Você está na página 1de 8

Operating Room Rescheduler

Dept. of CIS - Senior Design 2014-2015


Albert Shu Indu Subbaraj
salbert@seas.upenn.edu subbindu@seas.upenn.edu
Univ. of Pennsylvania Univ. of Pennsylvania
Philadelphia, PA Philadelphia, PA

ABSTRACT ations running over time or surgeons being delayed on cases


Many hospitals, including the Pennsylvania Hospital, suf- are a huge source of monetary loss for the hospital as well
fer from inefficiencies in operating room scheduling. Often as inconvenience for surgeons, nurses, and patients. These
times, these inefficiencies arise because it can be difficult for inefficiencies in resource utilization and allocation can be im-
nurses to manually adjust operating room schedules in re- proved by optimizing the reshuffling of the operating room
sponse to unanticipated events. Our goal was to develop a schedule in response to unanticipated events.
software prototype, in the form of a web application, that au- The current scheduling approach the majority of hospi-
tomates the operating room rescheduling process. Given an tals adopt is block scheduling [1]. In this form of schedul-
initial operating room schedule, this software prototype sys- ing, blocks of time over a time period (anywhere from a
tematically generates efficient schedule adjustments based on week to a year) are assigned to surgeons or surgical depart-
the current state of a hospitals operating room system. The ments. Surgeons fill their assigned time blocks with their
schedule adjustments are generated with the goal of minimiz- cases as they see fit. This scheduling approach requires two
ing the length of the surgery day and with the constraint of steps: (1) the creation of a cyclic timetable, also called a
avoiding delays in surgery start times. This software pro- master surgical schedule (MSS), that determines the time
totype serves as a proof of concept that automation of the blocks and assigns them to different surgeons and (2) the
operating room rescheduling process is feasible and is worth population of the MSS with specific cases [1]. The problem
pursuing in real hospital systems. of operating room scheduling can be broken down into two
segments: production of an initial schedule for a given time
unit (day, week, or month) and generation of altered sched-
1. INTRODUCTION ules based on complications or conflicts that require changes
in the initial schedule.
1.1 Background There are several current implementations of the first seg-
The operating room is often referred to as the heart of ment, generation of an initial operating room schedule, for
a hospital. It accounts for about 60-70% of hospital ad- hospitals to use. A quick online search of the topic reveals
missions, and its operating costs can account for more than that there are a significant number of companies that offer
40% of total hospital expenses [1]. In light of these costs applications to produce operating room schedules. Many of
and the operating rooms central role in a hospital, increas- these focus on providing a GUI that makes manual schedul-
ing the efficiency of resource utilization and patient flow in ing, for example by a nurse, easier. The system used at
the operating room is crucial. One way this problem can be the Pennsylvania Hospital is a mostly manual form of block
addressed is by optimizing operating room scheduling. Op- scheduling; surgeons choose which cases they will be per-
erating room scheduling is the assignment of surgical cases forming given blocks of time they can operate in [3].
and surgical teams to specific dates and times in the oper- Most of the implementations that create an initial sched-
ating room schedule. ule are not optimized and require tedious work by nurses or
The inefficiencies of operating room scheduling arise from surgeons. Even the applications that claim optimization do
the unpredictability of resources used in operations, includ- not offer details on their method of optimization. Although
ing time, space, and personnel [2]. A single operation re- there have been advances in development of theoretical op-
quires the coordination of a surgeon, an operating room timization algorithms, there remains a need for an efficient
nurse team to prep the patient and aid the surgeon dur- and practical implementation of these algorithms that re-
ing the surgery, an anesthesia team to administer anesthe- tains the context of the operating room.
