Decision support system for the operating room rescheduling problem
Authors
 First Online:
 Received:
 Accepted:
DOI: 10.1007/s1072901292022
 Cite this article as:
 van Essen, J.T., Hurink, J.L., Hartholt, W. et al. Health Care Manag Sci (2012) 15: 355. doi:10.1007/s1072901292022
Abstract
Due to surgery duration variability and arrivals of emergency surgeries, the planned Operating Room (OR) schedule is disrupted throughout the day which may lead to a change in the start time of the elective surgeries. These changes may result in undesirable situations for patients, wards or other involved departments, and therefore, the OR schedule has to be adjusted. In this paper, we develop a decision support system (DSS) which assists the OR manager in this decision by providing the three best adjusted OR schedules. The system considers the preferences of all involved stakeholders and only evaluates the OR schedules that satisfy the imposed resource constraints. The decision rules used for this system are based on a thorough analysis of the OR rescheduling problem. We model this problem as an Integer Linear Program (ILP) which objective is to minimize the deviation from the preferences of the considered stakeholders. By applying this ILP to instances from practice, we determined that the given preferences mainly lead to (i) shifting a surgery and (ii) scheduling a break between two surgeries. By using these changes in the DSS, the performed simulation study shows that less surgeries are canceled and patients and wards are more satisfied, but also that the perceived workload of several departments increases to compensate this. The system can also be used to judge the acceptability of a proposed initial OR schedule.
Keywords
Operating rooms Rescheduling Integer programming Decision support system1 Introduction
The Operating Room (OR) department is one of the most expensive resources of a hospital. However, managing the OR department is hard due to conflicting priorities and preferences of stakeholders. Therefore, planning and scheduling methods are needed to increase the efficiency in OR departments. See Cardoen et al. [3] and Hulshof et al. [12] for an overview on OR planning and scheduling.
In this paper, we focus on the rescheduling of surgeries, or more precisely, on the rescheduling of surgeries throughout the day. On the one hand, emergency patients who need surgery arrive throughout the day. In many hospitals, these surgeries are scheduled in one of the elective ORs which disrupts the OR schedule. On the other hand, a change in the surgery duration of elective surgeries may also disrupt the OR schedule. Therefore, the initial OR schedule may have to be adjusted throughout the day to ensure that it is still possible to execute the schedule. The new OR schedule must fulfil quite a number of restrictions, and in addition, there are several stakeholders whose preferences and priorities must be met. Since it is hard for an OR manager to consider all these restrictions and preferences simultaneously, we develop a decision support system (DSS) which supports the OR manager with rescheduling the ORs. The approach proposed in this paper is general, but the realization and used preferences are based on data from a hospital in the Netherlands.
Most existing literature focuses on operational offline scheduling, which is done one or several days before surgery, instead of operational online scheduling, which is done on the day of surgery. The papers concerning operational offline scheduling mainly focus on two methods. The first method is reserving time for emergency surgeries to minimize overtime and maximize OR utilization (see e.g. [2, 10, 13]). The second method is sequencing the elective surgeries such that the overtime caused by surgeries with a longer duration than expected is minimized (see e.g. [4, 14, 16]). In addition, some papers concerning operational offline scheduling include the preferences of stakeholders. For example, Marcon and Dexter [17] consider the preferences of the holding and recovery departments and Marjamaa et al. [18] consider the preferences of the anesthetists.
One of the papers concerning operational online scheduling is the paper by Dexter [5] who examined whether moving the last surgery of the day to another OR could decrease overtime labour costs. The developed statistical strategy was based on historical data. However, in practice, often one surgeon operates in an OR on a day or part of the day and therefore, it is not allowed to move a surgery to another OR.
Dexter et al. [8] introduce four ordered priorities on which an OR management decision for changing the OR schedule can be based. The first and most important priority is patient’s safety. The second priority states that a surgery can only be canceled if the patient safety is not endangered. The third priority is to minimize overtime, and the fourth and last priority is to reduce patient waiting times. These priorities, however, put minimizing overtime above the patients timing preferences. For patients it is not preferred to schedule their surgery earlier and certainly not later in the day. In addition, no priority considers the workload level on other departments like wards, the holding department, and the recovery department.
