The internet can be a scary place when it comes to medical searches. And while going into the black hole of symptom searches on the web is easy, finding the right physician to actually treat the problem is a more difficult feat.
Since the company was founded 10 years ago, Zocdoc has helped millions of patients find and book appointments with healthcare for more than 1,000 different procedures – from annual physicals and physical therapy to cardiology screenings and counseling.
Now, the company wants to help patients find the right doctor based on their exact symptoms, which are often lost in translation between colloquial language and medical jargon. This month, Zocdoc's Patient-Powered Search came out of beta so patients can use their own terms, like "gyno," "hurt wrist," or "post-election stress disorder," as well as a handful of emoijs, in order to find care. The feature is powered by artificial intelligence, and even accounts for common misspellings like "hemroids" or "diarea."
Zocdoc’s Director of Product Zahra Ladak spoke with Built In NYC about the technology and how it can work to ultimately provide valuable insights about population health.
[video:http://gph.is/2oieJKI autoplay:0]
When did your team realize there was a need for the Patient-Powered Search feature and how did it come to fruition?
We have long wanted to address the disconnect between the way patients speak about their care and the medical jargon of the industry, but we also needed to be sure there was minimal disruption to patients as we evolved our search experience. Free-text search is common across industries, but in medicine – more so than in travel, retail or research — getting matching right is of the utmost importance.
Over the past year, our team started to tackle this challenge by building a machine learning algorithm that is able to map colloquial terms to existing medical specialties and Zocdoc taxonomy. We coupled a natural language processing model with string-matching technology to ensure our search suggestions for patients could not only translate natural language but account for misspellings in the context of healthcare. Last fall, in a beta release, we started to overlay these two foundational search elements with real patient search data, along with another machine-learning algorithm that helps prioritize and rank search results to align with user preferences.
The product has gone through multiple rounds of testing across desktop and mobile in the lead up to our launch to ensure we were creating a better experience for patients.
What was a lesson your team learned while building the product?
In our testing phase, we learned quite a bit about how to apply sophisticated search technology to improve search and booking without degrading our user experience. We were keeping an eye on experience and conversion throughout the process, and testing each incremental change to make sure we didn’t cause confusion or decision paralysis for patients. We ultimately found that the experience was consistent or better for most patients.
What lessons have you learned since launching the product?
We’re continuing to keep a close eye on patient search and booking trends, but our learnings to date reflect much of what we uncovered in testing. As one example, we were pleasantly surprised by the broader range of specialties and procedures surfaced via patient inputs, which means we can make online booking a reality for more providers in the future. Previously, a pre-populated drop-down list of specialties on Zocdoc limited breadth of discovery, but during our testing, patients surfaced 135 percent more specialties and 80 percent more procedures by means of free-text search.
What have been some of the most popular searches so far?
Many of our most common searches connect patients with the specialist or procedure they’re looking for in four letters or less: ‘derm,’ ‘eye,’ ‘ent’ and ‘psy’ are among our top 20 specialist search terms and lead to dermatologists, ophthalmologists, ear nose and throat practitioners and psychiatrists, respectively. But what really sets Zocdoc’s search experience apart is when a patient wants to find care for a specific symptom, but doesn’t know what type of specialist to see. Some interesting symptom searches we’ve seen are colloquial terms like ‘hurt wrist’ for sports medicine specialist and misspellings like ‘hemroids.’
The best part is that through machine learning, the more our patients search, the better and smarter our results will get. In other words, patient searches are truly powering our feature, along with some input from medical experts to ensure our mapping remains accurate.
What type of traction have you seen with the product?
We have seen that Patient-Powered Search is simpler and more intuitive — it’s helping patients in their healthcare journey. While we don’t share conversion or booking information, we are moving this product out of beta and rolling it out across our mobile platforms because of the great response and positive user behavior we’ve seen thus far.
Is Zocdoc creating any other features in the machine learning space?
This year, we’re focused on making improvements to our core patient experience, especially on mobile, and innovating around pre-appointment experiences. Patient-Powered Search is just the beginning. While we don’t have specific plans to incorporate machine learning into other products and features on our current roadmap, it’s certainly not out of the question. We’ll continue to pursue development plans and approaches that empower patients to get the care they expect and deserve.
What's most exciting to you about working in the health tech space?
One of the things that attracted me to Zocdoc and keeps me excited about working in this space is the opportunity we have to improve the healthcare experience for patients. We can make a real impact. As risk continues to shift to the patient, Zocdoc, along with others in the tech industry, have an opportunity and responsibility to provide more tools and resources to help Americans access care when they need it.
From this feature, what types of insights are you hoping to learn about population health? Have you seen any trends so far?
While there’s still more to be done to improve our search experience, the core elements of Patient-Powered ensure that we’re constantly learning from patients searching and booking with us. Over time, we can imagine a future where search and booking data from millions of patients could generate important insights about trending symptoms and illnesses across the country. For example, because Zocdoc activity is centered around appointment booking, searches for ‘flu’ or ‘Zika’ on Zocdoc could be more closely correlated to real-time symptoms and intent to seek care, versus general web searches that could relate more broadly to interest.
As one more localized example, we recently looked into search trends on our platform around the Presidential Inauguration. When compared to the two weeks prior, we saw a 30 percent increase in patient searches for ‘therapist’ in the days leading up to and directly after the Inauguration. In looking at these ‘therapist’ searches overall within that time frame, we found that the majority were performed by New York City-area patients.
Image via Zocdoc.
Have a tip? Let us know or tweet us @builtinnewyork.
