This suggests that integrating personalization driven by AI into clinical practice could enhance treatment outcomes for hypercholesterolaemia, according to the researchers at the Center for New Medical Technologies in Russia.
“Traditional pharmacotherapy, such as statins, is widely used to manage high cholesterol levels,” said the researchers. “However, there is increasing interest in the use of [dietary supplements] DS as adjuncts or alternatives to conventional treatments.”
They added that prescribing supplements based on biochemical markers, patient history and lifestyle factors “often neglects genetic variations that can significantly impact the efficacy of different supplements”.
The AI system used in the study was GenAIS, which is designed to integrate and analyse complex biological data, such as genetic data, metabolomic profiling, biochemical markers and patient history, to predict the most effective DS regimen. Participants in two groups took one to four capsules per day and the supplements were supplied by S.Lab (Soloways, Novosibirsk, Russia).
Seventy participants aged 40-75 years with LDL-C levels between 70 and 190 mg/dL were enrolled in this randomized, parallel-group study, which lasted for three months. Thirty-five participants were randomised to AI-guided dietary supplement prescriptions and 35 to standard physician-guided dietary supplement prescriptions by a computer-generated random sequence. Sixty-seven participants completed the study.
Fasting lipid panels, complete metabolic panels and high-sensitivity C-reactive protein (hsCRP) levels were assessed at the start and again at day 90.
The primary endpoint was the percent change in LDL-C levels from baseline to the end of the study. Secondary endpoints included the percent change in hsCRP, high-density lipoprotein cholesterol (HDL-C), total cholesterol and triglycerides between the two groups.
The AI-guided group experienced a 25.3% reduction in LDL-C levels, compared with a 15.2% reduction in the physician-guided group. Total cholesterol decreased by 15.4% in the AI-guided group compared with 8.1% in the physician-guided group.
Triglycerides were reduced by 22.1% in the AI-guided group compared with 12.3% in the physician-guided group. HDL-C and hsCRP changes were not significantly different between groups.
The researchers noted that the AI-guided prescriptions included a wider variety of supplements, including plant sterols, omega-3 fatty acids, red yeast rice, coenzyme Q10, niacin and fibre supplements, which may have enhanced overall efficacy.
They also noted the influence of genetic profiles. “Patients with specific gene variants exhibited more substantial improvements in lipid levels when receiving AI-tailored supplement regimens. This underscores the importance of incorporating genetic information into personalised treatment plans to optimise clinical outcomes.”
Studies with larger cohorts and a more diverse participant pool are needed to confirm the findings, they concluded, along with a longer duration and follow-up period. The practical implications of integrating AI-guided interventions into a clinical setting also need further investigation.
“Studies such as these are important to test the feasibility of AI algorithms where diet and lifestyle intervention are not the first choice, which is rare,” said Dr Mariette Abrahams, a consultant, speaker and expert in the personalised nutrition industry.
However, there are issues to overcome for AI. “There are many challenges when dealing with personal, health, activity, behavioural, wearable and self-reported data,” added Abrahams. “These include lack of transparency, such as in this study, where it is not clear whether the AI recommended the supplement product combination or whether it created the dosage of each supplement based on the participant’s data.”
“The AI model employs deep learning techniques to identify patterns and correlations that may not be immediately apparent to human clinicians,” said the researchers. “By analysing genetic variants that influence lipid metabolism and integrating these with metabolomic profiles, the AI can suggest personalised DS regimens that are more likely to be effective based on an individual’s unique biological makeup.”
However, Abrahams pointed to another challenge – a lack of representative and complete data sets. “For example, in this study genetics was used to identify genetic variants associated with cardiovascular risk,” she said. “In general, genetics databases overrepresent Europeans and North Americans, which means that the algorithm would not necessarily be relevant to other ethnic groups. This study does not provide ethnicity data, which means that the AI system may have been trained on bias data. There is also the concern around privacy, especially where data-sharing is concerned in order to combine different datasets.”
All participants were told to maintain their usual diet and lifestyle. “The most concerning thing is that diet and lifestyle are first-line treatment when it comes to lowering cholesterol especially in an overweight population, not supplements,” said Abrahams. “The AI system should have guided participants towards specific foods that contain the ingredients (fish, psyllium, oats) and compared this to doctor recommendations. Considering that doctors receive minimal training on nutrition, the natural next step is not supplements, it should be a referral to a nutrition expert.
“The key message is therefore for AI systems to mimic current clinical guidelines and it would be good to have systems in place that can analyse longitudinal data as well as dosages of supplements to understand which supplements and in what dosages worked for which individuals. Another concern is that one company developed the supplements – for transparency and trust, people should be guided to several brands per ingredient.”
Source: Nutrients
https://doi.org/10.3390/nu16132023
“Efficacy of AI-Guided (GenAISTM) Dietary Supplement Prescriptions versus Traditional Methods for Lowering LDL Cholesterol: A Randomized Parallel-Group Pilot Study”
Authors: Pokushalov, E.; Ponomarenko, A.; Smith, J.; Johnson, M.; Garcia, C.; Pak, I.; Shrainer, E.; Kudlay, D.; Bayramova, S.; Miller, R.
