Smartphones have quite literally placed the world at our fingertips. Hence, channelising technological interventions naturally comes up in any discussion on scale and efficiency of public health work as well. A somewhat new entrant to this discussion is Artificial Intelligence (or AI), which could have a positive impact on the Indian healthcare scenario, which faces an acute shortage of human resources, variable accessibility of health services, and rising Out Of Pocket (OOP) expenditure for patients. Some tout AI as the tool required to take healthcare away from a reactive system onto a preventive one. However, a closer look at the application of AI in healthcare shows that at present its use is limited to clinical settings that are more sophisticated and less accessible.
In India, tertiary and secondary care units consisting of specialised services and advanced diagnostics are easily distinguishable from Primary Healthcare Centres (PHCs). These centres and their extended arms, the Community Health Workers (CHWs), are the cornerstone of last mile service delivery and serve as the first level of contact between the health system and patients. However PHCs are understaffed, with insufficiently-trained CHWs and retrofitted with inadequate infrastructure.
The closest that technology has gotten to addressing these challenges is to equip CHWs with tablets and apps that help them collect data for district, State and national level analysis, but don’t serve to streamline work.
However, most AI investments in healthcare are limited to the higher-end tertiary care units. AI startups deal with niche areas like regenerative medicine, radiology-based diagnostics, advanced screening and testing and so on. The more widely used AI solutions involve the use of Natural Language Processing (NLP) chatbots to connect urban patients with doctors. Other than this, AI is being used for expediting clinical trials and drug discovery. However, none of these solutions is rapidly transforming primary health systems.
Besides the predictive and classification functions that AI can perform through machine learning, AI has imaging, identification and screening abilities, natural language processing and speech recognition.
Many of these abilities can be leveraged for primary health too. For example there are AI birth weight measurement tools, AI-powered heat mapping to screen for breast cancer. However, while the former is still in development, the latter has faced hurdles in truly injecting itself into public health.
There are still many CHW and PHC functions that can be made more effective using AI. For instance, PHCs perform routine laboratory tests for haemoglobin screening, urine testing, blood grouping for expectant mothers — many of which can be prime use cases for AI solutions for the frontline. Moreover, now the Government aims to upgrade PHCs to Health and Wellness Centres (HWCs) with an expanded mandate of screening for non-communicable diseases like cancer, diabetes, cardiovascular ailments as well. Traditional methods of testing require expensive equipment and specialist human technicians, yet are unable to be highly accurate. This is an open sesame for many AI innovators to develop quicker technology- enabled ways of testing and screening.
A recurrent woe at the frontlines is the demand for enhanced efficiency and AI could potentially be the answer here. Effective triage tools like Babylon, rolled out by the UK’s National Health Service, can help streamline referrals and optimise the use of the limited human resources on cases that most require their attention. Chatbots can help solve superfluous or simple queries and refer most urgent cases first. Similarly, studying existing patterns of dropouts from care management and developing predictions for patients at highest risk of dropouts can perhaps take us further on the quest for adherence management in areas like TB, antenatal care and HIV.
The good news is that AI for frontline health work can be made possible. The Government’s insistence on digitising paper-based records into web portals has two outputs, viz. inducing a comfort with using basic technologies like tablets by CHWs and the ready availability of immense service delivery data. The issues, however, are data gaps and inaccuracies caused by transferring data from, for instance, an ASHA’s diary to the portal by a third person called the Data Entry Operator. As data changes hands, there is a greater risk of inaccuracies creeping in. Even when data is of good quality, the Government is still deliberating on the means to share it with private innovators while preserving privacy.
These challenges are acknowledged quite often by innovators but they forget to consider that most AI applications might not require that much data. Deep diving to understand patterns in existing data aside, easy screening tools like haemoglobin and blood groups have vast clinical literature present. In the presence of a defined clinical model, the task that remains is just training the algorithm on limited data sets to automate the model. As and when more data is collected through a prototype, the algorithm can be trained again to achieve better results. Here, the volume of reliable data ceases to be an inhibitor for innovation at the frontline.
Advanced clinical AI applications, like those functioning as decision support systems for doctors, are very often operating in a regulatory grey area due to lack of appropriate certification guidelines. Clinical applications require the highest accuracy and perhaps will face the most stringent liabilities (when such laws are developed). Most primary healthcare applications, though, are not clinical in nature. Most of them are routine classifications, administrative tools, and chatbots to enable healthier lifestyles — outside the purview of clinical certifications. Even screening tools on frontlines are basic triage tools meant to facilitate referrals and not to be the final diagnosis. The expectation of accuracy is lower in the face of the utility it can bring in areas where doctors aren’t even available.
Therefore, while private innovators might be not be interested in frontline healthcare solutions, this vacuum can be filled by civil society organisations, the Government, academia and social enterprises. Most private firms are now boasting their own AI for social good initiatives, mobilising forces to develop proofs of concept for specific social problems. These, too, can and should explore opportunities for using technologies to build a stronger base of public health by working extensively on frontlines. Once proven, these solutions can be adopted by the Government.
Technology to the front line worker is no longer a miracle solution to address problems of inaccuracy, inconsistency and lack of efficiency. If anything, it’s made the role of the CHW more data-centric, where we find them spending as much as half their work hours reporting on the services they deliver in order to make trends of service delivery more understandable to a District/Block Officer.
Here, the obligation on innovation should be to marry user-centricity (focussing on the CHW) with greater efficiency (focussing on the system). Both the CHWs and the patients would need to develop a certain level of trust in technology being able to perform a task they do, and trust in the results thereof, respectively. This would involve a higher emphasis on developing explainable AI and integrating solutions with workflows of CHWs.
The potential of AI in revolutionising India’s healthcare system is slightly exaggerated. There is immense hope generated with the idea that CHWs in two different parts of the country perform the same task of measuring a newborn baby’s weight with the same accuracy, followed by their District Officer receiving a system-generated alert on failing nutrition metrics in their geography, leading to resource-efficient health service delivery that is timely — all made possible on a smartphone. But, it’s wise to be cautious about the stage in the value chain at which innovation is introduced and how its influence can burden the user.-Daily Pioneer