Professional Documents
Culture Documents
To cite this article: Yu-Li Huang , Walton M. Hancock & Gary D. Herrin (2012) An alternative outpatient scheduling
system: Improving the outpatient experience, IIE Transactions on Healthcare Systems Engineering, 2:2, 97-111, DOI:
10.1080/19488300.2012.680003
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
IIE Transactions on Healthcare Systems Engineering (2012) 2, 97–111
Copyright
C “IIE”
Patient wait time has long been a recognized problem in modern outpatient health care delivery systems. Despite all the efforts to
develop appointment rules and solutions, the problem of long patient waits persists. Regardless of the reasons for this problem, the
fact remains that there are few implemented models for effective scheduling that consider patient wait times as well as physician idle
time and are generalized sufficiently to accommodate a variety of outpatient clinic settings. This paper presents a solution of designing
appointment slots for scheduling appointments in outpatient facilities that both patient wait time and physician idle time meet the
declared scheduling policies without overbooking and double-booking. Furthermore, this paper provides the implementation results
from three case studies to support the approach. The results confirm that the system can effectively reduce patient wait time as
much as 56% without significantly increasing physician idle time per patient and still allow physician to see and schedule exactly
same number of patients per clinic session. Consequently, this research improves the outpatient experience for both patients and the
medical professions, changes the perception of long waits in a physician’s office and ultimately enhances the quality of care.
Keywords: Outpatient scheduling, patient flow, patient wait time, physician idle time, process improvement
1948-8300
C 2012 “IIE”
98 Huang et al.
definitions; reported by AMA (2009). However, the CPT Also, actual case studies with implementation results were
did not finally become used nationwide by all carriers un- not available to support the feasibility of the theoretical ap-
til the end of the 1980s when the Resource Based Value pointment rules. This leaves a significant gap between the
System (RBVS), developed by Harvard University under theoretical constructs and the realities of practical appli-
the leadership of William Hsiao, was institutionalized in cations. Furthermore, regardless of how sophisticated the
the Omnibus Budget Reconciliation Act of 1989. With this proposed appointment rules are or how significantly they
Act, visits defined under the AMA’s different CPT codes have impacted the current scheduling systems, the vari-
were assigned RBVS units, which in turn dictated Medicare ability in physician treatment time has not been addressed
reimbursement of outpatient services, setting the precedent from the patient’s standpoint under clinical conditions. In
for all insurers and indelibly tying physician time to cost. essence, this simply means that patients are scheduled to
Together with the introduction of the RBVS, there was a accommodate the physician’s schedule.
concerted effort on the part of government carriers to re- The basic model, employed by most of the approaches
duce reimbursement for more expensive procedures, while and models presented in the literature (Ho and Lau,
increasing reimbursement for outpatient visits. Not sur- 1992; Ho, Lau, and Li, 1995; Klassen and Rohleder, 1996;
Downloaded by [University of Saskatchewan Library] at 09:45 11 January 2015
prisingly, there was a resurgence of interest in scheduling Yang, Lau, and Quek 1998; Vanden Bosch and Dietz,
efficiency, particularly in terms of cost. 2000; Denton and Gupta, 2003; Cayirli, Veral, and Rosen,
Shortly after the implementation of the Omnibus Bud- 2006; Wijewickrama, 2006; Kaandorp and Koole, 2007)
get Reconciliation Act, a number of studies developed focuses on minimizing total cost using the cost ratio
and evaluated appointment rules for efficient scheduling. approach while still trying to accommodate variation.
Ho and Lau (1992) formulated the patient wait time and The cost relative to physician idle time tends to be much
physician idle time categories associated with cost. They higher than to patient wait time due to the perception
evaluated various appointment rules using the “frontier” of higher physician compensation and cost of medical
approach and selected the eight best appointment rules ac- facility (Keller and Laughhunn, 1973; Yang, Lau, and
counting for the major environmental factors which impact Quek, 1998). Hence, this basic model inherently favors the
scheduling, such as the probability of no-shows, the coeffi- reduction of physician idle time. Moreover, to the best of
cient of variation of service time and the number of patients our knowledge, clinical policy or constraints have been
per session. Ho, Lau, and Li (1995) evaluated appointment discussed by many researchers on developing a schedule,
scheduling rules and introduced the concept of variable- but limited, such as the time at which the last patient
interval rules to reduce cost and improve service quality. should be seen in a session or the session should end.
Klassen and Rohleder (1996) evaluated the appointment Some approaches presented in the literature are case-
rules using a design of experiment approach for differ- specific, which precludes generalization. One group used
ing patient characteristics and concluded that scheduling simulation approaches to design for a specific clinic such as
a ‘low-variation patient type’ in the beginning of a ses- Swisher et al. (2001), Guo, Wagner, and West (2003), and
sion was the best solution among all appointment rules. Wijewickrama and Takakuwa (2005). Those simulation
Yang, Lau, and Quek (1998) developed a new appoint- studies, regardless of their focus, arrived at similar conclu-
ment rule taking into account variations in service time, sions regarding the factors that influence patient flow and
the percentage of no-shows, the number of appointments admission policies. Another group tended to rely on trial
per session, and the cost ratio of physician and patient and error to select the best schedule for a particular clinic
wait time. Harper and Gamlin (2003) identified and evalu- from a range of possible solutions such as Meza (1998),
ated 10 different scheduling policies for a specific clinic and Harper and Gamlin (2003), Klassen and Rohleder (2004).
