Improving Clinical Access and Continuity through Physician Panel
Ritesh Banerjee, Ph.D.1,*, Hari Balasubramanian, Ph.D.2, Brian Denton, Ph.D.3,
James Naessens, Sc.D.1, Douglas Wood, M.D1, James Stahl, M.D.4
1Division of Health Care Policy and Research, Department of Health Sciences Research,
Mayo Clinic, Rochester MN; 2Department of Mechanical and Industrial Engineering,
University of Massachusetts, Amherst MA; 3Department of Industrial and Systems
Engineering, North Carolina State University; 4Institute of Technology Assessment,
Massachusetts General Hospital.
* Dr. Banerjee completed this paper while working at the Mayo Clinic. His current
affiliation is Analysis Group, Inc., 111 Huntington Ave., 10th Floor, Boston, MA, 02199
Correspondence to: Hari Balasubramanian
Department of Mechanical and Industrial Engineering
University of Massachusetts at Amherst
160 Governors Drive
Amherst, MA 01003
413.577.3208 (Tel) 413.545.1027 (Fax)
Running Title: Improving access through panel redesign
Key Words: Primary Care Access, Continuity of Care, Systems Engineering
Word Count: 3391
Number of references: 32 Number of Tables: 3 Number of Figures: 2
Acknowledgement: Authors have no financial disclosures. We would like to thank Liang
Wang, Jason Egginton, and Sara Hobbs Kohrt for help with preparing this manuscript.
Improving Clinical Access and Continuity through Physician Panel
Running Title: Improving access through panel redesign
Word Count: 213
Population growth combined with the increasing prevalence of chronic disease due to
aging is projected to increase the demand for primary care services in the United States.
To use systems engineering methods to design physician panels that improve access to
and continuity of care.
We use numerical and simulation techniques to design physician panels based on
appointment and capacity data for 2004-2006 from a primary care group practice of 39
physicians with over 20,000 patients at the Mayo Clinic in Rochester, MN.
Patient waiting time and patient/clinician continuity, i.e., the number of times patients
redirected to see a provider other than their primary care physician (PCP).
Waiting time decreases by 57% [95% CI: 54.5%-59.5%] and continuity increases by 54%
[95% CI: 50.8% - 57.2%] in simulations that use the optimal panel design produced by
our numerical technique. The new panel design remains more efficient (less waiting,
more continuity) than a standard practice up to adding an additional 3500 more patients
to the new system.
Our simulation results indicate that redesigning primary care physician panels using
numerical techniques that trade-off access to and continuity of care has the potential to
increase the efficiency of primary care practices and may therefore help mitigate the
expected shortage of PCPs.
Recent estimates have suggested that the US faces an impending shortage in the
number of primary care physicians (PCPs).(1) These projections are based on forecasts of
both the needs of an aging population, and the number of PCPs who are likely to be in
practice in the near future.(2) An increasing number of Americans have at least one
chronic condition.(3) Current projections estimate that, due to an aging population, 25%
of Americans will have multiple chronic conditions by 2020.(4) PCPs, typically the first
point of contact between patients and the health system, play a particularly important role
in the management of chronic health problems, like back pain, asthma, diabetes,
hypertension and heart disease. Older patients, who often have multiple chronic ailments,
need careful management by a physician who has broad expertise and can spend the time
necessary to understand and manage their care.
Some authors find a strong positive relationship between primary care and greater
utilization of preventive services.(5) More effective use of primary care reduces
unnecessary and inappropriate specialist care. Moreover, greater access to primary care
reduces mortality-rates and is associated with system-wide positive effects, including a
more equitable distribution of health within populations.(5)
Insufficient primary care access appears to be rising and has perverse consequences 40%
of emergency department visits are reported to take place because patients are not able to
access their PCP.(6) From 1997 to 2001, the percentage of people reporting an inability
to obtain a timely appointment rose from 23% to 33% and in 2001, 43% of adults
reporting an urgent condition were unable to receive care when they wanted. (9)
Equally as important as access is continuity of care. Patients who regularly see
their own PCPs are 1) more satisfied with their care 2) more likely to take medications
correctly and 3) less likely to be hospitalized.(7-10) Conversely, not seeing your own
PCP can have an impact on efficiency and effectiveness of care provided (11) including
increasing the number of follow-up appointments.(12) Patients without a regular health
care provider also may be at higher risk of injury or adverse outcome.(13) The use of
unnecessary invasive diagnostic tests, or studies with risk, increases with physician
unfamiliarity with a patient.(14) Mainous et al. (2001) (15) found that roughly only 65%
of US patients in their study reported seeing their own PCP when they needed care.
Several proposals have been advanced to solve the problem of shortage in access
to primary care. Chief among these is payment reform which would reward physicians
for the quality rather than the quantity of care they provide (pay-for-performance) (16,17)
or payments for care coordination (medical home).(18,19) Some have proposed
incentives for patients to choose primary care providers.(22) The method we explore to
increase access and continuity of care is to improve the efficiency of physician panels –
the set of patients cared for by a physician. While others have proposed management of
patient demand, such as implementation of the advanced access paradigm (26), we
consider an alternative approach that simultaneously optimizes supply and demand
through the restructuring of patient panels. For a more technical discussion of related
work, see Balasubramanian, Banerjee et al. (32).
