  • Demographics: The modal age group of respondents was 41-50 years (32.0%), and the top roles were physicians (26%), laboratory managers (22%), and pathologists.

Limitations and Future Work

  • Generalizability: The study acknowledges that the participants did not represent the entire population of practitioners who will use AI in a diagnostic setting, limiting the generalizability of findings to all laboratory medicine practitioners across various settings.
  • Sample Size: The relatively small study population (n=128) might have limited the statistical power to detect significant differences in quantitative analyses.
  • Future Work: The paper suggests a continued need for AI education in medical communities and emphasizes the importance of companies providing extensive training for their AI tools. Further research into clinical evidence and implementation strategies is also recommended. The authors also highlight the need for a similar approach to patient communication regarding AI as used for other established algorithms.

Summary updated

Abstract

This survey demonstrates that specific knowledge of AI in the medical community is poor and that AI education is much needed. One strategy could be to implement new AI tools alongside existing tools.
BibTeX
@inproceedings{paranjape2020value,
  author = {Ketan Paranjape and Michiel Schinkel and Richard Hammer and Bo Schouten and Rishi Panday and Paul Elbers and Mark H.H. Kramer and Prabath W.B. Nanayakkara},
  title = {{The Value of Artificial Intelligence in Laboratory Medicine}},
  booktitle = {American},
  year = {2020},
  doi = {10.1093/ajcp/aqaa170},
}
Figure