Maintenance use of non-steroidal anti-inflammatory drugs and risk of gastrointestinal cancer in a nationwide population-based cohort study in Sweden.

OBJECTIVES: Aspirin and other non-steroidal anti-inflammatory drugs (NSAIDs) are potential candidates for chemoprevention of gastrointestinal cancer. We aimed to assess the association between contemporary NSAID use (≥180 days) and gastrointestinal cancer.
DESIGN: Nationwide Swedish population-based cohort study (2005-2012).
SETTING: Sweden PARTICIPANTS: All adults exposed to maintenance NSAIDs use (aspirin, n=783 870; unselective NSAIDs, n=566 209, selective cyclo-oxygenase (COX)-2 inhibitors, n=17 948) compared with the Swedish background population of the same age, sex and calendar period. OUTCOME MEASURES: The risk of different gastrointestinal cancer types expressed as standardised incidence ratios (SIR) and 95% CIs, taking into account concurrent proton pump inhibitors (PPIs) and statins usage.
RESULTS: The SIR for gastrointestinal cancer for aspirin use was 1.02 (95% CI 1.00 to 1.04), with clearly reduced risk for long-term users (SIR=0.31, 95% CI 0.30 to 0.33 for 5.5-7.7 years), but an increased risk for short-term users (SIR=2.77, 95% CI 2.69 to 2.85), and stronger protective effect for low-dose aspirin (SIR=0.86, 95% CI 0.85 to 0.88). Users of non-selective NSAIDs showed an overall decreased risk of gastrointestinal cancer (SIR=0.79, 95% CI 0.77 to 0.82), in particular for cancer of the stomach, colorectum and oesophagus, and the SIRs were further decreased among long-term users. Users of selective COX-2 inhibitors showed a SIR=0.89 (95% CI 0.73 to 1.09) for gastrointestinal cancers. Both aspirin and unselective NSAIDs users who also were using PPIs, had higher risks for all gastrointestinal cancer types; and lower risk if using statins.
CONCLUSION: Long-term use of (low-dose) aspirin and non-selective NSAIDs was associated with a decreased risk of all gastrointestinal cancer types. © Author(s) (or their employer(s)) 2018. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ.

Population-based and nationwide design based on contemporary use of aspirin and other non-steroidal anti-inflammatory drugs (NSAIDs)—resulting in sufficient power to assess less common types of gastrointestinal cancer, and different formulations of NSAIDs.

Concurrent maintenance use of statins and proton pump inhibitors is assessed.

This study is based on real-life user information because of the population-based design, which leads to inherent problems of confounding by indication and reverse causality that were taken into account in the design and analyses.

The findings are standardised for age, sex—which are often described as the major confounding factors in epidemiological studies—and calendar time.

Other confounders could not be taken into account because the information was not available for the total background population.

Exposure information is based on the Swedish Prescribed Drug Registry, which was initiated in July 2005 and has a complete nationwide coverage.

Introduction

Inflammatory processes in tumour tissue are likely to contribute to tumour progression, immunosuppression and facilitate tumour growth, and cancer susceptibility and severity may also depend on different inflammatory responses.1 Therefore, aspirin and other non-steroidal anti-inflammatory drugs (NSAIDs) are among the most promising candidates for chemoprevention of cancer, in particular tumours of the gastrointestinal tract. Despite the different indications for maintenance use of aspirin and non-aspirin NSAIDs, the underlying mechanisms are similar.2 3 NSAIDs inhibit cyclo-oxygenase (COX), an enzyme responsible for the formation of thromboxane (a lipid acting as a vasoconstrictor, which also facilitates platelet aggregation) and prostaglandins (a messenger molecule in the inflammatory pathway); yet only aspirin permanently inhibits platelet formation.4 5 There are two types of NSAIDs, inhibiting both COX-1 and COX-2, or only COX-2. COX-1 is expressed in most tissues regulating many physiological processes.6 By inhibiting prostaglandin synthesis, NSAIDs compromise gastroduodenal defence mechanisms, including reducing blood flow and mucus and bicarbonate secretion, which may lead to dyspepsia and peptic ulcers, for which proton pump inhibitors (PPIs) are often prescribed as prevention or treatment.5 6 COX-2 is expressed at sites of inflammation, and is the actual target of NSAIDs.6 In contrast to non-selective COX inhibitors (ie, aspirin and most other NSAIDs), COX-2 selective inhibitors or cyclo-oxygenase-2 inhibitors are also weakly acidic, and therefore avoid substantial accumulation in (and damage of) the gastric mucosa.6 Clinical studies have shown similar anti-inflammatory effects, a lower risk of gastrointestinal toxicity, yet a higher risk of cardiovascular morbidity for COX-2 selective inhibitors compared with non-selective COX inhibitors.3 7 8 Some of the older NSAIDs are ‘relatively selective COX-2 inhibitors’, that is, nabumetone, meloxicam, etodolac and nimesulide.3

