Main

Cancer is the second leading cause of death among women in the United States, with over 270 000 cancer deaths in 2015 (Siegel et al, 2015). The use of 3-hydroxy-3-methyl-glutaryl-CoA reductase inhibitor medications (‘statins’ or ‘HMG Co-A reductase inhibitors’) for cholesterol reduction has been hypothesised to interfere with cancer growth and metastasis through multiple mechanisms (Fenton et al, 1992; Herold et al, 1995; Deberardinis et al, 2008; Gauthaman et al, 2009; Mannello and Tonti, 2009). Current literature on statin use and cancer survival has been mixed (Dale et al, 2006; Cholesterol Treatment Trialists’ (CTT) Collaborators et al, 2012; Murtola et al, 2014; Cardwell et al, 2015; Desai et al, 2015; Wu et al, 2015; Zhong et al, 2015), although a retrospective nationwide Danish study found a statistically significant 15% reduction in all-cancer mortality among patients who used statins before cancer diagnosis (Nielsen et al, 2012).

Given the widespread and rapidly growing statin use in the United States (Stone et al, 2014), we aimed to investigate the relationship between statin use and all-cancer survival in the Women’s Health Initiative (WHI). To our best knowledge, this is the first prospective study to investigate statins and all-cancer survival.

Materials and Methods

Design, setting, and participants

The WHI is a large, multi-centre study designed to study major causes of morbidity and mortality in postmenopausal women. The WHI includes a clinical trial (CT) and an observational study (OS) cohort, with details described previously (Hays et al, 2003). Women meeting eligibility criteria (age 50–79, postmenopausal, minimum life expectancy 3 years) were recruited at 40 US clinical centres between 1 September 1993 and 31 December 1998. Primary analyses focused on current statin users among the N=23 067 women who experienced an incident cancer during follow-up (Supplementary Figure 1).

Exposures, confounders, and classification of cases

WHI implementation details have been previously published (Anderson et al, 2003). The medication inventory was repeated at years 1, 3, 6, and 9 for the CT, and year 3 for the OS during the initial study period, which ended on 31 March 2005. The overall follow-up period for the study was through 20 September 2013. Cancers were initially verified at the local clinical centre, and then confirmed by centrally trained physician adjudicators (Curb et al, 2003).

Statistical analysis

Multivariable-adjusted Cox regression techniques were used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) by modelling two-ordered events: time from enrolment to incident cancer (secondary end point) and time from incident cancer to cancer death (primary end point). Statin use was modelled as a time-dependent categorical variable (Therneau and Grambsch, 2000), with the following levels: (0) never use, (1) current use (at the time of the latest medication inventory), (2) past use, and (3) out-of-date medication inventory (Gong et al, 2016). The primary analysis compared current (1) vs never used statins (0), to investigate the effect of regular statin use.

The Cox proportional hazard analyses were adjusted for potential confounders and included baseline covariates age, race/ethnicity, education, smoking, body mass index, physical activity, family history of cancer, current health-care provider, oral contraception use, prior unopposed oestrogen use, prior oestrogen plus progestin use, solar irradiance (latitude), prior CHD history, prior diabetes history, randomisation into the CaD trial, and age at menarche. Participants who did not die of cancer were censored at death due to other causes, last contact, or out-of-date medication collection.

All analyses were conducted using SAS software, version 9.3 (SAS Institute, Cary, NC, USA) and R software version 2.15 (R Foundation for Statistical, Vienna, Austria).

Results

In our analysis, 146 326 participants contributed 1 805 759 person-years, median (interquartile range, IQR)=14.6 (8.1–16.2) years of follow-up. A cumulative 24 404 women were diagnosed with an incident cancer (Supplementary Figure 1). Of this group, 23 067 women had additional follow-up, median (IQR) of 4.8 (1.8–9.4) years, with 7411 all-cause mortalities: 5837 cancer deaths following a diagnosis of incident cancer (78.8%), 613 cardiovascular deaths (8.3%), and 961 other causes (12.9%). After censoring the follow-up of women with out-of-date medication inventories, 3152 cancer deaths were included in the primary analysis of current vs never statin users (709 current statin users and 2443 non-users). Table 1, and Supplementary Tables 1 and 2 display baseline characteristics.

Table 1 Participants’ characteristics in the Women’s Health Initiative Clinical Trial and Observational Study by statin use (current vs never)a at time of cancer diagnosis (N=17 285b)

Current statin use was associated with lower risk of cancer death compared with never-use (HR, 0.78; 95% CI, 0.71–0.86; P<0.001; Figure 1), and lower risk of all-cause mortality (HR, 0.80; 95% CI, 0.74–0.88). The lower risk of cancer death associated with statin use did not depend on statin potency (P-interaction (int)=0.22), category (P-int=0.43), type (P-int=0.34), or duration (P-int=0.33). Other lipid-lowering medications, used alone, were associated with a similar reduction in cancer deaths compared with monotherapy statin use (P-int=0.57). Prior statin users were not at lower risk of cancer death compared with never-users (HR, 1.06; 95% CI, 0.85–1.33).

