Business Background and Motivation: Our main client is the newly elected director of the hospital who found that they are losing money in the operating room department due to the improper management on the surgery scheduling. Thus, he arranged a special committee to solve this hard task. The business problem we face is that for surgery scheduling it is often subjectively assigned by the operating doctors. For the doctors, each of them has different preference of time schedule and even the operation rooms which results in the difficulty of proper scheduling. Our contribution is by accurately predicting the duration to help improve the existing scheduling system both from the space utilization and the human resource allocation perspective. We are able to give practical advice based on the numerical evidence. Under the premise of patients’ safety and the quality of the surgery, the medical team can arrange their time resource efficiently and the patients can be aware of the potential duration lasting intervals.

Source of Information: The raw data is the electronic version of the surgery agenda which is authorized by the administration department and is handled by the information department due to the privacy protection protocol. The data includes surgery records from a hospital during the years 2010 and 2011.

Analytics Solution: Once the reservation begins, the required predictors can be gained and will be plug in the specific model to predict the duration under different combination of variables. Based on the best interests of the patients, the scheduling can avoid extreme cases involving dramatic delay of the surgeries due to improper surgery arrangements. The analysis can also improve the utility of each individual operation room to prevent from idling. The data mining project encounters the variable selection difficulty which is solved by using the Lasso Regression and Tree methods to select important variables. By comparing the existing method (use average surgery duration of individual division) and the proposed method, our method outperformed the naïve method in predictive power. For better interpretation, the rules generated by trees are also provided for the administration department to have a general concept of time duration given the diagnosis and surgery type.

Recommendations: The prediction result can provide the scheduling team to build up a reliable appointment in the system. By inserting the required variables the model will reveal the predicted output. With this outcome, the people in charge can better arrange in the situation dealing with conflict of multiple surgeries and availability of rescheduling. In the future, the system can integrate with the shift arrangement system and the accounting system to better control the operation expense and reduce the possible waste of medical resource. All affected members can respond instantly and give improvement advice on a regular basis. If we divided the operation time into intervals, the only time that can be regulate will be the waiting time before a patient enters the surgery room. By controlling the waiting time to within 15 minutes, the operation time can be reduced by 5%. This result implies that the potential overtime expense can be reduced and the allocation of human resource can be more flexible. As for the scheduling rule, according to the simulation result using Arena (simulation software), by properly allocating the surgery room, the capacity can be increased by 7%. It shows that (instead of sticking to one specific operation room) if the surgery is scheduled in any available room, the cycle time can be significantly reduced.