However, epidemiological evidence to support a chemopreventive effect is still limited, mainly because large numbers are needed with a long follow-up, in particular for relatively rare cancer types. Meta-analyses have pooled the evidence of the gastrointestinal cancer preventive potential of aspirin and other NSAIDs.9–14 A large meta-analysis15 and another detailed scientific assessment16 concluded that a preventive effect on colorectal cancer was especially pronounced in daily and long-term users (>5 years) in both interventional and observational studies,15 16 with similar findings in recent studies on gastric cancer.17 18 Yet, these studies used several different definitions of exposure, ranging from a single prescription of aspirin to daily use for >5 years, with too few studied reporting stratified analyses per dosage (or indication, eg, low-dose anticoagulants vs high-dose analgesics) to draw reliable conclusions (although low dose has been recommended by individual studies).15 The statistical power was too low to identify associations with many other types of (gastrointestinal) cancer, and more, large original studies are needed to assess the potential preventive effect of other NSAIDs.15

The role of PPI use on the association between NSAIDs with gastrointestinal cancer is insufficiently understood yet increasingly investigated, with growing evidence of carcinogenic and other long-term side effects of PPIs19–22 as also shown by our group.23–25

The objective of this study was to assess the association of aspirin and other NSAIDs on the risk of different gastrointestinal cancer types, while also assessing the potential influence of concomitant PPI use.

Materials and methods

This nationwide Swedish population-based cohort study assessed the risk of gastrointestinal cancer in adult NSAIDs users,26compared with the risk in the entire Swedish background population of the corresponding sex, age and calendar year (7.1–7.6 million adults) as provided by Statistics Sweden.27 27 27 Participants were enrolled during the study period from 1 July 2005 (the start of the Swedish Prescribed Drug Registry) to 31 December 2012, as described in more detail elsewhere.23 25 28 29 The cohort members were followed up until the occurrence of any cancer (excluding non-melanoma skin cancer), death or 31 December2012 (ie, the end of data collection for the Swedish Cancer Registry), whichever occurred first. Individuals with a history of any cancer were excluded, as well as individuals with a cancer diagnosis within 12 months after inclusion (to avoid reverse causation). The unique 10-digit personal identity number, assigned to each Swedish resident, was used for identification of all participants and for linkages of their individual data between registries. This study was conducted according to a detailed and a priori established study protocol.

Data collection

The data for the present study were derived from our chemoprevention of cancer cohort. The study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki, without need for informed consent (2014/1291-31/4, approved 27 August 2014) (see online supplement 1).25 26 30 This data collection originates from the nationwide complete Swedish Prescribed Drug Registry, and includes all individuals residing in Sweden who have collected at least one dispensed prescription of any commonly prescribed drug between 1 July 2005 and 31 December 2014 (approximately 85% of all Swedish residents); with follow-up for cancer until 31 December 2012. This cohort has been linked to two other high-quality and complete nationwide Swedish registries, that is, the Swedish Cancer Registry (>96% completeness of all cancers, originated in 1961),31 and the Swedish Causes of Death Registry (>99% completeness, originated in 1952), by means of the personal identity number.

