- Email: [email protected]
Statin use and kidney cancer outcomes: A propensity score analysis
Urologic Oncology: Seminars and Original Investigations ] (2016) ∎∎∎–∎∎∎
Original article
Statin use and kidney cancer outcomes: A propensity score analysis Madhur Nayan, M.D., C.M.a,b, Antonio Finelli, M.D., M.Sc.a,b, Michael A.S. Jewett, M.D.a,b, David N. Juurlink, M.D., Ph.D.c, Peter C. Austin, Ph.D.d,e,f, Girish S. Kulkarni, M.D., Ph.D.a,b, Robert J. Hamilton, M.D., M.P.H.a,b,* b
a Department of Surgery, Princess Margaret Cancer Centre, University Health Network, University of Toronto, Toronto, Canada Department of Surgical Oncology, Princess Margaret Cancer Centre, University Health Network, University of Toronto, Toronto, Canada c Department of Internal Medicine, Sunnybrook Health Sciences Centre, University of Toronto, Toronto, Ontario, Canada d Institute for Clinical Evaluative Sciences, Toronto, Canada e Institute of Health Management, Policy and Evaluation, University of Toronto, Toronto, Canada f Schulich Heart Research Program, Sunnybrook Research Institute, Toronto, Canada
Received 26 January 2016; received in revised form 18 April 2016; accepted 10 June 2016
Abstract Purpose: Studies evaluating the association between statin use and survival outcomes in renal cell carcinoma have demonstrated conflicting results. Our objective was to evaluate this association in a large clinical cohort by using propensity score methods to reduce confounding from measured covariates. Methods: We performed a retrospective review of 893 patients undergoing nephrectomy for unilateral, M0 renal cell carcinoma between 2000 and 2014 at a tertiary academic center. Inverse probability of treatment weights were derived from a propensity score model based on clinical, surgical, and pathological characteristics. We used Cox proportional hazard models to evaluate the association between statin use and disease-free survival, cancer-specific survival, and overall survival in the sample weighted by the inverse probability of treatment weights. A secondary analysis was performed matching statin users 1:1 to statin nonusers on the propensity score. Results: Of the 893 patients, 259 (29%) were on statins at the time of surgery. Median follow-up was 47 months (interquartile range: 20–80). Statin use was not significantly associated with disease-free survival (hazard ratio [HR] ¼ 1.09, 95% CI: 0.65–1.81), cancerspecific survival (HR ¼ 0.90, 95% CI: 0.40–2.01), or overall survival (HR ¼ 0.89, 95% CI: 0.55–1.44). Similar results were observed when using propensity score matching. Conclusions: The present study found no significant association between statin use and kidney cancer outcomes. Population-based studies are needed to further evaluate the role of statins in kidney cancer therapy. r 2016 Elsevier Inc. All rights reserved.
Keywords: HMG-CoA reductase inhibitor; Kidney neoplasms; Nephrectomy; Propensity score; Survival
1. Introduction Kidney cancer is the third most common urological malignancy and has the highest case fatality rate among all urological malignancies, with an estimated 337,860 new diagnoses and 143,406 attributable deaths worldwide in 2012 [1]. Furthermore, the number of incident cases has been rising worldwide [1], likely reflecting increasing use of diagnostic imaging and rising rates of obesity and hypertension, known
Corresponding author. Tel.: þ1-416-946-2909; fax: þ1-416-946-6590. E-mail address: [email protected] (R.J. Hamilton). *
http://dx.doi.org/10.1016/j.urolonc.2016.06.007 1078-1439/r 2016 Elsevier Inc. All rights reserved.
risk factors for the disease [2]. Despite earlier detection, survival rates have only improved marginally [1]. Statins are a commonly prescribed class of medications that reduce blood cholesterol levels by inhibiting the 3-hydroxy-3-methylglutaryl coenzyme A reductase enzyme [3]. They are generally well tolerated with the most common side effects being myopathy and an asymptomatic rise in hepatic transaminases, both of which occur infrequently [4]. Recently, statins have gained interest in the oncology community, as studies have shown that they may be associated with improved cancer survival outcomes [5]. Results from epidemiological studies have led to clinical trials evaluating the role of statins in combination or as a
2
M. Nayan et al. / Urologic Oncology: Seminars and Original Investigations ] (2016) 1–6
stand-alone therapy in the neoadjuvant and adjuvant setting in patients with cancer [6,7]. Though the exact mechanism by which statins may exert their antineoplastic effects is unknown, evidence suggests there are cholesterol-dependent and cholesterol-independent mechanisms [8]. In the 3-hydroxy-3-methylglutaryl coenzyme A pathway, statins inhibit the biosynthesis of mevalonate and also the formation of downstream lipid isoprenoid intermediates. These isoprenoid intermediates are important in regulating cell apoptosis, angiogenesis, and inflammation. Cholesterolindependent mechanisms include effects on cell adhesion, invasion, proliferation, and inflammation through interactions with lymphocyte-function antigen 1 and proteosomes [8]. To date, the association between statin use and survival outcomes in renal cell carcinoma (RCC), the most common type of kidney cancer [2], has been evaluated in a limited number of studies that have provided conflicting results [9–13]. Further research is needed to evaluate whether statins may play a role in therapy for RCC. The objective of this study was to evaluate the association between statin use at the time of nephrectomy for unilateral, sporadic, M0 RCC, and survival outcomes by using propensity score methods to reduce confounding from measured covariates. 2. Methods 2.1. Patients and data sources After obtaining institutional Research Ethics Board approval (REB 14-8273-CE), we retrospectively identified patients who underwent a nephrectomy for kidney cancer between January 2000 and December 2014 using our institutional database (eKidney, University Health Network, Toronto). Patients were excluded if disease histology was inconsistent with RCC, if they had hereditary renal cancer syndromes, presented with bilateral disease, underwent previous treatment for kidney cancer, or underwent prior radiofrequency ablation on the same mass for which they underwent nephrectomy. Using electronic record review, we ascertained clinical, surgical, and pathological variables for each patient. Exposure to statins was defined as active use of the medication at the time of surgery. We identified recurrence and mortality, including cause of death, through electronic medical record review and through the Princess Margaret Cancer Centre Registry, which uses data from the Ontario Cancer Registry, a population-based provincial registry that collects data from various sources for all Ontario residents diagnosed with cancer, including their death certificates [14]. The Ontario Cancer Registry that had been validated [15] is estimated to be more than 95% complete [14]. 2.2. Outcomes measures The outcome measures were disease-free survival (DFS), cancer-specific survival (CSS), and overall survival (OS).