sia and monitor the patient during the surgery, specialized The second segment of operating room scheduling, op-
equipment to perform the surgery, and the allocation of a timally accounting for unexpected changes to produce an
block of time and an operating room for the surgery. A altered schedule, has not yet been successfully addressed in
simple miscalculation or unforeseen complication can lead a clinical setting. There are few, if any, companies that
to the surgery running longer than expected, delaying all offer satisfactory software services to address this problem
following surgeries that require resources being used by the [3]. It is, however, an immensely important one. Current
prolonged surgery. The complications that arise from oper- hospitals rely on nurses to deal with cancellations or delays
and rearrange the schedule manually. However, the already

Advisor: Linh Thi Xuan Phan (linhphan@cis.upenn.edu), busy workload of the nurses and complicated nature of the
Ari Brooks (ari.brooks@uphs.upenn.edu)
task renders manual rescheduling inefficient. This results in Algorithms based on the tactical level focus on develop-
many wasteful gaps in the schedule as well as unnecessary ment of the master surgical schedule. Different linear pro-
overtime operations. gramming models have been developed to optimize genera-
tion of this schedule. Some algorithms focus on minimizing
1.2 Contributions difference between requests and allocations while others aim
We have addressed the problem of inefficient reschedul- to minimize hospital costs [1].
ing upon unexpected events by implementing an operating Algorithms based on the operational level focus on schedul-
room rescheduling system that consists of a rescheduling al- ing surgeries and can be divided into two steps: assigning pa-
gorithm, a web application that provides an easy-to-use user tients to operating rooms (advanced scheduling) and gener-
interface, and a backend database. The rescheduling system ating a sequence of cases (allocation scheduling). Algorithms
allows users to input an initial surgery schedule, monitor generally attempt to optimize one of the two steps presented.
the schedule and trigger the rescheduling algorithm when The problem is modeled as a resource constraint scheduling
a surgery is prolonged or finishes early, and accept or re- model that distinguishes between renewable resources (per-
ject the suggested schedule changes that are output by the sonnel) and nonrenewable resources (equipment). These al-
algorithm. gorithms account for the variability of the operating room,
The purpose of our product is two-fold: to increase ef- a characteristic modeled by stochastic models, which are de-
ficient use of operation room resources by automating dy- scribed further below in the dynamic scheduling section [1].
namic scheduling and to do so by delivering a product that
emphasizes ease of utility. The latter aspect is important 2.2 Algorithms that Optimize Dynamic Oper-
because hospitals are large systems and slow to change; the ating Room Schedule
clients of our product, operating room nurses and surgeons, An early approach to the scheduling problem was appli-
are busy people who will not accept a steep learning curve. cation of the newsvendor theory to the clinical context [4,5].
The initial translation was performed in the context of a sin-
gle operating room serving two surgeries [5]. The newsven-
2. RELATED WORK dor model is a mathematical model used in operations man-
The majority of research on operating room scheduling agement and economics to produce optimal product levels
has focused on developing algorithms to create optimized, for a product with variable demand and limited lifespan [6].
initial operating room schedules. That is, these algorithms In the context of the operating room, the product is surgery
are designed with the goal of producing an operating room and the uncertainty of the product is the variable amount of
schedule that optimally utilizes the resources that the algo- time required for different surgeries. In the initial analysis
rithm accepts as input parameters. We focus on a slightly of the scheduling problem modeled by the newsvendor the-
different problemaddressing the dynamic nature of the ory, the optimal scheduling was solved by convex ordering,
operating room, which often results in inevitable changes a type of stochastic ordering, which is a type of probabilistic
to the initially generated operating room schedule. Even ordering of random variables [5].
though our focus is to implement an algorithm that gener- A later group, Denton et al. [7], expanded the prob-
ates efficient schedule adjustments, the modeling methods lem to accommodate the multiple-patient case by assuming
of algorithms for both initial operating room scheduling and each surgerys length was independent and identically dis-
dynamic scheduling can serve as a basis for our algorithm. tributed. They created a two-stage stochastic linear model
that accounts for costs of waiting time for resources, idle
2.1 Algorithms that Optimize Initial Operat- time when no surgeries are being performed, and delays in
ing Room Scheduling surgeries [7]. The algorithm derived by Denton et al. was
The algorithmic approaches to generating optimized, ini- later expanded by Denton et al. [8] by removing the require-
tial operating room schedules can be divided into the fol- ment for sequences of cases to be known before optimization.
lowing three categories: strategic, tactical, and operational. Denton et al. provide a two-stage stochastic integer model
These categories represent the three main decision levels in as well as multiple heuristics, the best of which was found
operating room scheduling; they are interdependent but due to be ordering cases by smallest variance first [8].
