Learning curve for Robot Assisted Laparoscopic Radical Prostatectomy in a large prospective multicenter study

Abstract

Background:

Differences in outcome after radical prostatectomy for prostate cancer can partly be explained by inter-surgeon differences, where degree of experience is one important aspect.

Objective:

To define the learning curve of robot-assisted laparoscopic prostatectomy regarding oncological and functional outcomes.

Materials and methods:

Out of 4003 enrolled patients in the LAPPRO trial, 3583 met the inclusion criteria, of which 885 were operated by open technique. A total of 2672 patients with clinically localized prostate cancer from seven Swedish centers were operated by robot-assisted laparoscopic and followed for eight years (LAPPRO trial). Oncological outcomes were pathology reported surgical margin and biochemical recurrence at eight years. Functional outcomes included patient-reported urinary incontinence and erectile dysfunction at 3, 12 and 24 months. Experience was surgeon-reported experience prior to and within the study. The relationship between surgeon experience and functional outcomes and surgical margin status was analysed by mixed effect logistic regression. Biochemical recurrence was analysed by Cox regression with robust standard errors.

Results:

The learning curve for positive surgical margin was relatively flat with a rate of 21% for surgeons between 0–74 cases and 24% for surgeons who had performed >300 cases. Biochemical recurrence at 4 years was 11% (0–74 cases) and 13% (>300 cases). Incontinence was stable over the learning curve but erectile function improved (at 2 years from 38% (0–74 cases) to 53% (>300 cases)).

Conclusions:

Analysis of the learning curve for surgeons performing robot assisted laparoscopic prostatectomy showed erectile function improved with increasing number of procedures which was not the case for the oncological outcomes.

Introduction

Learning curve in surgery refers to how increasing experience of performing a specific operation affects outcomes. Prostate cancer surgery is a balance between the oncological and functional outcomes where the primary aim is cancer cure. As disease recurrence may occur more than ten years after surgery, short-term outcome measures associated with the definitive oncological outcome like surgical margin, biochemical recurrence and time to metastases are used as surrogate markers. A poor concordance between positive surgical margin and biochemical recurrence rate has been reported [1] and possibly biochemical recurrence is a more reliable surrogate variable for prostate cancer specific mortality.

The surgeon has to take into account tumor severity and local extension in planning and performing the operation. The decision to preserve of the neurovascular bundles or not is a balance between oncological control and avoiding side-effects as urinary incontinence and impotence ([2]). Studies have reported that overall risk for incontinence and erectile dysfunction (ED) after surgery is relatively high, even when performing surgery on small tumors ([3]) and that such functional impairments seriously affect quality of life ([4]). The oncological and functional outcomes after surgery are affected by a number of different factors, such as tumor characteristics and surgical technique but also by surgeon heterogeneity ([5]).

A cohort of 9076 patients treated by open radical prostatectomy between 1987 and 2003 at Memorial Sloan-Kettering Cancer Center, Cleveland Clinic and Wayne State has been analysed for various aspects of surgical experience and reported in several publications [1,6–9]. The overall conclusions from these studies were summarized [10] as follows: the risk of recurrence and positive surgical margin was reduced by increased experience of the surgeon [1,6]. However, the mechanisms of learning for the two different outcomes appeared to differ [9]. The effect of learning was independent of pre-operative risk or pathological T-stage [7], [8], and outcomes remained heterogenous across surgeons after accounting for experience [9].

The LAPPRO trial involves several surgeons, all of whom perform several surgeries over time. This hierarchical study design enables a decomposition of the observed variability of outcome into three different sources: between patient, between surgeon and within surgeon, where the latter can partly be explained by increasing experience, that is a learning curve. Previous analyses in the LAPPRO trial have shown that heterogeneity among surgeons was substantial regarding both functional and oncological outcomes and was to some extent explained by experience [5]. The aim of this paper is to investigate the effect of increasing surgeon experience on oncological and functional outcomes after robot assisted laparoscopic prostatectomy.