Survival times were measured from the date of nephrectomy. Disease relapse was defined as physical examination suggestive of recurrence that was then confirmed by imaging, or imaging demonstrating evidence of disease recurrence, whichever came first. Patients not experiencing relapse or death were censored at their last clinical visit. 2.3. Statistical analysis We compared categorical variables between groups using the chi-square test, or Fisher exact test when cell counts were 5 or fewer, whereas continuous variables were compared using the Wilcoxon Rank-Sum test. To estimate the effect of statins on RCC outcomes, we used propensity score methods to reduce confounding because of differences in the distribution of measured covariates between statin users and nonusers. We estimated the propensity score using a logistic regression model, in which statin use at the time of surgery was regressed on measured baseline characteristics. Covariates for inclusion in the propensity score model were chosen based on their potential to be associated with the outcomes [16] and included patient characteristics (including age, sex, and Charlson comorbidity score), surgical characteristics (including year and type of surgery), and pathological characteristics (including pathologic stage, nodal stage, and histology). All variables were modeled as categorical variables, except for age was which was modeled as a continuous variable. The effect of age on the log-odds of statin use was modeled using restricted cubic splines with 5 knots. For the primary analysis, we computed inverse probability of treatment weights (IPTW), defined as the reciprocal of the probability of receiving the treatment that was actually received [16], using the derived propensity score for all patients. To improve the precision of estimated effects from the IPTW analyses, we stabilized weights by multiplying them by the marginal probability of receiving the exposure that was actually received [17,18]. Stabilized weights aim to improve precision by reducing the variance of the weights [17]. In a secondary analysis, we attempted to match each patient who was a statin user at the time of surgery to a nonusers who served as a control. Patients were matched on the logit of the propensity score using greedy nearest neighbor matching with a caliper width of 0.2 of the standard deviation of the logit of the propensity score [19]. This matching strategy led to a large number of statin users being unable to be matched to a nonuser. In this setting, strategies to improve matching were followed [20], a new variable was created by categorizing age (o60 vs. Z60). Patients were then matched on equal weights of the logit of the propensity score (which included age as a continuous variable along with the restricted cubic splines) and age as a categorized variable. This strategy led to an adequate number of matches. To ensure that propensity score methods had achieved adequate balance, we compared baseline covariates by
M. Nayan et al. / Urologic Oncology: Seminars and Original Investigations ] (2016) 1–6
calculating standardized differences for both the primary and secondary analyses [21,22] (in the primary analysis, weighted standardized differences were used). We constructed IPTW-weighted adjusted survival curves for all outcomes [23]. We calculated absolute differences in estimated survival at 2 and 5 years. IPTW-weighted Cox proportional hazard models with robust variance estimators were used to estimate the relative effect of statin therapy on the hazard of each outcome [16,18]. To do this, the hazard of the occurrence of the outcome was regressed on exposure status (statin users vs. nonusers) in the sample weighted by the IPTW. Similarly, marginal Cox proportional hazard models with robust variance estimators were used to estimate the hazard of each outcome in the propensity score matched cohort. The Cox proportional hazards assumption
3
was evaluated through the cumulative score statistic [24], and was found to hold for all outcomes. All tests were 2-sided with P o 0.05 considered statistically significant. Statistical analyses were performed using SAS v9.4 (SAS Institute Inc, Cary, NC).
3. Results 3.1. Cohort characteristics A total of 893 patients underwent nephrectomy for unilateral, sporadic, M0 RCC between January 2000 and December 2014. Of these, 259 (29%) were statin users at the time of nephrectomy. Cohort characteristics are
Table Baseline characteristics of 893 patients undergoing a nephrectomy for unilateral, sporadic, M0 kidney cancer Variables
Full cohort
Propensity score matched cohort
Statin Statin P value users nonusers (unadjusted (n ¼ 259) (n ¼ 634) comparison) Patient characteristics Age at nephrectomy, years (median [IQR]) Sex (n [%]) Male Female Charlson comorbidity index (n [%]) 0-1 2þ Surgical characteristics Type of surgery (n [%]) Radical nephrectomy Partial nephrectomy Year of surgery (n [%]) 2000–2004 2005–2009 2009–2014
66 (16)
57 (18)
187 (72) 72 (28)
387 (61) 247 (39)
499 (79) 135 (21)
140 (54) 119 (46)
337 (53) 297 (47)
27 (11) 107 (41) 125 (48)
101 (16) 253 (40) 280 (44)
66.2
7.8
0.002
23.8
0.1
0.81
58.1
1.8
Statin Statin users nonusers (n ¼ 231) (n ¼ 231)
65 (16)
64 (15)
161 (70) 70 (30)
152 (66) 79 (34)
Standardized difference (%)
1.0 8.3
9.4
0.9 133 (58) 98 (42)
134 (58) 97 (42)
122 (53) 109 (47)
118 (51) 113 (49)
0.5
3.5
0.10
Pathologic characteristics Pathologic tumor stage (n [%]) pT1 203 (78) pT2 18 (7) pT3 37 (14) pT4 1 (1) Pathologic nodal stage (n [%]) pNX 226 (87) pN0 26 (10) pNþ 7 (3) Disease histology (n [%]) Clear cell renal cell 187 (72) carcinoma Papillary renal cell 61 (24) carcinoma Chromophobe renal cell 7 (3) carcinoma Other renal cell carcinoma 4 (1) IQR ¼ interquartile range.
Sample weighted by IPTW standardized difference (%)
o0.0001
o0.0001 135 (52) 124 (48)
Standardized difference in unweighted cohort (%)
16.3 2.9 8.2
2.2 1.5 0.6
23 (10) 99 (43) 109 (47)
28 (12) 92 (40) 111 (48)
6.9 6.1 1.7
19.5 23.3 2.4 6.9
3.5 4.5 0 0.8
177 16 37 1
182 14 35 0
(79) (6) (15) (0)
5.2 3.5 2.4 9.3
13.7 15.0 0.1
9.6 12.0 2.6
199 (86) 25 (11) 7 (3)
204 (88) 22 (10) 5 (2)
6.5 4.3 5.4
466 (74)
2.9
12.0
173 (75)
172 (75)
1.0
94 (15)
22.3
2.5
48 (21)
49 (21)
1.1
60 (9)
28.6
14.6
7 (3)
7 (3)
0
14 (2)
4.9
0.2
3 (1)
3 (1)
0
0.011 443 89 96 6
(68) (14) (15) (1)
(77) (7) (16) (0.4)
0.15 522 (82) 95 (15) 17 (3) o0.0001
M. Nayan et al. / Urologic Oncology: Seminars and Original Investigations ] (2016) 1–6
described in the Table. Before propensity score weighting, statin users were significantly older, more likely to be male, had higher Charlson comorbidity scores, had lower pathologic stage, and were more likely to have papillary RCC. Year and type of surgery, and pathologic nodal stage did not differ between statin users and nonusers before weighting. After weighting, clinical, surgical, and pathological characteristics were well balanced between the 2 groups (Table). For the secondary analysis, 231 (89%) statin users were successfully matched to 231 nonusers on the propensity score. Propensity score matching also achieved good balance between the 2 groups (Table). Median follow-up was 47 months (interquartile range: 20–80) in the full cohort, and 46 months (interquartile range: 22–79) in the matched cohort. 3.2. Disease-free survival IPTW-adjusted survival curves comparing DFS between statin users and nonusers are shown in Fig. 1. DFS at 2 and 5 years were 93% and 87% for nonusers, and 90% and 89% for statin users, respectively. When comparing nonusers to statin users, the corresponding absolute risk differences in recurrence rates were 3.2% at 2 years, and 1.1% at 5 years, respectively. The Cox proportional hazard model fit in the weighted sample demonstrated no significant association between statin use at the time of nephrectomy and DFS (hazard ratio [HR] ¼ 1.09; 95% CI: 0.65–1.81; P ¼ 0.75). Results were similar in the propensity score matched cohort (HR ¼ 1.17; 95% CI: 0.69–1.99; P ¼ 0.55). 3.3. Cancer-specific survival IPTW-adjusted survival curves comparing CSS between statin users and nonusers are shown in Fig. 2. At 2 and 5 years, CSS was 96% and 95% for nonusers, and 98% and 96% for statin users, respectively. When comparing nonusers to statin users, the corresponding absolute risk differences in cancer-related death rates were 1.8% at 2 years, and 0.8% at 5 years, respectively. IPTW-adjusted marginal Cox proportional hazard models demonstrated no
100%
Survival Probability (%)
4
60% 40% 20%
Stan non-user Stan user
0% 0
20
40
60
80
100
Time (months)
Fig. 2. IPTW-adjusted survival curves comparing CSS among statin users and nonusers. (Color version of figure is available online.)