Another paper of Dexter et al. [7] considers the sequencing of urgent surgical cases. They proposed a sequencing which is based on the following three objectives: (i) minimize the average waiting time of surgeons and patients, (ii) sequence the surgeries in order of appearance, and (iii) schedule the surgeries in order of medical urgency. However, none of these objectives consider the preferences of the elective patients and other departments.
It seems that none of the existing papers on OR rescheduling considers the preferences and priorities of all the stakeholders simultaneously. This paper tries to fill this gap. Based on a given case in a Dutch hospital, we mainly focus on personal preferences and less on economical preferences, because patient and personnel satisfaction is highly important in the Netherlands due to shortage of personnel and competition between hospitals. In Section 2, we discuss the stakeholders and their restrictions and preferences which are based on a survey performed at the Isala Clinics, which is the above mentioned hospital in the Netherlands. Although these restrictions and preferences may differ between hospitals, the principle ideas of the method developed in this paper should be applicable for other hospitals too. Because we want to develop a DSS that generates adjusted OR schedules within a short amount of time, we first analyze the problem to determine which changes are preferred based on the preferences of the stakeholders. Therefore, we incorporate the restrictions and preferences in an Integer Linear Program (ILP), which has as goal to minimize the deviation from the preferences of the stakeholders. Although we prove that the problem is strongly NPhard for two or more ORs, we are able to solve this ILP due to the moderate size of the instance. The solutions to the ILP are used to determine the changes made in the optimal OR schedule for instances of the Isala Clinics. These changes are incorporated as decision rules in the DSS which is described in Section 4. As we incorporate the preferences of all stakeholders and as these preferences are given by the stakeholders themselves, we do not expect that the resulting changes are subject to psychological bias as mentioned in other papers (see e.g. [6, 15]). The developed DSS is tested by means of a simulation study to determine what improvements can be made to the ORschedule when the developed DSS is used in practice. The computational results of this simulation study are given in Section 5. Section 6 draws conclusions and gives recommendations for further research.
2 Problem formulation
In this section, we give an introduction to the OR rescheduling problem and we introduce an ILP model which can be used to determine a new OR schedule throughout the day. The ILP includes all relevant constraints that are imposed on the OR schedule; e.g. the availability of a patient, as well as the availability of an OR with OR assistants, a surgeon and an anesthetist. In addition, the capacity of the holding and recovery department are considered. A detailed description of the constraints is given in the following subsections.
The objective of the ILP is to minimize the deviation of the preferences for the involved stakeholders. When the OR schedule deviates from these preferences some penalty costs are incurred and the weighted sum of these penalty costs is minimized. There can be, for example, penalty costs for deviating from the scheduled start time of a surgery or for the amount of resulting overtime. In addition, we minimize the number of canceled patients, as this is not preferred by any of the stakeholders. The developed ILP can also be used to determine whether a proposed OR schedule is feasible or not, and in case it is feasible, to calculate the deviation from the preferences of the stakeholders.
For the modeling, we discretize an ORday into T time periods which have a length of δ minutes. The length of one ORday is therefore δT minutes. We denote by time t ∈ T the period ((t − 1)δ, tδ]. The set of ORs is given by set J and consists of M ORs. The start time of OR j ∈ J is denoted by S _{ j } and the end time by F _{ j }. The set of surgeries is given by set I and consists of N surgeries. The subset \(I_j\subseteq I\) denotes the surgeries that are scheduled in OR j ∈ J and O _{ i } ∈ J denotes the assigned OR for surgery i ∈ I.
The initial OR schedule, which is given at the beginning of the day and which has already been determined one or several days before, is defined by the assignment of the elective surgeries to an OR and the initially planned start times P _{ i } of the elective surgeries. Each surgery has an expected duration E _{ i } which includes the time for administering and reversing anesthesia, however, in practice, the duration of a surgery generally deviates from this duration and takes longer or shorter than expected. When a surgery takes less time than expected, and the next surgery starts at its assigned time P _{ i }, the initial OR schedule is not disrupted. However, it may be beneficial for the OR and other departments to schedule this next surgery earlier. When a surgery takes longer than expected, the next surgery may have to start later. This results in a shift of the not yet started surgeries in this OR. Because of this, some resource constraints may be violated. In addition to these deviations of the durations of the elective surgeries, emergency surgeries may arrive which also disrupt the initial OR schedule. Therefore, throughout the day, a new OR schedule may have to be created for all not started elective and emergency surgeries. In the following, we denote by set I the set of all these elective and emergency surgeries. The rescheduling is done by assigning a new start time to each surgery i ∈ I. Formally, this is expressed by binary variables s _{ it }, which are one when surgery i ∈ I starts at time t ∈ T, and zero otherwise. It is important to note that we do not allow the elective surgeries to be assigned to another OR, because each surgery has to be performed by the surgeon operating in the OR assigned to the surgery in the initial OR schedule. Thus, all elective surgeries have to be scheduled in the same OR as in the initial schedule. Because we only reschedule the not yet started surgeries, the start time of OR j ∈ J for the rescheduling problem is either given by the start time of the OR in the morning or the expected end time of the last started surgery in this OR.
