Tip:
Highlight text to annotate it
X
So in terms of the format, we'll hear from today's speakers first and after which will
open up the floor for discussion and any questions you might have. Now I'd like to introduce
you to our speakers: Leah Weber who is presenting today for Claire Ma and Emma Liu who will
be presenting on the utilization of operations research to improve the quality of oncology
care. Emma and Claire work as operations research analysts with the CIHR funded Operations Research
and Cancer Care Team. And they're at the BC Agency Research Centre. And Leah is an operations
research scientist with the team. And they've worked on important research that has really
changed the way that care is being delivered here at the BC Cancer Agency and we're delighted
they can be here with us today. So I will hand it off now to Leah and Emma.
Alright, good morning everybody. My name is Leah Weber, I'll be presenting on behalf of
Claire Ma who is a little bit under the weather but she is here today with us to answer some
questions on her research. Also with me is my co-presenter Emma Liu and we'll be talking
about improving access to quality cancer care through optimized oncologist practice patterns
and patient appointment scheduling. So today we're specifically going to be talking about
two separate case studies we did at the BC Cancer Agency. Then we'll be talking about
how we hope to link them together in our future research.
So we as Angela said, we call ourselves the ORICC Team and that stands for Operations
Research for Improved Cancer Care. And our team consists of a unique collaboration between
practitioners, oncologists and administrators at the BC Cancer Agency. And faculty, researchers
and students at the University of British Columbia Sauder School of Business. So at
the intersection of these two organizations sits our research team which we call the ORICC
Team and the mission of our team is the improve the efficiency of the cancer care team and
thereby enhance patient outcomes.
So at this point, you're probably wondering, "What is Operations Research?" I guess we've
said this term a few times already so I just want to briefly go over what this actually
is. Operations research is the application of advanced analytical methods to improve
decision-making. And it's sometimes is referred to as "The Science of Better". And what that
means is we're constantly trying to figure out ways to improve the efficiency of operations
and help managers and executives to make better decisions with more information. And we use
a number of techniques in operations research and here's some of the more popular techniques.
The one we will be talking most about today is simulation and we will go into a little
bit more detail as to what that actually is later on. However, despite how complicated
or the amount of math you see up there, operations research in our line of work really boils
down to two main problems. And those are trying to match supply and demand. Whereas demand
for us means patient demand for cancer services and supply means resources with which we provide
this cancer care such as oncologists, nurses and pharmacists etc. And we also deal a lot
with scheduling problems: How can we schedule things in the most efficient manner as possible?
Some past projects that we've worked on is chemotherapy scheduling and we're also hoping
to address radiation therapy scheduling in the future as well.
So just briefly want to cover some important definitions and terms that you're going to
hear throughout the presentation so that everybody's on the same page. And the first one is 'Practice
Patterns'. And essentially what this is, is the number and frequency of appointments and
oncology treatments associated with an oncologist for different types of patients. So essentially,
how often and how frequent patients are followed and who is actually following these patients
throughout their treatment. We're also going to be talking about 'Patient appointment scheduling'.
And that's simply just the process of assigning patients to an oncologist and a specific time
slot. And we're going to be using the term follow-up a lot today and I just want to make
this very clear about how we define follow up in our particular research. So when we
say follow up, we're referring to any appointment after the initial new patient consultation.
So we're including both appointments during treatment and follow-up appointments post-treatment
when we're using the term follow-up. And then lastly 'Practice size' and this just refers
to the total practice size of an oncologist. So it includes their new patients, patients
who are currently on treatment and patients who the physician is following post-treatment.
So here's the big picture research problem that our team is currently looking at. And
it all began when one of our centres was experiencing long wait times for new patient consultations.
So we started looking at this problem and at first the obvious solution seemed uh that
if we don't have enough capacity to see new patients, if you just increase oncologist
availability you're adding more new patient slots that'll help balance the supply and
demand. But then after we brought this up with executives, they said well then, wait
a second, we also have to consider what happens downstream of that. So we increase availability
for new patient consults by adding more new patient consultation slots but essentially
what it's going to do is increase the practice size for the oncologist and therefore create
lot's more downstream work. And so by creating this downstream work what's going to happen
is it's going to make it more difficult for oncologists to then be able to see new patients.
