Drug information is one of the top searches requested by both consumer and health professionals every day online. Currently, mobile apps can pull up drug details but they are slow and require too many manual steps to onboard and access. This is ok if you're looking for more in-depth information but for most cases, users just want the basic details such as Use, Purpose, Warnings, Side Effects, and Directions. Another inspiration is the low adult literacy rate. Here are some numbers:

  • 14% of the US population can't read
  • 19% of high school graduates can’t read

By using both Text and Voice to deliver information, we can help this demographic access drug information and improve safety and adherence that will ultimately lead to better outcomes.

What it does

Drug Facts skill uses your voice to quickly search drug information by Name or NDC number. Every drug has a unique National Drug Code that can be used to pinpoint the drug labels that were submitted directly by the Manufacturer to the National Library of Medicine. This is the same information that consumers normally see on product packaging and leaflets. Using the rich labels from DailyMed and pill photos from Pillsync, drug Images are shown in the alexa app for better association. Users can search by a single Name or NDC at a time to pull up the available sections or go straight to the combination search if they already know the section they're looking for like "What's the Directions for Tylenol." Side effects are fetched from FDA Most Reported Adverse Reactions API.

How I built it

Used a custom endpoint to accept requests from Alexa and return the correct responses. NDC numbers are formatted to ensure compatibility with NDC 10 and 11. For names and section search, there is a lot of parsing in place to match with the right intent. For drug information, a lot of conversion from HTML to SSML to ensure that it is read appropriately. For drug image, if there are multiple photos, a collage is produced to stay within the limit of 1 response card while helping users see the drug.

Challenges I ran into

The speech recognition engine has some difficulty recognizing common drug brand names like Zoloft and Celebrex. This makes it hard to search the right drug. Matching the right sections is another challenge as there are many closely related ones with highly technical terms. Another challenge is the conversion and formatting of drug SSML responses. More can be done to make the reading better.

Accomplishments that I'm proud of

Search by NDC has turned out pretty well. The numbers and dash are recognized and matched to the right intent and slot. I'd expect a lot of health professionals will love to search this way versus visiting DailyMed all the time.

What I learned

Once I was used to getting drug information from Alexa, I didn't want to go back to the app. I've learned a lot about how Sessions are looped to different Intents to create a more natural experience. I've also learned how Voice user interface is a whole new field that can be used to educate consumers and improve their access to technology.

What's next for Drug Facts

  • improve Name/Sections search
  • improve SSML parsing
  • log in to provide personal responses / results

Built With

Share this project: