Zocdoc turns ten today. In these ten years, we’ve focused on building products that empower patients by creating the healthcare experience they want and deserve. In doing so, we’ve observed a decade of patient behavior across the country and collected lots of data: on the order of millions of patient bookings on our platform and hundreds of millions of patient searches.
There is breadth in this data too: we work across state lines and across different practice management software (PMS) systems, which allows us to partner with large health systems and small practices in the U.S. and serve patients across all 50 states.
And because Zocdoc is a healthcare marketplace, we can take a look at both the demand (e.g. patient searches) and supply (e.g. provider availability) in a trillion-dollar industry. In this article, we use this data to show how Zocdoc matches supply and demand to provide better provider access to patients.
Figure 1: A map of Zocdoc patient searches in the United States in 2017. Darker colors indicate areas with higher patient demand. Note that these areas represent where patients are looking for doctors, and not where patients are located.
One consistent challenge in the U.S. healthcare system is the length of time a patient has to wait to get access to a doctor: it’s bad, and getting worse. The Merritt Hawkins 2017 Survey of Physician Appointment Wait Times showed that in 15 different large city markets, the average wait time for a physician appointment is 24.1 days – a 30% increase from 18.5 days in 2014.
Providers see greater no-show rates the longer their patients have to wait, as shown in a study from Geisinger Health System. That patients may suffer as well goes without saying – the same study goes on to state that longer wait times to specialty appointments can lead to increased emergency department utilization.
In the following analysis, we show that the Zocdoc marketplace does a better job in surfacing near-term provider availability and in keeping patient wait times lower than traditional booking methods. Among the many technologies that power the Zocdoc platform, here are two that help make that possible:
- Zocdoc’s proprietary sync technology allows us to connect with provider calendars – across many different PMS systems – in real-time, and surface open appointments to patients, including newly available slots that come open due to reschedules, cancellations and no-shows. (Based on an internal analysis, we believe up to 25% of appointments are impacted by these changes.)
- Zocdoc’s Patient-Powered Search uses machine learning to surface these open appointments – especially appointments within the next two calendar days – to patients.
These technologies help Zocdoc provide faster access to providers in the markets and specialties that were analyzed by Merritt Hawkins. Across Zocdoc searches in Q2 2017, the average number of Days to Third Next Available Appointment (DTTNA) for the first search result shown to a patient was 1.6 days. We dive deeper into this patient access metric in the “Supply: Provider Availability” section below.
This level of provider access allows patients to seamlessly book their appointments online with significantly less wait time, and patient wait times on Zocdoc are indeed significantly lower than the wait times cited in the Merritt Hawkins survey. Across Zocdoc appointments in Q2 2017, a new patient has an average wait time of 7.6 days (compared to the 24.1 days in ), with 11% of appointments booked within the same day and 39% booked within the next two calendar days (i.e. same day, next day, day after next). We dig into patient wait times in the “Demand: Patient Wait Time” section below.
Supply: Provider Availability
Provider availability – especially short-term availability over the next two calendar days – is an important factor for a Zocdoc patient when selecting a healthcare provider. In fact, the machine learning that powers Zocdoc Patient-Powered Search has organically learned, by analyzing millions of Zocdoc searches and patient interactions in the past, to display providers with more short-term availability (see Figure 2).
In the first result of a typical search, we see almost seven open appointments within the next two calendar days.. These availabilities are displayed prominently in each search result (see Figure 3).
Figure 2: Looking at the average number of available appointments in the first ten search results, across all searches on the Zocdoc platform in Q2 2017, we see that our results display providers with more availability in the next two calendar days.
Figure 3: Examples of two search results. Note that availability over the next two calendar days (i.e. same day, next day, day after next) is featured prominently to the right of each provider.
In the healthcare industry, provider access is often measured using a metric called Days to Third Next Available (DTTNA), which is the number of days between a patient appointment request and the third available appointment for that patient . Let’s adapt this metric to Zocdoc search by applying the metric to the first provider that appears each search. A methodology section, which includes a more detailed description of this metric, is included at the bottom of this article.
Given Zocdoc search data from Q2 2017, the average DTTNA for the first search result is 1.6 days. This means that Zocdoc is able to match a typical search with a provider with a DTTNA of 1.6 days – with just one search result. We can also see that, for more than half of these results, at least three appointments are available within one day (day of search [day 0] and day after [day 1]); and for more than a quarter of these results, at least three appointments are available on the same calendar day (Figure 4). Below, we plot this Zocdoc DTTNA for each of the five specialties analyzed in the Merritt Hawkins report (Figure 5).
Figure 4: Distribution of DTTNA for the first Zocdoc search result, across Zocdoc searches in Q2 2017. We see that more than a quarter of these search results have at least three availabilities within the same calendar day (DTTNA of 0 days).
Figure 5: Average DTTNA for the first Zocdoc search result, broken down by specialty, also over Q2 2017.