Exposures

Therapy with systemic NSAIDs was defined as at least 6 months (≥180 days) cumulative exposure during the study period. This was a cumulative exposure based on the defined daily dosage (DDD) per prescribed package, which takes into account the potency of the drug as well as the prescribed quantity. Three main types of NSAIDs were categorised based on their mechanisms of action (selective or non-selective COX inhibition) and drug class (aspirin or non-aspirin NSAIDs) with corresponding Anatomical Therapeutic Chemical classification codes (ATC): (1) aspirin (B01AC06, N02BA), (2) selective COX-2 inhibitors (cyclo-oxygenase-2 inhibitors, M01AH) and (3) non-selective non-aspirin NSAIDs (remaining M01A codes). Individuals with ≥180 days of exposure to two or three of these groups were excluded, so the three groups are mutually exclusive. Users of combination preparations including aspirin, that is, with corticosteroid (M01BA03), PPIs (B01AC56), statins (C10BX) as well as preparations for local (oral) use (A01AD05) were also excluded.

Additionally, the relatively selective COX-2 inhibitors, a subgroup of the non-selective NSAID users, containing meloxicam (M01AC06) and nabumetone (M01A×01), were also analysed separately. Aspirin users were also divided in two groups according to their ATC code (≥180 days): low-dose (B01AC06) and high-dose aspirin (N02BA) (those using both for ≥180 days were excluded).32 High-dose aspirin (N02BA) and some other NSAIDs (diclofenac, M01AB05 and ibuprofen, M01AE01) are also available over the counter in Sweden, but they are sold in only small packages and at higher prices per dose.32 33 Thus, we can assume that maintenance users had their doses prescribed, and were thus recorded in the present study.

Outcomes

The outcome was a first gastrointestinal cancer diagnosis recorded in the Swedish Cancer Registry according to the International Classification of Diseases 10th edition, including all cancers of the alimentary and hepatobiliary tract. Gastrointestinal cancers were categorised as follows: any gastrointestinal cancer (C15-C26) or cancer of the oesophagus (C15), stomach (C16), small bowel (C17), colorectum (C18-C21), liver, including intrahepatic bile ducts (C22), gallbladder or extrahepatic bile ducts (C23-24) or pancreas (C25). The category ‘other gastrointestinal cancer’ (C26) was not analysed separately. Additionally, the most common histological tumour types were analysed separately: adenocarcinoma (code 096) for oesophageal, gastric, gallbladder/biliary tract, pancreas and colorectal cancer; squamous cell carcinoma (code 146) for oesophageal cancer; hepatocellular carcinoma (code 066) and cholangiocarcinoma (code 076) for liver cancer and carcinoid (code 086) for small bowel cancer.

Statistical analyses

The relative risks of developing gastrointestinal cancer in individuals exposed to the drugs under study were standardised using the Swedish background population of the corresponding age, sex and calendar period. Standardised incidence ratios (SIRs) and 95% CIs were calculated, while accounting for changes in age and calendar categories when calculating years of follow-up.34 Follow-up time was counted from the dispense date of the first NSAID prescription to the date of a first cancer diagnosis, death or the end of the study (31December 2012), whichever occurred first. The expected incidence rates were calculated from cancer data recorded in the Swedish Cancer Registry and the age-stratified number of individuals per calendar year according to Statistics Sweden (Population Statistics). The overall SIR for gastrointestinal cancer was calculated, as well as SIRs for each anatomical location separately, including subanalyses for the most common histological types. The analyses were also stratified for sex and age for each cancer type. Subgroup analyses were performed for high-dose and low-dose aspirin, users of relatively selective COX-2 inhibitors, NSAID use with concurrent PPI (A02BC) or statin (C10AA) use (≥180 days), if the groups were sufficiently large. To assess the effect of PPI and statins, a multivariable Poisson regression model was fitted, adjusting for age at first prescription, sex and interaction between PPI and statins, and presented as incidence rate ratios (IRR) and 95% CI. The duration of the exposures was assessed by dividing the total cumulative dosage (sum of DDDs per package) received before the cancer diagnosis into four equally sized groups (quartiles), yet their total follow-up time was taken into account for the analyses. There were no missing data on exposures, outcomes or confounding variables. Effect estimates were only reported when at least five individuals developed the outcome.

Patient involvement

The Swedish patient organisation for cancer of the oesophagus, stomach, liver and pancreas was involved in supporting the present study (www.palema.org). The development of the research question and outcome measures were informed by patients’ priorities, experiences and preferences. The results will be disseminated to study participants by means of patient organisations.