Note that we do not consider the rescheduling of a canceled surgery, because we only focus on rescheduling within the day and not from day to day.
The decision variable s _{ it } and u _{ i } completely determine the new OR schedule. However, to model the other restrictions and preferences some extra variables have to be defined which are introduced at the places where they are needed.
The above constraints describe some of the hard constraints for the rescheduling process resulting from the situation in the OR. In the following subsections, we describe and model the involved stakeholders. For each stakeholder, we describe the tasks the stakeholder has to perform during the day, the restrictions they impose on the OR schedule, the impact a change in the OR schedule has on the stakeholder and the preferences of the stakeholder. The impacts are modeled by linear constraints, and penalty costs for deviating from the preferences are incorporated in the objective function. The chosen functions and corresponding parameters used to describe the preferences of the stakeholders and the penalty costs for deviating from these preferences are determined based on a survey at the Isala Clinics (see Appendix A). Applying the model to a different hospital asks for an analysis of the preferences of the stakeholders in this hospital and may lead to other penalty cost functions. However, the principle structure of the proposed approach does not have to be adapted.
2.1 Patient
Note that constraints (8) replace constraint (7) in the ILP.
Based on the survey in the hospital, we concluded that patients assign different penalty costs to different values of y _{ i }. To model this, a function f _{ PT }(y _{ i }), denoting the penalty costs when surgery i ∈ I is shifted y _{ i } time periods, is introduced. This function is also split into two parts, namely \(f^{\rm earlier}_{PT}(y^{\rm earlier}_i)\) and \(f^{\rm later}_{PT}(y^{\rm later}_i)\).
Note that the two penalty functions \(f^{\rm earlier}_{PT}\) and \(f^{\rm later}_{PT}\) in general have different parameters. Because we minimize the total penalty costs, this method is only applicable for nondecreasing step functions.
2.2 Ward
Prior to surgery, the patient is admitted to a ward. On this ward, the patient is prepared for surgery. The survey showed that when a surgery starts earlier than scheduled, the workload on the ward increases if the patient is not ready yet. When a surgery starts later than scheduled, the workload can also increase. Therefore, penalty costs are incurred when there is a change in the start time of a surgery. The total penalty costs are calculated in the same way as for the patient. Based on the outcome of the survey a step function f _{ W }(y _{ i }) is defined, which denotes the penalty costs for the wards if the start time of a surgery is shifted for y _{ i } time periods. Note, that we do not distinguish between a surgery being scheduled earlier or later. The total penalty costs for wards is then given by p _{ W } = ∑ _{ i ∈ I } f _{ W }(y _{ i }).
2.3 Holding department
The survey at the Isala Clinics showed that the holding department prefers a levelled amount of patients that are present at each point of time. Therefore, penalty costs given by the step function f _{ HD }(l _{ t }), are incurred when the number of patients l _{ t } exceeds a certain threshold. The penalty costs for different values of l _{ t } are specified by the manager of the holding department. As at the beginning of the day (until a prespecified time ψ), personnel from the recovery department assist on the holding department (no patients are present at the recovery department at this time), penalty costs are only incurred from time ψ on. The total penalty costs p _{ HD } are given by \(\sum_{t=\psi+1}^{T}f_{HD}(l_t)\).
2.4 Anesthetist
The anesthetist is responsible for administering and reversing anesthesia on one or more ORs. Therefore, he/she has to administer and reverse all anesthesias in these ORs. However, during the surgical procedure, the anesthetist does not have to be present in the OR, because the presence of an anesthesia nurse is enough. Therefore, similar to constraints (4) and (5), we include constraints which prohibits that more than one anesthesia is administered or reversed at a time in the ORs to which the anesthetist is assigned. For more details, see [11].