Figure 1
figure 1

Number of cancer deaths (annualised %) and multivariable-adjusted HR (95% CI) for statin use (current vs never).Annualised percentages by exposure group (time dependent) were computed by dividing the total number of cancer deaths, by the corresponding cumulative person-time since cancer diagnosis, for each exposure group. Cox regression models were adjusted for age at baseline, race/ethnicity, education, smoking, body mass index (BMI), physical activity, family history of cancer, current health-care provider, oral contraception use, prior unopposed oestrogen use, prior oestrogen plus progestin use, solar irradiance (latitude), prior CHD history, prior diabetes history, randomisation into the CaD trial, age at menarche, and stratified by age group, study groups (randomisation arms of the HT trials, DM trial, and OS enrolment) and enrolment in WHI extensions (I/II). *Annualised percentage. **Significance test of the main effect or test of heterogeneity between non-referent exposure groups. †Test of heterogeneity between atorvastatin, simvastatin, lovastatin, and pravastatin. ‡Analysis of statin duration included only CT participants; at the time of cancer diagnosis, among current statin users (n=2218), 893 (40.3%) used statins <3 years, 539 (24.3%) 3 to <5 years, 593 (26.7%) 5 to <10 years, and 193 (8.7%) 10+ years. Test of heterogeneity based on a 1 degree-of-freedom test for trend. ^Test of heterogeneity between statin use and other use.

Statin use was associated with a significantly lower risk of multiple, but not all cancer types (P-int=0.001; Figure 2). With the exception of current NSAID use (which attenuated the effect of statins), current statin use was not modified by any other subgroups (Supplementary Figure 2).

Figure 2
figure 2

Number of cancer deaths (annualised %) and multivariable-adjusted HR (95% CI) for statin use (current vs never) by cause of cancer death.Summary statistics are from a Cox regression model, using cause-specific baseline hazard functions, with the covariate adjustments described above. *Corresponds to a significance test of the main effect, or an 11-df test of heterogeneity for cause of cancer death. To avoid double counting, test of heterogeneity is between the main causes of death listed and does not include subtypes (i.e., cancer of the pancreas or other digestive organs, non-Hodgkins lymphoma, and leukaemia). ^Only participants without a baseline hysterectomy were used to compute the number of cases and annualised rates. There was one endometrial cancer case among the group of no statin use that reported having had a hysterectomy.

In a secondary analysis on cancer incidence, beginning at enrolment, statin use was not associated with a reduction of incident cancer (HR, 0.96; 95% CI, 0.92–1.001; P=0.056).

Discussion

In this prospective cohort study, we found that current statin use in postmenopausal women with cancer was associated with lower risk of cancer death. Use of other lipid-lowering medications was also associated with a lower risk of cancer death; this finding suggests that a reduction in circulating cholesterol levels may mediate increased cancer survival. However, a dose–response relationship was not found, suggesting that results should be interpreted cautiously. Multiple molecular mechanisms have been linked to statins and cancer, including the mevalonate pathway (Fenton et al, 1992; Herold et al, 1995; Deberardinis et al, 2008; Boudreau et al, 2010), G-proteins (Wong et al, 2002; Demierre et al, 2005), isoprenoid-mediated suppression (Wong et al, 2002), the RAF-mitogen-activated protein kinase 1 pathway (Wu et al, 2004), and anti-angiogenic properties of statins (Weis et al, 2002).

Comparison with other studies

Our study results are similar to findings from a retrospective nationwide Danish study, which found a statistically significant 15% reduction in all-cancer mortality among patients who used statins before cancer diagnosis (Nielsen et al, 2012). Our analysis is additionally strengthened by its prospective format, ethnically heterogeneous population, and covariate data. In addition, other studies of specific cancers in women (including breast and uterine) have suggested the protective effects of statins on cancer mortality and survival (Murtola et al, 2014; Cardwell et al, 2015; Nevadunsky et al, 2015). Also, some prospective cohort studies of cardiovascular disease have found associations between lower cholesterol levels and lower risk of death from several cancers (Cambien et al, 1980; Kagan et al, 1981; Keys et al, 1985).

However, several smaller studies and RCTs have found no significant associations. A meta-analysis of 27 RCTs in the Cholesterol Treatment Trialists’ (CTT) Collaborators (2015) database did not find an association with cancer mortality or incidence; however, the study used the time from randomisation rather than the time from incident cancer. A meta-analysis of 26 RCTs also found no effect of statin use on cancer incidence or survival (Dale et al, 2006). These conflicting results suggest the need for additional prospective studies and larger RCTs, particularly with cancer survival as the primary outcome.

Cancer incidence has been studied more extensively than cancer survival in relation to statin use, but results on the subject have been mixed (Bjerre and LeLorier, 2001; Bonovas et al, 2005; Browning and Martin, 2007; Kuoppala et al, 2008; Haukka et al, 2010; Jagtap et al, 2012; Simon et al, 2012; Desai et al, 2013; Singh and Singh, 2013; Tan et al, 2013). Similarly, biomarker-based studies of statins as candidate breast cancer chemoprevention agents have shown mixed results (Higgins et al, 2012; Bjarnadottir et al, 2013; Vinayak et al, 2013).

Strengths and limitations

The strengths of this study include its prospective nature, large size and geographic distribution, adjudication of cancer cases, and detailed information on confounders and exposures. Limitations of the study include the fact that medication use was not continuously updated, observational format, and majority Caucasian participants. The study may not be generalisable to populations other than postmenopausal women with similar age at cancer diagnosis. Residual confounding or reverse causation bias may be present.

Healthy user bias due to socioeconomic factors may also present in this analysis; however, this should be reduced as WHI data allow for rich covariate adjustment. In addition, the effect of statin use on cancer survival remained significant even with sensitivity analyses including tumour stage, physical functioning, and mammogram use.

Conclusions

In conclusion, in a prospective cohort of postmenopausal women, current use of statins and other cholesterol-lowering medications was associated with increased all-cancer survival, as well as increased survival of multiple cancer types. These findings, along with the previous Danish cohort study, suggest that statin use and/or lower cholesterol levels may have a protective effect on cancer death. Further research is needed in an RCT format to better control for healthy user bias and to study cancer as a primary outcome.