significant association between statin use at the time of nephrectomy and CSS (HR ¼ 0.90; 95% CI: 0.40–2.01; P ¼ 0.79). Results were similar in the propensity score matched cohort (HR ¼ 0.83; 95% CI: 0.38–1.84; P ¼ 0.65). 3.4. Overall survival IPTW-adjusted survival curves comparing OS between statin users and nonusers are shown in Fig. 3. At 2 and 5 years, OS was 92% and 86% for nonusers, and 95% and 90% for statin users, respectively. When comparing nonusers to statin users, the corresponding absolute risk differences in OS were 2.8% at 2 years, and 3.7% at 5 years, respectively. IPTW-adjusted marginal Cox proportional hazard models demonstrated no significant association between statin use at the time of nephrectomy and OS (HR ¼ 0.89; 95% CI: 0.55–1.44; P ¼ 0.63). Results were similar in the propensity matched cohort (HR ¼ 0.84; 95% CI: 0.54–1.31; P ¼ 0.43). 4. Discussion In this study, we found that statin use at the time of nephrectomy for RCC was not associated with significantly lower DFS, CSS, or OS. Statin use has been shown to be 100%
80% 60% 40% 20%
Stan non-user Stan user
0% 0
20
40
60
80
100
Time (months)
Fig. 1. IPTW-adjusted survival curves comparing DFS among statin users and nonusers. (Color version of figure is available online.)
Survival Probability (%)
100%
Survival Probability (%)
80%
80% 60% 40% 20%
Stan non-user Stan user
0% 0
20
40
60
80
100
Time (months)
Fig. 3. IPTW-adjusted survival curves comparing OS among statin users and nonusers. (Color version of figure is available online.)
M. Nayan et al. / Urologic Oncology: Seminars and Original Investigations ] (2016) 1–6
associated with improved survival outcomes in various malignancies [5], with randomized controlled trials currently underway evaluating the role of statins in combination or as a stand-alone therapy in the neoadjuvant and adjuvant setting in patients with cancer [6,7]. However, studies evaluating the association between statin use and survival outcomes in RCC have demonstrated conflicting results [9–13]. The results from our study do not support the use of statins to improve survival outcomes in RCC. To date, only a study has found a significant improvement in CSS and OS when comparing statin use vs. nonuse at the time of nephrectomy for RCC [10]. The analyses by Kaffenberger et al. included 666 patients and found that statin use at the time of nephrectomy was associated with a 52% reduction in cancer-specific mortality and a 38% reduction in all-cause mortality. However, it is difficult to generalize the estimates of association reported in this study. First, they found that there were significant differences in sex and age distributions between statin users and nonusers, as was found our study, but age was not adjusted for in the multivariable model evaluating CSS and sex was not adjusted for either multivariable analysis for CSS and OS. Given that age and sex [25] are associated with survival outcomes in RCC, these are important confounders to control for to minimize bias in estimates. Second, the study describes the median follow-up (42.5 months) in the full cohort of patients, but not those with complete data that were ultimately included in the multivariable analyses. As our study found, it may be possible that their study found a protective association because of early differences in survival, which may become attenuated over time. Last, this study used the American Society of Anesthesiologists Classification score to adjust for comorbidity. The utility of this risk adjustment method for comorbidities is not well known for predicting survival outcomes in RCC, compared with Charlson Comorbidity Index, which has been evaluated in several studies [26–28]. Although Kaffenberger et al. did not evaluate DFS, another study found that statin use at the time of nephrectomy was associated with a 33% improvement in DFS, but no improvement in OS. Although this analysis compared use vs. nonuse at the time of surgery, as was done in this study, they also performed time-varying analyses in which current use of statin was associated with a 29% reduction in all-cause mortality, but DFS was no longer significant. In their time-varying analyses, patients who started a statin after surgery contributed person-time to the nonuser group until starting the statin, after which they contributed to the user group. Time-varying analyses are more robust than ever vs. never analyses, as was done in all other studies evaluating statin use and outcomes in RCC [10–13], including our study, because it can account for intermittent exposure. However, with institutional data, it is difficult to obtain reliable data regarding time on and off medication. It is certainly possible that patients considered users may have missed time (e.g., 1 week) between
5
prescriptions, and should have contributed this missed person-time to the nonuser group. It is not expected that this data would be completely captured in institutional data; rather, databases in which there is validated, comprehensive information on prescription dates, and days supplied may be preferable for time-varying analyses [29,30]. Although our study is not the first to evaluate the association between statin use and survival outcomes in patients undergoing a nephrectomy for RCC, it is the first to use propensity score methods in this context. Propensity score methods allow one to reduce the effects of measured confounding and mimic some of the characteristics of the design of a randomized controlled trial as they compare outcomes in exposed and unexposed patients who have a balanced distribution of measured baseline characteristics [18]. As such, they are well suited to minimize bias for observational studies in pharmacoepidemiology [31]. Some limitations of our study merit emphasis. First, it is an observational study, and therefore cannot prove causality. However, observational studies such as these provide the rationale to proceed with randomized controlled trials. On the contrary, using propensity score matching allows us to mimic a randomized controlled trial [18] and compare outcomes in a population that is similar in all measured baseline characteristics, except for statin use. Nonetheless, it is important to emphasize that propensity score methods can only balance measured confounders, and residual confounding is still possible because of unmeasured confounders. Second, the cohort is limited by sample and this study may be underpowered; multiinstitutional or population-based studies may be required to obtain a significantly larger cohort. Third, exposure was classified based on use vs. nonuse at the time of surgery. It is possible that patients classified as users discontinued their statin after surgery, and nonusers were started on statins after surgery. Furthermore, we could not account for cumulative exposure. However, reliable data regarding intermittent use and cumulative exposure are difficult to obtain in institutional data. Rather, population-based studies involving administrate prescription data may be required to account for these, as has been done in other malignancies [29,30]. Last, the results of our study do not apply to patients with metastatic RCC; it may be possible that statins are beneficial in the setting of metastatic disease, as has been shown by others [32]. Despite these limitations, the use of propensity score methods in the present study strengthens the findings and, consistent with other prior studies [11–13], suggests that statin use at the time of surgery is not significantly associated with improved outcomes in patients undergoing nephrectomy for RCC.
