Why Belief in AI Doesn’t Translate to Better TB Care
In 2020, over a million people worldwide lost their lives to tuberculosis (TB). While modern medicine has developed effective treatments, eliminating TB has proven far more complex.
The COVID-19 pandemic further threw a spanner in the works, disrupting diagnostic and treatment services globally. The World Health Organization (WHO) recorded a 3.6% rise in new TB cases per lakh population between 2020 and 2021, following two decades of an almost steady 2% annual decline.
The resurgence poses a serious threat to countries, especially low- and middle-income countries (LMICs), with an already fragile health care system that struggles with underreporting, underdiagnosis, and limited access to care for TB patients. For instance, in India, nearly 29% of the cases were either missed or undiagnosed in 2021. As a result, there remains a significant gap between the estimated and reported number of TB cases in the country.
Using tech to fulfil the promise of AI
Narrowing this gap through timely detection is critical to the WHO’s TB control and elimination strategy. In this context, artificial intelligence (AI) tools that can interpret chest X-rays (CXRs) for diagnosis offer promising support, especially for countries facing a shortage of skilled radiologists.
Recent technological advancements have enabled AI systems to match, or even surpass, the diagnostic capabilities of radiologists. Unlike human experts, these tools can be accessed online from anywhere. With just about one radiologist per lakh population, the advantage can be especially valuable in regions of India where access to radiology is limited.
However, despite its potential, the promise of AI rests on healthcare providers’ willingness to use it.
Belief vs willingness
The private sector in India is crucial in treating TB, as it diagnoses twice as many cases. However, it is fragmented and relies on informal healthcare providers (IPs) who often lack formal training. Alongside them are practitioners of Ayurveda, Yoga and Naturopathy, Unani, Siddha, and Homoeopathy (AYUSH), constituting 22.8% of the country’s formally trained medical practitioners. Collectively, we refer to these groups as AIPs.
Given their influence, it is crucial for AIPs to adopt AI systems for faster, cheaper, and more accurate TB diagnosis. But are they willing to do so?
We explored this in our recent survey-based study across two Indian states of Gujarat and Jharkhand, along with co-researchers Snehil Rayal, Navya Pratyusha Varsala, and Sirisha Papineni.
Out of nearly 300 AIPs surveyed, over 90% agreed that AI could reliably detect and improve the diagnostic accuracy of TB. Still, close to 30% said they would be unwilling to try AI. The data shows that most AIPs trust AI and many are ready to use it in their practice, but there’s a notable gap between recognising its benefits and actually being open to adopting it.
The difference in perspective
Some practitioners view AI as a threat that might replace them, while others consider it a complementary aid. Prior research, meanwhile, shows no clear consensus on this.
Our study reveals an important insight. The disconnect between belief and willingness depends on providers’ confidence in their diagnostic skills and trust in local radiologists.