However, there are a few exceptions. When a surgery is complex, for example when the patient is younger than 6 months, the anesthetist must be present during the complete surgery which includes the surgical procedure. This means that during this time no anesthesia can be administered or reversed in one of the other assigned ORs. This is also ensured by constraints similar to constraint (5).
2.5 Surgeon
2.6 OR assistants
The step function f _{ OS }(o _{ j }) provides the penalty costs for OR j ∈ J when overtime of o _{ j } time periods is incurred. The OR assistants specified the penalty costs for different values of o _{ j }. The total penalty costs for the OR assistants is given by p _{ OS } = ∑ _{ j ∈ J } f _{ OS }(o _{ j }).
2.7 Recovery department
After surgery, the patient is transported to the recovery department. Here, the patient is monitored while he/she recovers from surgery. The length of stay on this department varies with the expected duration of the surgery and is given by \(\max(U, \frac{1}{2} E_i)\), where U is the minimum length of stay on this department. The number of patients present at the recovery department at time t ∈ T is denoted by z _{ t } and can be determined in the same way as for the holding department. The capacity of the recovery department is restricted by the number of beds O _{2}.
Another restriction is given by the number of patients who can be treated simultaneously, which depends on the number of available nurses R _{ t } at time t ∈ T. Each nurse can monitor φ patients at a time, and therefore, φR _{ t } patients can be treated simultaneously. Combining this with the number of beds O _{2}, the capacity of the recovery at time t ∈ T is given by \(\min\left(O_2,\varphi R_t\right)\). The number of patients present on the recovery department at time t ∈ T should be less than or equal to this capacity. This is ensured by constraints that are of the same structure as constraints (12) and (13).
Like the holding department, the recovery department also prefers a levelled amount of patients that are present at each point of time. Therefore, penalty costs are incurred when the number of patients exceeds a certain threshold. This is modeled by the step function f _{ RC }(z _{ t }) which provides the penalty costs incurred when z _{ t } patients are present at time t ∈ T. The total penalty costs for the recovery is then given by p _{ RC } = ∑ _{ t ∈ T } f _{ RC }(z _{ t }).
2.8 Radiology department
For some surgeries, an Xray machine is needed during surgery. These surgeries are given by the set \(I_{RL}\subseteq I\). For these surgeries a radiology technician should be present during administering anesthesia and the surgical procedure. This means that he/she does not have to be present during reversing anesthesia. The set of radiology technicians is given by set K and consists of χ radiology technicians. We restrict the number of required radiology technicians d _{ t } at time t ∈ T to be smaller than or equal to the number of available radiology technicians. The constraints that ensure this are similar to constraints (4) and (5).
The survey showed that it is important for the radiology department that their employees at the OR department finish as early as possible such that they can carry out other work at the radiology department. Therefore, penalty costs are incurred when a radiology technician finishes later than needed, i.e., when the time the radiology technicians are present is longer than the time the radiology technicians are needed. In the following, we show how this is incorporated in the ILP.
Using the above constraints and the fact that we minimize the working time of the radiology technicians, τ _{ k } denotes the time the radiology technicians finish. However, these values do not equal the number of time periods they are actually present at the OR. To obtain this value, the start time and break time should be subtracted. The start time of the radiology technicians is given by min _{ j } S _{ j }, i.e., the start time of the OR department. For other hospitals, the start times of the radiology technicians may not be fixed. When this is the case, the start times can be determined with constraints similar to (15) and (16). In addition, all radiology technicians have a break of 45 min, i.e., \(\frac{45}{\delta}\) time periods. The amount of periods the radiology technicians are having a break is thus given by \(\upsilon=\frac{45\chi}{\delta}\). Therefore, the amount of time periods the radiology technicians are present at the OR is given by ∑ _{ k } τ _{ k } − χS _{ j } − υ. This is an underestimation in case one or more radiology technicians finish before their break. However, we expect that this will rarely happen in practice.
