Objective Non-steroidal anti-inflammatory drug (NSAID) use decreases both the incidence of colorectal cancer and recurrence of adenomas among patients with prior colorectal neoplasia. However, few studies have investigated the association between NSAID use and colorectal cancer-specific survival. The role of prediagnostic NSAID use was therefore examined in relation to colorectal cancer-specific survival among cases from the Seattle Colon Cancer Family Registry (Seattle Colon CFR).
Methods This was a follow-up study that included incident cases of colorectal cancer from the Seattle Colon CFR. Cases were aged 20–74, diagnosed from 1997 to 2002, and were identified using the population-based Puget Sound SEER registry. Detailed information on history of NSAID use, including type, recency and duration, was collected through an interviewer-administered questionnaire. Follow-up for mortality was completed through linkages to the National Death Index. The main outcome measure was death due to colorectal cancer after diagnosis. Cox proportional hazards regression was used to investigate the relationship between prediagnostic NSAID use and colorectal cancer-specific mortality among cases.
Results NSAID use prior to colorectal cancer diagnosis was associated with an ∼20% lower rate of colorectal cancer mortality after diagnosis compared with never use (HR 0.79; 95% CI 0.65 to 0.97). This relationship appeared to be duration dependent, with longer reported use prior to diagnosis associated with lower rates of colorectal cancer mortality among cases. The most pronounced reductions in mortality were observed among cases diagnosed with proximal disease (HR 0.55; 95% CI 0.37 to 0.82), whereas no association was observed between NSAID use prior to diagnosis and colorectal cancer-specific mortality among cases diagnosed with distal or rectal disease.
Conclusions The findings suggest that regular use of NSAIDs prior to diagnosis is associated with improved colorectal cancer survival, particularly among cases diagnosed with proximal disease and in longer term NSAID users.
- Colorectal cancer mortality
- epidemiology, colorectal cancer, non-steroidal anti-inflammatory druga
Statistics from Altmetric.com
- Colorectal cancer mortality
- epidemiology, colorectal cancer, non-steroidal anti-inflammatory druga
Significance of this study
What is already known about this subject?
Regular NSAID use is associated with a lower risk of developing colorectal cancer and a lower rate of adenoma recurrence.
NSAIDs are hypothesised to alter colorectal cancer risk through inhibition of prostaglandin synthesis and reductions in cellular inflammation.
Recent observational studies provide evidence that NSAID use may also be associated with improved survival after a diagnosis of colorectal cancer.
What are the new findings?
This study, the largest to date investigating the role of NSAID use prior to a colorectal cancer diagnosis in survival among cases, provides evidence that NSAID use is associated with improved colorectal cancer survival.
This association may be duration dependent, with longer term NSAID users experiencing lower rates of colorectal cancer mortality after diagnosis than short-term users.
The association between NSAID use prior to diagnosis and improved colorectal cancer survival among cases appears to be limited to cases diagnosed with proximal disease.
How might this impact clinical practice in the foreseeable future?
Regular NSAID use earlier in life in persons at risk of colorectal cancer may not only prevent disease but may also prevent more aggressive disease or alter the course of disease after diagnosis.
This should be considered when weighing the risks and benefits of NSAID use and making recommendations for use to patients in a clinical setting.
The National Cancer Institute reported a 5-year survival rate for colorectal cancer of 67.3% in 2002, representing an improvement of only ∼5% over the previous decade.1 Evidence-based recommendations are needed to identify factors that can improve survival among colorectal cancer cases. NSAIDs (non-steroidal anti-inflammatory drugs) are commonly used medications in the USA for treating pain and reducing inflammation, and are available without a prescription.2 NSAID use has been associated with decreased incidence of colorectal cancer3–8 and lower risk of developing fatal colorectal tumours.9–12 Use among individuals previously diagnosed with colorectal cancer is associated with a decreased risk of adenoma recurrence.13 14 However, only three observational studies to date have reported on the association between NSAID use and survival among colorectal cancer cases.15–17 Only one investigated non-aspirin NSAID use, and that study was only able to include women.17 Of the remaining two studies, one was restricted to patients with stage III colorectal cancer concurrently enrolled in a randomised chemotherapy trial.16 The other included both male and female patients with stage I–III colorectal cancer.15 Although findings regarding the importance of the timing of use were not consistent across these studies, all three provided evidence that use of NSAIDs was inversely associated with mortality among colorectal cancer cases.