found that mixing different appointment intervals over the One other group proposed the Open Access approach such
duration of a clinic session is better in terms of reducing pa- as Murray and Tantau (1999), O’Hare and Corlett (2004),
tient delay. Cayirli, Veral, and Rosen (2006) used a simula- Green and Savin (2008). This approach focused on reduc-
tion approach to build a two-factor experiment, sequencing ing no-shows and improving patient access to services.
rules and appointment rules, including a number of unpre- Similarly, some research such as Kim and Giachetti (2006),
dictable factors such as walk-ins and no-shows and found LaGanga and Lawrence (2007), Muthuraman and Lawley
the impact of the choice of sequencing rules is greater than (2008) suggested that overbooking and double-booking
the choice of appointment rules. Wijewickrama (2006) used would reduce physician idle time and increase profitability
a simulation approach to evaluate four appointment rules and patient access. These approaches focus on maximizing
and their possible combinations and concluded that a hy- the utilization of the clinic capacity especially physicians’
brid combination of appointment rules works better than a time to reduce health care cost.
single rule. Despite the insights generated by research, the Cayirli and Veral (2003) conducted a fairly inclusive sur-
appointment rules developed did not seem to be broadly vey of the literature on the topic of outpatient schedul-
adopted by outpatient clinics because they are difficult to ing appointments, which provides an extensive review
follow and include no clear explanation as to which rules are of the problems associated with the definitions and for-
most appropriate in which type of outpatient environments. mulations of outpatient appointments, the performance
Improving outpatient experience 99
measurements and evaluations, and the historically used time for the physician. The perception is that by scheduling
analysis methods. However, Cayirli and Veral concluded 15 minutes for a service that takes 25 minutes, there will
that the studies they reviewed had one or more major limi- always be enough patients to keep the physician occupied,
tations: 1) most studies are case-specific and, therefore, dif- even in the event of no-shows or late arrivals. In addition,
ficult to generalize; 2) the models of patient flow are pred- time estimations are often based on a physician’s perception
icated on unrealistic assumptions; 3) the studies focus on of the ‘ideal’ visit or what should be accomplished in a given
developing appointment rules to accommodate additional type of visit, without considering the actual variations that
clinical scenarios as opposed to exploring the variation of occur. This again reflects a ‘physician-centric’ approach to
patient flow; 4) studies do not successfully implement ap- time estimation. Yet, whatever the underlying cause of the
pointment rules, presumably due to a lack of understanding unrealistic estimation, a simple remedy is to conduct a de-
of how rules are implemented in practice and how decisions tailed time study to provide accurate data about the actual
are made regarding which rule to follow. time involved in each visit type. The impact of inaccurate
To develop a more generalized approach to effective estimates of treatment times is discussed in section 3.
scheduling that could be easily implemented to reduce pa- In addition to determining the actual time that should
Downloaded by [University of Saskatchewan Library] at 09:45 11 January 2015
tient wait time without significantly impacting physician be allocated to a particular physician service, it is also im-
idle time, we felt we needed to address a number of out- portant to define what is typically included in a particular
standing questions: How do treatment times that are not service besides strict face-to-face physician-patient contact.
based on actual data impact the success of a schedule? How For our purposes, we decided that to establish a realis-
can the actual data be most appropriately utilized to esti- tic time estimation for a visit type, it is critical to include
mate the treatment time distribution, to take into account any ancillary services, such as reviewing charts, dictating
the variation of the treatment time, and to finalize the best findings, studying labs and x-ray reports, that may be per-
scheduled treatment time interval? formed outside the exam room, but which are ‘triggered’ by
or directly related to the service for a particular patient.
However, to counter physician-centric scheduling, we
1.2. Problem statement
needed to incorporate a patient arrival schedule into the
The problems faced in designing a model are how to, first, model together with the physician service time. The dis-
best model the scheduling of procedures and appointments tinction between patient arrival time and the actual ap-
so that both patient and physician wait times meet the pointment time, which we defined as the encounter time or
declared policy of a facility, such as clinic duration or start service start time, is often blurred in practice. Therefore, it
and finish times, and, then, second, generalize that model is not uncommon for a patient to arrive on time for the
so that it can be effectively used in a variety of clinic settings scheduled appointment, but be delayed at the front desk
with differing policies. completing necessary paperwork. Even though the patient
In designing a model, our specific objectives were to enlist may not perceive this as wait time, physicians often regard
a number of outpatient clinics, collect accurate data about this unplanned delay as idle time. It is not uncommon for
the actual time physicians spend with patients for each type physicians to fill this idle time with activities which, in turn,
of visit, then assign a realistic time interval to each type of may often extend beyond the initial delay causing a cas-
visit under clinic policy or constraints without overbook- cade that compounds patient wait time throughout a given
ing, and finally implement the model in the participating session. Hence, to eliminate the discrepancy between what
clinics to validate performance in realistic settings. From patients understand as arrival time and clinic staff regard
the outset, our primary goal was to develop a practical and as encounter time, we made a distinction between arrival
easy-to-use approach to scheduling that would equitably time and appointment time in our model.
