We observed an inverse association between NSAID use and death due to colorectal cancer among Caucasian women diagnosed with proximal disease. Women who regularly used NSAIDs before diagnosis experienced approximately half the risk of colorectal cancer mortality compared to never users. These women experienced an approximate halving of risk of colorectal cancer mortality compared with never users. This association was observed to be dose-dependent, with reductions in colorectal cancer mortality of greater magnitude associated with increasing amounts of NSAID use. We did not, however, observe any evidence of an association between HT use before diagnosis and colorectal cancer survival, regardless of hormone preparation type, duration of use or tumour characteristics at diagnosis.
Previous studies provide evidence that NSAID use may be important for colorectal cancer survival after diagnosis (Fuchs and Heseltine et al, 2005
; Chan et al, 2009
; Zell et al, 2009
). The report from the California Teachers Study observed a dose-dependent improvement in colorectal cancer survival associated with NSAID use before diagnosis (Zell et al, 2009
). The NHS and Health Professionals Follow-up Study (Chan et al, 2009
) reported that any association with improved survival was limited to post-diagnostic NSAID use. However, cases from the NHS and Health Professionals Follow-up Study cohorts did not experience any significant survival benefit associated specifically with post-diagnostic use if they also reported aspirin use before their colorectal cancer diagnosis (HR: 0.89; 95% CI: 0.59–1.35). These findings suggest that NSAID use both before and after diagnosis may have an important role in altering colorectal cancer survival.
Previous studies have not reported potential differences in the association between NSAID use and colorectal cancer mortality according to tumour sub-site. The NHS and Health Professionals Follow-up Study examined potential differences according to whether the tumour was located in the colon vs
the rectum and observed no heterogeneity. Our study is the first to report results according to proximal vs
distal or rectal tumour location, making comparison of our results to previous survival studies difficult. However, previous literature has reported a site-specific effect for NSAIDs on the risk of developing incident colorectal cancer (Rosenberg et al, 1991
; Smalley et al, 1999
; Mahipal et al, 2006
), with a stronger inverse association being observed for proximal disease.
Our findings for HT are consistent with results from the Women's Health Initiative, as well as data from a large population-based study of large bowel cancer, both of which showed no overall association between HT use and colorectal cancer survival (Ritenbaugh et al, 2008
; Newcomb et al, 2009
). Previous studies that observed a statistically significant reduction in colorectal cancer mortality associated with HT often failed to account for important covariates, such as colorectal cancer screening or NSAID use in their analyses (Slattery et al, 1999
; Mandelson et al, 2003
). We did in fact observe different prevalences of both screening and NSAID use among HT users compared with never users in our study, highlighting the importance of accounting for these potential confounders in analyses to obtain an unbiased estimate of the effect of HT use before diagnosis on colorectal cancer survival.
There are several plausible mechanisms by which NSAID use before diagnosis could affect colorectal cancer mortality among women. Prostaglandin synthases COX-1 and COX-2 are directly inhibited by NSAIDs, altering prostaglandin production and cellular inflammatory responses (Sheng et al, 1997
; Brown and DuBois, 2005
; Rodriguez-Moranta and Castells, 2005
; Wang and Dubois, 2010
). This alteration of prostaglandin production results in the promotion of apoptosis, inhibition of angiogenesis and disruption of processes crucial to tumour growth (Sheng et al, 2001
; Chan, 2006
; Wang and Dubois, 2006
; Cha and DuBois, 2007
; Greenhough et al, 2009
). The anti-inflammatory effects of NSAID medications may also impact upon colorectal cancer survival directly as inflammation has been linked to both colorectal cancer incidence and cancer progression (Coussens and Werb, 2002
; Erlinger et al, 2004
; Kim et al, 2008
; Mantovani et al, 2008
), and multiple markers of systemic inflammation have been linked specifically to colorectal cancer prognosis (Ishizuka et al, 2007
; Shiu et al, 2008
; Roxburgh et al, 2009
The use of NSAIDs before diagnosis may not only decrease the likelihood of an inflammatory tumour micro-environment but also may alter the type of tumour that initially develops. The inhibition of the COX-2 pathway may be of particular importance; multiple studies have observed that COX-2 expression is a negative prognostic factor for patients with colorectal cancer and results in high colorectal tumour loads in animal studies (Masunaga et al, 2000
; Soumaoro et al, 2004
; Yamac et al, 2005
; Ogino et al, 2008
). An earlier report from the NHS observed that regular pre-diagnostic aspirin use resulted in a lower than expected incidence of tumours expressing high levels of COX-2 (Chan et al, 2007
). Use of NSAIDs before diagnosis may lead to the development of tumours expressing lower levels of COX-2
, resulting in the diagnosis of tumours that are less aggressive and have a molecular profile that improves survival.
