Abstract
Background:
To utilize a semi-competing risk model to predict perioperative and oncologic outcomes after radical cystectomy and to compare the findings with the univariate Cox regression model.
Methods:
We reviewed the Institutional Review Board approved database of radical cystectomy of 316 patients who had undergone robot-assisted radical cystectomy (RARC) or open radical cystectomy between 2006 and 2016. Demographic data, perioperative outcomes, complications, metastasis, and survival were analyzed. The Bayesian variable selection method was utilized to obtain models for each hazard function in the semi-competing risks.
Results:
Of 316 patients treated, 48% and 18% experienced any or major complication respectively within 30 days. Intracorporeal RARC was associated with decreased metastasis risk. Extracorporeal RARC was associated with marginally decreased risks of overall complications or major complications. Patients with advanced cancer had an increased risk of metastasis, death after metastasis and death after complication. Positive nodes were associated with an increased risk of death without overall or major complications and increased risk of death after metastasis occurs. When a serious complication was taken into account there was no significant difference in mortality, irrespective of disease stage.
Conclusions:
A semi-competing risk model provides relatively more accurate information in comparison to Cox regression analysis in predicting risk factors for complications and metastasis in patients undergoing radical cystectomy.
Keywords: bladder cancer, radical cystectomy, robot-assisted radical cystectomy, intracorporeal, extracorporeal, semi-competing risk model, Bayesian statistics
Introduction
Bladder cancer is the fourth most common cancer and eighth leading cause of cancer death among men in the United States. With 79,000 new cases and almost 17,000 deaths, it is the most expensive and morbid of the urologic malignancies to treat.1 For patients who have disease invading the muscle wall of the bladder, radical cystectomy is the gold standard of care. However, radical cystectomy is the most morbid and complex urologic procedure performed today. Indeed the procedure can have operative times of up to 8 h, with complication rates as high as 58–77% within the first 30 days, and readmission rates in 27% of patients.2–5 The cause of such outcomes is multifactorial, in part due to the population affected by the disease and in part due to the disease process and treatments themselves. Therefore, above all else, it is important to understand the factors that lead to readmission and put patients at risk for these complications, and then how exactly these influence overall survival.
Studies have evaluated rates of complications, recurrence, and metastases in patients following radical cystectomy. In many of these studies, those factors that have continually proven important in predicting risk of complications include pathologic stage, operative time, estimated intraoperative blood loss, transfusion requirements, frailty indices, age, and sex.6–8 Rates of recurrence and future metastases are often determined by stage, lymph node involvement, and margin status.9,10 Furthermore, many of these studies have gone on to determine how these factors affect overall survival.
The majority of published literature utilizes standard survival models like Cox regression for overall survival, risk of distant metastasis, and complications. This model was originally proposed by Cox and colleagues in 1972 and has since been accepted as the mainstay for statistical analysis.11 Various problems arise when dealing with the more complex event causation structures seen in medical analyses. Many papers in the current literature violate a basic assumption of the Cox model by assuming that death is an independent censoring time for distant metastasis. This is incorrect because the patient may have had a distant metastasis but died prior to returning to the hospital for diagnosis of such an event.
After patients are treated, they can experience metastasis or a complication from treatment and then die, but they cannot experience one of these nonterminal events after death. This scenario represents a semi-competing risk structure, where the risk of death competes with the risk of a nonterminal event but not vice versa. This semi-competing risk structure more accurately models patients undergoing radical cystectomy for urothelial cancer where complications and recurrences affect overall survival.
Therefore, we use the model introduced by Lee and colleagues and used in the study by He and colleagues, as well as the subsequent variable selection procedure of Chapple and colleagues to determine what patient covariates impact the risk of surgical complications and metastasis, and the risk of death with or without one of these three nonterminal events.12–14 This model contains patient-specific frailty terms in the hazard of nonterminal and terminal events, which allows us to gain strength among the two time to event dependent variables and remove the effects of extremely weak or robust patients from the results for covariate significance.
By using our radical cystectomy database, we have performed simultaneous analyses utilizing both the semi-competing risk model along with a standard survival model to determine factors that affect the complication rate and overall survival, specifically stratifying the extracorporeal and intracorporeal robot-assisted radical cystectomy (RARC) for comparison with open radical cystectomy (ORC).15 These two models serve to highlight the fundamental differences in the statistical analyses, and demonstrate how a semi-competing risk model with a new insight brings us closer to understanding the effects of our therapeutic approaches.
