Operating Room Delays at a Hospital
Overview
A 300 bed general hospital in the greater Boston area had been unable to reduce backups arising when operations ran longer than scheduled. Although a new automated scheduling system had been installed, numerous operations were still being canceled because the rooms were not available on time. The hospital was forced to maintain costly additional operating units to handle operations that were displaced. We were asked to address two questions: what was the cause of the delays, and why was the new system failing.
Approach
For legal and administrative reasons, hospitals maintain records of the timing of operations, who was present and in what capacity, and the diagnosis and procedures performed. Complications, if any, are noted. Combined with the scheduling data these detailed records made it possible to build ANCOVA models (regression which handles both continuous and categorical variables) to establish the factors determining the length of operations. (These were measured as Z scores, a measure of how far away from the mean value of their type in standard deviations.) Once the model was constructed, exploratory data analysis tools were used to identify operations that were unusually long or short, and identify the special characteristics of those operations.
The models showed that in general, even before considering any specific factor that affects the length of the operations, the scheduling formula used by the automated system was assigning time slots nearly an hour too short—if the goal was allowing a high percentage of the operations to finish within their allotted time slot. This large error arose, not from a miscalculation, but because the average duration of a type of operation was an extremely poor predictor of the duration of that operation. The real problem was variability. The system scheduled operations by their average duration, while many ran significantly shorter or longer than the average. For example, it was not uncommon for some operations to take 200 or 300 percent longer than average. When operations ran shorter, the operating unit was under utilized; when operations ran longer, the entire system was often backed up for the day. Further, the models could account for most of the variability in duration. The length of time required for an operation depended, primarily, on who was performing the operation. The range of time taken for specific types of operations by surgeon was narrow, but across surgeons it was wide.
A second reason the system was so far off was that it could be overridden by doctors, who could block out a time slot for themselves and perform as many or as few operations as they chose. If they were likely to run over, the system would not make an adjustment, as it only adjusted the times it assigned.
The automated system could be adjusted to add extra time, including manually adding to the blocked times. This would reduce backups, but at the price of extreme under utilization. The more direct solution, assigning time slots by the individual surgeon’s record, ran into “political” problems. There was a general feeling that the faster and less variable surgeons were also the more skilled, and thus different time allocations would be public airing of a measure of their competence. Since the hospital’s customers were the doctors who practiced in it (they bring in the patients), and the Boston area had a surplus of medical facilities, there was a fear that the hospital could not afford to alienate its surgeons either by assigning time slots based on their records or restricting blocking privileges. Taking such actions, the hospital feared, would result in doctors choosing to practice elsewhere. As a result, the hospital chose to handle the problem in an informal manner, and for the most part, the pattern of under utilization and delays was allowed to continue.
This pattern of analytic success and practical failure, although not strictly the analyst’s responsibility, is sufficiently common that we now make it a standard practice to work with clients from the outset to be ready to utilize the results of studies.