Materials and methods

The basis for this study was the LAPPRO-study, a prospective, non-randomized multicenter trial comparing robot assisted laparoscopic prostatectomy (RALP) with retropubic open prostatectomy ([11]). Patients were included at 14 urological centers in Sweden from September 2008 to December 2011, where seven centers performed RALP([11], [12]). In this analysis patients had to meet the following inclusion criteria: age<75 years, clinical tumor stage ≤ T3, PSA <20 ng/ml, no signs of distant metastasis and operated by robot assisted laparoscopic technique ([12–14]). The LAPPRO study was approved by the Regional ethical review board, Gothenburg (number 277–07) and registered in the Current Controlled Trials database (ISRCTN06393679).

Surgeon experience

Surgeon experience was defined as the surgeon-reported experience prior to the study and the total number of surgeries within the study that is the surgeon’s total number of operations also included those performed before LAPPRO. The perioperative clinical report form documented the number of surgeon-reported procedures performed before the day of each surgical procedure according to the following intervals “0–50”, “51–100”, “101–150” and “above 150”. This information was recorded each time the surgeon performed an operation within the trial, as was the date of the procedure. Surgeons indicating “above 150” at first procedure within the trial were later contacted and asked for an exact number of performed procedures at the time of the first procedure registered in the trial. An integer-valued counting variable of experience was derived using a set of three rules:

1. Experience can only increase (e.g. a surgeon first reporting 100–150 and then 50–100 will have the lower value recorded).

2. A maximum of 50 procedures can be reported in each interval. If >50 procedures are reported in the same interval the recoding will involve a simple counting (one by one) to result in consistent numbering of procedures. Surgeons reporting all the surgeries within a single interval were defined as starting at the lowest values of that interval, for example, for the interval “51–100” counting starts at 51.

3. The variable counting the number of past surgeries hence start at the estimated number of operations performed before LAPPRO thereafter includes those performed within LAPPRO.

Clinical data

Clinical data was collected before, perioperatively and at 6–12 weeks, 12 and 24 months after surgery using clinical report forms, collecting information on prostate-specific antigen (PSA), clinical and pathological T-stage, Gleason score on biopsy and surgical specimen, prostate weight, length of cancer in biopsy, presence of a positive surgical margin and Body Mass Index (BMI). Information on residual and recurrent disease was documented by information on PSA, any adjuvant or salvage treatment (androgen deprivation therapy, anti-androgens, radiation, chemotherapy) and diagnosis of metastases.

Patient reported outcomes

At baseline and at 3, 12, 24 months and 8 years postoperatively patient-reported outcomes were collected through questionnaires mailed to the patients and returned to a third party [11]. The questionnaires included questions on education, marital status, comorbidity, urinary and erectile function as well as quality of life. Six years postoperatively, information on recurrent disease were collected through structured telephone interviews with the patients [15]. The construction of the questionnaires have been described in detail in ([12], [14], [5], [13]).

Outcomes

Oncological outcomes were surgical margin status as described in the pathology report of the prostatectomy specimen and biochemical recurrence within eight years, defined as either residual or recurrent disease, see Table 1 for definitions. Biochemical recurrence was derived as time to any event with follow-up times of 3, 12, 24 months, 6 and 8 years. The exact date of the events (PSA increase or initiation of adjuvant/salvage treatment) or censoring (dropout or no recurrence at 8 years) are interval-censored between the current and previous follow-up dates. For this reason, the event time was set to the mid-point between the two consecutive follow-up dates. For example, a PSA increase reported at 12 months but not at 3 month follow-up will be assumed to have occurred at 7.5 months. Differences between this approach and an interval-censored approach were found to be negligible [16]. Functional outcomes were urinary incontinence and erectile dysfunction at 3, 12 and 24 months after surgery. The different follow-up occasions (6–12 weeks, 3, 12 and 24 months and 6 and 8 years) where the different outcomes were assessed are presented in Table 1. The prespecified statistical analysis plan stated all four outcomes as equally important.

Table 1.

Definitions of study outcomes.

