Book Review: Better Health Care Through Math
LeanTaaS is a healthcare analytics company that works to apply Artificial Intelligence and Machine Learning (AI/ML) within healthcare to improve operations; specifically (as of this writing) operating rooms, infusion centers, and inpatient bed access.
As a healthcare junkie, I am always trying to broaden my knowledge of how we can help redesign our health system such that it is more consumer oriented. Yet, after obtaining my engineering degree in health systems, I have felt that there is significantly more that can be done to address the logistical and operational problems seen in healthcare, which is why I was pleasantly surprised when the founders, Mohan Giridharadas and Sanjeev Agrawal, announced they were publishing this book.
What problem does this address?
While it is amazing to see the advancements in scientific discovery, delivering this care has become more and more difficult over the last few decades.
People are living longer than previous generations; yet younger generations are having fewer children (see image below). Because of this, we end up with a couple of problems:
1) There are less working age individuals around to care for those who have retired or left the workforce; and
2) There are less working individuals contributing to healthcare coverage, which in turn means there is less money to support the care of these older generations.
With less revenue coming into the system, this prevents organizations from expanding by way of hiring more staff, or purchasing more expensive resources (e.g. MRI machines). Because of this healthcare organizations continuously need to deliver more with less.
In today’s market (at least pre-COVID), if you come down with some unknown, non-urgent ailment it may take 20 days to initially see a primary care doctor, and an additional 20 days to a specialist. 40 days of waiting just to see a physician (assuming your insurance requires a primary care referral first) [1]!
Imagine if you needed to wait 40 days to book a flight across country, hail a cab, or order a package online. We do not accept that in other parts of our lives, why should we when it comes to our most important part?
One method to improve this poor access is through improved scheduling of healthcare visits and resources. Which is exactly what the authors discuss.
The Scheduling Problem
To understand the scheduling problem a bit more, I like to think of an hour, or 60 minute long youth hockey game, which generally looks like the following.
This gives us a time of 49 minutes, yet as with any sport there may be stoppages due to points being scored, penalties, etc. Some games have more stoppages than others, and while most of the time this arbitrary 11-minute buffer is sufficient, for others it is not; and when the ice time is up the game ends early. While not fun in a close game, everyone accepts that they paid for an hour of ice and moves on to the next one.
However, this type of thinking cannot be applied in a healthcare setting. You cannot just stop a surgery, infusion, or emergency treatment and say, ‘next patient.’ Yet this is the EXACT way things are usually scheduled. It is not anyone’s fault, it is just the intuitive way for us to schedule things where we look at one another’s calendar and say which day works best and boom, scheduled appointment. It’s just not the best approach when it comes to scaling across an enterprise.
In addition to the variation in time for appointments, healthcare has an additional layer of complexity where we have variation in day-to-day patient demands, as well as a need to ‘link’ components of our care [4]. Just like flying, you cannot expect to show-up for your flight at the boarding time and expect to hop on to your plane. You likely need to schedule in time for each step from the point you leave your door to the time your plan departs; and each of these steps needs to take place in a specific order (you cannot board prior to traveling to the airport).
On our own, planning for a flight is usually straightforward, but imagine now needing to schedule thousands of these trips per day trying to slot people in with 15-minute increments. You would likely end up with a log jam somewhere and would have a lot of unhappy people waiting to get onto their plane.
Summary
The authors look at 5 different areas where we see scheduling problems today in healthcare and propose solutions around them. Surgical procedures, emergency departments, inpatient bed use, infusion centers, and ambulatory scheduling. Here are a few of my favorites:
Surgical Operating Rooms
In hospitals and surgery centers, keeping an OR open and staffed can cost as much as $100 per minute. Because of this, it is extremely important to maximize a room’s utilization when open, however most organizations parse out OR time in the form of blocks whereby Specialty A gets 4 blocks per week, Specialty B gets 3 blocks per week, etc. While this is helpful in maintaining a regular schedule for surgeons, the problem arises when small 15–20-minute increments go unused, or worse, when an entire procedure just is not scheduled. Much like airplanes landing on a runway, we do not expect pilots to say ‘give me the runway between 11:00–11:15’ we shouldn’t be doing that in healthcare [4].
Emergency Departments
Emergency departments are almost a perfect example of a random (aka stochastic) process encountered within healthcare. For the most part, we cannot predict what the next person through the door is going to need, or when they will arrive. Because of this, we need to rely on historical data to help us predict what demands may be both to staff an emergency department appropriately, ensure that ancillary services throughout the hospital can be made available for emergency patients (lab, imaging, etc), and have an idea of what types of inpatient beds are available should a patient need to be admitted. When all these factors are not managed in a systematic manner, we end up with a system with a wait time to see a physician can be as long as an hour [2], and length of stay can be upwards of several hours [3].