5. Conclusion The present study found no significant association between statin use and kidney cancer outcomes in patients
6
M. Nayan et al. / Urologic Oncology: Seminars and Original Investigations ] (2016) 1–6
undergoing a nephrectomy for unilateral, sporadic, and M0 RCC. Population-based studies are needed to further evaluate the role of statins in kidney cancer therapy. References [1] Cancer IAfRo, Organization WH. GLOBOCAN: Estimated cancer incidence, mortality, and prevalence worldwide in 2012. IARC; 2014. [2] Chow WH, Dong LM, Devesa SS. Epidemiology and risk factors for kidney cancer. Nat Rev Urol 2010;7:245–57, http://dx.doi.org/10. 1038/nrurol.2010.46. [3] Gazzerro P, Proto MC, Gangemi G, Malfitano AM, Ciaglia E, Pisanti S, et al. Pharmacological actions of statins: a critical appraisal in the management of cancer. Pharmacol Rev 2012;64:102–46. [4] Law M, Rudnicka AR. Statin safety: a systematic review. Am J Cardiol 2006;97:S52–60. [5] Nielsen SF, Nordestgaard BG, Bojesen SE. Statin use and reduced cancer-related mortality. N Engl J Med 2012;367:1792–802. [6] Manoukian GE, Tannir NM, Jonasch E, Qiao W, Haygood TM, Tu SM. Pilot trial of bone-targeted therapy combining zoledronate with fluvastatin or atorvastatin for patients with metastatic renal cell carcinoma. Clin Genitourin Cancer 2011;9:81–8. [7] Han J-Y, Lee S-H, Yoo NJ, Hyung LS, Moon YJ, Yun T, et al. A randomized phase II study of gefitinib plus simvastatin versus gefitinib alone in previously treated patients with advanced non– small cell lung cancer. Clin Cancer Res 2011;17:1553–60. [8] Demierre M-F, Higgins PD, Gruber SB, Hawk E, Lippman SM. Statins and cancer prevention. Nat Rev Cancer 2005;5:930–42. [9] Hamilton RJ, Morilla D, Cabrera F, Leapman M, Chen LY, Bernstein M, et al. The association between statin medication and progression after surgery for localized renal cell carcinoma. J Urol 2014;191:914–9. [10] Kaffenberger SD, Lin-Tsai O, Stratton KL, Morgan TM, Barocas DA, Chang SS, et al. Statin use is associated with improved survival in patients undergoing surgery for renal cell carcinoma. Urol Oncol 2015;1:[21.e11–21. e17]. [11] Krane LS, Sandberg JM, Rague JT, Hemal AK. Do statin medications impact renal functional or oncologic outcomes for robot-assisted partial nephrectomy? J Endourol 2014;28:1308–12. [12] Viers BR, Thompson RH, Psutka SP, Lohse CM, Cheville JC, Leibovich BC, et al. The association of statin therapy with clinicopathologic outcomes and survival among patients with localized renal cell carcinoma undergoing nephrectomy. Urol Oncol 2015;33:[388-e11]. [13] Choi S-K, Min GE, Jeon SH, Lee H-L, Chang S-G, Yoo KH. Effects of statins on the prognosis of local and locally advanced renal cell carcinoma following nephrectomy. Mol Clin Oncol 2013;1: 365–8. [14] Clarke E, Marrett L, Kreiger N. In: Jenson OM, Parkin DM, MacLennan R (eds.). Appendix 3 (c) Cancer registration in Ontario: a computer approach. Cancer Registration: Principles and Methods, vol. 95. Lyon, France: IARC Publication; 246–57. [15] Brenner DR, Tammemagi MC, Bull SB, Pinnaduwaje D, Andrulis IL. Using cancer registry data: agreement in cause-of-death data between the Ontario Cancer Registry and a longitudinal study of breast cancer
patients. Chronic Dis Can 2009;30:16–9. [16] Austin PC, Stuart EA. Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Stat Med 2015;34:3661–79. [17] Brookhart MA, Wyss R, Layton JB, Stürmer T. Propensity score methods for confounding control in nonexperimental research. Circ Cardiovasc Qual Outcomes 2013;6:604–11. [18] Austin PC. The use of propensity score methods with survival or time-to-event outcomes: reporting measures of effect similar to those used in randomized experiments. Stat Med 2014;33:1242–58. [19] Austin PC. Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies. Pharm Stat 2011;10:150–61. [20] Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behav Res 2011;46:399–424. [21] Austin PC. Assessing balance in measured baseline covariates when using many-to-one matching on the propensity-score. Pharmacoepidemiol Drug Saf 2008;17:1218–25. [22] Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med 2009;28:3083–107. [23] Cole SR, Hernán MA. Adjusted survival curves with inverse probability weights. Comput Methods Programs Biomed 2004;75:45–9. [24] Lin DY, Wei L-J, Ying Z. Checking the Cox model with cumulative sums of martingale-based residuals. Biometrika 1993;80:557–72. [25] Aron M, Nguyen MM, Stein RJ, Gill IS. Impact of gender in renal cell carcinoma: an analysis of the SEER database. Eur Urol 2008;54: 133–42. [26] Ather MH, Nazim SM. Impact of Charlson's comorbidity index on overall survival following tumor nephrectomy for renal cell carcinoma. Int Urol Nephrol 2010;42:299–303. [27] Gettman MT, Boelter CW, Cheville JC, Zincke H, Bryant SC, Blute ML. Charlson co-morbidity index as a predictor of outcome after surgery for renal cell carcinoma with renal vein, vena cava or right atrium extension. J Urol 2003;169:1282–6. [28] Arrontes DS, Aceñero MJF, González JIG, Muñoz MM, Andrés PP. Survival analysis of clear cell renal carcinoma according to the Charlson comorbidity index. J Urol 2008;179:857–61. [29] Margel D, Urbach DR, Lipscombe LL, Bell CM, Kulkarni G, Austin PC, et al. Metformin use and all-cause and prostate cancer-specific mortality among men with diabetes. J Clin Oncol 2013;31:3069–75, http://dx.doi.org/10.1200/JCO.2012.46.7043. [30] Lega IC, Austin PC, Gruneir A, Goodwin PJ, Rochon PA, Lipscombe LL. Association between metformin therapy and mortality after breast cancer: a population-based study. Diabetes Care 2013;36:3018–26, http://dx.doi.org/10.2337/dc12-2535. [31] Austin PC, Mamdani MM. A comparison of propensity score methods: a case-study estimating the effectiveness of post-AMI statin use. Stat Med 2006;25:2084–106. [32] McKay RR, Lin X, Albiges L, Fay AP, Kaymakcalan MD, Mickey SS, et al. Statins and survival outcomes in patients with metastatic renal cell carcinoma. Eur J Cancer 2016;52:155–62.