to the complexity of the problem, most approaches primar- The algorithm derived by Denton et al. was later ex-
ily address one of these levels. The strategic level deals with panded by Denton et al. [8] by removing the requirement for
the assignment of time blocks in block scheduling to different sequences of cases to be known before optimization. The up-
surgical departments. The tactical level focuses on develop- dated system of computing optimal sequence and schedule of
ment of the master surgical schedule. The operational level surgeries is a combinatorial stochastic optimization problem.
involves assignment of surgical cases to the schedule [1]. Denton et al. provide a two-stage stochastic mixed integer
Algorithms focusing on the strategic level analyze factors model as well as multiple heuristics to estimate the optimal
that determine the distribution of time blocks to surgical de- solution. The heuristics proposed are based on mean and
partments. These factors include the frequency of different standard deviation of surgery duration from historical data,
surgeries to be performed in the time period, the number of an improvement on the independent and identical distribu-
staff available in each department, and cost analysis. Plan- tion used in Denton et al. [7]. The maximum optimization
ning at this level is primarily a resource allocation problem was found by ordering cases by smallest variance first [8].
and research addressing it has focused on the similarity to The results are logically sound because schedule disruptions
manufacturing systems. Linear programming models based due to early cases exceeding their allotted times are more se-
on economic analysis to maximize profits and minimize costs vere and require more extensive rescheduling than schedule
and optimization models to maximize the number of cases disruptions from later cases exceeding their allot- ted times.
have been developed [1]. Mancilla et al. [9] improved the algorithm further by de-
veloping a two-stage stochastic integer programming (2-SIP) very difficult to implement. Consequently, there is lack of
approximation that optimized sequencing of surgeries and transition from proposed theoretical solutions to results in
arrival times. The model allowed for a revision of future the real world. This is particularly evident at Pennsylvania
surgeries given the need. The 2-SIP model was extended Hospital, where the scheduling software fails to address even
to the problem of dynamic scheduling in a single-operating basic inefficiencies [3]. We attempt to address this gap by
room system, in which surgery cancellations were accounted proposing a more practical schedule adjustment tool that
for [9]. has the potential to be used in a real clinical setting.
Dynamic scheduling in a multiple-operating room system
was also modeled with a 2-SIP model given information
about the bounds on surgery durations [4,9]. A problem After considering these limitations of mathematical mod-
encountered was optimization of both the scheduling and els of the rescheduling problem, we decided to approach
rescheduling, but the L-shaped algorithm and hedging-based the problem in a different manner. We identified that the
heuristics allowed an ideal case of reassignment to be solved operating room environment is analogous to the job shop
[9]. problemjobs translate to surgeries and machines translate
The most recent work, by Zhang et al. [4] develops a to operating rooms. Modeling the operating room reschedul-
multistage stochastic program to model the dynamic prob- ing problem with the job shop scheduling problem is not
lem of operating room scheduling. The system proposed an approach that has been attempted previously. We de-
performs a rescheduling decision at each surgery comple- cided to follow this approach because it yields an algorithm
tion event. To simplify the large state set and complexity that is implementable without simplifications and matches
of the problem, a two-stage stochastic integer programming our model and specific assumptions, detailed in the system
approximation is used to optimize the costs for the given model section of this paper. This approach also clearly ful-
state of the dynamic scheduling problem. The first stage fills the goal of shortening schedules by more efficient usage
cost of the 2-SIP approach is calculated as the sum of wait- of operating rooms, while also allowing for customization
ing time of the surgeon and idle time of the operating room with other metrics (see evaluation section for more details).
released at a given surgery completion event. The remaining The background required to understand our algorithm is
cost, or cost of all other rescheduling decisions, is estimated presented below.
