AI-assisted HIV case finding in Ukraine

When Russia launched its full-scale invasion of Ukraine in February, 2022, there were an estimated 245 000 Ukranians living with HIV and it was feared the fighting would lead to an increase in HIV incidence as access to treatment and case finding would be heavily affected. 2 years later, it seems that this has not come to pass. According to latest WHO figures, the Ukrainian Ministry of Health reported 9769 new HIV cases between January and October, 2023, essential HIV services has remained operational, and HIV data continues to be reported to global surveillance initiatives. This sentiment was supported by a situational report published by UNAIDS in February, 2024 showing that HIV testing throughout 2023 had increased on previous years. The Alliance for Public Health (APH)—Ukraine’s largest HIV advocacy and service organisation—announced that they have increased HIV screening by implementing innovative approaches to their work, including the use of artificial intelligence (AI) to identify individuals at high risk of HIV. In their statement released on Feb 23, 2024, APH emphasised that effective case finding has been one of the key factors in keeping the HIV epidemic in Ukraine under control during the war and says machine learning has been highly effective at identifying individuals in need of HIV testing, support, and medication.

Between 2016 and 2019, APH gathered data from screening questionnaires from 140 000 people tested using an optimised case-finding (OCF) strategy APH had developed. This strategy involved a HIV-positive index person who had been referred for counselling and recruited people in their transmission risk network for testing. APH staff wanted to use this data to identify factors that could help them recruit people more likely to have a positive person in their network, “for example someone who injects drugs, or had been in prison before, or had relatively recently tested positive having previously tested negative for HIV,” said Pavlo Smyrnov, Deputy Executive Director at APH (Kyiv, Ukraine). In November, 2019, APH began using a machine learning algorithm developed for the organisation by a local mathematician to analyse the collected data to determine the probability of HIV infection within an index person’s network. The algorithm then suggested whether a person in the network of someone who had been tested would be at high risk of HIV and indicated whether that person should also be tested. “This, in turn, allowed us to recruit more people with an undiagnosed HIV infection for testing,” explained Smyrnov.

The method, according to Smyrnov, is adaptable to changes in HIV risk factors brought about by the effects of the war. He explained, “In areas which have suddenly become high prevalence for HIV following an influx of internally displaced people, we have been able to use it to identify risks in social networks [of people with HIV] and get more people in for testing, helping to quickly establish our testing operation there”. The machine learning case-finding model demonstrated 37% better results in HIV case finding than a non-machine-learning approach in the period 2022–23, with the former registering a 5·2% HIV detection rate as opposed to a 3·2% rate for the latter, according to APH.

“The OCF algorithm allows testing services to be focused on those [people] who are at high risk of HIV. The referral system [for people tested] helps specialists recruit new project participants based on the client’s behavioural practices, rather than subjective assessments of a social worker,” Anna Cherednichenko, a member of APH’s mobile testing team in the Dnipro region of Ukraine, told The Lancet Microbe. However, as emphasised by Smyrnov, the machine learning algorithm remains a “work in progress”, with APH continuing to refine its algorithm in pursuit of better results. “Machine learning is about using data and what’s most important for the correct prediction is the quality and reliability of this data. Self-reported information usually doesn’t make [findings] more reliable. Our analysis shows the best predictor is the result of an HIV test.”

A growing body of research has been carried out into the use of AI and machine learning in HIV treatment and prevention in recent years, including in social network interventions. “Traditional methods rely on predefined assumptions and models, machine learning algorithms learn directly from the data, making them more flexible and potentially more accurate in identifying risk factors,” explains Bradley Segal, Health Data Scientist at Phithos Technologies (Johannesburg, South Africa). “The ability to quickly adapt testing strategies and identify high-risk individuals in new areas, as demonstrated in Ukraine, showcases AI and machine learning’s potential to accelerate early diagnosis and prevention efforts. This agility is particularly beneficial for responding to changing patterns of disease spread and targeting interventions more effectively,” he added.

“AI has great potential in HIV prevention and case finding because you don’t have to wait years for study results to be published, by which time that information is probably out of date. Situations can change—there can be disasters or, like here [in Ukraine], a war—and if you need to make changes quickly to respond to that, AI is the best way to analyse the data [to help make those changes],” said Smyrnov.

By Ed Holt