This article is part two of a four-part series analyzing a new report on artificial intelligence (AI) in healthcare. Part two explains four groupings of AI use cases that experts believe will largely impact AI’s global impact on healthcare.
The use of artificial intelligence (AI) has significant potential to shift healthcare on a global scale. To determine the most efficient and impactful ways to leverage AI in this sector, a recent report—a collaboration between The Rockefeller Foundation and United States Agency for International Development’s (USAID) Center for Innovation and Impact, in conjunction with the Bill and Melinda Gates Foundation—evaluated and stratified AI use cases based on health quality, cost, access in low- and middle-income countries (LMICs), and potential scale/impact.
After grouping and prioritizing potential AI use cases, researchers established four AI use case groups they believe could have the most significant scale and impact in LMICs: AI-enabled Population Health, Frontline Health Worker (FHW) Virtual Health Assistant, Patient Virtual Health Assistant, and Physician Clinical Decision Support Tools. For each use case, the researchers shared examples of existing technologies or tools they believe are in the realm of possibility in the near future, along with illustrative profiles of individuals to show how these technologies could be applied to a real-world setting.
AI-enabled Population Health
While not considered a widely used practice compared to other AI technologies, the researchers believe AI-enabled Population Health could be beneficial globally and in LMIC contexts. It pertains to the use of AI to evaluate current population health data—including disease presence and burden—predict future outbreaks and the spread of existing ones, and ultimately make recommendations. Key focal points include:
- Surveillance and prediction
- Population risk management
- Intervention selection
- Intervention targeting
The report highlights Dhesi, a hypothetical worker for the Ministry of Health (MoH) in Malaysia who focuses on controlling infectious diseases. Before Dhesi implemented predictive population health tools, the MoH was unable to protect its population because there was no way to prepare for a disease outbreak before it happened. Access to health data from different regions was minimal, and non-health data could not be integrated into the existing health data.
To incorporate AI into its protective strategy, the MoH went digital, using electronic medical records (EMRs) so that nationwide data could communicate with one another; AI was then applied to the use of this data. By implementing AI, the MoH was able to identify nationwide patterns amid the complex data in order to target risk factors and, ultimately, help prevent diseases from spreading. Non-health data—weather patterns, wind speed, and roof angles—also proved helpful; natural language processing helped interpret critical information from news reports and social media.
By using AI in conjunction with EMRs, Dhesi, and the MoH were able to interpret and act on health data as situations were unfolding, as opposed to analyzing data after an event has already taken place. Predictive machine-learning (ML) algorithms also allowed the MoH to predict exactly when and where an outbreak would take place—three months beforehand; these algorithms also shed light on which interventions would have the greatest impact, and when and where to implement them.
Frontline Health Workers (FHW) Virtual Health Assistant
FHW virtual health assistant use cases were identified as advanced technologies that have made their way across more advanced locations but have not yet been augmented for use in LMICs. The goal is to allow FHW workers to more efficiently direct patient care, including triage and symptom-based diagnostics and care recommendations, outside of a healthcare facility. By accessing patient data in real time, FHWs can make timely recommendations for whether patients should seek care—and, if so, how and where they may do so. This technology benefits both parties, as patients do not have to travel to a facility, and FHWs can spend more time directly with their most critical patients. Key points in this use case group include:
- Personalized outreach
- Behavior change
- Data-driven diagnosis
- AI-facilitated care
- Medical records
To illustrate this use case, researchers introduce Anita, a community health worker in a rural Kenyan village, where Nairobi is six hours away, and the closest hospital is two hours away on dirt roads. Anita travels through her community selling basic health products and providing health advice to the locals. Anita uses apps on her smartphone to input symptoms patients are experiencing along with any other information she has on their condition. The apps, powered by AI, provide a diagnosis treatment advice, and health and self-care recommendations.
Anita’s AI technologies are applied to the case of two-year-old Eric, who has a rash and a fever; his mother isn’t sure what is wrong or whether it would be safe to travel two hours to the hospital. Anita is able to input Eric’s symptoms and a photo of his rash into one of her apps. Advanced ML algorithms determine that Eric does not have dengue, but he could have malaria and suggests that the rash could be a spider bite. An app instructs Anita to conduct a malaria rapid diagnostic test with her phone camera and a blood test. An app confirms the results of the test are negative for malaria, that Eric does not need to be referred for further treatment, and provides Anita with information to relay to Eric’s mother about when to bring Eric for treatment if his condition changes.
Without the AI apps on Anita’s phone, she may have misdiagnosed Eric—leading her to provide incorrect medical advice or treatment recommendations. AI also saved Eric’s mother from taking a long, potentially dangerous trip to the hospital, as the app determined that no further care was required at the time; not only that but from a provider’s perspective, Eric did not take care away from a potentially more critically ill patient.
Patient Virtual Health Assistant
The patient virtual health assistant use case has the same key factors as FHW virtual health assistant but instead focuses on tools that allow patients to take their care and wellness into their own hands. Patients are able to receive diagnoses or recommendations without traveling to a healthcare facility. This also saves time and resources for patients who need to be in a healthcare facility.
The report shares the story of 21-year-old Kehinde, a vocational school student in Kumasi, Ghana, who is sexually active and has questions about sexual and reproductive health. She is experiencing pelvic pain and other gynecological symptoms, but isn’t sure where to seek information—she comes from a conservative family she is not comfortable asking—or care. Also of concern is the potential cost of a trip to the doctor. She downloads an app that allows her to ask a chatbot all of her questions. The app uses speech recognition and generation to process and answer questions, while ML is used to evaluate Kehinde’s questions and respond. The app also shares additional resources with Kehinde based on what her most pertinent concerns are. She inputs her pelvic and gynecological symptoms into the app, and the app determines that she does not need to seek additional care at this time. Kehinde also receives recommendations in the event her symptoms worsen, a phone number for a nurse she can text if she has additional questions, and information on what annual appointments she should schedule and where she can do so. Kehinde also finds out that there are free women’s health clinics near her home where she can receive contraceptives, women’s hygiene products, and a confidential consultation with a female health worker. Without the app, Kehinde’s questions about reproductive and sexual health—and her symptoms—would have remained unanswered. Not only were Kehinde’s current questions answered, but she now knows where to go in the future to seek care.
Physician Clinical Decision Support
The fourth use case showcases how general physicians (GPs) can provide more specialized care with the help of AI—for instance, by allowing a GP to read diagnostic images. Key areas of this use case include:
- Image-based diagnosis
- Clinical decision support
- Quality assurance and training
As with the previous three use cases, AI technology is not meant to replace the physician—it serves as a tool to help physicians optimize care.
The researchers share the story of Jacinta, a radiologist working at a public breast cancer clinic in a Quito, Ecuador, hospital. She sees hundreds of women every day, and there are not enough radiologists available to provide the best care for these patients.
Jacinta’s hospital purchases an AI tool that allows Jacinta to diagnose tissue samples from previous-day patients who reported lumps in their breasts. The tool provides a timely analysis of the images, identifies possible cancerous areas, and gives a potential cancer diagnosis—as well as its certainty percentage. The AI radiology tool allows Jacinta to work faster, even though she is still carefully evaluating each image herself; she also feels the AI tool serves as a double-checker for her own work. Jacinta now accomplishes three to five times as much work as she did before the AI tool came around.
With the new AI tool, Jacinta not only provides more diagnoses in a single day, but women also receive diagnoses faster than before—reducing the wait time from one to two weeks to one to two days—and can, therefore, start a course of treatment sooner.