by two distinct heuristic models. The first, one-period look
ahead heuristic (OPLA), estimates the remaining cost as 2.3 Job Shop Scheduling Problem
the cost observed at the next stage. The second, rule-based The job shop scheduling problem is an optimization prob-
multiperiod look ahead heurisitc (MPLA), calculates the re- lem in computer science that attempts to assign ideal jobs to
maining cost as the total cost, as opposed to the cost at the machines in such a way that the time required to complete
next stage as in OPLA. MPLA approximates the total cost the jobs in minimized [11]. The problem has been presented
by arranging the surgeries in a nondecreasing order of pre- in many forms, but we present in this paper the form most
defined surgery start times. Simulations comparing static relevant to our rescheduling algorithm.
scheduling, dynamic scheduling with OPLA, and dynamic Assume there are n atomic jobs and m identical machines.
scheduling with MPLA indicate that dynamic scheduling, A job is considered atomic if it is not divided into multiple
especially dynamic scheduling with MPLA, is a substantial operations. Each machine can only process a single job at a
improvement on static surgery scheduling [4]. time and all the jobs can be processed on any machine. The
Essen et al. [10] take a distinct approach to the operat- goal of the job shop scheduling problem is to assign the n
ing room rescheduling program by placing an emphasis on jobs to the m machines in an order such that the total time
stake- holder preferences. The authors define stakeholders required for all the jobs to complete is minimized. One al-
as the following: surgeons, patients, anesthesiologist, nurses, gorithm designed to solve this problem, the least processing
and multiple departments in the hospital, including recovery time algorithm, is the basis for our rescheduling algorithm
department and radiology department. The key stakeholder and is described in detail below.
in the paper is the patient and his availability preference
is weighted the most in the model. The authors modeled 2.4 Least Processing Time Algorithm
the system as an integer linear program (ILP). Constraints
The least processing time (LPT) algorithm is designed
representing the different preferences of stakeholders were
to solve the job shop scheduling problem presented above.
imposed on the model and optimization was performed to
The first step of the algorithm assigns to each job a weight
minimize the deviation of these preferences. Simulating the
equal to the length of the job. Next, the jobs are sorted in
model produced two primary schedule adjustments: shift-
descending order by weight. Each machine m is initialized
ing a surgery and scheduling a break between two surgeries.
to have a load of 0, and then jobs are sequentially removed
The model and optimization method led to fewer cancelled
from the sorted queue and placed in the machine with the
surgeries and a perceived reduction in workload for each of
shortest load. As a job is added to a machine, the length of
the stakeholders involved. The ILP model is developed im-
the job is added to the machine load.
proves the operating room schedule for only one operating
This algorithm guarantees a schedule that completes all
room at a time, which renders it inefficient for larger, real
the jobs in the shortest time possible. It is a 4/3 approx-
world systems. Another limitation is that due to changing
imation algorithm. We have used the LPT algorithm as a
priorities of stakeholders, the constraints and decision rules
basis for our rescheduling algorithm, with appropriate mod-
extracted as optimal also change [10].
ifications to address our specific problem.
While these theoretical approaches to optimizing operat-
ing room scheduling are well defined mathematically, many
are too complicated to solve without simplifications and are 3. SYSTEM MODEL
3.1 Overview 3.3 Rescheduling Algorithm
Our system consists of a software prototype that will gen- One of the key components of the operating room reschedul-
erate operating room schedule adjustments in response to ing software is the rescheduling algorithm. The goal of the
unanticipated events in the operating room system. In par- algorithm is to schedule the remaining surgeries to the op-
ticular, the software handles cases in which a surgery is un- erating rooms in such a way that the total time required to
expectedly prolonged and cases in which a surgery finishes execute the remaining surgeries is minimized. While trying
early. Users of the software can input an initial operating to achieve this goal, the algorithm is restricted by the con-
room schedule and update the schedule when an unantici- straint that it should avoid avoid delaying surgeries. The
pated event occurs. When such an event occurs, the software extent to which the algorithm avoids delaying surgeries is
will systematically generate schedule adjustment recommen- specified by user input for each class of surgeries (specified
dations for the user to evaluate. The user can then choose to as alpha, as mentioned in the Web Application section).
accept or reject the schedule adjustment recommendations The rescheduling algorithm used in the operating room
and continue to update the schedule accordingly. A block rescheduling software is an extension of an already existing
diagram of this system workflow is depicted in Figure 1. algorithm, called the Longest Processing Time (LPT) algo-
rithm. The LPT algorithm is used to solve the atomic job
shop scheduling problem. In this problem, the inputs are
n jobs with varying processing times and m machines. The
goal is to assign the jobs to the machines in such a way that
the total time required to process the jobs is minimized (see
Related Works for more details).