Results

Among all 1 368 027 users of NSAIDs, there were 783 870 (57.3%) aspirin users, 566 209 (41.4%) non-selective NSAIDs users and 17 948 (1.3%) COX-2 users (table 1, online supplement 1). Aspirin users were more likely to be male (53.8%) and older than 70 years (54.9.2%), while non-selective NSAID users and COX-2 users were predominantly female (62.8% and 59.9%, respectively) and between 40 and 70 years of age (68.2% and 70.8%, respectively). Use of PPIs was found in 25.6%, 26.2% and 31.2% of the aspirin users, non-selective NSAIDs users and COX-2 users, respectively; and use of statins in 55.2%, 13.7% and 14.4%, respectively. The majority of the population received their first prescription during the first half year of the study period (2005), 54.9% of aspirin users and 42.5% of non-selective NSAIDs users (see online supplement 2).

Table 1

Characteristics of the study cohort on therapy with aspirin, selective cyclo-oxygenase-2 (COX-2) inhibitors and non-selective non-steroidal anti-inflammatory drug (NSAIDs)

	Aspirin only	Non-selective non-aspirin NSAIDs	Selective COX-2 inhibitors
	Number (%)	Number (%)	Number (%)
Total	783 870	566 209	17 948
Sex
Men	421 609 (53.8)	210 705 (37.2)	7201 (40.1)
Women	362 261 (46.2)	355 504 (62.8)	10 747 (59.9)
Age at first prescription (years)
<40	12 189 (1.6)	110 592 (19.5)	2720 (15.2)
40–49	32 743 (4.2)	125 977 (22.3)	3849 (21.5)
50–59	108 683 (13.9)	146 981 (26.0)	4941 (27.5)
60–69	200 154 (25.5)	112 682 (19.9)	3914 (21.8)
≥70	430 101 (54.9)	69 977 (12.4)	2524 (14.1)
Calendar period at first prescription
2005–2006	557 023 (71.1)	387 443 (68.4)	10 393 (57.9)
2007–2009	156 790 (20.0)	145 208 (25.7)	5500 (30.6)
2010–2012	70 057 (8.9)	33 558 (5.9)	2055 (11.5)
Proton pump inhibitors use (≥180 days)
Yes	200 828 (25.6)	148 586 (26.2)	5602 (31.2)
No	583 042 (74.4)	417 623 (73.8)	12 346 (68.8)
Statins use (≥180 days)
Yes	432 996 (55.2)	77 514 (13.7)	2589 (14.4)
No	350 874 (44.8)	488 695 (86.3)	15 359 (85.6)
Gastrointestinal cancer	10 969 (1.40)	3428 (0.61)	100 (0.56)
Oesophageal cancer	539 (0.07)	134 (0.02)	7 (0.04)
Adenocarcinoma	319 (0.04)	75 (0.01)	4 (0.02)
Squamous cell carcinoma	203 (0.03)	50 (0.01)	2 (0.01)
Gastric cancer	1079 (0.14)	260 (0.05)	7 (0.04)
Adenocarcinoma	949 (0.12)	212 (0.04)	3 (0.02)
Small bowel cancer	253 (0.03)	94 (0.02)	5 (0.03)
Carcinoid	122 (0.02)	43 (0.01)	2 (0.01)
Colorectal cancer	6919 (0.88)	2017 (0.36)	60 (0.33)
Adenocarcinoma	6608 (0.84)	1887 (0.33)	59 (0.33)
Liver cancer	645 (0.08)	232 (0.04)	3 (0.02)
Hepatocellular carcinoma	358 (0.05)	100 (0.02)	1 (0.01)
Cholangiocellular carcinoma	81 (0.01)	41 (0.01)	0 (0.00)
Gallbladder and biliary tract cancer	385 (0.05)	190 (0.03)	5 (0.03)
Adenocarcinoma	288 (0.04)	149 (0.03)	2 (0.01)
Pancreatic cancer	1114 (0.14)	490 (0.09)	13 (0.07)
Adenocarcinoma	835 (0.11)	402 (0.07)	11 (0.06)
Other gastrointestinal cancer	35 (0.00)	11 (0.00)	0 (0.00)
Duration of follow-up in person-years
Total	3 776 237	3 376 275	82 733
Mean (SD)	4.82 (2.40)	5.96 (1.67)	4.61 (2.21)