Original article
Statin use and kidney cancer outcomes: A propensity score analysis Madhur Nayan, M.D., C.M.a,b, Antonio Finelli, M.D., M.Sc.a,b, Michael A.S. Jewett, M.D.a,b, David N. Juurlink, M.D., Ph.D.c, Peter C. Austin, Ph.D.d,e,f, Girish S. Kulkarni, M.D., Ph.D.a,b, Robert J. Hamilton, M.D., M.P.H.a,b,* b
a Department of Surgery, Princess Margaret Cancer Centre, University Health Network, University of Toronto, Toronto, Canada Department of Surgical Oncology, Princess Margaret Cancer Centre, University Health Network, University of Toronto, Toronto, Canada c Department of Internal Medicine, Sunnybrook Health Sciences Centre, University of Toronto, Toronto, Ontario, Canada d Institute for Clinical Evaluative Sciences, Toronto, Canada e Institute of Health Management, Policy and Evaluation, University of Toronto, Toronto, Canada f Schulich Heart Research Program, Sunnybrook Research Institute, Toronto, Canada
Received 26 January 2016; received in revised form 18 April 2016; accepted 10 June 2016
Abstract Purpose: Studies evaluating the association between statin use and survival outcomes in renal cell carcinoma have demonstrated conflicting results. Our objective was to evaluate this association in a large clinical cohort by using propensity score methods to reduce confounding from measured covariates. Methods: We performed a retrospective review of 893 patients undergoing nephrectomy for unilateral, M0 renal cell carcinoma between 2000 and 2014 at a tertiary academic center. Inverse probability of treatment weights were derived from a propensity score model based on clinical, surgical, and pathological characteristics. We used Cox proportional hazard models to evaluate the association between statin use and disease-free survival, cancer-specific survival, and overall survival in the sample weighted by the inverse probability of treatment weights. A secondary analysis was performed matching statin users 1:1 to statin nonusers on the propensity score. Results: Of the 893 patients, 259 (29%) were on statins at the time of surgery. Median follow-up was 47 months (interquartile range: 20–80). Statin use was not significantly associated with disease-free survival (hazard ratio [HR] ¼ 1.09, 95% CI: 0.65–1.81), cancerspecific survival (HR ¼ 0.90, 95% CI: 0.40–2.01), or overall survival (HR ¼ 0.89, 95% CI: 0.55–1.44). Similar results were observed when using propensity score matching. Conclusions: The present study found no significant association between statin use and kidney cancer outcomes. Population-based studies are needed to further evaluate the role of statins in kidney cancer therapy. r 2016 Elsevier Inc. All rights reserved.
Keywords: HMG-CoA reductase inhibitor; Kidney neoplasms; Nephrectomy; Propensity score; Survival
1. Introduction Kidney cancer is the third most common urological malignancy and has the highest case fatality rate among all urological malignancies, with an estimated 337,860 new diagnoses and 143,406 attributable deaths worldwide in 2012 [1]. Furthermore, the number of incident cases has been rising worldwide [1], likely reflecting increasing use of diagnostic imaging and rising rates of obesity and hypertension, known
Corresponding author. Tel.: þ1-416-946-2909; fax: þ1-416-946-6590. E-mail address: [email protected] (R.J. Hamilton). *
http://dx.doi.org/10.1016/j.urolonc.2016.06.007 1078-1439/r 2016 Elsevier Inc. All rights reserved.
risk factors for the disease [2]. Despite earlier detection, survival rates have only improved marginally [1]. Statins are a commonly prescribed class of medications that reduce blood cholesterol levels by inhibiting the 3-hydroxy-3-methylglutaryl coenzyme A reductase enzyme [3]. They are generally well tolerated with the most common side effects being myopathy and an asymptomatic rise in hepatic transaminases, both of which occur infrequently [4]. Recently, statins have gained interest in the oncology community, as studies have shown that they may be associated with improved cancer survival outcomes [5]. Results from epidemiological studies have led to clinical trials evaluating the role of statins in combination or as a
2
M. Nayan et al. / Urologic Oncology: Seminars and Original Investigations ] (2016) 1–6
stand-alone therapy in the neoadjuvant and adjuvant setting in patients with cancer [6,7]. Though the exact mechanism by which statins may exert their antineoplastic effects is unknown, evidence suggests there are cholesterol-dependent and cholesterol-independent mechanisms [8]. In the 3-hydroxy-3-methylglutaryl coenzyme A pathway, statins inhibit the biosynthesis of mevalonate and also the formation of downstream lipid isoprenoid intermediates. These isoprenoid intermediates are important in regulating cell apoptosis, angiogenesis, and inflammation. Cholesterolindependent mechanisms include effects on cell adhesion, invasion, proliferation, and inflammation through interactions with lymphocyte-function antigen 1 and proteosomes [8]. To date, the association between statin use and survival outcomes in renal cell carcinoma (RCC), the most common type of kidney cancer [2], has been evaluated in a limited number of studies that have provided conflicting results [9–13]. Further research is needed to evaluate whether statins may play a role in therapy for RCC. The objective of this study was to evaluate the association between statin use at the time of nephrectomy for unilateral, sporadic, M0 RCC, and survival outcomes by using propensity score methods to reduce confounding from measured covariates. 2. Methods 2.1. Patients and data sources After obtaining institutional Research Ethics Board approval (REB 14-8273-CE), we retrospectively identified patients who underwent a nephrectomy for kidney cancer between January 2000 and December 2014 using our institutional database (eKidney, University Health Network, Toronto). Patients were excluded if disease histology was inconsistent with RCC, if they had hereditary renal cancer syndromes, presented with bilateral disease, underwent previous treatment for kidney cancer, or underwent prior radiofrequency ablation on the same mass for which they underwent nephrectomy. Using electronic record review, we ascertained clinical, surgical, and pathological variables for each patient. Exposure to statins was defined as active use of the medication at the time of surgery. We identified recurrence and mortality, including cause of death, through electronic medical record review and through the Princess Margaret Cancer Centre Registry, which uses data from the Ontario Cancer Registry, a population-based provincial registry that collects data from various sources for all Ontario residents diagnosed with cancer, including their death certificates [14]. The Ontario Cancer Registry that had been validated [15] is estimated to be more than 95% complete [14]. 2.2. Outcomes measures The outcome measures were disease-free survival (DFS), cancer-specific survival (CSS), and overall survival (OS).