Our rescheduling algorithm results from adapting the LPT
algorithm to the operating room environment. The jobs
in the atomic job shop scheduling problem naturally trans-
late to surgeries, and the machines in the atomic job shop
scheduling problem naturally translate to operating rooms.
However, a few crucial adjustments were also necessary to
fully adapt the LPT algorithm to the operating room reschedul-
ing problem. First, in our rescheduling algorithm, each
surgery is assigned a weight corresponding to its surgery
length and its original start time before the rescheduling oc-
curs. The user input parameter alpha, which specifies the
importance of avoiding surgery delays, is also incorporated
when assigning each surgery its weight. In our algorithm,
the surgeries are now sorted in descending order of weight
instead of descending order of processing time. Another
crucial adjustment that had to be made was that our algo-
rithm must ensure that surgeries involving the same surgeon
are not scheduled to be performed concurrently. Therefore,
if the surgery that is currently being scheduled involves a
surgeon that is currently performing another surgery, the
Figure 1: Logical flow diagram of model surgery being scheduled must be placed in a delayed queue
to be scheduled later. The general steps of our rescheduling
algorithm are as follows:
3.2 Assumptions
In our system model, the following assumptions are made Let there be n surgeries with lengths t1 to tn
about the operating room environment. All of the assump-
tions were confirmed by our medical advisor as acceptable Let there be m identical operating rooms
and reasonable for our goals.
Let W[i] be the weight assigned to surgery i
All operating rooms are relatively similar and can ac-
commodate all kinds of surgeries Let L[i] be the current load on room i
Surgeries can be grouped into classes and handled ap- Let J[i] be the set of jobs assigned to room i
propriately by class
Let be the parameter that specifies the importance of
Surgeons are always available during the surgery day
avoiding surgery delays (user input)
except when they are scheduled to perform surgeries

Nurse teams and anesthesiologist teams are versatile 1. For each surgery i = 1 to n
and can perform any type of surgery
W[i] = ti - (start time of surgery i prior to current
Nurse teams and anesthesiologist teams are assigned rescheduling - current time)
to a single operating room for the duration of their
shift 2. Sort the surgeries in descending order of weight (using
priority queue) order for the schedule recommendations to be meaningful, it
was necessary to incorporate two considerations into our al-
3. For each room i = 1 to m gorithm. First, we had to ensure that the algorithm avoided
L[i] = sum of the lengths of surgeries that have surgery delays, perhaps even at the cost of a longer surgery
already started in room i schedule. This was especially important because surgery de-
lays significantly inconvenience both surgeons and patients.
J[i] = {} Second, we had to ensure that surgeons were not being
For each surgery j = 1 to n (in sorted order) scheduled to perform concurrently occurring surgeries. For
clear reasons, this was also very necessary.
Ensure that surgeon associated with surgery j is Another challenge was improving the usability of the soft-
not scheduled to perform another surgery at the ware itself. Since the hospitals are very hectic environments
same time and nurses are very busy, it was important to ensure that the
if (surgeon associated with surgery j is currently application is easy to use. This included creating a simple
performing a surgery) and clear user interface, and ensuring that all of the user in-
puts would be easily available to the nurses. Furthermore, it
Set current surgery j to be scheduled after surgery
was crucial to ensure that that the runtime of the scheduling
j+1
algorithm was efficient.
Continue
Let i = operating room with smallest load 4. SYSTEM IMPLEMENTATION
J[i] = J[i] U j The operating room rescheduling software was implemented
as a web application using the client/server model. The pri-
i = L[i] + tj mary components of the web application are the database,
the server-side functionality, and the user interface.