Patients and methods
Patients
We performed a retrospective review (Institutional Review Board approved) of our prospectively maintained radical cystectomy database from 2006 to 2016 at Wake Forest. Charts were reviewed for age, body mass index (BMI), estimated blood loss (EBL), operating time, total lymph nodes, sex, American Society of Anesthesia score (ASA), diversion type (intracorporeal versus extracorporeal), intraoperative blood transfusion, advanced pathologic stage (⩾T3 diseases), lymphovascular invasion (LVI), positive lymph node status, positive surgical margins, carcinoma in situ (CIS), upper tract involvement, and neoadjuvant chemotherapy. Additionally, we analyzed the surgical effects of extracorporeal and intracorporeal RARC compared with ORC. Mortality information was acquired via chart review and confirmed using the Social Security Death Index.
Statistical analyses
We used the Bayesian variable selection method of Chapple and colleagues (2017) to determine what patient covariates were important predictors of complication, metastasis, and the risk of death with or without one of these nonterminal events. We compared our results with univariate analyses with the standard Cox model and performed using backward selection to remove nonsignificant covariates.
Prior to performing statistical analyses, we examined the patient characteristics for the treatment groups and performed Fisher exact tests for binary covariates and analysis of variance for continuous covariates. For the semi-competing risks model, we used the package SCRSELECT found on CRAN with the shrinkage parameter c = 20 and β parameters (0.4, 1.6). This package performs Bayesian variable selection, obtaining a final model for each hazard function in the semi-competing risks structure and giving final results for interpretation of the covariates. Patients were assigned to available clinicians as they entered the hospital and treated according to their expertise, essentially making this assignment random. Therefore, we did not include a propensity score as a covariate in the model.
Results
Patient characteristics
Of the 316 patients treated, 48% experienced any complication (Clavien 1–5) within 30 days following surgery with 18% experiencing a major complication (Clavien 3–5). Seventy-five percent of patients experienced metastasis following surgery, and 53% of patients died while being followed. 61% of patients died after experiencing metastasis compared with 28% for patients who did not have metastasis.
Patients in the three surgical groups had much closer rates of complications: 57%, 40%, and 45% for the ORC, extracorporeal RARC (eRARC), and intracorporeal RARC (iRARC), respectively. The means of the patient variables broken down by treatment along with the p values associated with tests for all group differences are displayed in Table 1.
Table 1.
Patient clinicopathologic data.
| Variable | ORC | Extracorporeal RARC | Intracorporeal RARC | p value |
|---|---|---|---|---|
| Age | 66.68 | 67.6 | 65.74 | 0.52 |
| Sex | 0.76 | |||
| Male | 107 | 98 | 28 | |
| Female | 42 | 31 | 10 | |
| Race | 0.35 | |||
| White | 139 | 117 | 33 | |
| Other | 10 | 12 | 5 | |
| BMI (kg/m2) | 27.05 | 26.89 | 27.34 | 0.90 |
| ASA score | 0.14 | |||
| Poor score | 116 | 105 | 35 | |
| Better score | 33 | 24 | 3 | |
| Operating room time | 388.5 | 443.1 | 411.6 | <0.01 |
| Estimated blood loss | 926.8 | 375.6 | 231.8 | <0.01 |
| Blood transfusion | <0.01 | |||
| Yes | 81 | 114 | 0 | |
| No | 68 | 15 | 38 | |
| Advanced stage | 0.16 | |||
| Yes | 76 | 51 | 17 | |
| No | 73 | 78 | 21 | |
| LVI | 0.56 | |||
| Yes | 53 | 46 | 10 | |
| No | 96 | 83 | 28 | |
| Positive nodes | 0.69 | |||
| Yes | 39 | 39 | 9 | |
| No | 110 | 90 | 29 | |
| Positive margin | 0.28 | |||
| Yes | 20 | 20 | 2 | |
| No | 129 | 109 | 36 | |
| CIS | 0.31 | |||
| Yes | 48 | 44 | 8 | |
| No | 101 | 85 | 30 | |
| Upper tract invasion | 0.78 | |||
| Yes | 22 | 16 | 4 | |
| No | 127 | 113 | 34 | |
| Neoadjuvant chemotherapy | 0.90 | |||
| Yes | 17 | 13 | 3 | |
| No | 132 | 116 | 35 | |
| Total nodes | 17.37 | 20.28 | 19.39 | 0.21 |
ASA, American Society of Anesthesiologists; BMI, body mass index; CIS, carcinoma in situ; LVI, lymphovascular invasion; ORC, open radical cystectomy; RARC, robot-assisted radical cystectomy.
Between the three treatment groups, most of the covariates are distributed similarly except in operating room (OR) time, EBL, and blood transfusion. No patients who received an iRARC had a blood transfusion, and those patients had significantly less EBL than other patients. Patients with an eRARC had significantly longer OR times than patients in the other surgical groups. Of the 149 ORCs, 129 were performed with an ileal conduit, 7 with an Indiana pouch, 12 with an orthotopic neobladder, and 1with ureterosigmoidostomy. Of the 129 eRARCs, 123 were performed with an ileal conduit, 3 with an orthotopic neobladder, and 3 with cutaneous ureterostomy. Of the 38 iRARCs, 37 were performed with an ileal conduit, and 1with an orthotopic neobladder.
Metastasis analysis
We analyzed the effects of patient covariates and surgical techniques on the risk of metastasis and the risk of death with or without metastasis and the results are presented in Table 2. In this table, hazard ratio (HR) is the median posterior HR for each covariate with CI indicating the 95% credible interval for this quantity. Covariates with credible intervals that do not contain 1 are considered to be significant for effecting the hazard, with credible bounds of 0.99, 1, or 1.01 considered to be marginally significant. p(ß > 0) denotes the posterior probability that a covariate is hazardous with large values of p(ß > 0) along with a significant covariate indicating that the variable significantly increases that event risk and small values indicating a decreased risk. This differs from the usual p value seen in univariate analyses, which is also discussed. In general, a covariate included in a specific hazard is significant for that event if the credible interval does not contain 1, which generally corresponds to values of p below 0.02 and above 0.98.
Table 2.
Metastasis analysis, final model inference for each hazard.
| Variable | H1: metastasis |
H2: death with metastasis |
H3: death after
metastasis |
|||
|---|---|---|---|---|---|---|
| HR (CI) | p(ß > 0) | HR (CI) | p(ß > 0) | HR (CI) | p(ß > 0) | |
| Age | – | – | – | – | – | – |
| Sex (male/female) | – | – | 1.17 (0.88–1.56) | 0.86 | – | – |
| Race (white/other) | – | – | 1.14 (0.87–1.54) | 0.81 | – | – |
| BMI | – | – | – | – | 0.82 (0.70–0.96) | <0.01 |
| ASA score | – | – | 1.14 (0.87–1.51) | 0.82 | 1.19 (1.03–1.39) | >0.99 |
| Operating room time | – | – | – | – | – | – |
| Estimated blood loss | – | – | – | – | – | – |
| Blood transfusion (yes/no) | – | – | – | – | – | – |
| Advanced stage (yes/no) | 1.26 (1.12–1.42) | >0.99 | – | – | 1.39 (1.18–1.66) | >0.99 |
| LVI (yes/no) | – | – | – | – | – | – |
| Positive nodes (yes/no) | – | – | – | – | 1.22 (1.04–1.43) | >0.99 |
| Positive margin (yes/no) | – | – | 1.37 (1.06–1.74) | >0.99 | – | – |
| CIS (yes/no) | – | – | – | – | – | – |
| Upper tract invasion (yes/no) | – | – | – | – | – | – |
| Neoadjuvant chemotherapy (yes/no) | – | – | – | – | – | – |
| Total lymph nodes | – | – | – | – | – | – |
| Extracorporeal RARC | – | – | – | – | – | – |
| Intracorporeal RARC | 0.87 (0.74–0.99) | 0.02 | – | – | – | – |
ASA, American Society of Anesthesiologists; BMI, body mass index; CI, credible interval; CIS, carcinoma in situ; H, hazard; HR, hazard ratio; LVI, lymphovascular invasion; p(ß > 0), posterior probability that covariate increases the event risk; RARC, robot-assisted radical cystectomy.
For the semi-competing risks analysis featuring time to metastasis as a nonterminal event, the final model for the risk of metastasis only contained advanced stage, shown to significantly increase metastatic risk (HR 1.26, CI 1.12–1.42, p > 0.99), and iRARC, which significantly decreased metastasis risk (HR 0.87, CI 0.74–.99, p = 0.02). Sex, race, ASA score, and positive surgical margins were included in the risk of death without metastasis, with positive margins associated with an increased risk of death (HR 1.37, CI 1.06–1.74, p > 0.99). After a patient’s cancer metastasized, the only important variables were increasing BMI (HR 0.82, CI 0.70–0.96, p < 0.01), significantly decreasing risk of death after metastasis, and increasing ASA score (HR 1.19, CI 1.03–1.39), advanced stage cancers (HR 1.39, CI 1.18–1.66) and positive lymph nodes (HR 1.22, CI 1.04–1.43), all of which significantly increased the risk of death after metastasis (p > 0.99).
In the Cox regression for time to metastasis, we cannot assume that death is a noninformative censoring time and instead can only analyze the time to death or metastasis, whichever comes first, the results of which are seen in Supplemental Table 1. In the final model, for the risk of metastasis or death, positive lymph nodes and positive surgical margins were additionally included as covariates. One of the strengths of the semi-competing risks method is that we are able to draw inference directly on the risk of metastasis, showing that patients undergoing iRARC had a reduced risk. The Cox model for time to death without metastasis does not contain race and ASA score as seen previously. For the risk of death after metastasis, the Cox model only considers the patient data of those who had metastasis, including OR time and LVI in addition to those in the semi-competing risks model, both of which are significant (p = 0.03 and 0.02, respectively), which may be a result of ignoring patient frailties.
All complication analysis
The final model for any complication (Clavien 1–5) contained EBL, advanced stage, upper tract involvement, neoadjuvant chemotherapy, and eRARC as important covariates for the risk of a complication Table 3. Increased EBL significantly increased the risk of complications (HR 1.37, CI 1.18–1.55, p = 1.00), while advanced cancer (HR 0.83, CI 0.72–0.97, p < 0.01) or upper tract involvement (HR 0.84, CI 0.72–0.97, p = 0.01) had a decreased risk of complications after surgery. Patients who received neoadjuvant chemotherapy or eRARC had marginally reduced risks of postsurgical complications.
Table 3.
All complication analysis, final model inference for each hazard.
| Variable | H1: complication |
H2: death without
complication |
H3: death after
complication |
|||
|---|---|---|---|---|---|---|
| HR (CI) | p(ß > 0) | HR (CI) | p(ß > 0) | HR (CI) | p(ß > 0) | |
| Age | – | – | – | – | 1.09 (0.90–1.33) | 0.81 |
| Sex (male/female) | – | – | – | – | 0.90 (0.75–1.10) | 0.14 |
| Race (white/other) | – | – | – | – | – | – |
| BMI | – | – | – | – | 0.90 (0.73–1.12) | 0.17 |
| ASA score | – | – | – | – | 1.12 (0.93–1.37) | 0.87 |
| Operating room time | – | – | – | – | 0.85 (0.69–1.05) | 0.07 |
| Estimated blood loss | 1.37 (1.18–1.55) | 1.00 | – | – | – | – |
| Blood transfusion (yes/no) | – | – | – | – | 1.08 (0.88–1.33) | 0.77 |
| Advanced stage (yes/no) | 0.83 (0.72–0.97) | <0.01 | 1.40 (1.13–1.73) | >0.99 | 1.42 (1.13–1.76) | >0.99 |
| LVI (yes/no) | – | – | – | – | 1.22 (0.98–1.49) | 0.97 |
| Positive nodes (yes/no) | – | – | 1.28 (1.06–1.53) | >0.99 | – | – |
| Positive Margin (Yes/No) | – | – | 1.16 (0.95–1.38) | 0.93 | – | – |
| CIS (yes/no) | – | – | – | – | – | – |
| Upper tract invasion (yes/no) | 0.84 (0.72–0.97) | 0.01 | – | – | 1.11 (0.90–1.35) | 0.84 |
| Neoadjuvant chemotherapy (yes/no) | 0.87 (0.74–1.02) | 0.04 | – | – | 0.97 (0.77–1.19) | 0.39 |
| Total lymph nodes | – | – | – | – | 1.06 (0.86–1.29) | 0.73 |
| Extracorporeal RARC | 0.87 (0.74–1.02) | 0.04 | – | – | 1.18 (0.93–1.50) | 0.91 |
| Intracorporeal RARC | – | – | – | – | 0.91 (0.71–1.16) | 0.24 |
ASA, American Society of Anesthesiologists; BMI, body mass index; CI, credible interval; CIS, carcinoma in situ; H, hazard; HR, hazard ratio; LVI, lymphovascular invasion; p(ß > 0), posterior probability that covariate increases the event risk; RARC, robot-assisted radical cystectomy.