Variable	Question	Answer categories	Follow-up	Outcome
Surgical margin	“Growth in resection margin”	1. “no information”, 2. “negative”, 3. “focal”, 4. “extensive”, 5. “other”	Pathology report	Dichotomization: 3 or 4 defined as positive margin. 2 defined as negative. 1 and 5 defined as missing.
Biochemical recurrence	PSA level and additional/salvage treatment documented in CRF and questionnaires.	Residual disease (PSA >0.25 at 6–12 weeks) or recurrent disease (PSA <0.25 at 6–12 weeks and PSA >0.25 at a later follow-up or additional/salvage treatment at 1, 2, 6 or 8 years with or without detectable PSA before start of treatment)	3, 12, 24 months, 6 and 8 years	Time to residual or recurrent disease defined as biochemical recurrence.
Urinary incontinence	How often do you change pad, diaper or sanitary aid during a typical day (24 h)?”	(0) “Not applicable, I don’t use any protective pad”, (1) “Less than once/day”, (2) “About once/day”, (3) “About 2–3 times/day”, (4) “About 4–5 times/day”, (5); “About 6 times or more/day”	3, 12, 24 months	Dichotomization: Cut off between response 1 and 2 (continent and incontinent, respectively)
Erectile dysfunction	When you had erections with sexual stimulation, how often was your erection hard enough for penetration during the past six months?”	(0) “Not applicable. Not been sexually active”, (1) “Never been sufficiently stiff for intercourse”, (2) “My penis was stiff enough for intercourse at less than half of the times”, (3) “My penis was stiff enough for intercourse at more than half of the times”, (4) “Always sufficiently stiff for intercourse”	3, 12, 24 months	Dichotomization: Cut off between response 2 and 3 (dysfunction and function, respectively)

Statistical analysis

For surgical margin status, urinary incontinence and erectile dysfunction the effect of surgeon experience was quantified using a hierarchical logistic regression model with experience measured by number of past surgeries (before and within the study) included as a natural cubic spline with two knots and surgeon as a random intercept. Each follow-up was analysed separately. For recurrence a Cox proportional hazard model was used. Surgeon was included as a cluster and standard errors were estimated using robust variance estimation. Including a random effect/cluster for surgeon ensures that the dependency structure in the data is accounted for such that the uncertainty in the data is not underestimated.

Variables considered to be confounding were included in the regression models for adjustment. For all outcomes, adjustment was made for clinical T-stage, pre-operative PSA, biopsy Gleason score and length of cancer. For surgical margin status and biochemical recurrence, prostate weight was also adjusted for. For urinary incontinence, age at surgery, prostate weight, BMI and diabetes were adjusted for. For erectile dysfunction the models also adjusted for pre-operative erectile dysfunction, age at surgery, diabetes and cardiovascular disease. Nerve sparing technique is considered a mediator on the causal path between experience and outcome and is hence not adjusted for. However, to assess the direct effect of the learning curve on the outcome adjusted for nerve sparing approach, a sensitivity analysis was performed where nerve sparing (yes, no) was added to the set of adjustment variables (Supplementary analysis 1). All continuous variables, including surgical experience were standardized prior to analyses. Patients with pre-operative incontinence were excluded from the analysis of incontinence. To which extent the relationship between experience and outcome is modified by the intensity the surgeries are performed that is annual volume or case load, was addressed by an additional analysis (Supplementary analysis 2).

The bias that missing values may induce were seeked to be reduced by 10-time multiple imputations using predictive mean matching [17] and were subsequently pooled using Rubin’s rule [18]. The results were presented graphically as predicted (fixed effect) curves and associated confidence intervals of the chance of favourable outcome conditional on fixed levels of the adjustment variables. The levels used were median and most frequent values in the cohort values for continuous and categorical variables, respectively. For biochemical recurrence the cumulative incidence (1-survival) at four years was presented. Both adjusted and unadjusted curves were presented as well as raw proportions for experience categorized as 1–74, 75–149, 150–299 and≥300, respectively. The extent to which experience can explain the total variability (both between and within surgeon) in outcome was assessed by pseudo R-square. The models were compared with reduced models without experience included by a likelihood ratio test. For the Cox model with robust variance a Wald test was used. This is different from what was done in [5] where the part of between-surgeon heterogeneity that experience accounted for was quantified. The R software was used for the analyses, with the packages lme4 [19] and survival [20] for parameter estimation, mice [21] for multiple imputations and ggeffects [22] for calculation of predicted mean curves using the functions ggpredict and pool_predictions.