2.9 Pathology department
During some surgeries, tissue is removed from a patient which needs to be examined by a pathologist. These surgeries are denoted by the set \(I_{PA}\subseteq I\). After the surgical procedure, the tissue is transported from the OR to the pathology department. When tissue arrives after closing time, overtime is incurred. Let q _{ i } be an integer variable which denotes the amount of overtime which would be created by a single surgery i ∈ I, i.e., the number of time periods the tissue arrives late plus the examination duration W.
When two sets of tissue arrive really late at approximately the same time, this number is a lowerbound on the amount of overtime. However, this situation is not likely to occur in practice.
Again, a step function f _{ PA }(q _{total}) is used to express the penalty costs incurred for a value of q _{total} and represents the total penalty costs p _{ PA } incurred for the pathology department.
2.10 Logistic department
The logistic department is responsible for preparing materials needed during surgery. The materials are laid out in the order in which the surgeries are scheduled. When two surgeries are interchanged, the logistic assistant incurs penalty costs, because they have to change the order in which the materials are laid out. Two surgeries i ∈ I and \(\hat{i}\in I\) can only be interchanged when they are scheduled in the same OR, i.e., when \(O_i=O_{\hat{i}}\). These two surgeries are interchanged when \((P_iP_{\hat{i}})(w_{\hat{i}}w_i)>0\), where P _{ i } is the start time in the initial OR schedule and w _{ i } is the start time in the new OR schedule. When this holds, we either have that both \((P_iP_{\hat{i}})\) and \((w_{\hat{i}}w_i)\) are positive or that both are negative. When both are positive, we have that \(P_i>P_{\hat{i}}\) and \(w_i<w_{\hat{i}}\). This means that in the initial OR schedule, surgery i ∈ I was scheduled later than surgery \(\hat{i}\in I\) and that in the new OR schedule, surgery i ∈ I was scheduled earlier than surgery \(\hat{i}\in I\). When both are negative, we have the opposite case.
2.11 Objective function
The goal of our model is to minimize the deviation from the preferences of the stakeholders. We denote the set of stakeholders by Π and we consider each stakeholder to be equally important. Since the order of magnitude of the cost functions introduced for the different stakeholders may differ, we have to introduce a weighted sum of the penalty costs p _{ π } to compensate these differences. In this function, the priority β _{ π } assigned to stakeholder π ∈ Π is determined such that all stakeholders contribute approximately the same amount to the objective function value. The general concept of the weighted sum of penalties has furthermore the advantage that by varying the priorities, we can develop, for example, also a more patient centred OR schedule.
2.12 Problem complexity
The problem introduced in the previous subsections has been modeled as an ILP. The following theorem justifies this approach, since it shows that efficient exact approaches are unlikely to exist.
Theorem 1
The OR rescheduling problem is strongly NPhard for two or more operating rooms.
Proof
We prove the theorem by reducing 3partition to the OR rescheduling problem. The 3partition problem can be formulated as follows. Given positive integers a _{1},...,a _{3t }, and b with \(\sum_{j=1}^{3t} a_j=tb\), do there exist t pairwise disjoint subsets \(R_k\subset \lbrace 1,\ldots,3t\rbrace\) such that \(\sum_{j\in R_k}a_j=b\) for k = 1,...,t? The 3partition problem is proven to be strongly NPhard (see Garey and Johnson [9]).
The capacities of the holding and recovery departments are assumed to be larger than 6t − 2, thus we do not have to consider the given constraints for these departments. Furthermore, only one radiology technician is available, and therefore, we have to consider the given constraints for the radiology department. Our goal is to create an OR schedule with objective value less than or equal to zero.
The proof of this theorem shows that already a very restricted version of the OR rescheduling problem is strongly NPhard.
3 Computational results ILP
We tested our ILP on data from the Isala Clinics, a hospital in the Netherlands. The data consists of 1168 surgeries scheduled over 27 days. The surgeries consist of 354 emergency surgeries, 193 surgeries who need Xray, 79 surgeries during which tissue is removed, and 7 complex surgeries. The average expected duration of the surgeries is 103 min, and the average realized duration of the surgeries is 91 min. Because rescheduling is only performed during working hours, we removed the emergency surgeries that start before 07:30 and after 18:00. We implemented our model in AIMMS 3.10 and solved it with CPLEX 12.1 on an AMD Ahtlon X2 Dual Core L310 1.2 GHz processor with 4 GB RAM.