reduce, rather than try to eliminate, systemic wait times In short, our research objective was to develop a method-
for both physician and patient. Given the large number of ology to best schedule procedures and appointments in
variables that can affect scheduling, we felt it was more im- accordance with clinical policy or constraints so that pa-
portant to provide a tool for managing the schedule rather tient and physician idle times are limited or, on occasion,
than controlling it. Furthermore, since the true benefit of even eliminated. Specifically, our aim was to provide an ap-
any model or approach can only be realized if the principals proach based on simulation that could be easily adapted
are willing to use it, it was imperative that the approach be to any clinical setting and that would reliably provide best
neither difficult nor too complex to implement. scheduling practices responsive to both physician and pa-
We believe that much of the wait time experienced by pa- tient interests.
tients is the direct result of unrealistic estimations of treat-
ment time or unrealistic solutions to chronic problems. Un-
derestimating visit times, overbooking or double booking 2. Proposed solution
all represent a ‘physician centric’ solution to the problem
of ‘no-shows’ or late patients: events that can unexpectedly The goal is to provide a solution of designing a prede-
affect any daily schedule and create sudden unplanned idle termined scheduling template that accounts for patient
100 Huang et al.
wait time and physician idle time without overbooking and search balances patient wait time and physician idle time
double-booking. There are steps for the proposed solution. by determining the Wait Ratio between them. This in turn
allows us to determine the best treatment or service time
1. We first collected data on the treatment times for all
interval for each type of visit for a given Wait Ratio; dis-
services rendered by the clinic providers, including the
cussed in 2.1.1. Determining the best treatment time in-
physician, residents, registered nurses (RNs), and medi-
tervals across different visit types by the same wait ratio
cal assistants (MAs) to estimate treatment time param-
is to insure every patient is treated equally in terms of
eters and distribution for each visit type.
waiting.
2. Then, we determined the best (i.e., maximum) sched-
The best scheduled time interval is defined as the max-
uled time interval for each visit type under any declared
imum scheduled time interval allowed for each visit type
clinic policy or constraints. Simulation was chosen as the
that satisfies medical and clinic constraints. This section
solution technique since Cayirli and Veral (2003) sum-
provides the mathematical model of how the best sched-
marized from numerous literature that the advantage of
uled time intervals are generated and discusses policy or
simulation modeling is the ability to present the complic-
constraints that impact the decision.
ity and the difference of outpatient clinic environments.
Downloaded by [University of Saskatchewan Library] at 09:45 11 January 2015
Two simulation models were developed. First, a model 2.1.1. Definitions and formulations
for a single visit type was built to determine the best
scheduled time intervals for any given ‘wait ratio’ (the For our purpose, treatment or service time is defined
ratio between patient wait time and physician idle time) as the time from which a physician enters the exam
for each visit type. Second, an extended model sequenc- room to the time at which the physician finally exits the
ing the various visit types was constructed for the clinic exam room and includes any ancillary physician service
to finalize the optimal best scheduled time intervals un- for that patient such as reading charts or dictating. The
der clinical policy or constraints, assuming perfect clin- model is focused on patient flow in the physicians’ treat-
ical conditions such as punctual patients, no physician ment process; see Figure 1 for the graphical representa-
delay, no conflict in availability of staff or equipment tion of the model. The parameters defined in the model
(e.g., for x-rays, EKGs, or vital signs) for the purpose of are:
designing scheduling template. Ti = physician’s service time to treat patient i where i =
3. The provider schedule was subsequently developed from 1, 2, 3, . . . , n. Ti ∼ D(µ, σ 2 ), D is a probability distribution
the optimal best scheduled time interval established for with mean µ and standard deviation σ .
each visit type. n = the number of patients scheduled per session. A
4. Once the provider schedule had been determined, the session could be a day, a morning or an afternoon.
patient arrival schedule was constructed based on any d = number of standard deviations away from the mean
ancillary patient activities associated with a given visit. µ.
X = the scheduled time interval for a patient to see the
physician in minutes
2.1. Determination of the best scheduled time interval X = µ + dσ (1)
A successful appointment system should minimize patient Si = the scheduled time to start patient i where i =
delays while fully utilizing medical resources. However, 1, 2, 3, . . . , n and let S1 = 0
there is a tradeoff in that reducing patient wait time may
increase physician idle time and vice versa. Hence, this re- Si = Si −1 + X = (i − 1) X (2)
Fig. 1. The graphical representation of patient flow model in physician service process.
Improving outpatient experience 101
Fi = the finish time for patient i where i = 1, 2, 3, . . . , n Table 1. The average scheduled time intervals for each visit type
Ai = the actual time to start patient i where i = with various wait ratios
1, 2, 3, . . . , n. Assuming a punctual patient, from (2), Ai
Visit type (k)
and Fi are:
Wait ratio (R) NP RV FU ... HP
Fi = Ai + Ti (3)
Si if Fi −1 < Si 1 X1NP X1RV X1FU ... X1HP
Ai = (4) 2 X2NP X2RV X2FU ... X2HP
Fi −1 Otherwise
3 X3NP X3RV X3FU ... X3HP
Wi = the wait time for patient i in minutes, where i = .. .. .. .. ..