Results

Twenty-five surgeons at seven urological centers in Sweden performed robot assisted laparoscopic prostatectomy on 2672 patients between 2008 and 2011 (Figure 1). Before and within LAPPRO the median number of operations performed by a surgeon was 21(min; max: 0; 228) and 185 (min; max; 8;1026), respectively. Of all 25 surgeons, 21, 9, 3 and 3 had performed at least at total of 200, 500, 750 and 1000 surgeries, respectively, as depicted in Figure 2, 3 and 4. The median annual volume was 40 procedures (5; 84).

Figure 1.

Patient flow chart

Figure 2.

Left panel depicts percentage of patients with a negative surgical margin (no cancer cells in the surgical margin of specimens) as found in the pathology report. Right panel depicts predicted cumulative chance of freedom from biochemical recurrence 4 years after surgery. Unadjusted curve (blue dashed line) represent estimates from regression with no covariates included. Adjusted curve (red solid line) represent estimates from regression with covariates included. Grey area is 95% CI for adjusted estimates. Black dots depicts raw proportions with 95% CI. Number of surgeons performing at least 1, 50, 200, 500 and 750 surgeries are given at the bottom of the left panel.

Figure 3.

Percentage of patients reporting urinary continence (defined as change of sanitary pad less than once per 24 hours) at 3, 12 and 24 months postoperatively. Unadjusted curve (blue dashed line) represents estimates from regression with no covariates included. Adjusted curve (red solid line) represents estimates from regression with covariates included. Grey area is 95% CI for adjusted estimates. Black dots depict raw proportions with 95% CI. Number of surgeons performing at least 1, 50, 200, 500 and 750 surgeries are given at the bottom of the left panel.

Figure 4.

Percentage of patients reporting erectile function (defined as having a stiff enough penis for penetration after sexual stimulation at least 50% of times) at 3, 12 and 24 months postoperatively. Unadjusted curve (blue dashed line) represents estimates from regression with no covariates included. Adjusted curve (red solid line) represents estimates from regression with covariates included. Grey area is 95% CI for adjusted estimates. Black dots depict raw proportions with 95% CI. Number of surgeons performing at least 1, 50, 200, 500 and 750 surgeries are given at the bottom of the left panel.

Patient characteristics were in most aspects not different between surgeons in the early part of training and experienced surgeons, but a larger proportion of patients operated on by an experienced surgeon had a higher level of education. Pre-operative incontinence was reported by 26 patients (Table 2).

Table 2.

Patient, tumor and surgeon characteristics.