Aspirin

As presented in table 1, 10 969 (1.40%) aspirin users developed some type of gastrointestinal cancer during the follow-up. The most common cancer sites were colorectal (n=6919; 0.88%), gastric (n=1079; 0.14%) and pancreatic (n=1114; 0.14%). There was no association with gastrointestinal cancer based on the overall SIRs for aspirin users (SIR=1.02, 95% CI 1.00 to 1.04) (table 2). Shorter duration of use (<5.5 years) seemed to be associated with an increased risk for all gastrointestinal cancers. Yet, longer duration of aspirin use was followed by a decreased SIR for gastrointestinal cancer (SIR=0.31, 95% CI 0.30 to 0.33 for those with an estimated use between 5.5 and 7.7 years, and SIR=0.37, 95% CI 0.35 to 0.40 for >7.7 years) (table 3) and long-term aspirin users had clearly decreased SIRs for each gastrointestinal cancer type (table 3). The subgroup analyses including only the low-dosage aspirin users (n=668 305, 85.3% of the aspirin cohort) showed lower SIRs for all cancer locations, with significantly reduced risks for all locations except for oesophageal, gastric and liver cancer (see online supplement 3).

Table 2

The risk of different types of gastrointestinal cancer (and the major histological subtype) among users of aspirin and non-aspirin non-steroidal anti-inflammatory drugs (NSAIDs), expressed as standardised incidence ratios (SIRs) and 95% CIs and stratified by age and sex.