.
..
. . . . .
1, 2, 3, . . . , n. From (2) and (4), Wi is:
/ XlNP XlRV XlFU ... XlHP
Wi = Ai − Si (5)
W̄ = the average patient wait time based on (5): k = visit type, where k = NP (new patient), RV (return
Downloaded by [University of Saskatchewan Library] at 09:45 11 January 2015
n visit), FU (follow up), HP (pre-operation) . . .
Wi Therefore, XRk is the average scheduled time for visit
i =1
W̄ = (6) type k given Wait Ratio R and this is the end of the first
n
simulation; see Table 1.
Pi = the physician idle time waiting for patient i in min- Once the average scheduled time interval of each visit
utes, where i = 1, 2, 3, . . . , n. From (3) and (4), type for each wait ratio are established, the second simula-
Pi = Ai − Fi −1 (7) tion is run with an extended simulation model sequencing
the various visit types for a clinic. The sequence is decided
P̄ = the average physician idle time based on (7): based on the clinic preference. However, the suggested se-
n quence is to distribute each visit type evenly throughout a
Pi clinic session. This provides options for patients to accom-
i =1
P̄ = (8) modate their personal schedule. With the extended simu-
n lation model, the average scheduled time interval for each
R = the Wait Ratio, which is the degree to which patient visit type at any given wait ratio can be inputted and evalu-
wait time (6) exceeds R times of physician idle time (8), ated to determine the best wait ratio (R ∗ ) that satisfies the
where R = 1, 2, 3, . . . , l and l ∈ R+ , assuming patient wait clinic policy or constraints (discussed in 2.1.2) assuming
time is normally longer than physician idle time on average: perfect clinical conditions, which gives the optimal sched-
uled time interval ( X̄ ∗ ) for each visit type. An example is
W̄ presented in 2.4 to demonstrate the approach.
= R ⇒ P̄ × R = W̄ (9)
P̄
Use (9) as the objective function to find the value for the 2.1.2. Clinic policy and underlying constraint
decision variable (d) and scheduled time interval (X) at any To assure that our model does reflect actual clinic opera-
given R. tion, certain common clinic policies and underlying con-
Then a number of simulation runs are needed to conclude straints should be accounted for in order to generate the
average schedule time interval X̄ at any given R. Let: best treatment time interval:
m = the number of simulation runs (based on desired
1. Clinic or session finish time: Most clinics have either
error bounds or confidence interval widths)
4-hour session (half day) or 8-hour session (full day).
d j = the decision variable of simulation run j where j =
The normal finish time could be at noon or at 5:00 p.m.
1, 2, 3, . . . , m
However, some physicians would like to finish earlier, for
X j = the scheduled time interval of simulation run
example, finishing at 11:30 a.m. for a 4-hour morning
j where j = 1, 2, 3, . . . , m
session.
Equation (1) can be rewritten to:
2. Time of last appointment: Most clinics have their own
Xj = µ + d j σ (10) preference as to when the last scheduled patient should
be in for the appointment. For example, the last patient is
Therefore,
scheduled to come in at 10:30 a.m. for a 4-hour morning
m
session.
Xj
j =1
3. Number of patients to be seen (or number of desired ap-
X̄ = (11) pointment slots) in a given session: This decision should
m
be made by a clinic management team to best serve
Then, we need to find X̄ for each visit type with various patients’ needs based on the seasonal demands. This
wait ratios (R). Let: decision also highly depends on specialty.
102 Huang et al.
4. Average patient wait time and maximum patient wait to replace conventional cost ratios, in large part because
time: Some clinics use this as a measurement to evalu- physicians tend to overestimate the cost of their time as
ate their service quality. As mentioned in Huang (1994), opposed to patients’ time, most likely due to their lack of
the survey results that patient tolerance for delay dimin- criteria for accurately evaluating time cost for the patient.
ishes after approximately 30 minutes. Clinics could use This method effectively eliminates the cost of patient time
this 30-minute as a benchmark for their performance and the bias inherent in cost ratios from the model in favor
measurement. of well defined policy and constraints. In the case where a
5. Average physician idle time and maximum physician idle solution cannot be found based on these constraints, clinics
time: Physicians are the most costly element in the entire should either consider loosening up clinic constraints or
clinic visit process. Most clinics would like to utilize evaluate their capacity whether clinics have scheduled more
their physicians at their full capacity or minimize their patients than they can possibly handle on a regular basis.
physicians’ idle time.