We investigated the association between prediagnostic aspirin and ibuprofen use (individually and combined) and colorectal cancer survival after diagnosis in a population-based sample of cases identified through the Seattle Colon Cancer Family Registry (Seattle Colon CFR). This study was the largest to date addressing this question, enrolled cases representing all stages of disease as well as both men and women, and collected data on tumour characteristics not investigated in previous studies, including microsatellite instability (MSI) status and tumour subsite. It was hypothesised that NSAID use prior to diagnosis would be inversely associated with the rate of colorectal cancer-specific mortality among cases.
We investigated our hypothesis using colorectal cancer cases ascertained from the Seattle Colon CFR. Details of case recruitment have been presented elsewhere.18 Incident, invasive colorectal cancer cases were identified from the Puget Sound Surveillance, Epidemiology and End Results (SEER) Registry. All cases aged 20–74 diagnosed between 1997 and 2002 in the Puget Sound SEER counties were eligible for study participation (n=2551). Permission to contact cases from physicians and patient consent to be interviewed was obtained for 90% of the eligible cohort (n=2290). A total of 553 cases were excluded due to: death prior to interview (n=169), withdrawal from the study (n=77), completion of only a partial interview (n=21) or loss to follow-up prior to interview (n=286). Therefore, 1737 colorectal cancer cases were eligible and completed the interview. Cases were enrolled an average of 8 (95% CI 3 to 13) months after diagnosis. Informed consent was obtained from all participants, and the study was approved by the Institutional Review Board at the Seattle Colon CFR study site, the Fred Hutchinson Cancer Research Center.
Exposure and covariate assessment
Information on known and potential colorectal cancer risk factors was collected through an interviewer-administered questionnaire at study enrolment. Interviewer-collected data included family history of colorectal cancer, lifetime smoking and alcohol consumption history, height and weight 2 years prior to diagnosis, colorectal cancer screening history, existence of prior inflammatory or co-morbid conditions, and demographic characteristics such as age, race and sex. The questionnaire collected information on type of NSAID (aspirin or ibuprofen) used, recency of use relative to a reference date (current, former, never) and amount used (average daily dose, duration of use in years).
For exposure definitions, a reference date ∼2 years before the case's diagnosis of colorectal cancer was selected. Regular use was defined as use at least twice per week for ≥1 month. ‘Ever use’ was defined as regular use at any point prior to the selected reference date. For cases who reported ever using NSAIDs regularly, each patient was asked whether he/she was taking NSAIDs at the time 2 years prior to his/her colorectal cancer diagnosis. If yes, the case was recorded as using NSAIDs at the time 2 years before diagnosis (current user); if no, the case was considered to only have used NSAIDs at a point in time >2 years prior to diagnosis (former user). ‘Never users’ reported no use or use below the regular use threshold prior to the reference date. Information on average daily dose (measured in ‘times pills taken daily’) among regular users and the duration of regular use in years was also collected. The duration of regular NSAID use was divided into quartiles (first, 0–6 months; second, 6 months–2.5 years; third, 2.5–7 years; fourth, ≥7 years).
First-course treatment and tumour characteristics at the time of diagnosis, including stage, subsite and MSI status, were obtained from SEER reports. Advanced disease was defined as colorectal cancer with distant metastasis (n=202); non-advanced disease included localised and regional stage disease (n=1526). Subsite of disease was categorised using ICD10 (International Classification of Diseases, 10th revision) codes: proximal disease (C18.0–C18.5); distal disease (C18.6–18.7); and rectal disease (C19.9, C20.9 and C21.8). MSI status was assessed using 10 separate markers. Based upon established guidelines,19 cases were classified as MSI-stable if 0% of loci were unstable, MSI-low if 0–40% of loci (≤3 markers) were unstable and MSI-high if ≥40% of loci (≥ 4 markers) were unstable, with unequivocal results for at least four markers required to characterise MSI status.
