ICTworks

Artificial intelligence (AI) has potential to drive game-changing improvements for underserved communities in global health. In response, The Rockefeller Foundation and USAID partnered with the Bill and Melinda Gates Foundation to develop AI in Global Health: Defining a Collective Path Forward.

Research began with a broad scan of instances where artificial intelligence is being used, tested, or considered in healthcare, resulting in a catalogue of over 240 examples. This broad catalogue of examples was distilled into a framework of 27 use cases, which were further grouped into four use cases, including:

• AI-enabled population health: Tools for ingesting, analyzing, and providing recommendations on population health data
• Physician clinical decision support: A tool providing specialized expertise to physicians, for example, by enabling a GP to read diagnostic images
• Frontline health worker virtual health assistants: Tools augmenting FHW expertise to direct patient care, such as triage and symptom-based diagnostics and care recommendations
• Patient virtual health assistants: Tools helping patients direct their own care and wellness, including data-driven diagnostics and recommendations

The analysis yielded the following insights across the four groupings of AI use cases:

• The AI-enabled precision public health use cases were identified as technologies that are relatively nascent and less widely used (in both high- and low-income contexts) than many other AI technologies—but that are well suited for global health and LMIC contexts in the future.
• The patient and frontline-health worker (FHW) groupings of AI use cases were identified as relatively more advanced technologies that are increasingly common in developed markets, but that need to be tailored and scaled in LMICs.
• Clinical decision support (CDS) AI tools are relatively mature technologies that have greater levels of penetration in high-income markets than other AI technologies in question, but that need greater scaling and adaptation in LMIC contexts. In addition, some of these CDS use cases are relatively less suitable for low-resource settings at present due to broader resource gaps in these markets (i.e. image-diagnostics tools that rely on expensive radiology equipment that is rare in LMICs).

AI-enabled population health

This grouping involves tools that leverage AI to monitor and assess population health, and select and target public health interventions based on AI-enabled predictive analytics.

It includes AI-driven data processing methods that map the spread and burden of disease while AI predictive analytics are then used to project future disease spread of existing and possible outbreaks.

It also includes risk management tools that use AI to better understand risk across different groups of a given population and stratify these groups according to risk levels.

One example of the potential of tools in this opportunity area is Artificial Intelligence in Medical Epidemiology (AIME), an AI-enabled platform that helps a country’s Ministry of Health predict future outbreaks of diseases like Zika and dengue in a specific geography months before their possible occurrence, and helps the Ministry select the most appropriate vector control method to prevent the outbreak.

While AIME is a very early stage venture and its technologies and tools have not been validated at scale, its work reflects the potential of AI predictive analytics tools in global health.

These AI-enabled population health tools can provide value to populations, governments, and health systems across LMIC contexts. These tools help governments better understand health burdens and potential disease outbreaks across their geographies, and thus enable them to allocate their resources more effectively to prevent and manage outbreaks.

These AI-enabled tools can also help diverse stakeholders (beyond a country’s MoH) determine which communities are most in need of care and public health interventions and to optimize their resources accordingly

Physician clinical decision support

This grouping includes AI use cases which support and improve the decisions of clinical physicians.

Examples of AI tools in this grouping are:

• Image-based diagnosis support for radiologists and pathologists
• Decision support tools for clinicians
• Quality assurance and training to provide insights for clinicians on past performance and indicate where errors may have been made.

Just as with the patient virtual health assistant use case, it is important to note that AI tools in this use case are not intended to replace the physician.

Overall, the AI use cases in this CDS grouping provide value by augmenting physicians’ roles and their capacity to serve their patients, helping them provide faster and more accurate diagnoses to patients, and widening a significant bottleneck in provision of care in LMIC contexts, enabling them to focus on their patients most in need of care.

Given the often extreme scarcity of health providers in LMIC contexts, and how overburdened providers in these contexts are, this function of helping doctors optimize their time and focus on those most at risk can save patients’ lives and provide catalytic impact across health systems.

As one clinician put it, AI-tools like this “give super powers” to health providers and can greatly improve the quality of care they provide to their patients.

Frontline health worker virtual health assistants

This grouping of use cases involves placing AI in the hands of frontline health workers (FHWs), enabling them to better serve—and bring top-notch medical technology and advice to—their patients.

FHWs in LMICs use AI-enabled tools to triage and diagnose patients (often outside of health facilities), to assist with clinical decision support, and to monitor compliance of their patients. Rapid and accurate triage and diagnosis functions are enabled when AI is applied to real-time patient data collected by FHWs. FHWs are then able to provide targeted health recommendations for patients on whether, where, and how to seek care.

Overall, the AI uses cases in this opportunity area provide value by strengthening FHWs’ abilities to serve their patients by providing health information and advice (and eventually even possibly diagnoses), without them having to visit a facility.

This, in turn, reduces the patient burden on already overburdened facilities, and enables FHWs to focus on their most at-risk patients, helping them to optimize their time and effort.

This grouping of tools illustrates how AI technologies can overcome prior constraints of access, cost and quality. Patients without easy access to health facilities and with little ability to pay may be able to better access quality health advice and avoid unnecessary trips to health facilities.

Patient virtual health assistants

The use cases in this opportunity area put AI in patients’ hands for:

• Self-referral
• Behavioral change
• Data-driven self-diagnosis
• Personalized outreach
• Medical record collection
• AI-facilitated self-care functions

Through the collection of real-time data at the patient level, these AI-enabled tools can help identify the type and severity of a patient’s condition and provide health recommendations directly to the patient.

Recommendations may include how and where to seek care if it is needed, or guidelines for self-care and behavioral changes to address health issues outside of the health system. It is important to note that these AI tools are not intended to replace humans in the provision of diagnosis and care.

Rather, these tools can provide helpful recommendations on if, how, and where someone should seek formal care from a health professional—and what they can do in the meantime to best manage the situation. This can be critical for patients who may have to wait days to see a doctor or reach quality care.

These AI use cases provide tremendous value to patients by enabling them to access medical information, behavioral and lifestyle recommendations, care routing advice, and even potential diagnoses without having to go to a health facility, which can be time-consuming and expensive in LMIC health systems.

They can also provide value and efficiency gains to the broader health system by ensuring that only patients who truly need to go to health facilities do so, freeing up health providers’ time for acute patients, and by remotely collecting ongoing patient data which can be linked to a patient’s broader medical record.

An example of a patient-facing AI tool reshaping healthcare in LMIC contexts is Babyl, a subsidiary of Babylon, operating in certain LMIC geographies such as Rwanda. Developed in the UK, Babylon provides an integrated AI platform for patients, including an AI triage symptom checker, health assessment, and virtual consultations with a physician when referral is needed.

In Rwanda, Babyl aims to build toward the AI-powered services available in the UK, but has started with virtual consultations, prescriptions, and lab tests through mobile phones, including special phone options to reach underserved populations.

The Babyl experience in Rwanda also provides broader value to the global health field by studying how AI-enabled technologies can be ported from developed country markets to LMIC markets.