3.4 Web Application
The operating room rescheduling software is in the form 4.1 Server-Side Functionality
of a web application. The application will initially take in as The web application backend was implemented using a
user input the initially planned operating room schedule for request-response programming model using Java servlets.
a given surgery day along with basic information about the Each URL in the application is mapped to a servlet, and
operating rooms in the operating room system. When in- the servlets send the appropriate response back to the client.
putting the initially planned schedule, the user will have the The Java servlets interact directly with the database and the
option to assign surgeries to different classes. Each class of rescheduling algorithm.
surgeries can, in turn, be assigned a different priority level. The operating room rescheduling algorithm was imple-
Surgeries belonging to classes with higher priority levels will mented in Java.
be less likely to be delayed when the web application gen- The web application was hosted on a Jetty server. This
erates schedule adjustments. This feature of assigning surg- was done by packing the source code into a WAR file and
eries to different classes was included to account for the fact placing the it in the appropriate directory on the Jetty server.
that, in reality, surgeries have different priorities and that
some surgeries should not be delayed if possible. 4.2 Database
After the web application is initialized with the initially
The database was implemented using Oracle Berkeley DB,
planned schedule, it will continue to receive as user input
Java Edition. The primary database entities stored were the
information regarding the current status of the operating
surgeries and operating rooms.
room system. When an unanticipated event has occurred,
the user can update the schedule and generate schedule ad-
justment suggestions. When generating the schedule adjust-
ment suggestions, the user can input a value for alpha for
each class of surgeries. A surgery class alpha indicates the
importance of avoiding surgery delays for the surgeries in
that class. Then, based on the alpha values input by the
user, the web application will generate schedule adjustment
suggestions and display them to the user. The user will be
able to evaluate the schedule adjustment suggestions and
determine whether he/she would like to accept them.
Figure 2: Application Architecture
3.5 Technical Challenges
Many technical challenges were encountered throughout
the process of creating the rescheduling software. In par- 4.3 User Interface
ticular, the most difficult challenges were related to making The user interface was implemented using HTML and
the software practical in a real hospital setting. CSS. The web application has four primary pages. The first
One of the primary challenges was yielding schedule ad- page allows the user to initialize the operating rooms and in-
justment recommendations that would be reasonable for the put an initial schedule. The second page allows the user to
surgeons and patients. It is necessary for the rescheduling monitor the schedule and indicate an unanticipated event.
algorithm to output results that are not only mathemati- The third page allows the user to update the schedule and
cally efficient, but also useable by the parties involved. In specify the alpha parameters for the rescheduling algorithm.
If so, the move is made. This algorithm reflects the typical
response to unexpected events in the operating roomaAT-
surgeries are pushed back and occasionally moved to another
room.
We randomly generated different schedules that are rep-
resentative of a typical hospital operating room schedule,
based on input from Dr. Brooks. We generated schedules
instead of using actual hospital schedules, because HIPPA
privacy issues made it difficult to obtain complete surgery
schedules.
The characteristics of our schedules are as follows:
The surgery day of our schedules extends from 7 am
to 5 pm
There are 10 operating rooms

Figure 3: Screenshot of home page Surgeries were on average 1 hour long


About 50-65 surgeries were scheduled initially
The fourth page displays the schedule adjustment recom- The operating room loads are all approximately equal
mendations for the user to evaluate, and it gives the user to each other
the option to accept or reject the recommendations.
Operating rooms were filled almost completely, with at
most 2 hours between the end time of the last surgery
5. RESULTS in the room and the end of the surgery day

5.1 Evaluation Metrics Approximately 7% of surgeries were delayed an average


of 1 hour in each schedule
Evaluation of system performance is based on two primary
metrics. The first metric is the change in the length of the Our rescheduling algorithm, at each value of alpha, and
surgery day, calculated by subtracting the end time of the the baseline algorithm were run on each of the schedules and
last surgery in the original schedule from the end time of the the metrics were averaged to obtain the single data points
last surgery in the rescheduled schedule. By this calculation, displayed on the graphs below.
a negative value indicates a shortening of the length of the
surgery day while a positive value indicates a longer surgery 5.3 Aggregate Results
day. Refer to the two graphs on this page summarizing our
The second metric is the average delay between resched- evaluation results.
uled start times and original start times for all surgeries
rescheduled by our system in a given day. This metric is
calculated by summing the time differences between resched-
uled and original start times for all surgeries and dividing by
the total number of surgeries in a given day. If the surgery
is rescheduled to start later, it contributes a positive value
to the metric. A surgery rescheduled to start earlier con-
tributes a negative value, or decrease in average delay, to
the metric, and a surgery rescheduled to start at the same
time contributes a value of zero.

