Use AI to build an advanced drug interaction checker for people who take multiple drugs

Summary: traditional drug interaction checkers are not adequate to study the complex drug interactions of people who take multiple drugs. eHealthMe uses AI to create an advanced checker to better solve the problem.

Taking multiple medications at the same time is called polypharmacy. Polypharmacy is common among older adults or those with multiple chronic conditions. In the USA, the percentage of patients greater than 65 years-old using more than 5 medications increased from 24% to 39% between 1999 and 2012.

When more than two drugs are combined, the overall impact on the body might amplify, neutralize, or even negate the intended effects, leading to unexpected outcomes. For example, a combination of ACE inhibitors (like lisinopril) and diuretics (such as hydrochlorothiazide) is commonly used to lower blood pressure and reduce fluid retention. When a calcium channel blocker (like amlodipine) is added to the combination, it can cause excessive blood pressure lowering, leading to symptoms like dizziness, fainting, or even kidney failure in some patients.

Ideally every possible combination of multiple drugs should be analyzed. However, traditional drug interaction checkers use pairwise models, which can only focus on two drugs at a time. The checkers are not adequate to study the complex drug interactions from polypharmacy. Below is a table that shows the number of drug combinations that can be studied between pairwise models and the ideal all-combinations models:

Also, in the real world, reactions to drug combinations can vary significantly because of patient factors such as age, gender, genetics, comorbidities, lifestyle, and other treatments being used. As a result, traditional clinical trials, which focus on controlled environments with a smaller set of variables, often can't capture the full range of real-world drug interactions. Traditional drug interaction checkers are built on the results from these clinical trials and thus can not personalize the results at will.

The complexity of these interactions requires real-world drug data. The data is hard to study due to several challenges, primarily the noise, sheer complexity and volume of the data. Additionally, data from real-world settings may be unstructured or come from various sources, making it difficult to process and analyze effectively.

AI/ML models are highly effective at cleansing, categorizing, and matching real-world drug data, especially when dealing with the complexity and variability inherent in healthcare data.

To better study polypharmacy, eHealthMe has built an advanced drug interaction checker by using public and proprietary AI/ML techniques:

• Natural language process (NLP): convert unstructured information into structured form for analysis
• Handling missing data: impute missing values that are common in real-world data
• Clustering and Classification: categorize real-world drug data into meaningful groups for further analysis
• Entity Matching and Deduplication: identify duplicate records for the same drug, which may be recorded under different names or formats across multiple datasets.
• Drug Combination Matching Optimization: determine optimal drug combinations for study, because sometimes including all combinations of drugs could introduce unnecessary noise.

By applying AI/ML algorithms and models to real-world drug data, 23 million patients and 1.5 billion data points as of Feb. 2025, the advanced drug interaction checker is able to study all combinations of drugs, and personalize the results.

With ever-increasing data collected and more advanced AI models available, eHealthMe continues to monitor and make necessary corrections to the checker. The advanced drug interaction checker will improve patient safety and lead to more effective drug regimens for people who take multiple medications.