The model for the risk of death without any complications only contained advanced stage (HR 1.40, CI 1.13–1.73), positive nodes (HR 1.28, CI 1.06–1.53), and positive margins as covariates, with the first two associated with an increased risk of death without a complication (p > 0.99). After a patient had a complication, most covariates were included as predictors of death but only advanced stage was a significant covariate (HR 1.42, CI 1.13–1.76, p > 0.99).
The Cox model, seen in Supplemental Table 2 ignores the semi-competing risks structure and as a result only contains EBL as an important predictor of complication risk. We do not see eRARC or neoadjuvant chemotherapy as being marginally significant. Advanced stage and upper tract involvement were also not included as covariates, both of which were associated with a significantly decreased risk of complication in the semi-competing risks model.
The Cox model for the risk of death without a complication did not contain positive margins and had fewer covariates included in the risk of death after complication. OR time was associated with a decreased risk of death after a complication (p = 0.05), while advanced stage (HR 1.65, CI 1.27–2.14, p < 0.01), LVI (HR 1.35, CI 1.07–1.71, p = 0.01), and iRARC (HR 1.38, CI 1.09–1.74, p < 0.01) were associated with an increase in this risk.
Major complication analysis
When we analyze the semi-competing risks structure consisting of only the major complications (Clavien 3–5), with the results presented in Table 4, we see six covariates included as predictors for the risk of major complications, with only advanced stage (HR 0.81, CI 0.65–0.99, p = 0.02) significantly reducing this risk. Without a serious complication the death hazard model contains only the covariates BMI (HR 0.85, CI 0.73–1.00, p = 0.02), LVI (HR 1.45, CI 1.20–1.76, p > 0.99), advanced stage (HR 1.20, CI 0.98–1.46, p = 0.96), and positive lymph nodes (HR 1.20, CI 1.01–1.41, p = 0.98)
Table 4.
Major complication analysis, final model inference for each hazard.
| Variable | H1: major complication |
H2: death without major
complication |
H3: death after major
complication |
|||
|---|---|---|---|---|---|---|
| HR (CI) | p(ß > 0) | HR (CI) | p(ß > 0) | HR (CI) | p(ß > 0) | |
| Age | – | – | – | – | 1.12 (0.87–1.45) | 0.81 |
| Sex (male/female) | 1.19 (0.96–1.49) | 0.95 | – | – | – | – |
| Race (white/other) | – | – | – | – | – | – |
| BMI | – | – | 0.85 (0.73–1.00) | 0.03 | – | – |
| ASA score | 1.14 (0.92–1.43) | 0.88 | – | – | 1.35 (1.04–1.78) | 0.99 |
| Operating room time | – | – | – | – | 0.87 (0.67–1.12) | 0.15 |
| Estimated blood loss | 1.17 (0.95–1.41) | 0.93 | – | – | 1.12 (0.87–1.41) | 0.82 |
| Blood transfusion (yes/no) | – | – | – | – | – | – |
| Advanced stage (yes/no) | 0.81 (0.65–0.99) | 0.02 | 1.45 (1.20–1.76) | >0.99 | 1.27 (0.98–1.63) | 0.96 |
| LVI (yes/no) | – | – | 1.20 (0.98–1.46) | 0.96 | – | – |
| Positive nodes (yes/no) | – | – | 1.20 (1.01–1.41) | 0.98 | – | – |
| Positive margin (yes/no) | – | – | – | – | – | – |
| CIS (yes/no) | – | – | – | – | 1.12 (0.86–1.43) | 0.81 |
| Upper tract invasion (yes/no) | – | – | – | – | – | – |
| Neoadjuvant chemotherapy (yes/no) | 0.86 (0.68–1.07) | 0.09 | – | – | – | – |
| Total lymph nodes | – | – | – | – | – | – |
| Extracorporeal RARC | 0.83 (0.66–1.04) | 0.05 | – | – | – | – |
| Intracorporeal RARC | – | – | – | – | 0.84 (0.63–1.09) | 0.10 |
ASA, American Society of Anesthesiologists; BMI, body mass index; CI, credible interval; CIS, carcinoma in situ; H, hazard; HR, hazard ratio; LVI, lymphovascular invasion; p(ß > 0), posterior probability that covariate increases the event risk; RARC, robot-assisted radical cystectomy.