And then on also another effect of this is we have to consider the quality of care and
as well as patient safety when oncologists practice sizes are becoming quite large and
they have lots more downstream work. So here are some of the research questions that we're
hoping to address. So we want to look at efficient scheduling so we want to look at what oncologists
we should book to and when should that slot be booked. And then we also want to help answer,
what are the optimal practice patterns, how should patients be followed, what kind of
specialization mixes should our oncologists be working on and also what's the optimal
capacity to balance that supply and demand also considering downstream work. And then
for the last aspect, this is probably an area that some of you are more interested in, we
also need to look at what the minimum and maximum number of cases oncologists see each
year. We want them to see a minimum number so they retain their proficiency in their
designated tumour groups and we also want to make sure that They're not seeing too many
cases where it's affecting the quality of care and putting patient safety at risk. And
then we also have to ask when and how should patients transition to survivorship programs.
And this is a very important question because when patients transition to these survivorship
programs, what it does is it's freeing up additional capacity for oncologists and making
it easier for them to be able to treat new patients. And so the work that we've completed
to date is that we've kinda looked at these two sides of the slides separately. So we
started problem to look at the new patient wait times and then we've also worked on a
separate problem for downstream workload. However in our future research, we want to
look at this as a system perspective and consider and try and find the equilibrium between the
two sides.
So for the rest of the presentation you're going to see lots of graphs and lots of numbers
but at the end of the day, there's three main takeaways that we'd like you to get out of
this presentation. And the first is that you can use advanced analytical methods to make
better decisions and solve complex problems such as operations research techniques. The
second one is that the ability to see new patients in a timely manner is an important
problem. And this is something that's difficult for us to quantify because we work more on
the operations and business side and we're not really it's difficult for us to quantify
the effect of patient health outcome. However, we know it's an important problem because
if patients have longer wait times to see an oncologist, their condition could potentially
worsen during this time, their tumours could grow and it's also stressful for the patients
to wait and not know what's going to come next. So lastly the most important takeaway
is that we have to system perspective when approaching this problem. We can't just look
at new patient wait times in isolation because increasing capacity to see new patients in
a timely manner adds downstream work. However, too much downstream work then makes it harder
to see new patients in a timely manner.
As I said, our case study today was done at the BC Cancer Agency and the BC Cancer Agency
is a provincial agency made up of six regional cancer centres. And our study focused specifically
on the Centre for the Southern Interior (CSI) which is located in Kelowna, British Columbia.
And as I alluded to earlier, the motivation for this particular study is that executives
came to us and were saying that they were experiencing long wait times for new patient
consultations with medical oncologists. And addition to that, they were also having to
work patients into their schedules especially for the most urgent patients when they couldn't
be seen within the most maximum wait times. And this was adding overtime to the oncologists'
workday as well as making it difficult to manage the daily schedule. And so you can
see from our graph here this represents the median wait time from referral to new patient
consultations. So there's a few things that happen in between this process. We're mostly
concerned with the total time from when the patient enters the system to when they begin
their consultation. So the green bar in the graph represents the median wait times in
Kelowna and the red bars represent the Vancouver median wait times and the line represents
the median for all of BCCA. So as you can see from this graph for Kelowna compared to
all of BCCA, the median wait times are a lot longer. And this was the problem that we began
to look at.