Characteristic	Overall, N = 2,6721	0–74, N = 6061	75–149, N = 6401	150–299, N = 8251	300–, N = 6011
Age at surgery(yr)	64 (59, 67)	65 (60, 68)	64 (60, 67)	63 (59, 67)	62 (57, 66)
Body Mass Index	26 (24, 28)	26 (24, 28)	26 (24, 28)	26 (24, 28)	26 (24, 28)
Missing	352	77	83	97	95
University education	927(39%)	178(33%)	199(35%)	301(41%)	249(49%)
Missing	313	64	69	88	92
Residence
Abroad	9(0.4%)	0(0%)	1(0.2%)	3(0.4%)	5(1.0%)
Rural	321(14%)	72(13%)	80(14%)	106(14%)	63(12%)
Urban	2,028(86%)	470(87%)	490(86%)	628(85%)	440(87%)
Missing	314	64	69	88	93
In a relationship	2,140(91%)	493(91%)	521(91%)	661(90%)	465(92%)
Missing	314	64	70	87	93
Diabetes	142(6.0%)	26(4.8%)	35(6.2%)	48(6.5%)	33(6.5%)
Missing	322	68	74	87	93
Cardiovascular disease	822(35%)	184(34%)	204(36%)	262(36%)	172(34%)
Missing	324	66	74	90	94
COPD	54(2.3%)	14(2.6%)	10(1.8%)	21(2.8%)	9(1.8%)
Missing	325	67	77	88	93
Psychiatric morbidity	87(3.7%)	25(4.6%)	20(3.6%)	26(3.5%)	16(3.1%)
Missing	327	65	78	91	93
Previous abdominal surgery	509(22%)	102(20%)	130(24%)	170(24%)	107(22%)
Missing	395	88	89	108	110
Abdominal hernia	169(7.2%)	45(8.3%)	42(7.5%)	50(6.8%)	32(6.3%)
Missing	331	67	81	89	94
Preoperative incontinence. Change of pads
Not applicable	2,323(98%)	534(98%)	563(99%)	719(98%)	507(99%)
<1 times	12(0.5%)	4(0.7%)	0(0%)	7(1.0%)	1(0.2%)
approx 1 times	8(0.3%)	2(0.4%)	2(0.4%)	4(0.5%)	0(0%)
2–3 times	12(0.5%)	1(0.2%)	3(0.5%)	6(0.8%)	2(0.4%)
4–5 times	4(0.2%)	1(0.2%)	2(0.4%)	0(0%)	1(0.2%)
>5 times	2(<0.1%)	2(0.4%)	0(0%)	0(0%)	0(0%)
Missing	311	62	70	89	90
Erection sufficiently stiff for intercourse
No sexual activity	467(20%)	121(23%)	120(22%)	139(19%)	87(17%)
Never or rarely	61(2.6%)	18(3.4%)	8(1.4%)	24(3.3%)	11(2.2%)
<50% of times	98(4.2%)	22(4.1%)	27(4.8%)	33(4.6%)	16(3.2%)
approx 50% of times	121(5.2%)	24(4.5%)	29(5.2%)	45(6.3%)	23(4.6%)
>50% of times	289(13%)	74(14%)	69(12%)	86(12%)	60(12%)
Almost always or always	1,273(55%)	272(51%)	304(55%)	390(54%)	307(61%)
Missing	363	75	83	108	97
Preoperative PSA(ng/ml)	6 (4, 9)	6 (4, 9)	6 (4, 8)	6 (4, 9)	6 (5, 9)
Missing	9	1	0	0	8
Clinical T stage
T1	1,541(59%)	331(56%)	353(57%)	481(60%)	376(63%)
T2	993(38%)	244(41%)	254(41%)	299(37%)	196(33%)
T3	76(2.9%)	14(2.4%)	15(2.4%)	24(3.0%)	23(3.9%)
Missing	62	17	18	21	6
ASA
I	1,648(63%)	386(65%)	399(64%)	515(64%)	348(60%)
II	912(35%)	201(34%)	219(35%)	268(33%)	224(38%)
IV-V	54(2.1%)	9(1.5%)	9(1.4%)	24(3.0%)	12(2.1%)
Missing	58	10	13	18	17
Gleason Score on biopsy
1	1,349(51%)	290(48%)	312(49%)	429(52%)	318(53%)
2	914(34%)	228(38%)	227(36%)	259(32%)	200(34%)
3	241(9.1%)	55(9.1%)	59(9.2%)	76(9.3%)	51(8.6%)
>3	153(5.8%)	30(5.0%)	41(6.4%)	55(6.7%)	27(4.5%)
Missing	15	3	1	6	5
D’Amico classification (biopsy)
Low	770(29%)	154(26%)	175(28%)	245(30%)	196(33%)
Intermediate	1,649(63%)	404(67%)	403(64%)	494(61%)	348(58%)
High	219(8.3%)	44(7.3%)	53(8.4%)	70(8.7%)	52(8.7%)
Missing	34	4	9	16	5
Nervesparing approach
No neurovascular dissection	445(17%)	108(18%)	120(19%)	159(19%)	58(9.7%)
Uni/bi-lateral partial NS	397(15%)	120(20%)	106(17%)	110(13%)	61(10%)
Uni-lateral inter/intra NS	510(19%)	126(21%)	116(18%)	143(17%)	125(21%)
Bi-lateral NS, partially one side	510(19%)	103(17%)	138(22%)	157(19%)	112(19%)
Bi-lateral NS, one inter, one intra	550(21%)	121(20%)	109(17%)	184(22%)	136(23%)
Bi-lateral NS, both sides intrafascial	144(5.4%)	10(1.7%)	30(4.7%)	37(4.5%)	67(11%)
Bi-lateral NS, both sides interfascial	114(4.3%)	18(3.0%)	20(3.1%)	35(4.2%)	41(6.8%)
Missing	2	0	1	0	1
Lymph node dissection
Extended	227(8.5%)	40(6.6%)	56(8.8%)	68(8.3%)	63(10%)
Limited	85(3.2%)	11(1.8%)	25(3.9%)	33(4.0%)	16(2.7%)
No performed	2,346(88%)	553(92%)	553(87%)	719(88%)	521(87%)
Missing	14	2	6	5	1
Total length of cancer in prostate biopsy(mm)	8 (4, 16)	8 (4, 17)	8 (4, 15)	7 (4, 16)	8 (4, 15)
Missing	159	44	39	40	36
pT stage
T2	1,890(72%)	440(75%)	456(73%)	592(72%)	402(69%)
T3	710(27%)	145(25%)	167(27%)	221(27%)	177(31%)
T4	10(0.4%)	0(0%)	5(0.8%)	4(0.5%)	1(0.2%)
Missing	62	21	12	8	21
Prostate weight(gr)	42 (34, 53)	43 (35, 53)	42 (34, 53)	42 (34, 53)	42 (34, 53)
Missing	32	5	9	5	13