In the first two subsections, we discuss the parameter settings for the ILP model and the achieved results which are used to derive the decision rules for the DSS. In the last subsection, we determine the penalty costs for the initial OR schedule used at the Isala Clinics and the OR schedule realized at the end of the day. In addition, we optimize both the initial and realized OR schedule to show what improvements potentially can be realized when the developed method is used.
3.1 Parameter settings
In this subsection, we discuss the parameter settings for the time periods and the priorities for each stakeholder.
For each of the stakeholders, we have to determine its priority in the objective function. In the objective function, the total penalty costs of each stakeholder is multiplied by this priority. Our goal is that each stakeholder has approximately the same contribution in the objective function. In the following, we describe how we have determined the priority of each of the stakeholders.
Total penalty costs and priorities

Patient 
Ward 
Holding 
OR assistants 
Recovery 
Radiology 
Pathology 
Logistics 

Total penalty costs 
7.45 
5.26 
0.00 
1.11 
1.91 
1.00 
0.56 
0.91 
Priority 
0.08 
0.11 
1.00 
0.50 
0.29 
0.56 
1.00 
0.62 
Weighted costs 
0.56 
0.56 
0.00 
0.56 
0.56 
0.56 
0.56 
0.56 
Note, that the holding department does not incur any penalty costs, because penalty costs are only incurred when four patients are treated simultaneously. The constraints imposed, however, limit the number of patients present on the holding department to three. This means that the holding department is not a bottleneck in the current situation.
We conducted some further tests where we varied these priorities slightly. The results from these tests show that patients, wards, and OR assistants have opposite interests compared to the recovery, radiology, pathology, and logistic department.
3.2 Deriving decision rules
 Scenario 1
After 10 a.m.: In this scenario, the realized durations of the surgery that started before 10 a.m. are known. An emergency surgery is only included if it arrived before 10 a.m..
 Scenario 2
After 12 p.m.: In this scenario, the realized durations of the surgery that started before 12 p.m. are known. An emergency surgery is only included if it arrived before 12 p.m..
 Scenario 3
After 2 p.m.: In this scenario, the realized durations of the surgery that started before 2 p.m. are known. An emergency surgery is only included if it arrived before 2 p.m..
Results scenario 1, 2, and 3

10 a.m. 
12 p.m. 
2 p.m. 

Rescheduled surgeries 
566 
416 
213 
Shifted surgeries 
375 
297 
176 
Exchanged surgeries 
1 
0 
0 
Canceled surgeries 
0 
0 
1 
No break 
264 
183 
71 
Break 15 min 
166 
112 
66 
Break 30 min 
62 
33 
21 
Break 45 min 
22 
19 
11 
Break > 45 min 
38 
37 
13 
Mean break 
15.84 
18.71 
19.78 
Table 2 shows that shifting a surgery is the most frequent adjustment used, and often we see that a break is scheduled between two surgeries. The average length of a break is 15–20 min. When we only consider OR utilization, this may not seem to be optimal, however, these breaks can improve the perceived workload of other departments or may be necessary to fulfil the resource constraints. From the results we conclude that only two types of adjustments are preferred to be used. A surgery can be shifted or a break can be scheduled between two surgeries. This means that the order of surgeries stays the same during the day. So when we only allow these two adjustments, the number of feasible solutions decreases significantly, because we only have to consider one sequence of the surgeries instead of all possible sequences. Based on this, it is possible to develop a simple heuristic to determine a good OR schedule. A further benefit of this is that we do not need an expensive ILP solver to implement our approach.
Following the above considerations, we have incorporated this simple heuristic in a DSS which is described in Section 4.
3.3 Potential improvements
 Scenario 4
Initial OR schedule: In this scenario, we compare the initial OR schedule used at the Isala Clinics with a new OR schedule determined by the ILP model, where we rescheduled all elective surgeries. Therefore, there are no emergency surgeries to be scheduled, and only the expected duration of each surgery is given.