And so here's a typical cancer patient's pathway within the BCCA and I'm sure it's quite similar
in other centres throughout Canada. But it all begins outside the BCCA with diagnosis,
then they're referred and then after the referral process, they go through some sort of triage
and during the triage process, what we're looking at is determining the patient's urgency
level. And so for the purposes of our study, the most urgent patients are considered to
be urgency level 1 and the least urgent patients are considered urgency level 4. So we go from
urgency level 1 to urgency level 4. In addition to that, we also have to figure out well the
physicians have to figure out if there's any additional tests that are required and what
has to be done before the first consultation. Once that's all complete, the patient can
be booked for their first consult and then have that consult with the oncologist and
then most likely they're going to start some form of chemotherapy treatment. And until
a given stage in their chemotherapy, they're going to have a number of follow-up visits
and then post-treatment, they also need to have some follow-up visits with the oncologist
as well. And then at some point, depending on the tumour group, the position or the standard
practice, they're going to be discharged to the community where they may be seen by an
alternative care provider such as a nurse practitioner or maybe a general practitioner
physician who specializes in oncology. And so for the problem here, we have certain levers
that we can control. And the first one is demand management. And it may not seem that
likely that we can control demand because it's assumed that pretty much any patient
that's referred to the BCCA has to be treated there. However, one of the executives brought
up when we were approaching this problem was that it might be possible to divert some of
the least urgent demands which we would be considered to be urgency level 4 to maybe
a general physician who specializes in oncology or some other specialist outside of the BCCA
in order to reduce our demand. And so that's one thing we needed up looked at. We also
wanted to look at the booking rules and figure out the most efficient way to conduct patient
appointment scheduling. So we're using our resources as efficiently as we can. And then
we want to look at capacity management. And this sort of thing like how many oncologists
do we need, what's the optimal number of new patient slots that each oncologist has and
then at what point do we discharge these patients to the community to free up additional capacities
for our medical oncologists.
So here's the challenge that we face with this problem, as it's a quite complicated
problem. First of all, the demand is highly variable. We don't know how many patients
are going to arrive for a referral each day, we don't know what tumour groups they're going
to have and then we also don't know what their urgency level is going to be. And as I mentioned
earlier we do have multiple urgency levels to consider. Then we have to look at the capacity
side. And medical oncologists is a very expensive resource in healthcare and we also have a
limited amount and limited number of patient slots with which they can treat indistinct
their new patient consultation. In addition to that, each oncologist has a certain set
of tumour groups or tumour sites that they specialize in. And sometimes, it's difficult
to ensure that these specializations of oncologists is matching the distribution of tumour groups
that we see come in with demand. And so there's a number of managerial levers or strategies
that we can try in order to try and match demand and capacity. And I'll go into those
in a little bit more detail later on. However as I mentioned, one of our operations research
techniques is simulation. And that's the one we selected for this study.
So what is simulation? It essentially lets us build a model of the actual system. And
the reason that we do that is because we're not necessarily interested in the actual system
as it is. But what we want to try and do is experiment with different strategies and configurations
in order to see what happens to the system without making actual changes. And there's
a number of reasons for this. First it can be very expensive to make those changes. Second
of all, they can be very impractical or complicated to make. And then we also, it's going to take
a very long time to see the results from the study. So if we ran an experiment, we might
have to wait one, two or even three or four years to see what happens to the system when
we make those changes. However with simulation, we can build a model that represents the actual
system, we can make different configurations, changes, strategies in our model and then
we can run the simulation very quickly and then we can see what happens to the system.
So we call these "What if" scenarios. In our example picture, this would be an example
of the simulation or passengers going through airport security. And so a typical "What if"
scenario for this simulation would be oh what if we add an extra X-ray machine. How much
is that going to improve our system performance? And basically we can determine if that change
is worth it without having to invest in additional machine.
So here's some of the "What if" scenarios we looked at for our CSI study. And the first
thing that we looked at were the booking rules. So overall, we ended up looking at six different
booking rules but the main two ones we're going to be discussing today are 'first come
first serve' and 'first available slot'. I'm going to go more into detail into those a
little bit later on. Next we looked at how to handle overtime oncologists in the form
of add-on patients. What should the policy be? Should we allow the oncologists to use
add-ons designated for all patients so that all patients can be seen within their maximum
recommended wait time? Or should we reserve add-ons for the most urgent patients only?
Now we can also consider well what happens if we don't use add-ons at all, how will that
affect patient wait times and what happens to the system? Then we can also test what
happens if we divert demand. Diverting the leas urgent patients and comparing that to
the current system where we're not diverting any patients and compare if the benefit from
that change might be worth experimentation in practice. And lastly, we can adjust capacity.
