Statistical analysis

Cut-off values for surgeon and hospital volume were based on the Swedish national guidelines for prostate cancer that recommend that a RP surgeon should perform at least 25 RP’s per year and that there should be at least two surgeons at each department who perform this number of RP’s [14]. Multiples of these numbers were used to define cut-off values for each volume group: very low (surgeon <13/ hospital <50), low (surgeon 13-25/hospital 50–100), intermediate (surgeon 25-50/hospital 100–150), high (50-75/ hospital 150–200) and very high (surgeon ≥75/ hospital ≥200). The mean number of RARP’s per year was calculated by dividing the accumulated number of RARP’s per surgeon and hospital with the study period of 4 years. The grouping of hospital volume was performed before the exclusion of RARP’s with missing data on surgeon identity.

The Multiple Imputation Chained Equations (MICE) method [15] was used to impute missing data on PSA, prostate volume, number of positive biopsy cores, cT stage, Gleason score and extent of lymph node dissection. The following additional variables were included to improve predictions: region, patient age at surgery, year of surgery, total mm of cancer in biopsy cores, number of biopsies with cancer, vital status, and time from surgery to death or 31^st of December 2018, whichever event came first. Data were imputed 20 times and results from the models fitted to each dataset were pooled using Rubin’s rules [16].

Outcome variables were dichotomized; thresholds for short operative time and low blood loss were based on the median. Univariable logistic regression was used to calculate odds ratios (OR) for the outcomes. In the multivariable logistic regression we included age at RARP, PSA, prostate volume, PSA density, number of positive biopsy cores, cT stage, Gleason score, and extent of lymph node dissection. Surgeon and hospital volume were analysed in two separate models. The multivariable logistic regression models were subsequently used to construct covariate-adjusted funnel plots using the method proposed by Spiegelhalter [17].

All statistical analyses were performed in R Statistical Software (Version 3.6.1; Foundation for Statistical Computing, Vienna, Austria).

Strength and limitations

The strengths of this study include the nation-wide, population-based study group with a virtually complete capture of all men in Sweden who underwent RARP at all types of hospitals. In contrast to many studies based on administrative registries, we had access to detailed patient and cancer characteristics enabling us to adjust for potential confounders in our analyses. Another strength is inclusion of both surgeon and hospital volume. Exclusion of surgeons and hospitals with <5 RARP/year could have introduced a bias. We argue for the opposite, that including these observations would have led to erroneous estimates since it is likely that many of these registrations were of poor quality. With this threshold, we avoided erroneous registration of RARP at hospitals where there is no surgical robot or registration of RARP performed by visiting high volume surgeon at a small hospital. The proportion of missing data was small for all variables in the model except for number of positive biopsy cores. We applied multiple imputation to handle missing data, however, results were virtually identical when we analyzed these factors in a complete case analysis. Limitations of our study include lack of data on functional and long-term oncological outcomes and surgical experience for individual surgeons prior to the study period. Given this lack of data we had to assume that all surgeons practiced during the entire study period, which would “falsely” lower the average number of RARP’s/year for some surgeons.

A systematic review on the association between volume and outcome after RP reported 49 original publications of which 11 studies examined both surgeon and hospital volume [4]. High surgeon and hospital RP volume were associated with better outcome, however, the association varied between outcomes. For example, cost and mortality were more dependent on institutional factors whereas incontinence and length of hospital stay were more related to individual surgeon [4].

We found a clear dose-response association between surgeon volume and short operative time and low blood loss, which could not be observed for hospital volume, indicating that these outcomes are more dependent on the individual surgeon’s technique than on hospital factors. Other studies have reported similar associations [18, 19]. Surgeons and hospitals in the highest volume groups had higher odds of a nerve sparing intent compared to very low volume surgeons and hospitals. However, nerve sparing intent was based on the surgeon´s self assessment and is not an outcome and in order to investigate the validity of this variable, data on functional outcomes are needed.

We found that the highest volume surgeons had almost two-fold higher odds of negative surgical margins vs. the lowest volume surgeons. Surgical margin status in organ confined prostate cancer is an important measure of surgical quality and positive surgical margins are associated with increased risk of biochemical recurrence and secondary treatment [20]. Most previous studies have reported a higher proportion of negative margins with increasing RP volume [21–23]. For example, in a recent study based on the American National Cancer Database, there was a higher proportion of negative margins in high volume hospitals [22]. Besides surgical technique, margin status is also affected by T stage, Gleason, quality of the pathology assessment etc, which could explain the lack of association with hospital volume and the wide range between hospitals in the same volume group.

We found no association between readmission and volume which is in line with a previous study from PCBaSe [24] but in contrast to a register-based US study [22].

The type of volume-outcomes association differed between outcomes. For example, for operative time and blood loss, we found a dose-response association with improved outcome with increased volume but for negative surgical margins, there seemed to be a threshold effect where only the very-high volume group had significantly better results. One potential explanation could be that the learning curve to reach improved oncological outcome with RARP is longer than the learning curve for operative time and blood loss.

We speculate that surgeon volume is the driver in the volume-outcome association. Compared to many previous studies, the association between hospital volume and perioperative outcomes was weak in our study [4], likely due to the fact there were very high volume hospitals with many low to intermediate volume surgeons and lower volume hospitals with high/very high volume surgeons.

We observed a wide range in outcomes also for surgeons and hospitals within the same volume group. However, for most of the outcomes, variation within a volume group was wider with decreasing volumes. For example, the largest variation in operative time was seen for surgeons and hospitals in the lowest volume groups.

The proportion of men who received a nerve sparing procedure spanned from less than 10% to close to 100% between very low volume surgeons and from less than 30% to more than 90% between high volume hospitals. Similarily, the proportion of negative surgical margins in very high volume surgeons varied from <70% to >80%. For readmission, the variation was smaller in all volume groups.

In these comparisons within a volume group, there should be limited influence by case-mix suggesting that there are other factors that affect outcome. Random effect can probably explain some variation, especially in the lower volume groups. There are numerous reports describing heterogeneity between surgeons, including those with high volumes [25–29]. In a recent population-based, Swedish study, significant variation in functional and oncological outcomes was seen, also for experiences surgeons. Even tough surgeons’ experience and annual volume could explain some of the observed heterogeneity in outcomes, most of the heterogeneity remained after adjusting for these factors. Individual surgeon skill and talent probably explain some variation. Some surgeons are more skilled than others are, but to what extent these skills can be improved by training is unknown and probably differs between individuals [29].

Our results do not answer whether prostate cancer surgery can or should be centralized but rather emphasize the need to register and report surgical performance in order to ensure quality control and quality improvement. When outcomes are registered, best practice can be identified and disseminated. Continuous registration and feedback from registers such as the NPCR can help raise the minimum quality standards for surgeons and hospitals regardless of volume. We argue that quality registration should be compulsory since. quality assurance of RP has been shown to improve outcome. For example, in a quality assurance program at the Pelvic cancer surgical center in London, UK complication rate fell from 13% to 7% and urinary continence 3 months after surgery increased from 57% to 67% [30].

In 2018, NPCR introduced electronic forms for patient reported outcome measures (ePROM) for men undergoing RP and radical radiotherapy. Patients are asked to fill in electronic questionnaires before treatment, and at 3 and 12 months after treatment. The results are immediately available online at the secured Information Network for Cancer Care (INCA) platform where comparisons between regions, hospitals, and individual surgeons within the same hospital can be performed. Capture was limited in 2018, so there is not enough data as of yet to investigate the association between volume and PROM.