6. The sequence of the various visit types: Some studies
2.2. Provider schedule
concluded that the scheduling sequence of the various
Downloaded by [University of Saskatchewan Library] at 09:45 11 January 2015
visit types in a clinic session has significant impact on pa- Based on the optimal scheduled time interval ( X̄ ∗ ) for each
tient waiting, such as Klassen and Rohleder (1996) and visit type, the provider schedule from equation (2),Si =
Harper and Gamlin (2003). However, an optimal se- Si −1 + X̄ ∗ , is constructed so that appointment slots are
quence may not provide clinics a flexible enough sched- consecutive, without overbooking or double-booking. For
ule to accommodate patients’ needs. Each clinic man- the physician, this schedule represents the actual time at
agement team should be able to decide its own sequence which each patient encounter should begin. This schedule
for individual physician without any restriction. can also provide a timeline for physicians to best utilize their
7. The underlying constraint for deciding the best treat- ‘idle’ time in between patients. For example, a single follow-
ment time interval is to ensure that the probability of up visit type with X̄ ∗ = 7.4 minutes, the physician schedule
patient waiting given this time interval is less than 50%. starts at 8:00 a.m. (S1 = 0), 8:07 a.m. (S2 = S1 + X̄ ∗ = 7.4),
This constraint defines the minimum value of what the 8:15 a.m. (S3 = S2 + X̄ ∗ = 14.8), and so on.
best treatment time interval should be for a good patient
flow. From the equation (4), the current patient delays
2.3. Patient arrival schedule
are highly influenced by the wait time (Wi −1 ) and treat-
ment time (Ti −1 ) from the previous patient. Since the Once a physician schedule is established, then the cor-
treatment time (Ti ) for each patient is a random vari- responding patient arrival schedule must be determined.
able, the only controllable factor here is the scheduled Again, the main concept behind the arrival schedule is
time interval ( X̄ ∗ ). It is desirable that the probability of to provide sufficient time between the patient arrival at
Ti − X̄ ∗ ≤ 0 is higher than at least 50%. the clinic and the actual examination time for the pa-
tient to complete activities such as signing in, filling out
In other words, given Ti ∼ D(µ, σ 2 ) and best time in- paperwork, having vitals taken, having an x-ray taken,
terval X̄ ∗ , the objective should be Pr(Ti ≤ X̄ ∗ ) ≥ 0.5; that providing a specimen, and moving between lab or x-ray
is the probability of a wait less than or equal to 50% (see room and exam room. Let Yi = arrival time for patient
Figure 2 where the area under the curve less than best time i and Bi = time scheduled for pre-visit activities for pa-
interval X̄ ∗ is greater than or equal to 0.5). tient i; therefore, Yi = Si − Bi . For example, assuming an
In short, our model does not aim at forcing the clinic 8-minute x-ray (Bi = 8) is needed for some patients and
setting to fit the solution, but rather aims at fully utilizing a patient is scheduled to see the physician at 8:13 a.m.
the available resources and capacity to achieve the best (S3 = 12.8), if he/she is required to have an x-ray taken
solution. Although clinic policy and constraints are limited before seeing the physician, the arrival time is 8:05 a.m.
in the literature, they are used here to determine a wait ratio (Yi = Si − Bi = 12.8 − 8 = 4.8), otherwise the arrival time
is 8:13 a.m. (Yi = Si − Bi = 12.8 − 0 = 12.8). The time as-
signed to pre-visit activities will differ from clinic to clinic
and between specialties. However, if the time needed for
these activities is not well defined, wait time will be com-
pounded for either physician or patient. Ideally, the physi-
cians should be able to maintain their schedules without
contributing significantly to patient wait.
2.4. An example
Fig. 2. The constraint of the probability less than the best time The example of a case study from an Orthopedic Surgery
interval is at least 0.5. Clinic is presented here to demonstrate how this approach
Improving outpatient experience 103
Table 2. Average scheduled time intervals for various wait ratios that the physician service times (Ti ) of all visit types can
be best fitted by Gamma distribution with averages of 10.6,
Average scheduled time intervals (in minutes)
7.3, 5.5 minutes and standard deviations of 4.5, 4.0, 3.4 min-
Wait ratio FU NP XR utes for NP, FU, and XR, respectively. Therefore, the aver-
age scheduled time interval ( X̄) for each visit type given wait
1 9.2 12.7 6.8 ratios (R) from 1 to 20 after 50 simulation runs (m = 50) are
2 8.6 12.1 6.4 shown in Table 2. The number of simulation runs taken was
3 8.4 11.8 6.1 based on the error rate to be within ±0.1 with probability
4 8.2 11.5 6.0 of 0.95 and the initial estimation of population standard
5 8.0 11.4 5.9
deviation to be 0.355 for the FU patient type. These as-
6 7.9 11.3 5.8
7 7.8 11.2 5.7 sumptions generated the final number of simulation runs
8 7.8 11.1 5.7 must be at least as large as 48.5, which 50 was chosen.
9 7.7 11.0 5.6 Then, after simulating the extended clinic model, the
10 7.6 11.0 5.6 results after 50 simulation runs are in Table 3.