We can determine how many slots should be assigned to each oncologist and to what oncologist
should we add extra slots to. So what we were able to do with the simulation model is try
different combinations of all these scenarios together and to see which ones were the best
performing ones overall.
So here's an animation that basically represents what our simulation model looks like. So when
patients arrive to the system, we consider patient arrival to be the point at which their
triage process is complete and they're ready to be booked. So as they arrive to our system
we're going to assign them a tumour group and urgency level and that's based off the
historical distribution that they saw in Kelowna. So for this example here, this patient is
going to be in the breast tumour group and they're going to be urgency level 1. So we
go into our oncologist calendar and we find an oncologist that treats the breast tumour
group and we look for any available slots to schedule that patient to. And so we'll
assign that patient in our simulation. When the next patient arrives, we assign them a
tumour group of melanoma and an urgency level 4. So let's say for example, in this configuration
of our simulation model, we're choosing to divert the least urgent patient. So in this
case, that patient would not be booked to our oncologist calendar. And then our last
example patient is in the GU tumour group and they're urgency level 2. And so again
we look in our oncologist calendar and see there's only one oncologist that specializes
in GU but they have no available slots within that patient's maximum recommended wait time.
So what we need to do is add this patient on to that oncologist's schedule and work
them into a slot in any way that we can. And so what we're able to do is we're able to
run this simulation for 5 years. So we run it for 260 appointment days per year and we
ended up just looking at the middle 3 years to account for some warm-up and termination
in the system. And then once that's all complete, we can repeat this whole process 30 times.
And what we're trying to do is we're trying to compare certain performance metrics to
see how each configuration performs. And here's the performance metric that we looked at for
our simulation model. And the first is service level and service level is the percentage
of patients who are seen within their maximum recommended wait times. So we looked at the
overall service level and we also looked at the service level for each urgency level.
Then we also looked at the number of add-ons. So we looked at how many add-ons were added
to each oncologist's calendar each week of the simulation. And then we looked at the
new patient slot utilization. So this is what percentage of new patient slots are we actually,
are actually being utilized or scheduled. And so for these performance metrics, we have
to consider that, we looked at three overall because we have to consider the trade-off
between them. For example, if we added a lot of new patients, added a lot of capacity,
it could result in very low numbers of add-ons but it would also result in low new patient
slot utilization. And so for service level, we want that number to be very high that would
be our target. But the number of add-ons, we want that number to be as low as possible
because add-ons are disruptive to the oncologist's schedule. And we want our new patient slot
utilization to be very high but not necessarily at 100% because we need some wiggle room for
that for certain scenarios.
So now I'm going to talk about some of the results of the simulation model. And the first
thing that we looked at was what was the best booking policy. So we looked at two different
performance measures which I'm going to show here and that was average weekly add-ons and
we also looked at new patient slot utilization. So the two policies that we ended up deciding
between were the 'first come first serve policy' which is the current policy they're using
in Kelowna and that essentially means that as patients arrive in to the system, they're
ready to be booked, they're just booked to the earliest available slot with an oncologist
that specializes in that tumour group as they arrive in to the system. The other policy
that we're considering was first available slot. And this is extremely similar to first
come first serve but the main difference is now that we are collecting all referrals until
the end of the day and then we're booking them starting with the most urgent patients
first and working our way down to the least urgent patients. And so you can see from our
graph as I said that we want our average weekly add-ons to be as low as possible. So you can
see the red and the blue line are very similar in this graph and they're also the lowest
out of all the other policies that we tried. And for new patient slot utilization again,
the blue and the red line are very similar but you can see that we ended up choosing
the first available slot policy for a specific reason and that was because it was reserving
capacity on a daily basis for the most urgent patients. And so that was one of our recommendations
based on the simulation model.
The next question we were trying to answer is what was the right number of new patient
slots that we should add for each week in Kelowna. And the performance measures that
we looked at here were service level and new patient slot utilization. And so you could
see on the x-axis there we're comparing the current number of slots that they're using
versus what happens if we add one to eight additional slots per week. And so if you look
at the service level graph, what we're most interested in is where is the slope of the
line the steepest? Because that means for each additional slot that we're adding, we're
getting the biggest improvement in our overall service level. However as I said, we have
to consider the trade-off because if we add slots all the way up to eight slots here,
we're going to be close to 100% service level but we're going to have extremely low utilization.