Survival times were measured from the date of nephrectomy. Disease relapse was defined as physical examination suggestive of recurrence that was then confirmed by imaging, or imaging demonstrating evidence of disease recurrence, whichever came first. Patients not experiencing relapse or death were censored at their last clinical visit. 2.3. Statistical analysis We compared categorical variables between groups using the chi-square test, or Fisher exact test when cell counts were 5 or fewer, whereas continuous variables were compared using the Wilcoxon Rank-Sum test. To estimate the effect of statins on RCC outcomes, we used propensity score methods to reduce confounding because of differences in the distribution of measured covariates between statin users and nonusers. We estimated the propensity score using a logistic regression model, in which statin use at the time of surgery was regressed on measured baseline characteristics. Covariates for inclusion in the propensity score model were chosen based on their potential to be associated with the outcomes [16] and included patient characteristics (including age, sex, and Charlson comorbidity score), surgical characteristics (including year and type of surgery), and pathological characteristics (including pathologic stage, nodal stage, and histology). All variables were modeled as categorical variables, except for age was which was modeled as a continuous variable. The effect of age on the log-odds of statin use was modeled using restricted cubic splines with 5 knots. For the primary analysis, we computed inverse probability of treatment weights (IPTW), defined as the reciprocal of the probability of receiving the treatment that was actually received [16], using the derived propensity score for all patients. To improve the precision of estimated effects from the IPTW analyses, we stabilized weights by multiplying them by the marginal probability of receiving the exposure that was actually received [17,18]. Stabilized weights aim to improve precision by reducing the variance of the weights [17]. In a secondary analysis, we attempted to match each patient who was a statin user at the time of surgery to a nonusers who served as a control. Patients were matched on the logit of the propensity score using greedy nearest neighbor matching with a caliper width of 0.2 of the standard deviation of the logit of the propensity score [19]. This matching strategy led to a large number of statin users being unable to be matched to a nonuser. In this setting, strategies to improve matching were followed [20], a new variable was created by categorizing age (o60 vs. Z60). Patients were then matched on equal weights of the logit of the propensity score (which included age as a continuous variable along with the restricted cubic splines) and age as a categorized variable. This strategy led to an adequate number of matches. To ensure that propensity score methods had achieved adequate balance, we compared baseline covariates by
M. Nayan et al. / Urologic Oncology: Seminars and Original Investigations ] (2016) 1–6
calculating standardized differences for both the primary and secondary analyses [21,22] (in the primary analysis, weighted standardized differences were used). We constructed IPTW-weighted adjusted survival curves for all outcomes [23]. We calculated absolute differences in estimated survival at 2 and 5 years. IPTW-weighted Cox proportional hazard models with robust variance estimators were used to estimate the relative effect of statin therapy on the hazard of each outcome [16,18]. To do this, the hazard of the occurrence of the outcome was regressed on exposure status (statin users vs. nonusers) in the sample weighted by the IPTW. Similarly, marginal Cox proportional hazard models with robust variance estimators were used to estimate the hazard of each outcome in the propensity score matched cohort. The Cox proportional hazards assumption
3
was evaluated through the cumulative score statistic [24], and was found to hold for all outcomes. All tests were 2-sided with P o 0.05 considered statistically significant. Statistical analyses were performed using SAS v9.4 (SAS Institute Inc, Cary, NC).
3. Results 3.1. Cohort characteristics A total of 893 patients underwent nephrectomy for unilateral, sporadic, M0 RCC between January 2000 and December 2014. Of these, 259 (29%) were statin users at the time of nephrectomy. Cohort characteristics are
Table Baseline characteristics of 893 patients undergoing a nephrectomy for unilateral, sporadic, M0 kidney cancer Variables
Full cohort
Propensity score matched cohort
Statin Statin P value users nonusers (unadjusted (n ¼ 259) (n ¼ 634) comparison) Patient characteristics Age at nephrectomy, years (median [IQR]) Sex (n [%]) Male Female Charlson comorbidity index (n [%]) 0-1 2þ Surgical characteristics Type of surgery (n [%]) Radical nephrectomy Partial nephrectomy Year of surgery (n [%]) 2000–2004 2005–2009 2009–2014
66 (16)
57 (18)
187 (72) 72 (28)
387 (61) 247 (39)
499 (79) 135 (21)
140 (54) 119 (46)
337 (53) 297 (47)
27 (11) 107 (41) 125 (48)
101 (16) 253 (40) 280 (44)
66.2
7.8
0.002
23.8
0.1
0.81
58.1
1.8
Statin Statin users nonusers (n ¼ 231) (n ¼ 231)
65 (16)
64 (15)
161 (70) 70 (30)
152 (66) 79 (34)
Standardized difference (%)
1.0 8.3
9.4
0.9 133 (58) 98 (42)
134 (58) 97 (42)
122 (53) 109 (47)
118 (51) 113 (49)
0.5
3.5
0.10
Pathologic characteristics Pathologic tumor stage (n [%]) pT1 203 (78) pT2 18 (7) pT3 37 (14) pT4 1 (1) Pathologic nodal stage (n [%]) pNX 226 (87) pN0 26 (10) pNþ 7 (3) Disease histology (n [%]) Clear cell renal cell 187 (72) carcinoma Papillary renal cell 61 (24) carcinoma Chromophobe renal cell 7 (3) carcinoma Other renal cell carcinoma 4 (1) IQR ¼ interquartile range.
Sample weighted by IPTW standardized difference (%)
o0.0001
o0.0001 135 (52) 124 (48)
Standardized difference in unweighted cohort (%)
16.3 2.9 8.2
2.2 1.5 0.6
23 (10) 99 (43) 109 (47)
28 (12) 92 (40) 111 (48)
6.9 6.1 1.7
19.5 23.3 2.4 6.9
3.5 4.5 0 0.8
177 16 37 1
182 14 35 0
(79) (6) (15) (0)
5.2 3.5 2.4 9.3
13.7 15.0 0.1
9.6 12.0 2.6
199 (86) 25 (11) 7 (3)
204 (88) 22 (10) 5 (2)
6.5 4.3 5.4
466 (74)
2.9
12.0
173 (75)
172 (75)
1.0
94 (15)
22.3
2.5
48 (21)
49 (21)
1.1
60 (9)
28.6
14.6
7 (3)
7 (3)
0
14 (2)
4.9
0.2
3 (1)
3 (1)
0
0.011 443 89 96 6
(68) (14) (15) (1)
(77) (7) (16) (0.4)
0.15 522 (82) 95 (15) 17 (3) o0.0001
M. Nayan et al. / Urologic Oncology: Seminars and Original Investigations ] (2016) 1–6
described in the Table. Before propensity score weighting, statin users were significantly older, more likely to be male, had higher Charlson comorbidity scores, had lower pathologic stage, and were more likely to have papillary RCC. Year and type of surgery, and pathologic nodal stage did not differ between statin users and nonusers before weighting. After weighting, clinical, surgical, and pathological characteristics were well balanced between the 2 groups (Table). For the secondary analysis, 231 (89%) statin users were successfully matched to 231 nonusers on the propensity score. Propensity score matching also achieved good balance between the 2 groups (Table). Median follow-up was 47 months (interquartile range: 20–80) in the full cohort, and 46 months (interquartile range: 22–79) in the matched cohort. 3.2. Disease-free survival IPTW-adjusted survival curves comparing DFS between statin users and nonusers are shown in Fig. 1. DFS at 2 and 5 years were 93% and 87% for nonusers, and 90% and 89% for statin users, respectively. When comparing nonusers to statin users, the corresponding absolute risk differences in recurrence rates were 3.2% at 2 years, and 1.1% at 5 years, respectively. The Cox proportional hazard model fit in the weighted sample demonstrated no significant association between statin use at the time of nephrectomy and DFS (hazard ratio [HR] ¼ 1.09; 95% CI: 0.65–1.81; P ¼ 0.75). Results were similar in the propensity score matched cohort (HR ¼ 1.17; 95% CI: 0.69–1.99; P ¼ 0.55). 3.3. Cancer-specific survival IPTW-adjusted survival curves comparing CSS between statin users and nonusers are shown in Fig. 2. At 2 and 5 years, CSS was 96% and 95% for nonusers, and 98% and 96% for statin users, respectively. When comparing nonusers to statin users, the corresponding absolute risk differences in cancer-related death rates were 1.8% at 2 years, and 0.8% at 5 years, respectively. IPTW-adjusted marginal Cox proportional hazard models demonstrated no
100%
Survival Probability (%)
4
60% 40% 20%
Stan non-user Stan user
0% 0
20
40
60
80
100
Time (months)
Fig. 2. IPTW-adjusted survival curves comparing CSS among statin users and nonusers. (Color version of figure is available online.)