The covariates age, ASA score, OR time, EBBL, advanced stage cancer, CIS, and iRARC were included as important predictors for the risk of death after a major complication, with only increased ASA score (HR 1.35, CI 1.04–1.78, p = 0.99) being associated with a significant increase in this risk.
Comparing this model to the univariate Cox model, as seen in Supplemental Table 3 we see that the univariate model for severe complication drops the covariates ASA score, advanced cancer, and eRARC. EBL (HR 1.34, CI 1.10–1.64, p < 0.01) and male sex (HR 1.50, CI 1.07–2.11, p = 0.02) both significantly increase the risk of major complications, a result not seen in the semi-competing risks model. The univariate Cox model for the risk of death without a major complication contains the same set of covariates as the semi-competing risks model, except in this case BMI significantly decreases the risk of death without a major complication (HR 0.79, CI 0.66–0.95, p = 0.01). The Cox model for the risk of death after a major complication, which only uses the data of 53 patients, contains only ASA score, advanced stage, and iRARC, where advanced cancer (HR 1.49, CI 1.04–2.13, p = 0.04) and ASA (HR 2.31, CI 1.05–5.08, p = 0.03) significantly increased this risk, and iRARC (HR 0.52, CI 0.27–0.99, p = 0.05) was associated with a decreased risk of death.
Discussion
We applied novel variable selection methodology in a semi-competing risks setting to properly analyze the covariate effects on the risks of any complication, major complication or metastasis, and the risk of death with or without one of these nonterminal events. The association between iRARC and decreased metastasis risk is particularly interesting. Recently, Bochner and colleagues found that there was a nonstatistically significant increase in metastatic sites for those undergoing ORC compared with RARC.16 In addition, they found a greater number of local or abdominal sites in the RARC-treated patients. It has been suggested that in theory the RARC may put patients at greater risk of port site seeding and peritoneal carcinomatosis. However, Nguyen and colleagues demonstrated that there was no difference when comparing open with eRARC.17 No study has specifically compared intracorporeal and extracorporeal diversions in an RARC series. However, a systematic analysis that evaluated the literature did not find a difference in oncologic outcomes to date.18 The discrepancy may in fact be due to the inappropriate model in use. It should be noted that while positive margins were not significantly different between the groups, the iRARC group had a positive margin rate of 5% versus 13% and 16% for ORC and eRARC, respectively.
We found that eRARC was associated with a marginally decreased risk of any complication or serious complication, and that patients with iRARC had a significantly decreased risk of metastasis. The differences in complication rates for RARC performed extracorporeally and intracorporeally contrast those found by Ahmed and colleagues, who did not demonstrate a significant difference between the diversions, despite a trend favoring extracorporeal diversion at 90 days.19 The differences observed could be the result of the differing analysis or of looking at 90 days versus 30 days. Both intracorporeal and extracorporeal diversions were shown to be important, but not significant, predictors of risk of death after a complication, while only intracorporeal diversion was included in the final model for the hazard of death after serious complication. This warrants further study of the two diversion techniques, as an increased patient sample size for intracorporeal diversion may lead to improved assessment of this procedure’s efficacy.