So we had to look at two indistinct for this. And so you could see what happens here when
we go from three to five additional slots. You can see that our utilization drops off
quite a bit there. So in the end, we ended up just deciding to recommend that they add
four additional slots for their oncologist's calendar per week in order to balance the
trade-off between service level and new patient slot utilization.
Next we looked at the patient diversion policy. Should we see all patients from the BCCA or
should we divert the least urgent patients to alternative care providers? And so here
we're looking at service level and we're also looking at the average weekly add-on. And
I just want to comment that when we divert the least urgent patients, this results in
approximately 7% of new patients being diverted to other practitioners. So the pink line in
our graph for service level represents the case where we are diverting the least urgent
patient and the blue line represents where we're seeing all patients within the BCCA.
And so you could see essentially this graph shifts from the right side to the left side.
So this means for reducing our number of new patient slots by two, we're achieving the
same overall service level. And then we look at add-ons, again the blue line is we're treating,
we're seeing all patients in the BCCA and pink is we're diverting the least urgent patients,
for the same number of additional new patient slots, the number of add-ons is reduced from
about 2.75 down to 1. So that's quite a significant reduction for the average weekly add-on. So
what this mean is if Kelowna decided that they wanted to divert the least urgent patients,
instead of having to add four additional slots, they would only need to add two.
And then the next question that we looked at was would it be possible to improve the
specialization mixes of the oncologists in the practice. And this question right now
is a little bit more theoretical than practical because in practice, it's actually difficult
to add a new speciality to the oncologist. It'll take them some time to learn about the
new tumour group and they also have a lot of time invested in the tumour group that
they're already working on. However, it's an interesting question and it does have some
practical application such as if you are hiring an additional oncologist, what should their
specialty be. So here in this graph our y --axis represents the percentage of new patient
slots not utilized and the x-axis represents the average weekly add-on. And so at the beginning
you get higher you go on the y-axis, this means that we have lots of excess capacity
in the form of slots not utilized in our system. And then when we go along the x-axis, this
means that we have a lot of overtime in our system because we're having a high number
of indistinct weekly add-ons into the calendar. So we compared the pink line which are the
current specialties in CSI and then we tried altering these specialties with revised specialties.
And that would be what if we added GI to one oncologist and maybe we added maybe lymphoma
to another oncologist, what would happen to the system? You could see from this graph
that the ideal point you want to be at is at (0,0). This means you have no slots not
utilized and you have no add-ons. So you want to get as close to that point as possible.
So you could see as you go from the pink line which is our current specialty to the blue
line which are revised specialties, we get closer to this ideal point. And so that would
be an improvement over the current specialties that they're using. Again, this is a little
bit more theoretical at this point but it does have some interesting implications for
the future planning.
So based on the results of our simulation model, we provided certain recommendations
to CSI. We ended up providing two different alternatives and really the alternatives differ
on whether or not we are diverting least urgent patients which is a strategic decision that
they need to carefully consider and make. However, the first alternative would be if
they did not divert any patients we would recommend based on our simulation model they
add four addition new patient slots per week. We would also recommend they use fast-booking
by booking patients to the first available slot in decreasing urgency level and then
for the add-on policy which I didn't talk about too much but we determined the best
policy from the simulation model would be to allow add-ons for all urgency levels and
this would result in approximately 1.1 add-ons per week. An alternative to this would be
if they chose to divert the least urgent patients, now they would only need to add two extra
new patient slots each week and they would use fast-booking to book patients to the first
available slot in decreasing urgency levels. The add-on policy would be the same but it
would result in an average of one add-on per week so the add-ons overall is pretty low.
And since we are using add-ons, I do want to comment that this does result in 100% service
level so all of the patients will be seen within their maximum recommended wait time
and our expected new patient slot utilization will be around 96%. So these are the recommendations
we make to Kelowna based on our simulation model.
