The other day my brother and I took our mother out for her birthday. She had recently come back from California where her best friend had been in the hospital for a while. Mom told us how her friend had some bleeding in her esophagus. When she got back home to Minnesota and turned on the TV she saw a commercial:
“If you or a loved one has experienced internal bleeding, hemorrhages or death related to [the drug her friend was taking] then call 1-800-555-SUED because we intend to sue them for every burrito they got, make gazillions and give you a few cents for your suffering.”
Or something like that, I may not quite remember it correctly, I’m pretty sure about the burrito thing though. My mother was worried about her friend, but fortunately everything was going fine now. As a typical Norwegian who is prone to overly sentimental displays of emotion I reached over and patted her on the shoulder exactly twice and said, “I’ll get you a report for that.” Like it was a salve for a wound. After all, isn’t data science a salve for every wound? The conversation ended there and I started immediately thinking about FAERS (the Federal Adverse Event Reporting System). This is public data. We can do something with that. My mother may have still been talking…I’m not sure about what though.
The Federal Adverse Event Reporting System
FAERS is a reporting system into which healthcare professionals, consumers, patients, family members, lawyers and pharmaceutical manufacturers report adverse events related to medications. Granted, I’d imagine not a lot of patients or family members take the time to officially put in an adverse event report. Reporting of adverse events and medication errors by healthcare professionals and consumers is voluntary in the United States. If a manufacturer receives an adverse event report, they are, in most cases, required to send the report to FDA. For instance, if you say to your pharmacist, “Man, aspirin really makes my elbows greasy.” They may report such a thing. It is likely they are prudent about such reports. (FDA FAQs)
FDA Public Data and Applications
There are big problems with FAERS. The data is messy. Some people report aspirin, some aspirin tablets, some aspirin coated tablets etc. It is hard to clean up the drug names. Machine Learning distance algorithms, like the ubiquitous Levenshtein, do not work in this case. However, the active ingredient is a bit cleaner. We wanted to create a dashboard that anyone could use to put in a drug name, toggle to the active ingredient and get a simple list of results. “Surely, this has been done before.” You may ask. And I’d answer, “Yes, of course. The FDA has a nice dashboard you can use.” But I wanted something you could easily use for the basics. Something my mom could pull up on her phone while in the waiting room. When the doctor says, “We’re putting you on drug x.” My mom could look that up to see the top 25 reported adverse events for 2014-2017. She could also filter it for sex and age group. Simple…nothing more.
It is important to note that this dataset is skewed. Your doctor or pharmacist finds out that you died or had a cardiac arrest from taking a prescription, you can bet they are reporting that. As for your greasy elbows, not as likely. (You should see a doctor about that by the way). This means the dataset is skewed in that severe events are more likely to be reported than non-severe events. This is important to note when seeing that death accounts for 15% of reported adverse events. That number is inflated.
Talk to your Doctor
Another very important note. Do not use this dashboard to decide whether to take a medication or not. That is up to you and your doctor to decide. I thought it was sad my mom saw a commercial about a class-action lawsuit that alerted her to a medication’s side-effects. This is a better way.
So here ya go mom, for now use the link below (intended for mobile). If people express interest…we’ll see about wrapping it as an app. Of course, it’s $19.99, you can send us a check. Just kidding. It’s free and there will be no in-app advertising. That’s cheap and classless. The day Blue Diesel turns to in-app advertising is the day we close the doors. Although…there is probably an algorithm*1 to predict when a company debases themselves in such a way…hmmmm.
*1 By “algorithm” I mean a recursive neural network for a time-series with features about a company’s start date, profit, cash-flow and other financials and then an indicator to show when they started in-app advertising. Probably a worthless endeavor, but fun none-the-less. If we can’t have fun with data science, then why bother?