JMIR Publications

ABSTRACT

Background:

Tuberculosis (TB) is a major global health concern, causing 1.5 million deaths in 2020. Diagnostic tests for TB are often inaccurate, expensive, and inaccessible, making chest X-rays (CXRs) augmented with Artificial Intelligence (AI) a promising solution. However, whether providers are willing to adopt AI is not apparent.

Objective:

The study seeks to understand the attitude of AYUSH (Ayurveda, Yoga and Naturopathy, Unani, Siddha, and Homoeopathy) and informal healthcare providers, who we jointly call AIPs, towards adopting AI for TB diagnosis. We chose to study these providers as they are the first point of contact for a majority of TB patients.

Methods:

We conducted a cross-sectional survey of 406 AIPs across the states of Jharkhand (162 participants) and Gujarat (244 participants) in India. We designed the survey questionnaire to assess the AIPs’ confidence in treating presumptive tuberculosis (TB) patients; their trust in local radiologists’ reading of the CXR images; their beliefs regarding the diagnostic capabilities of AI; and their willingness to adopt AI for TB diagnosis.

Results:

After removing incomplete responses, we found that 93.7% (270/288) of AIPs believed that AI improved the accuracy of TB diagnosis, and 69.4% (200/288) were willing to try AI. AIPs who were more confident in diagnosing TB and had greater trust in the local radiologists were more likely to believe in AI-based TB diagnosis. However, we found significant differences in AIPs’ willingness for AI adoption across the two states. Specifically, in Gujarat, a state with better and more accessible healthcare infrastructure, 73.4% (155/211) were willing to try AI; and in Jharkhand, 58.4.% (45/77) were willing to try AI. Moreover, AIPs in Gujarat who showed higher trust in the local radiologists were less likely to try AI. In contrast, in Jharkhand, those who showed higher trust in the local radiologists were more likely to try AI.

Conclusions:

While most AIPs believed in the potential benefits of AI-based TB diagnosis, many did not intend to try AI, indicating that the expected benefits of AI measured in terms of technological superiority may not directly translate to impact on the ground. Improving beliefs among AIPs with poor access to radiology services or those who are less confident of diagnosing TB is likely to result in greater impact of AI on the ground.