Vital status was ascertained through linkage to the National Death Index (NDI) records to obtain date and cause of death; causes of death were classified using ICD10 codes.20 The NDI identifies known deaths with a high degree of sensitivity, validity and completeness.21 The primary outcome of interest was mortality from colorectal cancer, assessed from underlying cause of death obtained from the NDI. Time to colorectal cancer mortality was evaluated from Seattle Colon CFR records of date of colorectal cancer diagnosis and NDI-recorded date of death. Patients alive and free of end point at the time of their last known vital assessment were administratively censored at that date, with the most recent vital status linkage occurring in January 2010. Patients dying of causes other than colorectal cancer were administratively censored at their recorded date of death.
Kaplan–Meier survival curves were generated comparing NSAID ‘ever use’ with ‘never use.’ Differences between survival curves were evaluated using the log-rank test. The proportional hazards assumption was evaluated both graphically and statistically, using Schoenfeld residuals.22 Cox proportional hazards regression was used to estimate HRs and 95% CIs for the association between prediagnostic NSAID use and colorectal cancer-specific mortality after diagnosis. To account for lag time between diagnosis and study enrolment, staggered entry was used such that cases did not contribute follow-up time to the analysis until completion of the baseline interview.
Cox regression models were restricted to Caucasian cases (n=1549) only due to sparse numbers of events among persons of other races that led to a lack of positivity in adjusted models. To increase comparability with previous studies that excluded metastatic disease,15 we also reported HR estimates restricted to cases diagnosed with non-advanced disease (n=1379). Cox regression models included covariates from the following list of potential confounders: age, sex, body mass index (BMI), smoking status, history of diabetes, prior inflammatory conditions (ulcerative colitis, Crohn's disease), receipt of preventive colorectal screening (sigmoidoscopy or colonoscopy received at least 2 years prior to diagnosis of colorectal cancer), first-course treatment, and stage of disease at diagnosis. Models without stage at diagnosis and first-course treatment are presented due to the potential for stage to be in the causal pathway between NSAID use prior to diagnosis and mortality after diagnosis. Results from models including both stage and treatment are also presented. The multivariable adjusted Cox regression models were evaluated across strata of tumour subsite, MSI status and BMI; potential statistical interactions between NSAID use and stratification variables were investigated by inclusion of interaction terms in the model. Additionally, exploratory analyses were run among non-Caucasian cases. All statistical analyses were performed using SAS software, version 9.1 (SAS Institute); all p values reported are two-sided.
Mortality occurring in the lag time between diagnosis and interview did not contribute to our regression models. Therefore, we conducted a sensitivity analysis to see if the association between NSAID use prior to diagnosis and colorectal cancer survival varied by time between diagnosis and interview. A series of analyses was performed to simulate failure to enrol cases in a timely manner by restriction of case groups to patients enrolled after 6 months, 8 months and 1 year, respectively. Results were used to determine if inclusion in our analyses of only cases that survived long enough after diagnosis to be interviewed could have resulted in enough lost information to introduce longevity bias. It is possible that cases who were not enrolled in a timely manner, possibly due to death from severe disease, were distinct from cases who survived long enough to be interviewed and that the effect of prediagnostic NSAID use therefore differed according to enrolment lag time.
Approximately 50% of cases reported ever using NSAIDs regularly prior to colorectal cancer diagnosis (table 1). ‘Ever users’, compared with ‘never users’, were more likely to be older, report ever smoking, have received preventive colorectal screening, have diabetes and be obese. A larger proportion of ‘ever users’ compared with ‘never users’ were also more likely to be diagnosed with localised colorectal tumours.
After an average of 8 years of follow-up after colorectal cancer diagnosis, 707 deaths from any cause, including 462 deaths due to colorectal cancer, were ascertained among eligible and enrolled cases. Colorectal cancer survival curves based on NSAID use are presented for all enrolled cases in figure 1. ‘Ever users’ experienced improved colorectal cancer survival compared with ‘never users.’ This association of NSAID use with improved survival was observed among cases with proximal disease (log rank p value=0.04) but not cases with distal or rectal disease (log rank p value=0.41). The proportional hazards assumption was not significantly violated.