But then as I alluded to earlier, we often have to consider the downstream work. So what
effect does adding these extra slots or altering an oncologist's specialization mix have on
downstream oncologist workload? And my colleague, Emma Liu, is going to talk to you about that
problem.
Thanks Leah.
So besides new patient utilizations, I've mentioned before another big chunk of oncologists'
time is spent on following with existing patients. And if we add one new patient slot to the
oncologist, that means the future practice size of the oncologist will increase which
for sure will translate into higher future workload. So in order to decide what is the
optimal number of new patients that oncologists should see per week, we first need to understand
how much downstream workload a new patient is likely to generate.
In order to do that, we conducted a data analysis using 10-year follow-up appointment information
from the BCCA information system and then we summarized and estimated patient follow-up
demand and also patients duration of stay by tumour site. Indistinct that beside tumour
site, there may be other factors that could impact on follow-up demand. For example, patients'
stage of disease, their age or medical oncologists' preferences or even the common follow-up practice
at the time. But for this preliminary analysis, we only focus on categorizing follow-up demand
by tumour site. And then utilizing the outcomes from our data analysis we developed a tool
that enables you to experiment with different practice patterns and specialization mixes.
Next I'll briefly talk about our data set and how we estimated patient demand for follow-up
appointments and their duration of stay. In our data set, we have over 8000 patients with
single cancer site who had their new patient consults with a medical oncologist at Kelowna
from 1999 to 2011. And this 13 years gives us 13 cohorts of patients with each cohort
have their new patient consult in each year. Taking breast cancer as an example, this table
gives the number of patients in each consult and we can see here so for the 1999 cohort,
we have 13 years of follow-up information. But for the 2011 cohort, we only have one
year of follow-up information available. The first column of this table shows the number
of patients in each cohort. And if we look at each cohort along time these numbers show
how many patients are there each year after their new patient consult. But we are mostly
look at the 1999 cohort, there's 109 patients who have their new patient consult in 1999
and 107 of them survive their first year after new patient consult and 102 of them survive
their first two years after the new patient consult and so no. So this table gives us
the group of patients that were in the system who generated or could have generated follow-up
workload for the oncologist. And similarly, we summarized the number of follow-up appointments
for each new patient appointment year and each year after their new patient appointment.
If we take the sum of each column from the patient table, that gives us the number of
patients who were there who could have generated follow-up workload and if we sum up each column
of the follow-up table, that gives us the number of follow-up appointments we generated.
Then if we divide the number of follow-up appointments by number of patients in each
year that gives us the estimated average of follow-up appointments each year after new
patient consults. Since we only have a limited number of patients to estimate for the demand
from year 11 to year 13, we decided to only estimate the demand and patients' duration
of stay for the first 10 years after the new patient consult.
So we did the same calculation for all tumour sites. The first outcome of our analysis is
the conditional death rates table showing the conditional death rate of each year after
the new patient consult for different tumour sites. As we can see here, the conditional
death rates for different tumour sites are quite different. For example, for breast cancer
patients, 4% of them died in the first year after new patient consult and compared to
lung cancer patients where 57% of them are deceased after their initial new patient consult.
In the second year, 4% of the breast cancer patients who survived the first year after
new patient consult are deceased in the second year compared to lung cancer patients, 38%
of the lung cancer patients who survived the first year are deceased in the second year.
Again, the estimates for year 8 to year 10 are more noisy due to the limited number of
patients we had especially for the more cancer sites.
Another output from our data analysis is a table showing the estimated average number
of follow-up appointments each year after new patient consults for different tumour
sites. Again, we can see here the estimated numbers are quite different across tumour
sites and we identified the top three tumour sites that generated the most follow-up workload.
Those are breast cancer, gastrointestinal cancer and lymphoma. Since the number of patients
in each tumour site are different...
...we took the weighted average of the number of follow-up appointments each year after
new patient consult. So on this graph, the averages shows the time after new patient
consult from year 1 to year 10. On the y axis, it shows the weighted average count of follow-up
appointment per patient in each year. We can see here, most of the follow-up appointment
workload were generated in the first year after new patient consult which averages to
around 3.6 follow-up appointments per patient. And this number goes down to around 0.5 follow-up
appointments per patient at year 10. And utilizing these two tables, the data analysis outcome...