In addition, we want to determine what improvements the DSS could make to the realized OR schedule. Filling in the actual surgery durations in the initial OR schedule will most likely result in an infeasible solution and because we do not know the exact decisions the OR manager will make, we can only guess what improvements can be achieved. To provide an upperbound on these improvements, and thus a lowerbound on the penalty costs, we optimized the realized OR schedule and compared it to the realized OR schedule provided by the Isala Clinics. The realized OR schedule of the Isala Clinics denotes how the surgeries were actually performed. This means that several changes have been made to the initial OR schedule, and therefore, we can determine the penalty costs achieved for these changes. For the optimal realized OR schedule, we assume that all realized durations of the elective and emergency surgeries are considered to be known in advance, and also the canceled surgeries are taken into account. In practice, this information is not known beforehand, and thus, this optimal realized OR schedule can not be achieved in practice. However, it provides a bound and, therefore, an indication of the room for improvement.
 Scenario 5
Realization: This scenario consists of all the elective surgeries scheduled in the initial OR schedule and all emergency surgeries that arrived during the day. The realized duration of all surgeries is known.
Results scenario 4 and 5
Total penalty costs 
Initial OR schedule 
Realized OR schedule  

Original 
Optimal 
Original 
Optimal  
Cancellation 
0.00 
0.00 
66.67 
22.22 
Patient 
0.00 
0.00 
40.91 
36.50 
Ward 
0.00 
0.00 
34.40 
28.56 
Holding 
0.00 
0.00 
0.00 
0.00 
OR assistants 
0.00 
0.68 
6.43 
4.20 
Recovery 
4.93 
0.74 
8.85 
3.11 
Radiology 
1.63 
0.56 
2.41 
1.07 
Pathology 
0.64 
0.15 
0.70 
0.51 
Logistics 
0.00 
0.00 
9.26 
5.93 
Total costs 
3.00 
1.03 
87.03 
35.82 
Table 3 shows that in Scenario 4, the initial OR schedule, the total penalty costs for the recovery and radiology department decreases. However, this can only be achieved by scheduling some surgeries in overtime. This follows from the slight increase of the total penalty costs for the OR assistants. For Scenario 5, the realization, the results show that the objective function value is reduced with approximately 60%. The major decrease is caused by the reduction of the number of cancellations. Also, the total penalty costs for the recovery department decreases significantly. In practice, the penalty costs for the realized OR schedule will lay somewhere between 35.82 en 87.03 when the DSS, discussed in the next section, is used.
Concluding, our model can potentially improve the initial and realized OR schedule significantly.
4 Decision support system
To make our method applicable in practice, we have developed a DSS which can be used by the OR manager. We incorporated the two decision rules that are derived from the results in Section 3.2. The first decision rule is that the order of surgeries must be maintained, but that a surgery can be shifted in time. In addition, the second decision rule states that it is allowed to schedule a break of at most one hour between two surgeries. This may help to decrease the perceived workload of several stakeholders and may be necessary to fulfil the resource constraints.
The DSS gives insight in how other departments are influenced by a change in the OR schedule by denoting the penalty costs and weighted costs incurred for each stakeholder. This can convince surgeons that it can be useful to schedule a break between two surgeries. In addition, the DSS can determine whether the initial OR schedule is feasible or not by checking all constraints given in Section 2. The system denotes for each constraint how many times it is violated in the proposed OR schedule. It furthermore can be used to adjust the schedule such that it is feasible. Also, the penalty costs for the proposed OR schedule are calculated which gives an indication of how good the schedule is. The last advantage of the DSS is that the realized OR schedule can be evaluated. This way, the OR manager can learn from his decisions made in the past.
5 Simulation study DSS
To determine the effect of using the DSS in practice, we tested the developed DSS on the instance given in Section 3 by means of a simulation study. In this study, we simulate the use of the DSS by rescheduling an OR immediately when it is disturbed. We choose to evaluate this rescheduling strategy, because we believe that this strategy will provide the best results.
At the beginning of each of the days in the considered instance, only the information that at that time is also known in practice is given. More specific, only the initial OR schedule is known, which is defined by the assignment of the elective surgeries to an OR and the initially planned start times P _{ i } of the elective surgeries together which their expected duration E _{ i }. This means that the emergency surgeries are not known in advance in contrast to the considered Scenario 5.
For each point in time during the simulation, we insert the arrived emergency surgeries to the OR schedule and we update the surgery duration for each of the ongoing surgeries. For each arrived emergency surgery, we set the planned start time equal to their time of arrival and set their duration equal to the expected surgery duration. In addition, each emergency surgery has to be performed in the same OR as it was performed in the realized OR schedule of the Isala Clinics.
