Background
Kenya has one of the largest HIV epidemics globally, with women constituting 62% of the 1.4 million people (15+ years) living with HIV. In 2023, Kenya has achieved 97% in 1st 95 recording 17 681 new HIV infections. Identifying the last 3% of individuals is becoming increasingly difficult. There’s a need to adopt targeted testing approaches hence growing interest in machine-aided artificial intelligence (AI). Studies have demonstrated that 4% of males and 11% of females are at high risk of infection using machine learning.
Methods
A cross-sectional data review was done on outcomes of machine-aided HIV risk stratification in 95 health facilities supported by USAID Tujenge Jamii project across 4 counties in Kenya. The program trained healthcare workers on the use of machine-aided HIV risk screening in health facilities using electronic medical records at the HIV testing points in the general population. This was done alongside counselors’ HIV screening, routinely comparing machine-lead outcomes and manual screening outcomes.
Results
Between April 2023 and June 2024 out of 100,000 clients tested for HIV 57,450 were screened using machine Learning where 2,374(4%) were classified as very high risk 10,683(19%) as high risk,26,455(46%) as moderate risk and 17,938(31%) low risk. From the very high risk, 217 (9%) tested HIV pos, 315 (3%) from the high risk,388 (1%) of the medium risk and 141(1%) from the low risk. The 9% and 3% positivity among the Very high risk and very risk categories were significantly higher compared to 14% client screened eligible with 530,431 clients tested by following a counselor manual guided screening tool where 5,649 (1.1% ) tested HIV positive.
Conclusions
Machine-aided HIV risk classification scaleup provides an opportunity to improve HIV identification and efficiency, There is a need to scale up the use of ML in HIV testing.
Table or Figure
