Studies reveal that taking medications drugs with dietary supplements can lead to serious complications in patients. In such situation, 54% of elderly patients and 31.5% of patients with cardio-vascular conditions across all ages experience interactions of the products (Redd et al., 2023). When prescribing medications, being able to predict accurately these interactions is challenging. By leveraging AI-powered technology, developers can build platform-based models to tackle this challenge and help guide clinicians in their decision-making process to avoid complications and save lives.

In fact, there are already a few existing AI-powered models on the market that are trying to address the problem. Unfortunately, a lot of them fall short and are facing huge criticism regarding their poor performances. Understanding this situation, improved models can, not only be built to predict food-to-drug interactions, but to guarantee greater reliability, with the integration of an actual patient’s microbiome sequencing approach, they can be, specifically, built to predict previously unknown negative, positive and neutral dietary supplements-to-drug interactions.  Using either Machine learning or Deep learning approach, these models can make accurate predictions; and can provide reliable insights to patients and healthcare providers: Avoid specific harmful associations of dietary supplements/drugs or Recommend specific beneficial associations of dietary supplements/drugs.

To build such models, there are needs to acquire sets data encompassing input features and labels from reliable sources, such as Natural Medicines (NatMed), National Institutes of Health (NIH), the American Academy of Family Physicians (AAFP), patients ‘records from healthcare centers. Particularly, these data will cover food or dietary supplement information, drug information, patient information, including microbiome sequence data, labelled negative interactions, labelled positive and labelled neutral interactions. To ensure reliable predictions, calibration and validation sessions must be consistently executed. But we acknowledge that collecting these data may be challenging, not only because of the scarcity of labels, but also because of privacy concerns when it comes to use patients’ information and potential bias regarding race/socio group or sex. The development teams must find ways to mitigate labels’ scarcity. And, regarding to bias and data privacy, incorporation of adversarial debiasing and encryption techniques must be required.

Implementation strategies should reflect a comprehensive approach to the healthcare workflows, including the IT systems. Particularly, they should outline a strong engagement of the stakeholders, such as physicians/nurses, pharmacologists/pharmacists, researchers, registered dietitian nutritionists (RDN), certified nutrition specialists (CNS), patients and payers in the key steps of deployment including trial run, model launch, monitoring & evaluation and upkeep planning. Proactively and continuously, regarding pre-defined threshold, specialized teams can measure the models’ performances through key metrics: Accuracy, Sensitivity, Specificity, Positive predictive value and Negative predictive value, etc.

In essence, these strategies are underpinned in fundamental mission to address stakeholders needs. Studies reveal that patients and healthcare providers, increasingly, show needs for more effectiveness of drugs in patients’ treatments (Huang et al, 2024). Between 40 and 60% of Americans adults, with chronic disease, use dietary supplements (Gary et al, 2017). Furthermore, around 50% of clinical pharmacist report that they don’t have enough knowledge of dietary supplement-to-drug interactions (Zawiah et al., 2020). Adoption of these models can reduce subjectivity in decision-making process significantly, reduce time needed to make decisions by 84-93%, increase accuracy of predictions to 96%, promote a more personalized medicine, empower patients, save on financial cost by 40-75% and increase treatment effectiveness by over 85% (Khanna et al, 2022).

Since these models are intended to guide healthcare providers in their decision-making process, it is required to get pre-market FDA approval and follow the healthcare system standards and regulations to guarantee patients’ safety and privacy.

“In conclusion, full adoption of these user-friendly models can secure satisfaction for all”

References

Crispo JAG, Thibault DP, Willis AW. Adverse Drug Events as a Reason for Adult Hospitalization: A Nationwide Readmission Study. Ann Pharmacother. 2019 Jun;53(6):557-566. doi: 10.1177/1060028018818571. Epub 2018 Dec 10. PMID: 30525918.

ForbesTechnologyCouncil. (2023,July14). AI&FairnessMetrics:Understanding&Eliminating Bias. Forbes. https://councils.forbes.com/blog/ai-and-fairness-metrics#:~:text=The%20Role%20of%20Fairness%20Metrics,%2C%20race%2C%20or%20socioeconomic%20status.

Gary NA, Amanda HC, Roy LH (2017): Common Herbal Dietary Supplement– Drug Interactions. American Family Physician. http://www.aafp.org/afp. Volume 96, Number 2.

Khanna NN, Maindarkar MA, Viswanathan V, Fernandes JFE, Paul S, Bhagawati M, Ahluwalia P, Ruzsa Z, Sharma A, Kolluri R, Singh IM, Laird JR, Fatemi M, Alizad A, Saba L, Agarwal V, Sharma A, Teji JS, Al-Maini M, Rathore V, Naidu S, Liblik K, Johri AM, Turk M, Mohanty L, Sobel DW, Miner M, Viskovic K, Tsoulfas G, Protogerou AD, Kitas GD, Fouda MM, Chaturvedi S, Kalra MK, Suri JS (2022). Economics of Artificial Intelligence in Healthcare: Diagnosis vs. Treatment. Healthcare (Basel). 10(12):2493. doi: 10.3390/healthcare10122493. PMID: 36554017; PMCID: PMC9777836.

Huang K, Chandak P, Wang Q et al (2024): A foundation model for clinician-centered drug repurposing. Nat Med 30, 3601–3613. //doi.org/10.1038/s41591-024-03233Redd, D., Workman, T. E., Shao, Y., Cheng, Y., Tekle, S., Garvin, J. H., Brandt, C. A., & Zeng-Treitler, Q. (2023). Patient Dietary Supplements Use: Do Results from Natural Language Processing of Clinical Notes Agree with Survey Data? Medical Sciences, 11(2), 37. https://doi.org/10.3390/medsci11020037

Ryu J.Y., Kim H.U., & Lee S.Y. (2018). Deep learning improves prediction of drug–drug and drug–food interactions, Proc. Natl. Acad. Sci. U.S.A. 115 (18) E4304-E4311, https://doi.org/10.1073/pnas.1803294115.

Zawiah M, Yousef AM, Khan AH, Al-Ashwal FY, Matar A, ALKhawaldeh B, Nassar R, Abduljabbar R, Abdo Ahmed AA. Food-drug interactions: Knowledge among pharmacists in Jordan. PLoS One. 2020 Jun 17;15(6):e0234779. doi: 10.1371/journal.pone.0234779. PMID: 32555684; PMCID: PMC7299397.