Downloaded by [University of Saskatchewan Library] at 09:45 11 January 2015
Table 3. Simulation results for wait ratios (1–20) for case of orthopedic surgery clinic
Finish Time with Various Wait Ratios Last Patient Scheduled Time with Various Wait
240 230 Ratios
220
230
220
Clinical Contraint Finish Time < 210 minutes (11:30 am) 200
210 190
180
200
Clinical Contraint Last Patient Scheduled
170 Time < 180 minutes (11:00 am)
190
160
Downloaded by [University of Saskatchewan Library] at 09:45 11 January 2015
180 150
1:1
2:1
3:1
4:1
5:1
6:1
7:1
8:1
9:1
10:1
11:1
12:1
13:1
14:1
15:1
16:1
17:1
18:1
19:1
20:1
1:1
2:1
3:1
4:1
5:1
6:1
7:1
8:1
9:1
10:1
11:1
12:1
13:1
14:1
15:1
16:1
17:1
18:1
19:1
20:1
Wait Ratios Wait Ratios
Fig. 3. The results of finish time with various wait ratios for the morning session shows that the best wait ratio is 7:1 for a finish time
of 11:30 a.m. or 210 minutes (left). For the last patient scheduled arrival time with various wait ratios, the results show that the best
wait ratio is 18:1 for a last patient visit before 11:00 a.m. or 180 minutes (right).
∗
8:13 a.m. (S3 = S2 + X18XR = 7.4 + 5.3 = 12.7), 8:23 a.m. uled at 8:03 a.m. (Y3 = 12.7 − 10 = 2.7), which rounds to
∗ 8:05 a.m. in reality.
(S4 = S3 + X18NP = 12.7 + 10.6 = 23.3), and so on; see
Table 4.
Finally, assuming 8 minutes for x-ray and 10 minutes
for other pre-visit activities, the patient arrival schedule is 3. Simulation result supporting importance of data
constructed as in Table 4. The first visit is FU and sched- collection
uled to see physician at 8:00 a.m. (S1 = 0), so this patient
should come in 10 minutes (B1 = 10) early for pre-visit This section is to demonstrate the impact of designing an
activities. Hence, this patient should be scheduled to ar- appointment template without considering the role of data
rive at 7:50 a.m. (Y1 = −10). The second visit is XR and collection. We actually demonstrated through our simu-
scheduled to see physician at 8:07 a.m. (S2 = 7.4), so this lation how critical even small errors in time estimation,
patient should come in 18 minutes early (10 minutes for such as one minute, can actually have on patient wait time,
pre-visit activities and 8 minutes for taking x-ray,B2 = 18). physician idle time and finish time in the course of a single
Hence this patient should be scheduled to arrive at 7:49 session. We believe that long wait times in current outpa-
a.m. (Y2 = 7.4 − 18 = −10.6), which rounds to the near- tient scheduling systems come primarily from inaccurate
est 5 minute increment to be 7:50 a.m. For the third visit, estimates of treatment time.
NP, scheduled to see physician at 8:13 a.m. (S3 = 12.7), A simulation was developed to demonstrate the im-
this patient should arrive 10 minutes (B3 = 10) early for pact of treatment time estimations on patient wait time,
pre-visit activities. Hence, the arrival time should be sched- physician idle time and total treatment time per day. The
Table 4. Patient arrival and physician schedule for case of orthopedic surgery clinic
Physician schedule The optimal scheduled Pre-visit activity scheduled Patient arrival
i Visit type (Si = Si −1 + X̄ ∗ , S1 = 0) time interval ( X̄ ∗ ) time (Bi ) (Yi = Si − Bi )
106
Table 5. Detail summary results for three case studies
Cases Case 1: Orthopedic Surgery Case 2: Plastic Surgery Case 3: Vascular Surgery
Clinic Type Private Teaching Teaching
Number of Residents None One Two
Scheduling Approach Scheduled by Physician Scheduled by Scheduled by Physician
Resident Physician
Clinical Constraints 1. Last patient schedule at 11 am 1. Finish clinic at 4:45 pm 1. Finish clinic at 5 pm
2. Finish clinic at 11:30 am 2. Schedule 26 patients a day 2. Schedule 32 patients a day
3. Schedule 25 patients a session (6 NP, 13 FU, 6 XR) 3. Resident sees all NPs (8 NP, 24 RV).
6 NP, 3 POP, 4 HP, 5 RV 3 POP, 5 RV
Parameters Ti ∼ Gamma(µ, σ 2 ): (10.6, 4.52) for NP, (7.3, 4.02) for FU, (5.5, n = 26; Bi = 10; m = 50; Ti ∼ Gamma(µ, σ 2 ): Ti ∼ Gamma(µ, σ 2 ): (19.7, 9.22) for NP,
3.42) for XR; n = 25; Bi = 10 or 18 (with x-ray); m = 50. (30.6, 12.22) for NP, (15.9, 6.52) for POP, (11.1, 7.22) for POP, (12.7, (14.1, 6.12) for RV; n = 32; Bi = 20 (with
(15.9, 8.42) for RV, (18.6, 7.32) for HP 7.42) for RV resident); m = 50.
