Productivity in relation to organization of a surgical department: a retrospective observational study

The operation planning and scheduling software Orbit (version 4.25.4, EVRY Healthcare System AB, Sweden) was used to extract perioperative patient records for patients with primary diagnoses corresponding to either elective surgery for breast cancer or malignant melanoma at Umeå University Hospital in Sweden between the years 2012 and 2016. We chose the time interval to maintain a consistent group of surgeons and personnel at each department. To prevent bias, data from the Orbit software from late 2016 were not included, as there had been a major update to the software with a completely new user interface, which may have affected the registration process. Orbit contains both patient- and procedure-specific information gathered according to a predetermined protocol, though with some freedom for comments and notes; therefore, it offers a good basis for data collection.

Patients undergoing surgical intervention for breast cancer or malignant melanoma were given surgery time slots at one of two surgical departments, namely at a general surgery (GS) department or a cardiothoracic surgery (CT) department. We provide an organizational comparison of these two departments in Table 1.

Personnel scheduling operations allocated the patients from a surgical waiting list with the aim of maximizing OR utilization. Thus, the selection to each department thus mainly depended on when the patient was referred for surgery, typically being allocated in the order in which they were referred, as well as the expected duration of the operation (shorter surgeries may fit into shorter time slots in the schedule).

The study design was chosen based on the assumption that the department to which a patient was assigned was unrelated to patient characteristics (i.e., that the assignment in practice could be considered randomized). However, after 2013, more available resources were prioritized for breast cancer surgery, which affected the referral of patients with malignant melanoma, decreasing the number of melanoma patients referred to the CT department. In addition, the CT department was utilized only 1–2 days/week for the included surgery, whereas the included procedures were, to some extent, performed 5 days/week at the GS department. For more information about the allocation process, see Additional file 1: Fig. S1. In order to keep the data set manageable, the time frame for the GS department was reduced to only allow inclusion between January and May of each year, whereas patients from the CT department (with a smaller patient base) were included during the whole study period. Therefore, the overall conclusions of the study were not altered by allowing the estimated effect to vary between procedure groups.

To be eligible for inclusion in the study, a patient had to be admitted for elective surgery under general anesthesia for either breast cancer or malignant melanoma. The surgical procedure was to be performed Monday to Thursday, excluding bank holidays, between January and May at the GS department or January to December at the CT department. Three criteria were used to classify the different procedure groups: (1) according to diagnostic and intervention codes, as classified by the ICD-10 standard [17], (2) the type of procedure they underwent, and (3) the Classification of Care Measures (i.e., CCM codes). The CCM codes include codes for surgical procedures used in Sweden, a classification generated by the Swedish National Board of Health and Welfare [18]. The procedures were divided into more extensive (Group 1) and limited (Group 2) procedures for breast cancer and operations for malignant melanoma (Group 3). For more information on the procedure code-classification, see Table 2. Note that, for each patient, the ICD-10 classification was only the primary classification in each of the procedure groups; that is, certain specific interventions, such as lymph node biopsy (PJA10) or reconstruction using tissue flap and suture (ZZR70), can be performed in each case without affecting which procedure group includes the patient. Any patients having diagnoses/interventions not listed in Table 2 were excluded from the analysis. Patients undergoing autologous or implant-based immediate breast reconstruction, cases lacking registered values for body mass index (BMI), and those with time registration errors were also excluded.

As listed in Table 1, the anesthesiologists at the GS department handle 2–3 ORs, whereas they are only responsible for one OR at the CT department. At the GS department, there is often a lack of personnel outside the OR assisting with transportation of pathology samples and equipment set-up. The GS department uses a team of one operating room nurse (ORN) and one enrolled nurse during the surgical procedure, whereas the CT department strives to always have two ORNs. During lunch breaks, the GS department usually switches the whole OR team, whereas the CT department has their auxiliary personnel switched in one member at a time. At the GS department, staff can start working at different hours (e.g., scrub nurse starts at 07:00 whereas the circulating nurse starts at 07:30), whereas the staff in the CT department always start at the same hour. There are also some differences in the total number of personnel at each department, whit the GS department having a staff of approximately 80 and the CT department having a staff of approximately 40. Both departments maintain the availability of resources for any emergency surgical procedures.

The outcome variable of interest was the total case time in minutes. This is defined as the time elapsed from when the OR team starts case preparation until the room is cleared and ready for the next operation. The difference in the average total case time between patients undergoing surgery at the CT and GS departments was the primary end point.

Several covariates were available, such as clinical, staffing and patient characteristics (Table 3). Only covariates measured before assignment to the specific department were included in the analysis, as adjustment for covariates determined after the assignment can introduce bias in the estimated effect of interest [19].