5.2 Methodology
We evaluated our algorithm using both of the described
metrics at six different values of alpha, ranging from 0 to
15. The performance of our rescheduling algorithm at these
different alpha values was compared to the performance of
a baseline rescheduling algorithm. We designed the base- Figure 4: Change in surgery day length
line rescheduling algorithm to simulate the process currently
used by nurses to reschedule surgeries upon unexpected events, For change in surgery day length, our rescheduling algo-
as described by our medical advisor Dr. Brooks. rithm performed better than the baseline algorithm for all
The baseline algorithm takes in as input the surgery that values of alpha. Our algorithm decreased the length of the
is prolonged or cancelled as well as the new end time. The surgery day for all values of alpha, but as the alpha value
baseline algorithm extends the appropriate surgery and pushes increased, the amount of shortening of the surgery day de-
back all other surgeries in the same operating room that fol- creased. The baseline always increased the length of the
low. After pushing back the surgeries, the algorithm checks surgery day, as is expected by a simple push-back process.
to see if moving the last surgery to the end of the operating For low values of alpha, high average delays are observed,
room with the smallest load, or the room whose last surgery as expected because the algorithm does not prioritize schedul-
ends the earliest, would result in the surgery ending earlier. ing algorithms close to their initial start time. As the value
The operating room rescheduler faces some potential eth-
ical issues. The primary concern is that the system must be
extremely secure because of the sensitive nature of its data.
Since it works with patient and hospital records, the resched-
uler must follow the federal Health Insurance Portability and
Accountability Act of 1996 (HIPPA). This act was formed
to ensure that the confidentiality and security of healthcare
information would be protected. If the reschedule system
were somehow compromised, sensitive patient and hospital
records would be compromised, which is unacceptable. The
repercussions would be multifold, including legal and mon-
etary trouble for the hospital and emotional stress for all
hospital staff and patients involved.
Another concern is the potential for our system to crash in
Figure 5: Average delay the middle of an operating day. If our system were the pri-
mary source of information about the surgery schedule, this
would result in a halt or significant delay in the surgery day
of alpha is increased, the average delay drops to be similar until the system could be rebooted, resulting in monetary
to the value of the baseline algorithm. loss for the hospital and emotional distress for the hospital
We observed a trade-off between the amount the surgery staff and patients dealing with the uncertainty. Our system
day can be shortened and the average delay of surgeries. would also have to ensure that data is stored in a persistent
This observation makes sense intuitively because higher al- state that can be recovered upon rebooting.
phas give importance to preventing surgery delays as much Incorporating our rescheduler with currently used systems
as possible, but at the expense of flexibility in shuffling surg- in hospitals can mitigate both concerns about HIPPA poli-
eries around. However, despite this trade-off we found that cies and fault tolerance. Integration with the hospital would
even at high alphas where average delay was low, there was involve removing the application from the internet, reducing
still a decrease in the total length of the surgery day. its vulnerability. The data would also be stored on secure
hospital servers and backups could easily be maintained in
6. FUTURE WORK case of crashes. It would also ensure we follow the same
Since our application is a prototype, there are many po- security protocols to honor HIPPA as the rest of the hos-
tential improvements that can be made. pital system. This would involve, on a high level, secure
One such improvement is addition of a notification system authentication to limit access to only appropriate person-
that informs hospital staff, including surgeons, nurses, and nel as well as strong encryption. We faced the implications
anesthesiology teams, of their new surgery schedules. Such of HIPPA when attempting to acquire real hospital data
a system could be implemented through text messaging or for our evaluation. The patient identifying information had
email and would increase the usability of our application. to be removed from the surgery schedules before delivered
This notification system could be extended further by re- to us and because there was no automation to process the
laxing our assumption that surgeons are always available un- records, the amount of hospital data we were able to acquire
less they are performing another surgery. Instead of using a was limited.
messaging system just to inform surgeons of their changed
surgery, it could be used to ask surgeons if they are avail- 8. CONCLUSIONS
able during the new suggested surgery time. The surgeons Operating room rescheduling is not a problem that has
responses could then be used to output the final reschedul- been addressed extensively. Most of the research that has
ing. This would increase the usability of our application been done on it is purely theoretical, with no focus on prac-
tremendously. tical implementations that can be used in a clinical setting.