Proximal colorectal tumours have a distinct molecular phenotype compared with distal or rectal tumours (Richman and Adlard, 2002
; Azzoni et al, 2007
; Benedix et al, 2010
) and may, therefore, represent a different form of colorectal disease. For example, variation in expression of the NSAID-targeted COX-2 enzyme according to the location of colorectal tissue has been described previously (Dimberg et al, 1999
; Wiese et al, 2003
; Nasir et al, 2004
), and may be a potential explanation for our observed tumour sub-site difference. Additionally, tumours that feature a high-degree of CpG island methylation are more likely to be found in the proximal colon, particularly among women (Jass, 2007
). Chronic inflammation has previously been hypothesised to accelerate the process of methylation in patients with ulcerative colitis (Issa et al, 2001
), providing a possible aetiologic link between inflammatory processes and proximal colorectal tumours. Although the exact molecular mechanisms behind this ‘proximal effect' are not presently clear, its implications have growing importance in light of recent reports that have shown that proximal tumours have inherently poorer prognosis than more distal tumours (Wray et al, 2009
; Benedix et al, 2010
This study's strengths included having a large, population-based sample of women and the collection of detailed exposure information for both NSAIDs and HT, including data on type, duration and frequency of NSAID medication used, as well as information on HT preparation type and duration of use. Follow-up of all enrolled colorectal cancer cases was complete and standardized. The inclusion of important covariates in our regression models allowed us to generate measures of association that were not biased by potentially strong confounders such as screening. Additionally, the availability of data on tumour characteristics at diagnosis allowed us to examine differences in hypothesised associations across tumour sub-site.
Limitations of the study include the availability of only pre-diagnostic information for NSAIDs and HT. It seems likely that pre-diagnostic medication use is correlated with post-diagnostic use, and future studies to examine these exposures independently are warranted. Information on medication use was collected after diagnosis, introducing the potential for incorrect recall, although this likely affected all cases equally and as such would result in non-differential misclassification and bias of the observed association to the null. Only limited treatment information was available in this population; however, SEER-reported first-course treatment type mirrored stage of disease at diagnosis for the majority of cases, and we accounted for extent of disease in our analyses. We did observe that after accounting for stage of disease at diagnosis, women diagnosed with rectal tumours were slightly more likely to receive treatment (data not shown). Finally, we were unable to enroll all cases immediately after diagnosis. Failure to enroll cases in a timely manner can result in patient loss that may lead to a lack of representativeness of the study cohort (Coghill et al, 2009
). However, we did not experience a high percentage of patient loss, as we were able to enroll approximately 80% of cases within 10 months of their diagnosis.
In addition to colorectal cancer-specific survival, we attempted to explore the association between the use of these common medications before diagnosis and all-cause mortality. We observed HRs similar in direction to those reported for colorectal cancer mortality, with no significant associations observed. However, the vast majority of deaths in our cohort were due to colorectal cancer, limiting our ability to investigate the relationship with mortality from other causes.
Despite the inverse association consistently observed between NSAIDs and HT in relation to colorectal cancer incidence in women, our results suggest that this same relationship did not uniformly extend to colorectal cancer mortality. Pre-diagnostic NSAID use was associated with significantly improved colorectal cancer survival among women diagnosed with proximal tumours. However, HT was not associated with colorectal cancer survival. Future studies should examine molecular characteristics of diagnosed tumours, including COX-2 expression and methylation status among regular NSAID users, and the potential interaction of NSAID use history with patient characteristics, such as genetic polymorphisms. Studies of the use of these common medications in relation to survival after a diagnosis of colorectal cancer should also be conducted among multi-ethnic cohorts of women. The practical implication of this study is that regular NSAID use earlier in life in persons at risk of colorectal cancer may not only prevent disease but also may prevent more aggressive disease. This should be considered when weighing the risks and benefits of NSAID use as a chemopreventive therapy.