No-show Rate 9% 10% 4%
∗ ∗
Scheduled Time Interval (X) (min) Visit Type Original X(R) Proposed Visit Type Original X(R) Proposed X̄ (R = 3) Visit Type Original Proposed
and Associated Wait Ratio (R) X̄ ∗ (R ∗ = 18) X(R) X̄ ∗ (R ∗ = 9)
NP 10(58) 10.6 NP 30(25) 33.3 NP 30(1/9) 20.7
FU 5(2000) 7.4 RV 15(33) 17.5 13.8 RV 15(6) 14.6
∗ 12.7
XR None 5.3 POP 15(53) 17.4
HP 15(1500) 20.4
Average Patient Wait Time (min): Before After Before After Before After
(95% C.I.) (95% C.I.) % Reduction (95% C.I.) (95% C.I.) % Reduction (95% C.I.) (95% C.I.) % Reduction
Data Collection (3 months, 3 weeks) 27.8 13.1 53% 15.0 7.5 50% 27.8 12.4 56%
(26.9, 28.7) (10.9, 15.3) (13.1, 16.9) (4.2, 10.8) (23.8, 31.8) (8.1, 16.7)
Simulation (50 runs) 26.9 13.0 52% 15.8 7.4 53% 25.2 12.6 50%
(23.7, 30.1) (9.4, 16.6) (12.7, 18.9) (5.9, 8.9) (18.1, 32.3) (7.9, 17.3)
Simulated Average Physician Idle Before After Before After Before After
Time (min)
0.2 0.8 5.4 5.3 2.0 1.1
∗
Physician considered XR as squeeze-in patient, hence no time was assigned.
Improving outpatient experience 107
4.2. Clinic conditions and constraints pedic Surgery prefers to finish each session in three and a
Certain common clinic conditions are considered in our half hours, i.e., 11:30 a.m., if possible and the last patient
simulation model for the performance evaluation such as should be scheduled at, for example, 11:00 a.m., to im-
no-show rate (including actual no-shows, cancellation and pose the 11:30 a.m. finish time. The participated physician
open slots), add-ins, patient and physician lateness (espe- prefers to use this last 30 minutes to dictate his patients. The
cially in the beginning of a session), x-ray conflicts, over- Plastic Surgery prefers to finish 15 minutes earlier if possi-
write one patient type to another, and the original schedul- ble for wrapping up a clinic session. The Vascular Surgery
ing templates (sequences) and times for these three clinics would prefer to complete a nine-hour session around 5 p.m.
(see Table 6). The most critical concern from these clinics
for redesigning the scheduling templates is the templates
must have the same number of slots and a similar number 4.3. Prior implementation performance and simulation
of slots for each patient type to ensure they are able to model development
meet the current demands. Two of the clinics, Orthopedic Three months of data collection indicated that patient wait
and Plastic Surgery, are using a four-hour session and the times are 28 minutes on average for both Orthopedic and
Downloaded by [University of Saskatchewan Library] at 09:45 11 January 2015
Vascular Surgery is using a nine-hour session. The Ortho- Vascular Surgery and 15 minutes on average for Plastic
Table 6. Appointment scheduling templates for three clinics (a) original (b) proposed
Surgery. In general, medical staff addressed their stress lev- including cancellation and open slots is about 9%. 2) The
els are very high due to patients’ complaint about waiting physician tends to be late for his first appointment by 18
and service quality and the pressure from the physician. In minutes on average. The physician explained that since the
addition, when physicians are behind their schedule, they first patients were not generally ready to be seen at the
tend to move patients along quicker, which reduces service assigned appointment time, he simply got in the habit of
quality. In addition, some patients complained about the starting late. More than likely the physician’s perception
flexibility of time slots, especially for the Plastic Surgery. results from a discrepancy between what he assumes to be
The clinic only scheduled new patients in the afternoon the actual appointment time and what is in fact the arrival
session, which supports our concerns about the optimal se- time of the patient. 3) There are two physicians working
quencing of patient slots. The optimal sequencing may be at the same time and there is only one available x-ray ma-
preferred by clinics but may not be flexible for patients to chine. The conflict of using x-ray has been an issue for this
schedule their appointments. Furthermore, in order to un- clinic. 4) Patients’ lateness for appointment during the first
derstand the current clinic situation, we built a simulation 30 minutes is about 9 minutes on average. 5) Three patient
model for each clinic that reflects current clinic conditions. types are used: FU (follow-up), XR (x-ray), and NP (new).
For the Orthopedic Surgery Clinic, the following condi- 6) Overwriting the designated slots is commonly occurred;
tions were considered: 1) The probability of no-show rate for example, schedule a FU patient in a NP slots.
Improving outpatient experience 109
In the case of the Plastic Surgery Clinic, the conditions be easily adjusted to the new templates. The only change
considered were: 1) The probability of no-show rate made is when patients are scheduled to come in. Medical
including cancellation and open slots is about 10%. 2) One staff including physicians continues to perform their nor-
or two new patients on average are added in the afternoon mal clinic functions. Hence, the impact is relatively minimal
session outside the regular template. 3) Four patient types when implementing the proposed solution.
are used: HP (pre-operation), POP (post-operation), RV
(return visit), and NP (new). 4) Overwriting the designated
4.5. Post-implementation performance
slots. 5) There is normally one resident available and the
physician prefers to have the resident evaluate patients The results shown in Table 5 indicate that there is no statis-
first, especially new patients (NP). Therefore, we proposed tical difference on average patient wait time between the re-
a schedule template based on resident’s time. Since all sults of simulations and actual data collections according to
pre-operation patients (HP) are only seen by the resident, 95% confidence intervals for each case. This validates that
so we placed a couple of scheduling slots, especially return our extended simulation models reflects clinical conditions
visit patients (RV) and post-operation patients (POP), in closely. The implementation results from a three-week data
Downloaded by [University of Saskatchewan Library] at 09:45 11 January 2015
conjunction with HP for the physician to see without the collection indicates that the percent reduction on average
resident. Hence, we proposed a scheduling template with patient wait time are 53%, 50%, and 56% for Case Stud-
6 NPs, 3 POPs, 4 HPs, and 5 RVs for the resident with the ies 1, 2, and 3, respectively; despite the estimated average
physician and 3 POPs and 5 RVs (bolded in Table 6(b)) physician idle time varies less than one minute per patient.