Among all colorectal cancer cases, ‘ever use’ of NSAIDs prior to diagnosis was associated with a 21% (95% CI 3% to 35%) lower rate of colorectal cancer mortality after diagnosis compared with ‘never use’ (table 2). Regular use 2 years prior to diagnosis (current use) was associated with a similar decrease in the hazard of colorectal cancer mortality among all cases (HR 0.79; 95% CI 0.62 to 0.99). The effect estimate for former use (use >2 years prior to diagnosis) was also consistent with an inverse association, although the 95% CI included the null value (HR 0.78; 95% CI 0.60 to 1.01). Inclusion of stage in regression models or restriction to cases with non-advanced disease did not change the interpretation of results, with reductions in colorectal cancer mortality after diagnosis associated with NSAID use prior to diagnosis of ∼20–30% reported consistently. However, HR estimates were attenuated after stage adjustment, and CIs for cases with non-advanced disease included the null value.
Investigation of duration of prediagnostic NSAID use in years (using quartiles of duration) suggested a duration-dependent relationship (table 3), with lower colorectal cancer-specific mortality observed among cases reporting longer durations of use (p trend=0.03), although very long-term users (>7 years) did not follow the trend and may instead represent very sick patients with colorectal cancer. In an exploratory model investigating ‘NSAID use in years’ and ‘average number of times pills taken daily’ as distinct exposures, with each centred around its mean, each additional year of NSAID use prior to diagnosis was associated with a significantly lower hazard of colorectal cancer mortality; however, after accounting for duration of use, the average daily dose taken did not appear to be associated with colorectal cancer survival (data now shown).
Estimates for aspirin use were remarkably similar to those reported for overall NSAID use across all stages of disease, with observed reductions in colorectal cancer mortality risk after diagnosis of ∼20–30%. In contrast, estimates for ibuprofen use were not consistent with an improvement in colorectal cancer survival. When each medication type was examined independently of the other, aspirin was observed to decrease colorectal cancer mortality risk significantly among all cases (HR 0.74; 95% CI 0.60 to 0.92), while ibuprofen use results were null (table 2). The patterns of regular use for aspirin and ibuprofen also differed substantially. Among cases using aspirin, ∼85% reported taking pills no more than once per day; >50% reported at least 2 years of regular use, with an average duration of 5.8 years. In contrast, only 30% of cases using ibuprofen reported use for >2 years, with an average duration of 2.8 years. Additionally, ∼60% of ibuprofen users reported taking pills at least twice per day.
Heterogeneity in the role of prediagnostic NSAID use in relation to colorectal cancer survival was observed across strata of tumour subsite (p interaction=0.08). ‘Ever use’ of NSAIDs prior to diagnosis was associated with a 45% (95% CI 18% to 63%) lower rate of colorectal cancer mortality among cases with proximal tumours (table 4). In contrast, no evidence of an association was observed among cases diagnosed with distal or rectal tumours (HR 1.00; 95% CI 0.72 to 1.41). Duration of NSAID use prior to diagnosis appeared to be important for proximal disease, with longer durations associated with greater reductions in the hazard of colorectal cancer mortality after diagnosis (p trend=0.03) (table 3). Estimates reported in table 4 were consistent with those unadjusted for stage as well as those observed for patients diagnosed with advanced disease (data not shown).
HR estimates were imprecise across strata of MSI status, particularly among MSI-low and MSI-high cases, due to limited sample size, but no statistical evidence of heterogeneity was observed (p interaction=0.57). No statistical significance for the association between NSAID use prior to diagnosis and colorectal cancer mortality after diagnosis was observed in any MSI subgroup. Exploratory analyses revealed some suggestion of heterogeneity between strata of BMI (data not shown). Reductions in colorectal cancer mortality after diagnosis appeared to be limited to non-obese cases. However, results were not consistent by type of NSAID, with this pattern only being observed among regular aspirin and not ibuprofen users.
Among non-Caucasian cases, we observed 30 total events, with a suggestion that NSAID users may actually experience elevated mortality, in contrast to results observed among the Caucasian cases (p interaction for NSAID effect and race=0.10). However, among African American cases, who accounted for 20 of the 30 observed events, the prevalence of obesity was 40%, high cholesterol 50% and smoking 80%, raising the possibility that these cases used NSAIDs for other serious indications that may in fact have led to elevated mortality rates. These subgroup results need to be confirmed in future studies designed to investigate these specific hypotheses.
The sensitivity analyses run with models restricted to cases with increasing lag times between diagnosis and enrolment did not reveal any evidence of substantive bias, producing effect estimates consistent with those reported here (data not shown).