	Aspirin users (n=783 870)		Non-selective non-aspirin NSAIDs users (n=5 66 209)
	Number of cases	SIRs (95% CI)	Number of cases	SIRs (95% CI)
Gastrointestinal cancer	10 969	1.02 (1.00 to 1.04)	3428	0.79 (0.77 to 0.82)
Men	6659	1.24 (1.21 to 1.27)	1390	0.78 (0.74 to 0.82)
Women	4310	0.99 (0.96 to 1.02)	2038	0.81 (0.77 to 0.84)
18–39 years	3	–	14	0.59 (0.32 to 0.99)
40–49 years	20	0.70 (0.43 to 1.08)	116	0.67 (0.55 to 0.81)
50–59 years	408	1.13 (1.03 to 1.25)	423	0.61 (0.55 to 0.67)
60–69 years	2286	1.13 (1.08 to 1.18)	1074	0.71 (0.67 to 0.75)
≥70 years	8252	0.99 (0.97 to 1.01)	1801	0.94 (0.90 to 0.98)
Oesophageal cancer	539	1.10 (1.01 to 1.19)	134	0.75 (0.63 to 0.89)
Adenocarcinoma	319	1.17 (1.04 to 1.30)	75	0.81 (0.64 to 1.01)
Squamous cell carcinoma	203	1.06 (0.92 to 1.22)	50	0.66 (0.49 to 0.87)
Men	415	1.09 (0.99 to 1.20)	90	0.88 (0.63 to 0.97)
Women	124	1.12 (0.93 to 1.33)	44	0.68 (0.49 to 0.91)
18–39 years	0	–	0	–
40–49 years	1	–	5	0.87 (0.28 to 2.03)
50–59 years	21	0.99 (0.61 to 1.51)	18	0.58 (0.34 to 0.92)
60–69 years	164	1.35 (1.15 to 1.57)	47	0.65 (0.48 to 0.86)
≥70 years	353	1.02 (0.92 to 1.13)	64	0.92 (0.71 to 1.18)
Gastric cancer	1079	1.08 (1.01 to 1.14)	260	0.70 (0.62 to 0.80)
Adenocarcinoma	949	1.07 (1.00 to 1.14)	212	0.66 (0.58 to 0.76)
Men	714	1.08 (1.00 to 1.16)	128	0.72 (0.60 to 0.85)
Women	365	1.08 (0.97 to 1.19)	132	0.69 (0.58 to 0.82)
18–39 years	0	–	1	–
40–49 years	3	–	13	0.71 (0.38 to 1.22)
50–59 years	51	1.48 (1.10 to 1.94)	37	0.61 (0.43 to 0.85)
60–69 years	208	1.20 (1.04 to 1.37)	73	0.61 (0.48 to 0.77)
≥70 years	817	1.03 (0.96 to 1.11)	136	0.81 (0.68 to 0.95)
Small bowel cancer	253	1.05 (0.93 to 1.19)	94	0.84 (0.68 to 1.02)
Carcinoid	122	1.11 (0.92 to 1.32)	43	0.84 (0.61 to 1.13)
Men	150	1.05 (0.89 to 1.23)	34	0.72 (0.50 to 1.00)
Women	103	1.06 (0.86 to 1.28)	60	0.93 (0.71 to 1.19)
18–39 years	0	–	1	–
40–49 years	1	–	3	–
50–59 years	18	1.53 (0.91 to 2.41)	18	0.77 (0.46 to 1.22)
60–69 years	64	1.16 (0.89 to 1.48)	33	0.81 (0.56 to 1.14)
≥70 years	170	0.99 (0.85 to 1.15)	39	0.98 (0.69 to 1.34)
Colorectal cancer	6919	1.00 (0.98 to 1.03)	2017	0.74 (0.71 to 0.77)
Adenocarcinoma	6608	1.00 (0.80 to 1.03)	1887	0.74 (0.70 to 0.77)
Men	4105	1.03 (1.00 to 1.06)	793	0.73 (0.68 to 0.79)
Women	2814	0.97 (0.93 to 1.00)	1224	0.74 (0.70 to 0.78)
18–39 years	2	–	10	0.61 (0.29 to 1.13)
40–49 years	9	0.52 (0.24 to 0.98)	51	0.47 (0.35 to 0.62)
50–59 years	241	1.16 (1.02 to 1.31)	232	0.55 (0.48 to 0.63)
60–69 years	1268	1.05 (0.99 to 1.11)	600	0.66 (0.60 to 0.71)
≥70 years	5399	0.99 (0.96 to 1.02)	1124	0.88 (0.83 to 0.94)
Liver cancer	645	1.11 (1.03 to 1.20)	232	0.96 (0.84 to 1.09)
Hepatocellular carcinoma	358	1.13 (1.02 to 1.25)	100	0.83 (0.77 to 1.01)
Cholangiocellular carcinoma	81	1.14 (0.91 to 1.42)	41	1.10 (0.79 to 1.49)
Men	449	1.12 (1.02 to 1.23)	130	1.00 (0.84 to 1.19)
Women	196	1.09 (0.94 to 1.26)	102	0.91 (0.74 to 1.10)
18–39 years	0	–	3	–
40–49 years	4	–	10	0.95 (0.46 to 1.75)
50–59 years	32	1.04 (0.71 to 1.47)	44	0.90 (0.65 to 1.21)
60–69 years	182	1.36 (1.17 to 1.57)	90	0.99 (0.79 to 1.21)
≥70 years	427	1.03 (0.94 to 1.14)	85	0.95 (0.76 to 1.17)
Gallbladder and biliary tract cancer	385	0.92 (0.83 to 1.01)	190	1.03 (0.89 to 1.19)
Adenocarcinoma	288	0.93 (0.82 to 1.04)	149	1.07 (0.90 to 1.25)
Men	181	1.00 (0.86 to 1.15)	50	0.98 (0.73 to 1.29)
Women	204	0.85 (0.74 to 0.80)	140	1.05 (0.88 to 1.24)
18–39 years	0	–	1	–
40–49 years	0	–	7	0.97 (0.39 to 2.00)
50–59 years	6	0.52 (0.19 to 1.13)	12	0.46 (0.24 to 0.80)
60–69 years	91	1.31 (1.05 to 1.60)	63	1.00 (0.77 to 1.28)
≥70 years	288	0.85 (0.76 to 0.96)	107	1.23 (1.00 to 1.48)
Pancreatic cancer	1114	1.04 (0.98 to 1.11)	490	1.00 (0.92 to 1.10)
Adenocarcinoma	835	1.00 (0.93 to 1.07)	402	1.02 (0.92 to 1.13)
Men	629	1.07 (0.99 to 1.16)	163	0.89 (0.76 to 1.03)
Women	485	1.01 (0.92 to 1.11)	327	1.08 (0.96 to 1.20)
18–39 years	1	–	1	–
40–49 years	3	–	26	1.65 (1.08 to 2.42)
50–59 years	37	0.91 (0.64 to 1.25)	68	0.86 (0.67 to 1.09)
60–69 years	307	1.19 (1.06 to 1.33)	174	0.85 (0.73 to 0.99)
≥70 years	766	1.00 (0.93 to 1.07)	221	1.18 (1.03 to 1.35)

Table 3

The risk of gastrointestinal cancer among aspirin and non-aspirin non-steroidal anti-inflammatory drug (NSAID) users, by estimated duration of use, expressed as standardised incidence ratios (SIRs) and 95% CIs.