for the physician only. The associated best scheduled time Table 6(b) shows how the proposed physician and patient
intervals are also different (see Table 5). It is ideal to keep arrival schedules are constructed. The proposed schedul-
the physician busy while the resident is seeing HP patients. ing sequences are based on distributing each patient type
Some variations in the proposed scheduling template due as evenly as possible throughout a clinic session. The re-
to resident’s involvement are justified in this case. sults also indicate that the clinics from Case Study 1 and
As for the Vascular Surgery Clinic, the conditions con- 2 underestimated the provider’s treatment time for all pa-
sidered are: 1) The probability of no-show rate is about tient types while the clinic from Case Study 3 overestimated
4%. 2) Two patient types are used: RV (return visit) and NP them given the clinical constraints. Besides the fact that the
(new). 3) Overwriting the designated slots. 4) The physician proposed solution significantly reduces average patient wait
is needed once or twice to help with the patients scheduled time, the physicians from each case study are still allowed
for the nurse practitioner. This occupied about 20 minutes to see and schedule exactly same number of patients within
of the physician’s time from his own clinic session. the same time frame as before without overbooking and
double-booking.
After the implementation, the participated medical staff
4.4. Implementation reasons and challenges
in the Orthopedic Surgery noted increased patient satis-
The underlying reasons for improvement for these three faction with the service, especially in respect to wait time,
clinics are: 1) The design of the scheduled time intervals as well as a significant drop in the stress the staff had ex-
is not based on actual treatment time, which leads to in- perienced when behind schedule. The attending physician’s
accurate estimated treatment time intervals that cause the assessment of the new schedule in the Plastic Surgery is pos-
waiting. 2) Clinics’ insensitivity to the wait ratio between itive. She said “I’m really shocked at what a difference this
patient and physician wait time results in long patient wait has made. I don’t think I’ve run behind since your sched-
time. For examples from Table 5, the original wait ratios ule has been implemented. Plus, patients have the option
from each visit type of Case Study 1 and 2 ranges from of morning or afternoon appointments, which they like.”
R = 25 to R = 2000 or not even assigned, which are rel- The participated physician and medical staff in the Vascu-
atively high compared to the proposed R ∗ . This indicates lar Surgery indicated that the stress level to the end of the
clinics are not sensitive to patients’ waiting and prefer new day has dropped significantly. All comments indicated the
patient over other patient types. 3) Clinics allow schedulers success of the proposed approach.
to overbook and double-book on the designed schedule
template to keep physician busy at all times (see Table
4.6. Sustainability guidelines and recommendation
6(a)). Case Study 3 is a typical example where wait ra-
tio is reasonable enough to create a good patient flow, but To sustain the proposed approach, some overwriting guide-
the patient wait time is still large because of overbooking lines were developed for schedulers to adjust for the demand
and double-booking. 4) Clinics’ current scheduling systems fluctuation of different visit types. For example, in the first
do not differentiate between the physician schedule and the case of the Orthopedic Surgery, when a slot is overwritten,
patient arrival schedule and are not flexible enough to take certain rules must be observed: 1. NP slot can be used to
into account pre-visit activities, which significantly com- schedule any other type of patient besides XR slot since
pounds the wait time. The challenges for implementation NP slot does not consider the x-ray time. 2. XR slot cannot
are very minimal. As mentioned earlier, the systems can be used for any other visit type, since the XR treatment
110 Huang et al.
time is shorter and most of the other visit types do not In the case of a Single Resident/Medical Student, if the
require the x-ray. If a clinic wishes to increase the num- physician prefers to have the resident assess all of the pa-
ber of time slots due to the increasing demand, the solution tients beforehand, then “Schedule by Resident” is prefer-
needs to be revaluate rather than overbooking patient slots. able, as presented in Case Study 2. Implementation results
The approach was demonstrated through the imple- indicate a 50% reduction in patient wait time and the resi-
mentation of three clinics. The results are positive and dent is 52% busier while physician idle time keeps consistent
it takes less than six month time for us to implement in (Table 5). On the other hand, if the physician prefers to keep
these three clinics. The implementation steps are data busy all the time, whether the patient is seen by the resident
collection (mainly focusing on the physician treatment or not, then “Schedule by Physician” is preferable.
time), simulation model development for the process flow, In the case of Multiple Residents/Medical Students,
determination of the optimal treatment time intervals since there will always be enough residents or medical stu-
for each visit type, development of physician schedule, dents, “Schedule by Physician” is always the best option,
development of patient schedule considering ancillary regardless of the physician’s preference, as illustrated by
services, and system adjustment. Case Study 3. Implementation results indicate that “Sched-
Downloaded by [University of Saskatchewan Library] at 09:45 11 January 2015