Regular NSAID use prior to a diagnosis of colorectal cancer was associated with an ∼20% lower rate of colorectal cancer-specific mortality among cases. This inverse association was most pronounced among cases with proximal tumours, with reductions in the hazard of colorectal cancer mortality of ∼45%. Our results also suggest that the improved colorectal cancer survival associated with prediagnostic NSAID use may increase with longer duration of use prior to diagnosis.
Our findings are consistent with previous studies that observed an inverse association between NSAID use and colorectal cancer mortality. A study done using patients with stage III colorectal cancer (n=846) observed that consistent aspirin use, measured both prior to diagnosis and after treatment initiation, was associated with a reduced risk of cancer recurrence and death.16 A recent report from the California Teacher's Study (n=621) observed a significant decrease in the rate of colorectal cancer mortality among cases associated with NSAID use prior to diagnosis.17 Chan and colleagues from the Nurses Health Study and Health Practitioners Follow-up Study (n=1279) observed that aspirin use after a colorectal cancer diagnosis was associated with a reduced risk of colorectal cancer-specific mortality among cases.15 These studies suggest that both prediagnostic and postdiagnostic NSAID use may be important for colorectal cancer survival.
Previous studies have either not investigated the role of non-aspirin NSAIDs in relation to colorectal cancer survival15 16 or have only reported results combined for overall NSAID use.17 Our results by type of NSAID appeared to suggest that aspirin use may be more important than ibuprofen use for colorectal cancer survival, but the limited number of cases regularly using ibuprofen precluded any meaningful inference regarding its role in colorectal cancer mortality among cases. Patterns of regular NSAID use by type observed suggest that ibuprofen users were shorter term, higher dose users compared with cases regularly taking aspirin, who medicated at lower doses on average albeit for substantially longer periods of time. The difference in the pattern of use and potentially not the type of NSAID may be the crucial factor in determining effectiveness against colorectal cancer mortality. The importance of considering the cumulative pattern of exposure is further supported by our observation that increasing duration of prediagnostic use in years was associated with greater reductions in colorectal cancer mortality after diagnosis.
Observed differences in the association of NSAID use with colorectal cancer survival by tumour subsite are consistent with literature suggesting that NSAIDs' chemopreventive effects against colorectal cancer incidence may be more pronounced for proximal disease.23–26 This potential site-specific NSAID effect for colorectal cancer survival has not been investigated to date, and replication in future studies is needed. Although the exact reasons for a subsite difference are not clear, proximal tumours do have a distinct molecular and genetic profile compared with distal or rectal tumours,27–29 and studies have suggested that levels of cyclo-oxygenase 2 (COX-2), a target for NSAIDs, may vary by location in colorectal tissue.30–32 The question of whether NSAIDs plays a different role in colorectal cancer survival depending upon tumour subsite is of increasing importance, with recent reports of inherently poorer prognosis for proximal tumours28 33 and ineffectiveness of colorectal screening in decreasing mortality for proximal disease.34 35
The stage distribution at diagnosis among NSAID users in our study included a higher proportion of localised disease compared with ‘never users’. NSAID users had higher screening rates than non-users, so we investigated whether the shift towards earlier stage of disease at diagnosis was due to screening practices. Among both cases who did and did not receive preventive screening, NSAID users presented with higher percentages of localised disease compared with ‘never users’. This suggests that the association of NSAID use with earlier stage at diagnosis may be mediated not entirely through screening practices but in part through slowing the progression of the developing tumour. Thus, stage may be in the causal pathway between NSAID use and colorectal cancer mortality, resulting in attenuation of the association between NSAID use and colorectal cancer mortality after accounting for stage in regression models.