DTTNA is a provider-level metric, in the sense that it’s generally used to measure an individual provider’s level of access. However, Zocdoc search can show up to ten doctors on one search results page, so additional numbers help us demonstrate the access provided by the whole set of doctors surfaced in a search.
When we look across all providers shown on a search results page, we see that 79% of Q2 2017 searches have surfaced at least one available timeslot in the next three days, with an average of 55.25 total time slots available in the next three days on a typical search page.
Demand: Patient Wait Time
Because Zocdoc is able to show provider availability to a patient with an average DTTNA of 1.6 days, patients can expect and demand a higher standard for their healthcare – and, unsurprisingly, they often choose to see a provider sooner rather than later. The average wait time in days (referred to here as booking lead time) on Zocdoc is 7.6 days for new patients in Q2 2017, compared to the benchmark average of 24.1 days .
Figure 6: Booking lead times on Zocdoc for new patient appointments in Q2 2017, across all specialties and markets. Note that 11% of new patient appointments are booked in the same day, and less than a third are booked 7 or more days in advance.
Let’s look at wait times for patients on the Zocdoc platform and see how they stack up against the statistics in the Merritt Hawkins survey. Note that this isn’t exactly an apples-to-apples comparison; whereas the researchers from Merritt Hawkins asked for the first available appointment, we have no way of knowing the intent of a Zocdoc patient – she may be looking for an appointment tomorrow, or trying to book one for next week or next month.
Across the board, Zocdoc has significantly shorter waiting times than our benchmark, in all specialties and major metropolitan areas cited:
Zocdoc serves millions of patients each month in different states and across different PMS systems, and we actively use our data and technology to create a healthy marketplace to better serve our patients. Two numbers are worth repeating: in Q2 2017, the top search result on Zocdoc had an average DTTNA of 1.6 days, and Zocdoc new patient bookings had an average lead time of 7.6 days.
That these numbers are so different from any benchmark we’ve seen – that they show not a slight shift but a threefold reduction in wait times – is a testament to the effectiveness of Zocdoc and its use of technology and economies of scale. And improvements on these numbers aren’t just abstractions – for our patients, the difference is in a same-day appointment with a PCP instead of an emergency room visit; not having to wait on hold on the phone while already in distress; diagnosing and treating a life-threatening disease before it’s too late.
So – if you’re an engineer, data scientist, doctor, policy wonk, or just someone who wants to improve healthcare, join us, and let’s work together to move our healthcare system a few steps closer to where it should be.
About the author
Andy Chen is a Machine Learning Engineer on the Data Science team at Zocdoc. Andy obtained his degree from Brown University in Mathematics and Computer Science, and he is passionate about NYC startups, public policy, and SoulCycle.
Andy can be reached at firstname.lastname@example.org.
Notes on methodology
As noted above, our methodology was necessarily different from that of the Merritt Hawkins study. Here, we outline some key differences:
- The Merritt Hawkins 2017 Survey of Physician Appointment Wait Times  surveyed wait times by calling providers and asking for the first available appointment, while we used actual booked Zocdoc appointments to calculate Zocdoc wait time. In fact, the Merritt Hawkins’ methodology can be seen as a more relaxed version of DTTNA; instead of Days to Next Third Appointment, the researchers were asking for the Days to Next First Appointment. It would be more than fair, then, to compare our DTTNA of 1.6 to Merritt Hawkins’ 24.1 days, but we decided on an even more conservative approach by using wait times from actual bookings.
- When calculating all Zocdoc numbers in this analysis, we set Q2 2017 as our time frame.
- When evaluating new patient wait times, we did not restrict our analysis to any individual procedure or set of procedures for a given specialty; furthermore (this is particularly relevant to Figure 7), we included all Zocdoc providers in our analyses instead of sampling 10-20 providers per specialty per market.
- Our provider access metric, DTTNA, was calculated using appointment availability for both new and existing patients.
- The number of days in both DTTNA and Booking Lead Time is calculated by looking at how many day boundaries are crossed: if a patient searches at 10 PM and a provider has their third next appointment at 12:30 PM on the next day, we say that the DTTNA is 1 even though the time difference is less than 24 hours because exactly one day boundary was crossed. A DTTNA of 0 means that there are at least three appointments available on the same calendar day as the search.
- We believe our adaptation of the DTTNA metric to be a representative measure of provider access that Zocdoc provides, given that the patient is immediately able to book with the top search result once it is surfaced. Even though Zocdoc shows many providers per search and we can, in theory, look at the DTTNA of the inventory of all providers shown in the results, we only look at the availability of the top search result so that DTTNA remains a provider-level metric, allowing for a fairer comparison between our DTTNA and that of an individual provider. Note that if the top result has no availability, we omit the result from our average since it has an undefined DTTNA; this happens in ~2% of Zocdoc searches.