Further, more of our assumptions can be relaxed to cre- However, the inefficiencies of the current manual method of
ate an environment more similar to the real world. For ex- rescheduling surgeries signal that such a tool is very neces-
ample, not all operating rooms are actually identical in a sary.
real hospital setting. Accounting for this in our application We have designed a prototype of such an application. Our
would allow assignment of special surgeries to those operat- system has a rescheduling algorithm based off of the job shop
ing rooms. Another improvement would be to account for least processing time algorithm, a web application user in-
emergency surgeries, as well as the prescheduled surgeries terface, and a back end that stores the necessary data about
we currently allow. surgeries and operating rooms. Upon a surgery prolongation
The long-term goal is to integrate our application with the or cancellation, a user can run the rescheduling algorithm
hospital system. This would allow easier access and usage of and accept or reject its suggested changes. We also allowed
hospital data, including the initial schedule, available hos- users to dictate how important it was to schedule surgeries
pital staff, and patient information. Integration would also as close to their initial start time as possible.
ensure that our application is as secure as the rest of the Our evaluation of the application showed a decrease in
hospital system, guaranteeing a higher level of compliance the total length of surgery day with our algorithm as com-
with HIPPA. pared to a baseline algorithm. We also observed a trade-off
between the decrease in surgery day length and the average
7. ETHICS delay time of surgeries. Overall, our application is a good
proof-of-concept prototype that illustrates that automated
rescheduling is possible and beneficial for use in a clinical
setting.

9. REFERENCES
[1] Guerriero F and Guido R. (2011).Operational research
in the management of the operating theatre: a survey. Health
Care Management Science 14(1), 89114.
[2] Cardoen, B., Demeulemeester, E., and BeliAnn, J.
(2010). Operating room planning and scheduling: A lit-
erature review. European Journal of Operational Research,
201, 921932.
[3] Ari Brooks, MD (Advisor)
[4] Zheng Zhang; Xiaolan Xie; Na Geng. Dynamic Surgery
Assignment of Multiple Operating Rooms With Planned
Surgeon Arrival Times. Automation Science and Engineer-
ing, IEEE Transactions on , vol.11, no.3, pp.680,691, July
2014. doi: 10.1109/TASE.2013.2267273.
[5] E. N. Weiss. Models for determining the estimated
start times and case orderings. IIE Trans., vol. 22, pp.
143150, 1990
[6] Anesth Analg. 2010 Jun 1;110(6):1698710. Epub
2010 May 6.
[7] B. Denton and D. Gupta. A sequential bounding ap-
proach for optimal appointment scheduling. IIE Trans., vol.
35, no. 11, pp. 10031016, 2003.
[8] B. Denton, J. Viapiano, and A. Vogl. Optimization of
surgery se-quencing and scheduling decisions under uncer-
tainty. Health CareManage. Sci., vol. 10, pp. 1324, 2007.
[9] C.MancillaandR.Storer.A sample average approxima-
tion approach to stochastic appointment sequencing and schedul-
ing. IIE Trans., vol.44, no. 8, pp. 655670, 2012.
[10] van Essen, J. T., J. L. Hurink, W. Hartholt, and B.
J. van den Akker. 2012. Operating Room Rescheduling.
Working paper, Beta Research School for Operations Man-
agement and Logistics, University of Twente. Accessed July
26, 2014.
[11] Jones, A., Rabelo, L. C. and Sharawi, A. T. 1999.
Survey of Job Shop Scheduling Techniques. Wiley Encyclo-
pedia of Electrical and Electronics Engineering. .

Você também pode gostar