The anti-inflammatory and chemopreventive effects of NSAIDs are mediated through direct inhibition of COX-1 and COX-2.36–41 Inflammation plays a well-documented role in the initiation of colorectal neoplasia42–45 and is also important for cancer progression,46–48 and levels of inflammatory markers have been demonstrated to be prognostic for survival in patients with colorectal cancer.43 49–51
Some of the earliest reports of the chemopreventive effects of NSAIDs noted the association of these medications with decreased incidence not just of colorectal tumours but specifically of fatal tumours.9–12 More recently, investigators from the Nurses Health Study (NHS) reported that regular prediagnostic aspirin use resulted in a lower than expected incidence of tumours expressing high levels of COX-2.52 Elevated COX-2 expression in tumours has been demonstrated to be associated with tumour metastasis and to be a negative prognostic factor in patients with colorectal cancer.37 41 53–58 Use of NSAIDs prior to diagnosis may therefore lead to development of less aggressive tumours expressing lower levels of COX-2, a molecular profile that improves survival. NSAID use may not only result in the formation of inherently less aggressive tumours but may also contribute to slower disease progression for tumours of all grades through direct effects on the tumour microenvironment. COX enzyme-mediated mechanisms have been linked to the ability of tumours to initiate vascularisation55 59 and angiogenesis,60 most probably via direct effects on prostaglandin production.38 61 The prostaglandin pathway may also regulate cellular apoptosis62 63 and responses to growth factors such as transforming growth factor β (TGFβ).64 Ultimately, NSAID use prior to diagnosis may facilitate the development of less aggressive tumours and anti-inflammatory effects on the microenvironment of the initiating tumour that can alter colorectal tumour progression.
Our study has several important strengths. Study participants were drawn from a population-based registry, suggesting good generalisability. Our study is the largest to date in the literature addressing NSAID use and colorectal cancer survival, and includes cases diagnosed at all stages of disease. Follow-up and outcome ascertainment among enrolled cases was complete and standardised. The collection of detailed exposure information, including type, dose and duration of use, allowed us to investigate patterns of exposure not previously addressed in the literature. In addition, our study did not suffer from long lag times between diagnosis and study enrolment that can result in patient loss and limit interpretation of study results, as we have recently suggested.65 Finally, inclusion of data on important tumour characteristics allowed us to investigate subsite-specific effects not addressed in previous studies.
This study is not without limitations. NSAID use histories were ascertained after the diagnosis of colorectal cancer, raising concerns of recall bias. However, we have no reason to believe that cases reported NSAID use differently based upon mortality status that was unknown at the time of study enrolment. This non-differential misclassification probably biased our results to the null rather than introduced a false association. Only prediagnostic NSAID use information was routinely collected in the study population; it was therefore not possible to examine effects of postdiagnostic use. In a sample of cases for which we had additional data at 5 years postdiagnosis, approximately one-third of cases changed their NSAID exposure status (data not shown), although the pattern was not predictable and further investigation is warranted.
In addition to colorectal cancer-specific survival, we attempted to explore the association between NSAID use prior to diagnosis and all-cause mortality among colorectal cancer cases. We observed HRs of similar direction to those reported for colorectal cancer mortality but reduced in magnitude, with no significant associations observed. However, our ability to investigate this relationship was limited because the vast majority of deaths in our cohort were due to colorectal cancer. Finally, a proportion of eligible cases were lost prior to enrolment. We did not have covariate information for the majority of these cases. However, among a subgroup of cases who completed a baseline interview but later withdrew from the study (n=42), the NSAID use pattern appeared similar to that of cases included in our analyses. Cases lost from the study however were younger and had a higher BMI on average compared with cases in the study. Lost cases also had a slightly higher percentage of proximal and localised disease at diagnosis.
Our results suggest that regular, prediagnostic NSAID use is associated with improved colorectal cancer survival, particularly among cases diagnosed with proximal tumours. This improvement in survival could be realised through NSAIDs altering the profile of the developing tumour or through anti-inflammatory effects on the developing tumour's microenvironment. The reductions in colorectal cancer mortality among cases appeared to be greater for cases reporting longer durations of NSAID use prior to their diagnosis. Future studies should investigate the biological changes that occur in tumours of NSAID users, including variations in COX-2 expression. The importance of the timing and dose of NSAID use should also be established so that evidence-based recommendations can be made to improve survival after diagnosis among colorectal cancer cases.
Funding This work was supported by the National Cancer Institute at the National Institutes of Health (grants R25 CA094880 to SVA, T32 CA09168-32 to ANBH, R03 CA137791 and U24 CA074794).
Competing interests None.
Ethics approval This study was conducted with the approval of the Fred Hutchinson Cancer Research Center/Colon Cancer Family Registry IRB.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.