significant association between statin use at the time of nephrectomy and CSS (HR ¼ 0.90; 95% CI: 0.40–2.01; P ¼ 0.79). Results were similar in the propensity score matched cohort (HR ¼ 0.83; 95% CI: 0.38–1.84; P ¼ 0.65). 3.4. Overall survival IPTW-adjusted survival curves comparing OS between statin users and nonusers are shown in Fig. 3. At 2 and 5 years, OS was 92% and 86% for nonusers, and 95% and 90% for statin users, respectively. When comparing nonusers to statin users, the corresponding absolute risk differences in OS were 2.8% at 2 years, and 3.7% at 5 years, respectively. IPTW-adjusted marginal Cox proportional hazard models demonstrated no significant association between statin use at the time of nephrectomy and OS (HR ¼ 0.89; 95% CI: 0.55–1.44; P ¼ 0.63). Results were similar in the propensity matched cohort (HR ¼ 0.84; 95% CI: 0.54–1.31; P ¼ 0.43). 4. Discussion In this study, we found that statin use at the time of nephrectomy for RCC was not associated with significantly lower DFS, CSS, or OS. Statin use has been shown to be 100%
80% 60% 40% 20%
Stan non-user Stan user
0% 0
20
40
60
80
100
Time (months)
Fig. 1. IPTW-adjusted survival curves comparing DFS among statin users and nonusers. (Color version of figure is available online.)
Survival Probability (%)
100%
Survival Probability (%)
80%
80% 60% 40% 20%
Stan non-user Stan user
0% 0
20
40
60
80
100
Time (months)
Fig. 3. IPTW-adjusted survival curves comparing OS among statin users and nonusers. (Color version of figure is available online.)
M. Nayan et al. / Urologic Oncology: Seminars and Original Investigations ] (2016) 1–6
associated with improved survival outcomes in various malignancies [5], with randomized controlled trials currently underway evaluating the role of statins in combination or as a stand-alone therapy in the neoadjuvant and adjuvant setting in patients with cancer [6,7]. However, studies evaluating the association between statin use and survival outcomes in RCC have demonstrated conflicting results [9–13]. The results from our study do not support the use of statins to improve survival outcomes in RCC. To date, only a study has found a significant improvement in CSS and OS when comparing statin use vs. nonuse at the time of nephrectomy for RCC [10]. The analyses by Kaffenberger et al. included 666 patients and found that statin use at the time of nephrectomy was associated with a 52% reduction in cancer-specific mortality and a 38% reduction in all-cause mortality. However, it is difficult to generalize the estimates of association reported in this study. First, they found that there were significant differences in sex and age distributions between statin users and nonusers, as was found our study, but age was not adjusted for in the multivariable model evaluating CSS and sex was not adjusted for either multivariable analysis for CSS and OS. Given that age and sex [25] are associated with survival outcomes in RCC, these are important confounders to control for to minimize bias in estimates. Second, the study describes the median follow-up (42.5 months) in the full cohort of patients, but not those with complete data that were ultimately included in the multivariable analyses. As our study found, it may be possible that their study found a protective association because of early differences in survival, which may become attenuated over time. Last, this study used the American Society of Anesthesiologists Classification score to adjust for comorbidity. The utility of this risk adjustment method for comorbidities is not well known for predicting survival outcomes in RCC, compared with Charlson Comorbidity Index, which has been evaluated in several studies [26–28]. Although Kaffenberger et al. did not evaluate DFS, another study found that statin use at the time of nephrectomy was associated with a 33% improvement in DFS, but no improvement in OS. Although this analysis compared use vs. nonuse at the time of surgery, as was done in this study, they also performed time-varying analyses in which current use of statin was associated with a 29% reduction in all-cause mortality, but DFS was no longer significant. In their time-varying analyses, patients who started a statin after surgery contributed person-time to the nonuser group until starting the statin, after which they contributed to the user group. Time-varying analyses are more robust than ever vs. never analyses, as was done in all other studies evaluating statin use and outcomes in RCC [10–13], including our study, because it can account for intermittent exposure. However, with institutional data, it is difficult to obtain reliable data regarding time on and off medication. It is certainly possible that patients considered users may have missed time (e.g., 1 week) between
5
prescriptions, and should have contributed this missed person-time to the nonuser group. It is not expected that this data would be completely captured in institutional data; rather, databases in which there is validated, comprehensive information on prescription dates, and days supplied may be preferable for time-varying analyses [29,30]. Although our study is not the first to evaluate the association between statin use and survival outcomes in patients undergoing a nephrectomy for RCC, it is the first to use propensity score methods in this context. Propensity score methods allow one to reduce the effects of measured confounding and mimic some of the characteristics of the design of a randomized controlled trial as they compare outcomes in exposed and unexposed patients who have a balanced distribution of measured baseline characteristics [18]. As such, they are well suited to minimize bias for observational studies in pharmacoepidemiology [31]. Some limitations of our study merit emphasis. First, it is an observational study, and therefore cannot prove causality. However, observational studies such as these provide the rationale to proceed with randomized controlled trials. On the contrary, using propensity score matching allows us to mimic a randomized controlled trial [18] and compare outcomes in a population that is similar in all measured baseline characteristics, except for statin use. Nonetheless, it is important to emphasize that propensity score methods can only balance measured confounders, and residual confounding is still possible because of unmeasured confounders. Second, the cohort is limited by sample and this study may be underpowered; multiinstitutional or population-based studies may be required to obtain a significantly larger cohort. Third, exposure was classified based on use vs. nonuse at the time of surgery. It is possible that patients classified as users discontinued their statin after surgery, and nonusers were started on statins after surgery. Furthermore, we could not account for cumulative exposure. However, reliable data regarding intermittent use and cumulative exposure are difficult to obtain in institutional data. Rather, population-based studies involving administrate prescription data may be required to account for these, as has been done in other malignancies [29,30]. Last, the results of our study do not apply to patients with metastatic RCC; it may be possible that statins are beneficial in the setting of metastatic disease, as has been shown by others [32]. Despite these limitations, the use of propensity score methods in the present study strengthens the findings and, consistent with other prior studies [11–13], suggests that statin use at the time of surgery is not significantly associated with improved outcomes in patients undergoing nephrectomy for RCC.