¹Median (IQR); n(%)

For the oncological outcome measure positive surgical margin, the learning curve was relatively flat, with an adjusted risk among the inexperienced surgeons (<75 operations) of 21% compared to 24% for the most experienced (>300 operations). Biochemical recurrence was stable across degrees of experience, with an adjusted risk of 11% for the inexperienced surgeons and 13% for the most experienced surgeons (Figure 2). The models had limited abilities to explain variability as demonstrated by the low values of R-square (6% and 11% for surgical margin and biochemical recurrence, respectively), and adding surgeons experience made no improvement (p-values 0.504 and 0.570), see Table S1). Adjusted and unadjusted predictions and raw proportions are presented in the Supplement (Table S2).

Urinary incontinence at 3 months was less if the operation was performed by the most experienced surgeons at 44% (>300 cases) compared with 54% among inexperienced surgeons (<75 cases) (Figure 3). At 24 months the adjusted risk for incontinence was 16% and 21% respectively for surgeons with experience above 300 cases compared with those with experience of less than 75 cases respectively. The models had limited abilities to explain variability (5%–6%), see Table S1.

Erectile dysfunction was less for those operated by experienced surgeons at all three time points, with an adjusted risk of 64% (>300 operations) versus 80% (< 75 operations) at 3 months, to 47% (>300 operations) versus 62% (< 75 operations) at 24 months follow-up. (Figure 4). The models had good abilities to explain variability (30%–38%) and experience had a significant improvement at 24 months (p-value 0.022), see Table S1.

In the supplementary analyses, adjusting also for nerve sparing surgery did not have any major impact on the predicted learning curve. Furthermore, variation in annual volume did not give rise to any pronounced modification of the learning curve.

Discussion

The analyses of the effect of surgeon experience in this prospectively followed, large, multicenter cohort of patients undergoing RALP showed that there was a learning curve regarding the functional outcomes, meaning that outcomes improved by increasing number of performed operations. For the oncological outcomes we did not observe such positive development. The learning curve for urinary continence was relatively short until a plateau was reached but for erectile function the upward slope was constant over the entire follow-up period. For urinary continence the effect of inexperience was attenuated after a longer period of follow-up, meaning that the negative effect of being operated on by an inexperienced surgeon was rather short-lived. The inter-surgeon variability is manifested in the width of the confidence intervals of the predicted learning curves. The relative contribution of the various sources to this variability was quantified in Nyberg et al ([5]) for the outcomes at 24 month follow-up. It was found that differences in experience explained 42%, 11% and 19% of the between-surgeon variability in incontinence, erectile dysfunction and recurrence, respectively. This is different from our approach, where we explore the impact of experience on the total (both within- and between-surgeon) variability.