...we build a capacity management tool that we call 'Oncologist Capacity Manager'. It's
an Excel-based tool. So the users can experiment with different practice considerations.
The goal of this tool is to quantitatively evaluate the impact of different patterns
of practices and the different specialization mixes of an oncologist's annual workload and
their practice sizes. Since we only estimated patients duration of stay or conditional death
rate and their demand for follow-up appointments for 10 years, our model horizon was 10 years.
And we assumed that all patients are followed until death. We realize that in real life,
after treatment and after several follow-up appointments, some patients are discharged
back into the community and they may or may not come back to the cancer agency, depending
on if there is a relapse of disease and some patients actually follow until death. And
for modeling purposes here, we assume that all patients are followed until death. And
this assumption is supported by the model input which are the two tables we just saw,
the conditional death rates and the follow-up appointment demand estimates. To use this
tool, the users can design their own scenarios by controlling the oncologist specialization
mixes and the number of new patients that oncologists see every year and also the percentage
of workload that can be diverted to an alternative care provider. The tool takes all this input
the user inputted and then compute and summarize the predicted number of follow-up appointments
for the first 10 years of the oncologist's practice and also the predicted practice size
for this medical oncologist.
So for each model run, the tool takes the user's input and generates or simulates a
group of new patients and assigns the patients with a tumour site. And also decide on if
the patient will be diverted to an alternative care provider. All of these decisions are
based on the perimeters that the user inputted. And if the patient is not diverted to an alternative
care provider, then they will go through a set patient path indicated by the data analysis
results. Each year they have a probability of exiting the system and if they stay in
the system, they will keep generating the set number of follow-up appointments each
year. At the same time, the tool keeps track of all the patients' paths and then summarizes
and sums up the number of follow-up appointments that are generated each year for this oncologist
and also the practice size, in terms of how many patients are there at the practice each
year after the oncologist started the practice.
Next I will go through a couple scenarios that we tested with the tool. The first scenario
looks at the impact of different specialization mixes. Consider two medical oncologists with
different specializations. Oncologist 1 spends 90% of the new patient slots in seeing breast
cancer patients and 10% on head and neck patients. Oncologist 2 spends 70% of the new patient
consults on genitourinary cancer patients and 20% on lung cancer, 10% on breast cancer.
Remember that all these tumour sites have very different conditional death rates each
year and they have very different demands for follow-up appointments. And we assume
that both oncologists see the same number of new patients per year and given this input,
we can use the tool to compare what's the predicted follow-up workload and what's the
predicted practice size for this oncologist. So the first output of the tool is the graph.
On the x-axis is the time of the first 10 years of the oncologists' practices. On the
y-axis is the number of follow-up appointments each year generated in the first 10 years
of the practice. And the red line represents oncologist 2, the blue line represents oncologist
1. As we can see here, even though these specialization mixes of these two oncologists are very different,
the predicted number of follow-up appointments of these oncologists are quite similar in
the first 10 years of their practices. With oncologist 1 ending up with a slightly higher
follow-up workload at the 10th year of their practices. The tool also compares the predicted
practice size of these oncologists. Again, the x-axis shows the time, the first 10 years
of the practices and the y-axis shows how many patients are there under the practices.
We can see here oncologist 1 at the 10th year of the practice will end up with a much higher
practice size compared to oncologist 2. This is may because oncologist 1 focuses on breast
cancer patients who have a much longer survival time compared to other kinds of indistinct.
So we can see here users can actually use this tool to test a combination of oncologist's
specializations and different practice patterns and to test the impact of these practice patterns
on future workload and practice size.
Now we go back to the question what if we add a new patient slot per week to an oncologist.