We saw that patients with advanced cancer had an increased risk of metastasis, death after metastasis, and death after complication. Positive nodes were also associated with an increased risk of death without any complication or major complication, and an increased risk of death after metastasis occurs. This is in line with other studies looking at the morbidity and mortality of radical cystectomy in this patient population. It is interesting to note that after a serious complication, there is no significant difference in mortality, irrespective of the stage of the disease. Only by way of a semi-competing risk model could we have demonstrated this finding. It implies that even for a patient who has localized disease that is presumably curable with cystectomy, a resulting serious complication negates any survival advantage that patient might have over someone who has advanced disease. Interestingly, patients with advanced cancer had significantly reduced risk of complication or serious complication. Kauffman and colleagues did not find a difference in rate of complications when looking at cystectomies for tumors that were <pT2 versus ⩾pT2 and those that were ⩾pT4 versus <pT4.20
This study has a few limitations. First, it is a single-institution database created at an academic center that may contain biases in terms of patient selection, operating standards, surgical techniques, postoperative management, and reporting of complications. Thus, it may not be representative of the larger scope of practice. Furthermore, the data were evaluated in a retrospective manner and subject to the inherent biases related to this approach. In addition, longer-term follow up of our metastatic analysis would better delineate difference between groups. However, due to this being a tertiary referral center, follow up is often referred back to local physicians. Nevertheless, we feel that this analysis demonstrates the importance of choosing the correct statistical model that accurately represents the semi-competing risk structure of those patients undergoing a radical cystectomy.
Overall, the semi-competing risks model contains more covariates in each hazard, but with fewer significant covariates in each analysis. These differences are due to several factors, including using all patient information for the semi-competing risks model, the inclusion of patient-specific frailty terms which minimizes covariate effects that are driven by a few weak or robust patients. In addition, the baseline hazards are fit and estimated in the Bayesian semi-competing risks approach while they are not in the Cox model. Furthermore, the Bayesian variable selection algorithm differs from the usual backward elimination approach in the Cox model both in the stepwise nature and the fact that the procedure is done for all three hazards simultaneously. While some similarities exist between the two methodologies used, when a semi-competing risk structure is of interest with some nonterminal and terminal event, these types of models should be used to avoid loss of patient information or to improperly consider death to be a noninformative censoring event or to look at time to either the nonterminal or terminal event, which does not give information on each event separately.
Supplemental Material
Supplemental material, Table_5,_6_and_7 for Semi-competing risk model to predict perioperative and oncologic outcomes after radical cystectomy by Taylor Peak, Andrew Chapple, Grayson Coon and Ashok Hemal in Therapeutic Advances in Urology
Footnotes
Funding: Andrew Chapple was partially supported by the NIH grant 5T32-CA096520-07.
Conflict of interest statement: The authors declare that there is no conflict of interest.
Supplemental Material: Supplementary material for this article is available online.
ORCID iD: Taylor Peak 
https://orcid.org/0000-0002-2984-232X
Contributor Information
Taylor Peak, Urology, Wake Forest Baptist Medical Center, Winston-Salem, NC, USA.
Andrew Chapple, Statistics, Rice University Wiess School of Natural Sciences, Houston, TX, USA.
Grayson Coon, Urology, Wake Forest Baptist Medical Center, Winston-Salem, NC, USA.
Ashok Hemal, Department of Urology, Wake Forest Baptist Medical Center, 1 Medical Center Blvd, Winston-Salem, NC, USA.