In a recent report from a single center, a risk reduction with increasing experience was found for surgical margin but not for biochemical recurrence ([23]). A single center - single surgeon study has reported that risk of positive surgical margin decreased with increasing experience in RALP, but the surgeon had extensive experience of open radical prostatectomy before turning to robot technique ([24]). Thus, external validity of those results is low, albeit a large cohort was studied. In a study on learning curve for open radical prostatectomy a poor concordance between a surgical margin status and recurrence rates was found ([1]). The authors suggested that the improvement in the two different oncological outcomes was driven by different mechanisms. Since the outcome regarding surgical margin status can be evaluated right after the operation there is an opportunity for immediate feedback, beneficial for learning. Biochemical recurrence may occur many years later, which gives less opportunity for feedback. It seems reasonable that there could be differences in learning curve between open and robot radical prostatectomy. Preferably a comparison of learning curve for open and robot assisted techniques in our study should have been made as presented by others [24], but this was not possible as all surgeons practising open technique within LAPPRO were already experienced at the start of inclusion. At that time (2008–2011) almost no surgeons were in training in the open prostatectomy technique.

Our results of functional outcomes are in line with an earlier report of learning curve for one surgeon turning from open to robotic technique [24]. The learning curve for preservation of erectile function was continuous without a clear plateau. One contributing factor might be that increasing experience not only adds dexterity but also insight into the planning of a procedure where many factors should be taken into account. Even though the primary aim of radical prostatectomy is cancer cure, analyses of functional outcomes are of importance in view of their negative impact on the patient’s quality of life. ([4]).

We observed that patients with a university education to a larger extent were operated on by experienced surgeons. One explanation could be related to demography; patients with higher educational level live in urban areas nearby tertiary academic hospitals, where experienced surgeons work. Others have reported on worse outcomes of treatment for colorectal cancer for the socio-economically deprived in both the Netherlands and England[25], [26].

The strengths of this study include the size of the cohort, the prospectively collected data, and the multicenter design. Patients from three tertiary referral (university) hospitals were included as well as four “county” hospitals giving high external validity. Since functional outcomes were patient reported to a third party, an interviewer effect was avoided [27], [28]. Limitations include that these analyses were secondary endpoints in the trial, and thus study sample size was not optimized for the current objective. Since the distribution of the surgeons total experience was skewed with a majority having limited experience, all surgeons will contribute to the characterization of the beginning of the learning curve (the left-hand part) but few will contribute to the right-hand part, making this part determined with greater uncertainty.

Our results suggest that the primary goal of prostate cancer surgery, cancer cure, is relatively unaffected by the surgeon’s experience but that the quality of life-affecting side effects after surgery decreased with experience.

Supplementary Material

Supplementar_analyses_REVISED

Supplementary analyses

Table_S1

Table S1. Comparison of model fit for models with and without surgeon experience included.

Table_S2

Table S2. Adjusted and unadjusted predictions and raw proportions averaged across categories of experience. For predictions 95% confidence intervals are presented within parentheses.

Biographies

David Bock, PhD, ass. Prof. is senior lecturer in biostatistics at Göteborg University, Göteborg, Sweden

Martin Nyberg, MD, PhD, is urologist and researcher at Lund University Cancer Center (LUCC), Lund, Sweden

Anna Lantz, MD, PhD, ass. Prof. is urologist and researcher at Karolinska Institute, Stockholm, Sweden

Sigrid V Carlsson, MD, PhD, MPH, is an Assistant Attending Epidemiologist at Memorial Sloan Kettering Cancer Center (MSKCC), New York, USA

Daniel D. Sjoberg is Senior Biostatistician at MSKCC and DrPH candidate in Biostatistics at Columbia University, New York, USA

Stefan Carlsson, MD, PhD, ass. Prof. is urologist and researcher at Karolinska Institute, Stockholm, Sweden

Johan Stranne, MD, PhD, ass. Prof. is urologist and researcher at Sahlgrenska University hospital, Göteborg, Sweden

Gunnar Steineck, MD, PhD, is professor at Sahlgrenska Academy, Göteborg University, Göteborg, Sweden

Peter Wiklund MD, PhD, is professor of Urology, Department of Urology at the Icahn School of Medicine at Mount Sinai, New York, USA

Eva Haglind, MD, PhD, is professor of Surgery at Sahlgrenska University Hospital, and pricipal investigator of LAPPRO

Anders Bjartell, MD, PhD, is professor at Division of Translational Cancer Research, Lund University