In this scenario, we have selected an oncologist and increased the new patient slot from 3
to 4 per week. And then we run the model and now the blue line shows the predicted number
of follow-up appointments before the new patient slots and the red line shows what happens
after we add the new patient slots. We can see here in the first 10 years of this oncologist's
practice, adding one new patient slot means that the oncologist needs to see an extra
400 follow-up appointments at the 10th year of the practice and this 400 follow-up appointments
plus the one new patient per week will translate into 248 clinic hours. And looking at this
graph, we can see here at the beginning of this practice, it is actually feasible for
the oncologist to see 4 new patients per week. But as time goes on, if the oncologist continually
sees 4 new patients per week, the practice will slowly become unsustainable. This is
an interesting question that we'd like to address in future research. That is, when
is the optimal time that oncologists should start decreasing their new patient intake
in order to maintain a sustainable practice? And we can see here that adding one new patient
slot per week to the selected oncologist would result in an extra 330 patients under the
oncologists practice at the 10th year.
In the first part of our analysis, another insight or recommendation we came up with
is adding new patient slots and diverting the least urgent patients. Given the specialization
mix of these oncologists, that means we are diverting around 10.2% of the new patients.
Now we're going to model again with the blue line representing before adding new patient
slots, red line meaning after adding new patient slots and diverting the least urgent patients.
We can see here now adding one new patient slot will result in such a big difference
from before. Now the oncologist will only end up with extra 98 clinic hours at the 10th
year of the practice. Similarly we compare the impact of adding new patient slots and
diverting least urgent patients, the impact on practice size. We can see here by adding
new patients and diverting least urgent patients, the practice size will increase around 100
patients at the 10th year of the practice.
And the last scenario we tested is changing oncologist specialization. Assume that oncologist
does 144 new patient appointments each year. And we added gastrointestinal cancer into
the pie chart. By changing the specialization mix, here is the red line after changing the
specialization and blue line represents before changing the specialization, we can see that
oncologist's predicted workload actually decreases from before. And if we look at the predicted
practice size of this oncologist, we can see here by adding gastrointestinal into the pie
chart, the predicted practice size will actually decrease as well at the 10th year of the oncologist
practice.
So besides using this tool to compare two different practices or compare practices of
two different oncologists, we can also use this tool to do capacity planning for the
whole centre if we aggregate demand and aggregate capacity. So assume the given capacity of
medical oncologists at Kelowna and assume the given new patients from the last and next
10 years, assume we use the same follow-up patterns that we estimated from the data analysis,
assume 20% of the follow-up work is diverted to alternative care provider which is the
current situation at the cancer centre and new patient consultation takes 90 minutes,
follow-up takes 30. Our analysis predicts that the projected clinical workload will
exceed the capacity of oncologist's time at year 2016. And imagine that you may consider
either adding more capacity or diverting part of the demand in order to have the two sides
meet.
So the next step with the indistinct grant and the CIHR, we plan to generalize our analysis
to include all BCCA centres and looking at longer time period more than 10 years, maybe
20 years. And we also want to incorporate the radiation part into the story. And in
this preliminary analysis, we only examined the variability in follow-up demand between
tumour sites. In the future, we would like to also look at the variability in follow-up
demand across oncologists, type of cancers, regions and over time. Then we would like
to compare what would actually happen with the follow-up guidelines indicated by BCCA
policy. And then using a more detail data analysis, we will be able to build a more
provincial, more comprehensive model to evaluate different patterns of practice and different
specialization mixes. And then we'll study different new patient booking processes at
different centres and look for the best booking rule for different centres. And now in this
analysis, we're looking at new patient booking and before demand separately using two analyses.
We can see here that new patient booking is studied on a daily basis while the follow-up
appointments is studied on an annual basis. And in the future, we'd like to explore the
possibility of combining parts together and build a system level model. And eventually
create a framework to support decision makers in making evidence-based decisions in capacity
planning.
After this presentation, I hope we can see that one can use advanced analytical methods
such as operations research to help guide evidence-based decision making and also help
solve complex problems. And operations research is applied into many different industries.
Our team specializes in applying operations research in healthcare, in particular cancer
care. So if you have any questions or if you have a business problem that you think that
can be solved with our technique, with our expertise, then please contact us, we'll be
happy to help. The problem with this