References
- 1. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2017. CA Cancer J Clin 2017; 67: 7–30. [DOI] [PubMed] [Google Scholar]
- 2. Djaladat H, Katebian B, Bazargani ST, et al. 90-Day complication rate in patients undergoing radical cystectomy with enhanced recovery protocol: a prospective cohort study. World J Urol 2017; 35: 907–911. [DOI] [PubMed] [Google Scholar]
- 3. Shabsigh A, Korets R, Vora KC, et al. Defining early morbidity of radical cystectomy for patients with bladder cancer using a standardized reporting methodology. Eur Urol 2009; 55: 164–174. [DOI] [PubMed] [Google Scholar]
- 4. Yuh B, Torrey RR, Ruel NH, et al. Intermediate-term oncologic outcomes of robot-assisted radical cystectomy for urothelial carcinoma. J Endourol 2014; 28: 939–945. [DOI] [PubMed] [Google Scholar]
- 5. Kader AK, Richards KA, Krane LS, et al. Robot-assisted laparoscopic vs open radical cystectomy: comparison of complications and perioperative oncological outcomes in 200 patients. BJU Int 2013; 112: E290–E294. [DOI] [PubMed] [Google Scholar]
- 6. Hollenbeck BK, Miller DC, Taub D, et al. Identifying risk factors for potentially avoidable complications following radical cystectomy. J Urol 2005; 174: 1231–1237; discussion 7. [DOI] [PubMed] [Google Scholar]
- 7. Chang SS, Cookson MS, Baumgartner RG, et al. Analysis of early complications after radical cystectomy: results of a collaborative care pathway. J Urol 2002; 167: 2012–2016. [PubMed] [Google Scholar]
- 8. Bostrom PJ, Kossi J, Laato M, et al. Risk factors for mortality and morbidity related to radical cystectomy. BJU Int 2009; 103: 191–196. [DOI] [PubMed] [Google Scholar]
- 9. Psutka SP, Carrasco A, Schmit GD, et al. Sarcopenia in patients with bladder cancer undergoing radical cystectomy: impact on cancer-specific and all-cause mortality. Cancer 2014; 120: 2910–2918. [DOI] [PubMed] [Google Scholar]
- 10. Stein JP, Lieskovsky G, Cote R, et al. Radical cystectomy in the treatment of invasive bladder cancer: long-term results in 1,054 patients. J Clin Oncol 2001; 19: 666–675. [DOI] [PubMed] [Google Scholar]
- 11. Cox D. Regression models and life tables. J R Stat Soc B 1972; 34: 187–220. [Google Scholar]
- 12. Lee KH, Haneuse S, Schrag D, et al. Bayesian semi-parametric analysis of semi-competing risks data: investigating hospital readmission after a pancreatic cancer diagnosis. J R Stat Soc Ser C Appl Stat 2015; 64: 253–273. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. He L, Chapple A, Liao Z, et al. Bayesian regression analyses of radiation modality effects on pericardial and pleural effusion and survival in esophageal cancer. Radiother Oncol 2016; 121: 70–74. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Chapple AG, Vannucci M, Thall PF, et al. Bayesian variable selection for a semi-competing risks model with three hazard functions. Comput Stat Data Anal 2017; 112: 170–185. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Peak TC, Hemal AK. Anatomic. Robot-assisted radical cystectomy. In Hemal A, Menon M, eds. Robotics in Genitourinary Surgery. 2nd ed London (UK): Springer; 2018. pp. 715–732 (In press) [Google Scholar]
- 16. Bochner BH, Dalbagni G, Marzouk KH, et al. Randomized trial comparing open radical cystectomy and robot-assisted laparoscopic radical cystectomy: oncologic outcomes. Eur Urol 2018. (In press) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Nguyen DP, Al Hussein Al, Awamlh B, O’Malley P, et al. Factors impacting the occurrence of local, distant and atypical recurrences after robot-assisted radical cystectomy: a detailed analysis of 310 patients. J Urol 2016; 196: 1390–1396. [DOI] [PubMed] [Google Scholar]
- 18. Yuh B, Wilson T, Bochner B, et al. Systematic review and cumulative analysis of oncologic and functional outcomes after robot-assisted radical cystectomy. Eur Urol 2015; 67: 402–422. [DOI] [PubMed] [Google Scholar]
- 19. Ahmed K, Khan SA, Hayn MH, et al. Analysis of intracorporeal compared with extracorporeal urinary diversion after robot-assisted radical cystectomy: results from the International Robotic Cystectomy Consortium. Eur Urol 2014; 65: 340–347. [DOI] [PubMed] [Google Scholar]
- 20. Kauffman EC, Ng CK, Lee MM, et al. Critical analysis of complications after robotic-assisted radical cystectomy with identification of preoperative and operative risk factors. BJU Int 2010; 105: 520–527. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplemental material, Table_5,_6_and_7 for Semi-competing risk model to predict perioperative and oncologic outcomes after radical cystectomy by Taylor Peak, Andrew Chapple, Grayson Coon and Ashok Hemal in Therapeutic Advances in Urology