How might the design of a physician’s panel affect the ―efficiency‖ of a practice?
Size alone is not the only important factor. Together, the number of patients in a panel
and their disease burden composition determine the panels aggregate demand for health
care services. Optimally choosing a size-composition combination will allow a practice to
improve the care existing patients receive by reducing wait times for appointments and
increasing the frequency at which these patients see their own provider, as well as, allow
new patients to enroll, since good panel design may open up previously unavailable
We examined the panels of a primary care group practice at the Mayo Clinic in
Rochester, Minnesota. The practice consists of 39 physician panels and covers
approximately 20,000 patients living in Olmsted and surrounding counties. Since a period
in our model corresponds to one week, we collected weekly appointment and capacity
data for each physician from 2004 to 2006. An example of the variation in appointment
request rates can be seen in Figure 1 where we plot the distribution of weekly visits for
three categories of patients. The three distributions illustrate how differences in
appointment request rates can be attributed to gender and age.
For our analysis, we divided patients into 14 age categories of five-year
increments starting at age 18 and going through age 83. We further divide patients by sex
so that we end up with 28 categories in total. We chose this classification for simplicity
as suggested by Murray. (20)
The 39-physicians in the Mayo (Primary Care Internal Medicine) PCIM practice
cover 20,000 patients in Olmsted County in Rochester, MN. After accounting for part-
time and other activities (e.g., education and research), the practice group is equivalent to
17 physicians working full time. The average panel size is approximately 1200 patients
per physician panel. To obtain panel sizes more representative of the typical practice (~
2000/provider, see (23)), we used the following method. We increased the total
empanelled population to 34,000, while keeping the proportion of people in the different
demographic categories the same. The FTE adjusted panel size thus increased from 1200
per physician to about 2000 per physician on average. Note that the composition of
patients in the new panels is unchanged relative to the original panels. We also removed
physicians who worked less than a day a week on average from our analysis. In
summary, our experimental clinic has 33 physicians who cover 34,000 patients. We use
this modified data set to compare the baseline design against the optimal and capacity
In order to evaluate whether panel design can increase effective capacity, we have
developed a model of patient access to a primary care group practice in which the guiding
principle is the trade-off between timely access and continuity of care. Thus our model
seeks to choose a panel design that balances these two key, but opposing, forces. On the
one hand, timely access to care often necessitates a need to see any available physician.
This, however sacrifices continuity of care and brings with it the attendant problems
described in previous sections. A prudent choice of panel design – the ―optimal panel
design‖ – is one that jointly maximizes timely access and continuity of care.
While we compare designs based on timely access and continuity, we also report
on the utilization of the clinic. Specifically, we report two measures: the number of
unfilled slots in any week for the entire clinic, and the number of extra slots (additional
capacity) that the clinic needed on a weekly basis to meet demand.
In essence, the panel design problem is an allocation problem: given a set of
health categories, and a number of physician panels in a group practice, how many
patients from each category should be assigned to each panel?
Consider a simple example with three patient categories and three physician
panels as shown in Figure 2. The three categories have a total of 220, 370 and 450
patients, respectively, while each physician can see a maximum of 35 patients in one
week. Demand from each category for the week is assumed to be 10% of category size.
The number of patients from each category assigned to each physician’s panel is
indicated on the arrows in the figure. The dashed lines indicate redirection of flow for the
single-week model. Redirection means that patients have to see another provider since
their PCP is unavailable. The panel design in this example is not optimal; however, if the
number of patients assigned from category 3 to panels 1, 2 and 3 is changed to 20, 12 and
13, respectively, all patients would see their own PCP.
With an additional week, redirections could either be to the same provider in the
next week (implying waiting 1 week for an appointment) or to a different provider in the
same week. These are shown using solid arrows in the figure.
The model’s complexity significantly increases when there are multiple weeks,
the number of appointment requests for each panel is a random variable, and physicians
have a different capacity in each week due to vacations or other commitments. In some
panel designs, physicians may have unfilled slots in one week and an over-full calendar
in another, which will adversely affect patient waiting time and continuity of care.
Physician and staff morale and satisfaction may also suffer. Optimizing the composition
and size of panels thus becomes important.
To find the optimal panel design, we formulate the model as a Stochastic Linear
Program,(21) which we solve using numerical techniques. Such methods are an important
methodological area within the field of systems engineering, and have been applied to
many problems in other service industries including the design of financial portfolios,
design of transportation systems, and airline management. (22) The computed optimal
panel design is evaluated over 52 weeks using discrete event simulation which allows us
to calculate summary statistics like average waiting time and the number of redirections
to other providers.
The simulation works in the following manner: physicians start with an empty
calendar in the first time period, which in our case is one week. In each week, patients
make appointment requests that are satisfied on a first-come-first-served basis; if a
request was made in an earlier week, it is filled first, and ties between requests in the
same week are broken arbitrarily. When a physician’s calendar in a week is full, patients
can either choose to wait for a future week to see their own provider, or they can see
another physician in the same week (provided capacity is available). If capacity is not
available, extra slots are added to accommodate these patients.