To increase the precision in the estimated effect of the department allocation, and to reduce bias because of differences in covariates between the two departments, we adjusted for covariates of clinical importance (Table 3). Missing values were handled through listwise deletion (i.e., the removal of observations with any missingness). A linear regression model was used to estimate the average difference in total case time between the two departments. In addition to patient-related covariates (patient age, BMI, sex, ASA score), the procedure group and three additional covariates were included in the regression model: the presence of a surgical trainee/assistant surgeon, whether the surgeon was a consultant, and the surgeon’s level of experience. These were pre-treatment variables because the staff schedules were set in advance, before allocation to the departments. In terms of the choice of patient related covariates, we argue that ASA score provides a rather well-summarized picture of the severity of the patient’s condition [20]. According to Luedi et al. [21], estimates of case duration can be improved by including patient age. BMI is a significant, albeit weak, predictor of difficulties with tracheal intubation and the prolongation of non-operative time [22]. Unfortunately, there were some missing values for this covariate. Although listwise deletion leads to unbiased estimates of the average treatment effect if the values are missing completely at random [23], additional efficiency can be gained in randomized studies by imputing the missing data. We investigated the sensitivity of our results to deviations from the missing completely at random assumption by systematically imputing higher BMI values at one department than at the other. If the data are missing completely at random for a covariate in a randomized study, then imputing the sample mean into the regression model for the missing data and adding an indicator variable for missingness typically leads to estimates with lower standard errors [24]. The results were largely unchanged by this imputation scheme. We argue that, even if the missing BMI values were systematically higher at one of the two departments, our qualitative conclusions are the same, at least if the missing BMI values are similar in magnitude to those observed. Therefore, we opted to include all these covariates in our model.

We carried out all statistical analyses in R (software version 3.5.1, R Foundation for Statistical Computing, Vienna, Austria, 2018) [25].

A total of 797 eligible patients were identified in the chosen time frame, but we excluded 142 of these patients due to not having one of the pre-specified diagnoses. Another 118 patients were excluded because of missing patient data (i.e., BMI). Ten more patients were excluded due to time registration errors in the Orbit program, where the most frequent error was two or more surgeries being registered as taking place in the same room at the same time. Finally, one patient was excluded because they were the only individual with an ASA score of 4. After these exclusions, the final study cohort comprised 526 patients (Fig. 1).

As mentioned previously, several covariates, including clinical, staffing, and patient characteristics, were available and used in the analysis. We summarized these covariates in Table 4. The differences between patients allocated to the two departments were small overall, except for the distributions of ASA scores and procedure groups.

The average unadjusted difference in total case time between the two departments was 40.39 min: 206.96 min (standard deviation 60.89) at the GS department and 166.57 min (49.35) at the CT department (p < 0.001). Table 5 gives a breakdown of the three main phases of the operations defined by the PTG terminology, and their unadjusted time consumption. Notably, in this setting, the first four steps outlined by the PTG occur in parallel with the clinical preparations (e.g., RSS) and are not accounted for in time measurement and not included in the table. They are also outside the scope of this study because the focus is on the perioperative part, and all the surrounding aspects are not affected by which department will perform the procedure.

Adjusting for covariates in a linear regression model, the estimated regression coefficients are shown in Table 6 with their corresponding standard errors and p-values. After adjusting for the selected covariates, the difference in total case time remained significant, with patients who underwent surgical procedures at the CT department having a 30.67-min shorter total case time on average than those at the GS department (p < 0.001, 95% confidence interval − 40.02 to − 21.32). Differences in average operation time are estimated in our model because dummy variables for procedure groups were included, and we investigated a possible interaction effect between procedure group and operating department. However, a separate subgroup analysis would allow for coefficients of the control variables (such as age or sex) to differ between procedure groups, which our model did not. The estimated effect of interest was similar in all three procedure groups (data not shown), though other coefficients changed in both magnitude and sign.

Only the BMI variable had missing data, but the results were similar when mean-imputation of BMI was performed and an indicator for missingness added to the regression model (estimate = − 31.94, p < 0.001), with a 95% confidence interval of − 40.81 to − 23.07 for the estimated effect of department assignment on total case time. In the two sensitivity analyses for the missing completely at random assumption, the results were similar to the results obtained through listwise deletion, with the 95% confidence intervals for the effect of department allocation being − 38.10 to − 21.01 (estimate = − 29.55, p < 0.001) and − 41.65 to − 24.69 (estimate = − 33.17, p < 0.001) for the two imputation schemes, respectively.

There are limitations in comparing the Swedish healthcare system to, for example, the US healthcare system. With a population of almost 360 million versus Sweden’s 10 million, the US spent $10,624 per capita, or 17% of its gross domestic production (GDP), on healthcare in 2018, whereas Sweden spent $5982 per capita, or 11% of its GDP [39, 40]. In regards to hospital ranking, the fact that 4 of the top 10 hospitals in the world are in the US [41] says a lot about the competence, quality, and potential of the US healthcare sector. Sweden, however, also ranks high with one hospital in 7th place [41]. In a comparison of healthcare performance by the Commonwealth Fund, the US was ranked 11th and Sweden 7th [42]. Although the US system heavily outranks most others when it comes to the metrics of care processes, they are ranked last on all others [42], showing that discussions about socialized medicine are complex and multifaceted. This leads us to believe that the present findings may be one small piece of the puzzle in promoting a more cost-efficient way of delivering healthcare in terms of OR department organization.