ADS Capstone Chronicles Revised
27
price information, like from the DrugBank database, could make the price data more robustanduseful.Calculatingdrugpriceper prescribedregimen,insteadofdrugpriceper unit, could also allow for more accurate financial impact. 6.3 Ethical Implications There are a few ethical considerations for theuseofFAERSdataandthetoolscreated from this project. The tools fromthissystemdonotstoreany informationordatafrom/abouttheend-user. The Streamlit application is view-only and its purpose is to provide a prediction of adverse drug reaction outcome based on historical training data from FAERS only. The tools should not be used to make healthcare decisions without input from licensed medical professionals. There are manyfactorsinvolvedwhendecidingtotake a medication that are beyond the scope of the current model. For example, the riskof a seriousADRordeathmaybeoutweighed by the risk of death or complications from the underlying targeted health condition. The use of AI models and algorithms in healthcare has been a growing concern, as they can perpetuate bias(Mittermaieretal., 2023) . Many factors can bias models, including the distribution of gender, race, age,andotherdemographicfactorsfromthe input data. This limitation may lead to a model that is generalized, and potentially biased towards a certain group as the true population distribution of patient demographics is uncertain in FAERS data due to underreporting. The data also does notaccountforunderlyinghealthconditions which maybemediatingfactorsforadverse event outcomes.
Another ethical concern includes overall reliability of the model. Since models are generatedbasedondatawithunderreporting bias, model performance may not be accurate and can predict a wrong outcome, especially if the drug in question was not included in the original training data. 6.4 Future Directions To improve this precision health system, more data is needed with larger computational resources. Expanding the insights from the most currentquarter-year, to the most recent year or fiveyears,could beevenmoreinsightful.Incorporatingmore accurate pricing data, demographics, and specific side effect terms could also be helpful. To increase the confidence in population frequency counts inFAERSdata,datafrom the MEPS (HHS, n.d.) has been used for population rates ofsocioeconomicvariables related to use of certain prescription drugs (Yueetal.,2024),butcouldbeexpandedto MEPS features like employment status, health conditions, and food security among people who take specific medications. Incorporatingpopulationnormratesintothe FAERS features could produce a more accurate and clinically useful algorithm. Medicaid data on healthcare expenditures and utilization, andinflationratesandother economic data from the Bureau of Labor Statistics API (2 016), could be useful as well. ItisrecommendedthattheFDAimprovethe FAERS system by adding additional data quality measures. It would be useful to increase individual differences in reporting by making certain demographic fields required, such as age, sex, and weight. Adding features like height, race, ethnicity, income, underlying healthconditions,stress levels, diet, exercise, smoking status, and
177
Made with FlippingBook - Online Brochure Maker