Machine Learning Capable of Predicting Hyperlipidemia in People With HIV

Predicting hyperlipidemia in people living with HIV (PLWH) may have become easier with new research indicating that machine learning is capable of predicting hyperlipidemia in those who have been taking highly active antiretroviral therapy (HAART) for 6 months. The study, published in AIDS,¹ shows promise in preventing cardiovascular diseases in PLWH.

PLWH number approximately 40 million around the globe,² with approximately 30.7 million using HAART.³ These patients have an increased risk of health complications brought on by HIV, including cardiovascular conditions.⁴ Predictive models using machine learning algorithms have been established in the past to predict the probability of a patient developing a certain condition, which can be helpful for PLWH given their predisposition to multiple conditions. This study aimed to assess the efficacy of a machine learning algorithm to predict hyperlipidemia in PLWH who have been receiving HAART for at least 6 months.¹

The study was conducted at the Beijing Ditan Hospital in China and included individuals who were naive to HAART. All participants included in this study visited the hospital between January 2015 and January 2023 and were separated in a 7:3 ratio to training and testing groups, respectively. All included participants were aged 18 years or older and had been on HAART for 6 months. Any individual who was pregnant, an infant, or younger than 18 years was excluded from the study. The hospital e-health system was used to collect clinical data.

Machine learning can help to predict hyperlipidemia in people with HIV | Image credit: putilov_denis - stock.adobe.com

The researchers created the predictive framework using other machine learning techniques. Accuracy, positive predictive value (PPV), negative predictive value (NPV), specificity, sensitivity, and kappa coefficient were all used to evaluate performance. Hyperlipidemia was defined as having higher than normal measures of at least 1 lipid in plasma, with any participant meeting this criteria considered to have hyperlipidemia.

There were 2479 participants in this study, of whom 96.01% were men. The mean age of the participants was 33 years. Hyperlipidemia was identified in 1196 of the participants included in this study.

The authors tested 5 different machine learning models to correctly predict the incidence of hyperlipidemia. The LightGBM model was found to have optimal clinical capability across all tests with an accuracy of 0.7219, a PPV of 0.7539, an NPV of 0.7004, a specificity of 0.8087, a sensitivity of 0.6289, a kappa of 0.44, and an area under curve of 0.780. All machine learning models had an accuracy above 50% and most of the models had an area under curve of more than 0.7. This indicated that any of the machine learning models would be capable of predicting hyperlipidemia, with the LightGBM model the most effective.

There were some limitations to this study. Only risk factors for each patient were considered in their overall risk of hyperlipidemia, which excludes other factors related to lifestyle, such as smoking and drinking. There has not been implementation or adaption for prediction models based on machine learning classification algorithms. Almost all of the participants were men, which limits the generalizability to women. External validation was not received for the predictive model that was studied.

The researchers concluded that machine learning can be a novel way to predict hyperlipidemia in PLWH, especially in those who were newly prescribed HAART. This can “[allow] for prompt adjustment of treatment regimens to decrease the incidence of cardiovascular diseases in this population.”

References

1. Ding Y, Li J, Gao C, et al. Machine learning algorithms to predict the risk of hyperlipidemia in people living with HIV after starting HAART for 6 months. AIDS. Published online May 21, 2025. doi:10.1097/QAD.0000000000004244

2. HIV. World Health Organization. Accessed May 21, 2025. https://www.who.int/data/gho/data/themes/hiv-aids

3. HIV – reported number of people receiving antiretroviral therapy. World Health Organization. Updated July 22, 2024. Accessed May 21, 2025. https://www.who.int/data/gho/data/indicators/indicator-details/GHO/reported-number-of-people-receiving-antiretroviral-therapy

4. HIV and heart disease. HIVinfo. Updated March 31, 2025. Accessed May 21, 2025. https://hivinfo.nih.gov/understanding-hiv/fact-sheets/hiv-and-heart-disease