A fraction of patients chooses not to wait additional weeks. In our model, we
assume 40% of patients decide to see other physicians in the same period. We base this
estimate on the rate observed at the primary care practice at Mayo Clinic. A fraction of
patients (50% in this case) who are redirected to other physicians are subsequently asked
to follow-up with their own PCP. This feature captures the idea that seeing another
physician generates additional follow-up appointments. (23)
In the discrete event simulation model, we sample randomly from historical visit
data for each of the demographic categories from 2004-2006. However, in order to factor
in seasonality, for any given week we restrict our sampling within an eight week window
around that week (four weeks on either side). Each physician has a weekly schedule that
we use to determine weekly capacity. The results we present are averages of 50
replications of the simulation for each design.
Our model is based on the following assumptions:
1. We divide patients into categories based on age and sex.
2. There are multiple time periods .
3. Physician capacity is fixed in a time period but may vary between time periods.
Assumption 1 is a simple way to categorize patients. We discuss alternative methods
below. Assumption 3 says that physician capacity is taken as given and is not a decision
variable in our model. This can also be modified to account for additions to staff or
changes in how a practice is managed.
The above information can be used to create panels that generate a relatively
stable appointment request rate which would allow practice managers to better plan their
staffing needs to optimize continuity while meeting expected demand for appointments.
One strategy is to populate each panel with patients from categories that generate
appointment request rates which are negatively correlated with each other. In a context
where there are two categories of patients, this means that when the request rate from one
group is high, the request rate from the other group tends to be low and vice versa. When
there are many groups, this is equivalent to choosing the panel composition such that the
total variation in request rates is minimized given an expected level of requests. This
principle was first proposed by Markowitz (1952) (24) in the context of designing a
diversified portfolio of assets. Our results indicate this concept can be applied to the
design of physician panels to maximize timely access to, and continuity of, care.
We compare the results from our optimal panel design arrived at through our
numerically solved Stochastic Linear Program with two other designs, also using discrete
event simulation. The first, which we call the baseline design, is the design currently used
by the PCIM practice. The baseline design allows a comparison of the optimal design
with current practice, and helps benchmark the improvements in timely access and
continuity that can be expected if the practice were to reallocate its patients optimally.
We also compare the optimal design with a simpler and easier to implement rule
of thumb that we call capacity-based panel design. We construct these panels as follows:
we tabulate each physician’s average share of the total average weekly capacity of the
group practice. So, if physician A sees patients on average for 40 hours a week out of a
total of 200 hours of patient-time by the group, her share is 20%. We use this proportion
to calculate her share of each category of patients, that is, physician A is assigned 20% of
each category. Thus, the capacity-based design is an intuitive allocation that allows us to
evaluate the performance of a practice in which panels are balanced based on average
Case Study: Adding New Patients to the Practice
Our age-sex categorization reveals that in this particular primary care practice,
younger adults of both sexes tend to ask for appointments less often. What would happen
to the access metrics if we increased the number of patients from these categories? As the
demand for primary care doctors increases in the United States, practices are routinely
faced with decisions regarding whether to empanel new patients. In panels that had young
patients, we increased the number of such patients by 25% which corresponded to 3500
more patients in the system. We analyze how the different panel designs perform under
As displayed in Table 1, based on the actual assignment of patients to PCPs, the
mean waiting time was 0.81 weeks and there were on average 400 redirections to other
physicians per week .
The bottom row in Table 1 shows that the capacity-based design produces an
average waiting time of 0.42 weeks and 218 redirections to other physicians per week.
Note that this design is 49% better in wait time and 46% better in the number of weekly
redirections relative to the baseline.
Table 1 shows that the optimal design reduces wait time by 57% [95% CI: 54.5%-
59.5%] and the number of weekly redirections by 54% [95% CI: 51.8% - 56.2%].
Table 2 shows the average number of unfilled slots (unused capacity) and the
additional or extra slots the clinic had to create to meet urgent requests. Note that in any
given week, if one of the measures is positive, the other has to be zero. The numbers in
the table are averages of the two measures over 52 weeks. The optimal design and
capacity-based design require fewer extra slots to be created on average. However, they
also have a higher number of unfilled slots on average per week.
Increasing Panel Size
Results are shown in Table 3. Waiting times and redirections increase under both
the current and optimal designs. In sensitivity analysis the optimal design does better on
both metrics with 2000 additional patients than the current design without additional
patients. Our simulation suggests that this finding remains valid with up to 2500
additional low-request patients in the system. Moreover, fewer numbers of high-request
patients, or some combination of the two, could also be added to the system.
Optimizing panel structure leads to reductions in wait times and maximizes
continuity. A closer look at the model results reveals that the stochastic linear program
algorithm better matches physician capacity with historical demand from each category
of patients. Thus, physicians who have less patient-time on their schedule are given
proportionally fewer patients from categories that tend to request more appointments and
vice versa. The capacity-based design also performs quite well relative to the base-case
for the same reason: physician capacities under this method are well matched with