5. Conclusion The present study found no significant association between statin use and kidney cancer outcomes in patients
6
M. Nayan et al. / Urologic Oncology: Seminars and Original Investigations ] (2016) 1–6
undergoing a nephrectomy for unilateral, sporadic, and M0 RCC. Population-based studies are needed to further evaluate the role of statins in kidney cancer therapy. References [1] Cancer IAfRo, Organization WH. GLOBOCAN: Estimated cancer incidence, mortality, and prevalence worldwide in 2012. IARC; 2014. [2] Chow WH, Dong LM, Devesa SS. Epidemiology and risk factors for kidney cancer. Nat Rev Urol 2010;7:245–57, http://dx.doi.org/10. 1038/nrurol.2010.46. [3] Gazzerro P, Proto MC, Gangemi G, Malfitano AM, Ciaglia E, Pisanti S, et al. Pharmacological actions of statins: a critical appraisal in the management of cancer. Pharmacol Rev 2012;64:102–46. [4] Law M, Rudnicka AR. Statin safety: a systematic review. Am J Cardiol 2006;97:S52–60. [5] Nielsen SF, Nordestgaard BG, Bojesen SE. Statin use and reduced cancer-related mortality. N Engl J Med 2012;367:1792–802. [6] Manoukian GE, Tannir NM, Jonasch E, Qiao W, Haygood TM, Tu SM. Pilot trial of bone-targeted therapy combining zoledronate with fluvastatin or atorvastatin for patients with metastatic renal cell carcinoma. Clin Genitourin Cancer 2011;9:81–8. [7] Han J-Y, Lee S-H, Yoo NJ, Hyung LS, Moon YJ, Yun T, et al. A randomized phase II study of gefitinib plus simvastatin versus gefitinib alone in previously treated patients with advanced non– small cell lung cancer. Clin Cancer Res 2011;17:1553–60. [8] Demierre M-F, Higgins PD, Gruber SB, Hawk E, Lippman SM. Statins and cancer prevention. Nat Rev Cancer 2005;5:930–42. [9] Hamilton RJ, Morilla D, Cabrera F, Leapman M, Chen LY, Bernstein M, et al. The association between statin medication and progression after surgery for localized renal cell carcinoma. J Urol 2014;191:914–9. [10] Kaffenberger SD, Lin-Tsai O, Stratton KL, Morgan TM, Barocas DA, Chang SS, et al. Statin use is associated with improved survival in patients undergoing surgery for renal cell carcinoma. Urol Oncol 2015;1:[21.e11–21. e17]. [11] Krane LS, Sandberg JM, Rague JT, Hemal AK. Do statin medications impact renal functional or oncologic outcomes for robot-assisted partial nephrectomy? J Endourol 2014;28:1308–12. [12] Viers BR, Thompson RH, Psutka SP, Lohse CM, Cheville JC, Leibovich BC, et al. The association of statin therapy with clinicopathologic outcomes and survival among patients with localized renal cell carcinoma undergoing nephrectomy. Urol Oncol 2015;33:[388-e11]. [13] Choi S-K, Min GE, Jeon SH, Lee H-L, Chang S-G, Yoo KH. Effects of statins on the prognosis of local and locally advanced renal cell carcinoma following nephrectomy. Mol Clin Oncol 2013;1: 365–8. [14] Clarke E, Marrett L, Kreiger N. In: Jenson OM, Parkin DM, MacLennan R (eds.). Appendix 3 (c) Cancer registration in Ontario: a computer approach. Cancer Registration: Principles and Methods, vol. 95. Lyon, France: IARC Publication; 246–57. [15] Brenner DR, Tammemagi MC, Bull SB, Pinnaduwaje D, Andrulis IL. Using cancer registry data: agreement in cause-of-death data between the Ontario Cancer Registry and a longitudinal study of breast cancer
patients. Chronic Dis Can 2009;30:16–9. [16] Austin PC, Stuart EA. Moving towards best practice when using inverse probability of treatment weighting (IPTW) using the propensity score to estimate causal treatment effects in observational studies. Stat Med 2015;34:3661–79. [17] Brookhart MA, Wyss R, Layton JB, Stürmer T. Propensity score methods for confounding control in nonexperimental research. Circ Cardiovasc Qual Outcomes 2013;6:604–11. [18] Austin PC. The use of propensity score methods with survival or time-to-event outcomes: reporting measures of effect similar to those used in randomized experiments. Stat Med 2014;33:1242–58. [19] Austin PC. Optimal caliper widths for propensity-score matching when estimating differences in means and differences in proportions in observational studies. Pharm Stat 2011;10:150–61. [20] Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behav Res 2011;46:399–424. [21] Austin PC. Assessing balance in measured baseline covariates when using many-to-one matching on the propensity-score. Pharmacoepidemiol Drug Saf 2008;17:1218–25. [22] Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med 2009;28:3083–107. [23] Cole SR, Hernán MA. Adjusted survival curves with inverse probability weights. Comput Methods Programs Biomed 2004;75:45–9. [24] Lin DY, Wei L-J, Ying Z. Checking the Cox model with cumulative sums of martingale-based residuals. Biometrika 1993;80:557–72. [25] Aron M, Nguyen MM, Stein RJ, Gill IS. Impact of gender in renal cell carcinoma: an analysis of the SEER database. Eur Urol 2008;54: 133–42. [26] Ather MH, Nazim SM. Impact of Charlson's comorbidity index on overall survival following tumor nephrectomy for renal cell carcinoma. Int Urol Nephrol 2010;42:299–303. [27] Gettman MT, Boelter CW, Cheville JC, Zincke H, Bryant SC, Blute ML. Charlson co-morbidity index as a predictor of outcome after surgery for renal cell carcinoma with renal vein, vena cava or right atrium extension. J Urol 2003;169:1282–6. [28] Arrontes DS, Aceñero MJF, González JIG, Muñoz MM, Andrés PP. Survival analysis of clear cell renal carcinoma according to the Charlson comorbidity index. J Urol 2008;179:857–61. [29] Margel D, Urbach DR, Lipscombe LL, Bell CM, Kulkarni G, Austin PC, et al. Metformin use and all-cause and prostate cancer-specific mortality among men with diabetes. J Clin Oncol 2013;31:3069–75, http://dx.doi.org/10.1200/JCO.2012.46.7043. [30] Lega IC, Austin PC, Gruneir A, Goodwin PJ, Rochon PA, Lipscombe LL. Association between metformin therapy and mortality after breast cancer: a population-based study. Diabetes Care 2013;36:3018–26, http://dx.doi.org/10.2337/dc12-2535. [31] Austin PC, Mamdani MM. A comparison of propensity score methods: a case-study estimating the effectiveness of post-AMI statin use. Stat Med 2006;25:2084–106. [32] McKay RR, Lin X, Albiges L, Fay AP, Kaymakcalan MD, Mickey SS, et al. Statins and survival outcomes in patients with metastatic renal cell carcinoma. Eur J Cancer 2016;52:155–62.