For the ongoing surgeries, we update the duration according to the following rules. When the realized duration of a surgery is less than the expected duration, we set the duration equal to the expected duration until the surgery is finished. At the moment in time the surgery is finished, we set the duration equal to the realized duration. When the realized duration of a surgery is larger than the expected duration, we set the duration equal to the maximum of the expected duration and the already executed duration thus far. When the already executed duration equals the realized duration, the surgery is finished and thus, the duration is set to the realized duration. In the next step of our simulation, we determine for each OR whether it becomes empty or whether an elective surgery which is scheduled to start at this point in time cannot be started. When an OR becomes empty, a new surgery may be started which could potentially improve the OR schedule for one or more of the stakeholders. In case an elective surgery which is scheduled to start cannot be started, we have the situation that an already started surgery assigned to the same OR takes longer than its expected duration. When one of these two cases occur, we reschedule this OR to decrease the penalty costs or to make the OR schedule feasible again. We reschedule this OR with respect to the fixed schedule in the other ORs as described in Section 4.
Results simulation study DSS
Total penalty costs 
Realized OR schedule  

Original 
Optimal 
DSS  
Cancellation 
66.67 
22.22 
0.00 
Patient 
40.91 
36.50 
29.50 
Ward 
34.40 
28.56 
21.44 
Holding 
0.00 
0.00 
0.00 
OR assistants 
6.43 
4.20 
12.21 
Recovery 
8.85 
3.11 
14.70 
Radiology 
2.41 
1.07 
2.78 
Pathology 
0.70 
0.51 
1.14 
Logistics 
9.26 
5.93 
0.00 
Total costs 
87.03 
35.82 
45.13 
6 Conclusions
In this paper, we considered the problem of rescheduling surgeries on the day of execution. We formulated an ILP which determines the best adjusted OR schedule at a given point in time. The results show that patients, wards, and OR assistants have opposite interests compared to the recovery, radiology, pathology, and logistic department. Furthermore, the achieved results show that, without a few exceptions, the only used adjustments are (i) shifting surgeries, and (ii) scheduling breaks between two surgeries. These two decision rules are incorporated in a developed DSS. This system determines the best adjusted schedule for one OR with respect to the given constraints and gives insight in how the workload of stakeholders is influenced by adjusting the OR schedule throughout the day. The simulation study shows that by using this DSS, less surgeries are canceled and patients and wards are more satisfied, but also that the workload of several departments increases to compensate this.
A drawback of the developed DSS is that the decision rules may not be applicable when the priorities of the stakeholders change. A change in these priorities for the ILP can result in, for example, more exchanges or cancellations of surgeries. However, this is not expected in hospitals that have a similar group of stakeholders as the Isala Clinics, because these two adjustments are less preferred than shifting a surgery.
Another aspect of the DSS, which may be seen as a drawback, is that it only improves the OR schedule for one OR at a time. The idea for such a type of approach comes from the Shifting Bottleneck heuristic [1], where in each step the schedule of the bottleneck resource is optimized with respect to the schedule for the other resources. As for the Shifting Bottleneck procedure, our approach is only a heuristic approach and thus will not result in optimal solutions. However, the results of the simulation study show that the DSS performs relatively good when compared to the optimal OR schedule. In addition, the Isala Clinics do not prefer to reschedule multiple ORs at once, because then the process of optimization may be unclear to the user and the necessary changes to the ORschedule at one point in time may be quite large resulting in a decrease of the acceptance of the achieved results.
Further research could focus on including the Central Sterile Supply Department (CSSD) into the model. This department prepares the instrument sets needed for a surgery. When a surgery is added to the OR schedule during the day, this may influence the workload on the CSSD. In addition, the CSSD may impose some extra constraints on the OR schedule.
There are several ways in which the developed DSS can be used, for example, reschedule an OR immediately when it is disturbed or reschedule all ORs at some moments in time. The last example also raises the question in what order the ORs should be rescheduled. Therefore, it would be interesting to investigate the best way to use the DSS.
Acknowledgements
This research is supported by the Dutch Technology Foundation STW, applied science division of NWO and the Technology Program of the Ministry of Economic Affairs. We would like to thank the anonymous referees for their useful comments which helped to improve the paper.
Open Access
This article is distributed under the terms of the Creative Commons Attribution License which permits any use, distribution, and reproduction in any medium, provided the original author(s) and the source are credited.