Ambulatory Office Visits
In a traditional ambulatory office, the demand on a specific day is likely unable to be predicted. One day we may have 5 calls from new patients, or we could have 6 calls from existing patients who have pressing issues. However, the way we schedule these is wholly inadequate. The general approach to scheduling is to roughly block off X number of new patient visits per week, Y number of short, existing patient visits each week, and Z number of long, existing patient visits each week. While conducive to setting a routine for the staff, but not always the best patient centered scheduling approach.
What I liked
The Author’s Utilization Analogy
While we need to do more with less, we also need to recognize we cannot expect 100% utilization of our resources. Much like our highways, when the road is 100% packed with cars you end up with bottlenecks everywhere. Like healthcare, we need to make sure that we develop scheduling methods to level load demand across the day to avoid traffic jams. Peaks in utilization lead to delays, rushing of care, and unnecessary overtime for the organization.
The Need to Improve Our Use of Data
In the last decade, healthcare organizations have gotten good at bringing data to the table to admire the problems they are dealing with. After months of hard work on behalf of data analysts, leaders may end up looking at how their resources are being used and then ask to see the data in several different fashions hoping to ask a question that leads them to some ultimate solution.
The main problem is that most of the time we are using historical data to summarize our problems. Rather we need to be able to use the enormous amount of data that has been collected over the last decade and apply it to predicting and then prescribing options for frontline workers.
Just as Google Maps can say how long it may take us to get from our homes to a destination during certain parts of the day, we need software solutions that can do the same in real time.
Need to Expand Beyond of Electronic Health Records (EHRs)
EHRs have come a long way over the last decade yet fall short when it comes to scheduling incredibly complex processes. The authors call out a few items:
EHRs which are built on flawed scheduling assumptions
Reverting back to my hockey analogy, appointments are usually time based in 15, 20, 30-minute intervals. Again, while intuitive, we should be able to harness the data that EHRs have been collecting to better understanding exactly how long a person may be when they are placed onto the schedule.
They employ a “Here’s an Open Slot” principle
Just imagine if tomorrow the COVID pandemic ended and you decided to book a trip to a place on your bucket list. Yet rather than looking up flights online that prescribe to you dates, flight times, and expenses; you call up each airline and stay on hold until you reach the pilot of each flight only to play a game of phone tag with them over the next several hours to find the best time that works for you and them.
EHRs Do Not Have the Right Math Engine
Imagine you have an 8-hour shift of work in which you will see 4 types of patients (new, returning, complex, and follow-ups), where each of these 4 patients has a different appointment time. This leaves us with FIVE THOUSAND TRILLION different calendar configurations [4]. Which further goes to show that the ‘here is an open slot’ principle is not a great scheduling solution. For context, even if we were to narrow that down by 99.99%, we’d still have 500 billion combinations, which if you tried to sketch out each of these every second it would take to 158 thousand years to map them all out.
What I would Have Liked to See More of
Case studies
Though there are plenty on their website, I would have enjoyed being able to see more real-life examples of how technology can be deployed to help with these problems. Whether that be from specific case studies of groups they have partnered with, or examples of the technology that the users interact with.
Different mathematical methods and how they help with these problems
As an engineer, who has completed coursework in simulation modeling, and data mining, I would have liked the opportunity to dive in more on how these mathematical approaches get built into software that can be used by front-line staff day-to-day as opposed to one-time analyses that are way to prevalent in our health systems today.
Who Should Read This?
From a logistics standpoint, this book does a great job of framing the problems that we see in healthcare operations today. Many times, I am asked by engineering students on how math is or could be used in healthcare. While we are seeing many advancements in the use of Artificial Intelligence and Machine Learning (AL/ML) when it comes to clinical diagnosing and treatment as one use, we need to find better ways to help people get access to this this amazing treatment.
This book frames up the opportunities and constraints quite well from a care delivery perspective and I would recommend for anyone who is interested in applying engineering or data science principles into healthcare.
If you’re interested in reading more, it can be picked up here: https://leantaas.com/bhctm/
Sources:
[1] https://www.mgma.com/data/data-stories/how-long-are-patients-waiting-for-an-appointment
[2] https://www.cdc.gov/nchs/products/databriefs/db102.htm
[3] https://focus.hqca.ca/emergencydepartments/total-length-of-patients-emergency-department-stay-los/
[4] Better Healthcare Through Math: Bending The Access and Cost Curves
