South African clinics use artificial intelligence to expand HIV treatment

Two doctors are using data analysis and predictive algorithms to stretch healthcare resources in South Africa and help millions of people live with HIV.

Dr. John Sargent and Dr. Ernest Darkoh co-founded BroadReach in 2003 to make the healthcare system more efficient and treat more patients. In 2010, the two developed Vantage, a data analysis platform and recommendation engine that runs on Microsoft Azure. 

The initial idea was to use the platform to manage and improve the public-private partnerships that support many healthcare services in Africa. The two realized that the analytic work could also improve access to healthcare in countries where there are many more people than doctors.

In Kwa-Zulu Natal on the east coast of South Africa, Sargent and Darkoh have worked with the Department of Health to hit the Triple 90s:

• Test 90% of the people at risk for HIV

• Put 90% of the people who test positive on a treatment plan

• Make sure 90% of those patients take their meds every day


Since 2014, BroadReach has used Vantage to help healthcare workers test 5.5 million people for HIV and help 530,000 people start on anti-retroviral therapy. Among the people taking HIV medications, 93% show no symptoms of the illness.

To help so many people live healthy lives, the BroadReach team had to first collect data from the clinics and patients into a database and then turn the information into specific recommendations.

Crunching data to make better decisions

When this process started, Sargent said the Vantage team identified two big sources of inefficiencies. The first was that significant amounts of important data was trapped in paper format. That included payroll data from healthcare clinics as well as lab data from patients. The BroadReach team used the Vantage platform to aggregate 100 unique data sources.

The second problem was staffing clinics. There was no way to allocate valuable resources to the clinics that needed them most. Because there are so few healthcare workers in the province, no one had the time to do this kind of analysis.

“One person runs the HIV program for the entire province, and it’s impossible to answer questions like: Are patients staying on treatment? Which docs are good/bad? Am I staffed for the next week? Which patients need extra attention?” Sargent said.
 
Sargent’s team uses machine learning to analyze this data and make prescriptive recommendations. A Vantage analysis assesses a clinic’s performance against the Triple 90 targets and makes staffing or operational recommendations about how to hit each metric. The platform does all the analysis in the background and uses natural language generation to make recommendations.

“Let’s say a clinic is doing really well on the first 90% but is way off target on the third 90%,” Sargent said. “Our analysis could show that the problem is that a clinic is short one nurse, and they are running out of medication to distribute.”
 
A unique source of data within the system is an app that doctors, nurses, and field workers use to track their daily routine.

“People are able to put in why they missed the target, things like, ‘There was no electricity in the clinic for a week,'” Sargent said.

When the app launched in April 2019, only 3% of staffers were recording their daily tasks. By October, the total rose to 65%.

“This is unprecedented in public health to be able to monitor and manage what your staff is doing on a daily basis,” Sargent said.

SEE: The ethical challenges of AI: A leader’s guide (free PDF)

Predicting non-compliance

Vantage has worked with HIV patients in South Africa since 2012. The team has used medication data from one million people to predict who is likely to have trouble staying on the treatment plan.

“Once people stop taking their meds, you start getting mutations in the virus that become resistant to the meds,” Sargent said.

He added that an in-house team of data scientists developed the adherence model with data collected over three years. The analysts built the model with data from two years and then tested the model with data collected in the third year.

“We have a risk rating, and we give that list to the clinic once a week and the nurse follows up with people who are at risk of not showing up,” Sargent said.

To make sure doctors, nurses, and community health workers trust and use the Vantage recommendations, the BroadReach team explains the algorithms, the data collection process, and the scientific validation of the recommendations. Sargent said that the Vantage team has built trust over time.
 
“As people see value in the data and the recommendations, it’s a self-fulfilling prophecy,” he said. “When we first started, only 20% of the staff gave us information, but now we see much higher usage rates and adoption rates.”

Expanding to chronic conditions

AIDS is a communicable disease, meaning one person can transmit the illness to another. Public health leaders around the world are dealing with a non-communicable disease pandemic as rates of obesity and diabetes rise. 

“A lot of what we’ve learned is going to help us conquer the next set of challenges that are coming from diabetes, high blood pressure, and high cholesterol,” Sargent said.

BroadReach, he continued, is in early stage discussions with a government in Africa that wants to start using the Vantage platform to manage multiple diseases. BroadReach will launch the next generation of Vantage early next year.  

Sargent said the Vantage team believes they can leapfrog back into developed markets.

 “This works well in government-run health systems where all data is in one place, like Medicaid.” 

Also see