	Aspirin only (n=783 870)			Non-selective non-aspirin NSAIDs (n=566 209)
	Categories (quartiles)	Number of cases	SIRs (95% CI)	Categories (quartiles)	Number of cases	SIRs (95% CI)
Gastrointestinal cancer
	0.5–2.5 years	4158	2.77 (2.69 to 2.85)	0.5–0.7 years	865	1.00 (0.93 to 1.06)
	2.5–5.5 years	4532	1.83 (1.77 to 1.88)	0.7–1.1 years	832	0.92 (0.86 to 0.98)
	5.5–7.7 years	1310	0.31 (0.30 to 0.33)	1.1–2.1 years	977	0.86 (0.80 to 0.91)
	>7.7 years	969	0.37 (0.35 to 0.40)	>2.1 years	754	0.54 (0.50 to 0.58)
Oesophageal cancer
	0.5–2.5 years	204	2.91 (2.52 to 3.33)	0.5–0.7 years	35	0.93 (0.65 to 1.29)
	2.5–5.5 years	216	1.83 (1.60 to 2.09)	0.7–1.1 years	32	0.84 (0.57 to 1.18)
	5.5–7.7 years	61	0.31 (0.24 to 0.40)	1.1–2.1 years	43	0.91 (0.66 to 1.23)
	>7.7 years	58	0.56 (0.42 to 0.72)	>2.1 years	23	0.41 (0.26 to 0.61)
Gastric cancer
	0.5–2.5 years	61	2.89 (2.62 to 3.19)	0.5–0.7 years	55	0.73 (0.55 to 0.95)
	2.5–5.5 years	466	2.00 (1.82 to 2.19)	0.7–1.1 years	61	0.78 (0.60 to 1.01)
	5.5–7.7 years	99	0.26 (0.21 to 0.31)	1.1–2.1 years	80	0.82 (0.65 to 1.02)
	>7.7 years	110	0.45 (0.37 to 0.55)	>2.1 years	64	0.54 (0.42 to 0.69)
Small bowel cancer
	0.5–2.5 years	96	2.78 (2.25 to 3.39)	0.5–0.7 years	22	0.94 (0.59 to 1.43)
	2.5–5.5 years	109	1.94 (1.59 to 2.33)	0.7–1.1 years	20	0.83 (0.51 to 1.29)
	5.5–7.7 years	26	0.28 (0.18 to 0.41)	1.1–2.1 years	25	0.85 (0.55 to 1.25)
	>7.7 years	22	0.39 (0.25 to 0.60)	>2.1 years	27	0.76 (0.50 to 1.11)
Colorectal cancer
	0.5–2.5 years	2658	2.78 (2.67 to 2.88)	0.5–0.7 years	540	0.99 (0.91 to 1.08)
	2.5–5.5 years	2844	1.79 (1.73 to 1.86)	0.7–1.1 years	489	0.86 (0.78 to 0.94)
	5.5–7.7 years	813	0.31 (0.29 to 0.33)	1.1–2.1 years	565	0.78 (0.72 to 0.85)
	>7.7 years	604	0.36 (0.33 to 0.39)	>2.1 years	423	0.47 (0.43 to 0.52)
Liver cancer
	0.5–2.5 years	222	2.63 (2.30 to 3.00)	0.5–0.7 years	63	1.23 (0.95 to 1.57)
	2.5–5.5 years	272	1.96 (1.74 to 2.21)	0.7–1.1 years	53	1.02 (0.76 to 1.33)
	5.5–7.7 years	100	0.44 (0.35 to 0.53)	1.1–2.1 years	70	1.10 (0.86 to 1.39)
	>7.7 years	51	0.40 (0.30 to 0.53)	>2.1 years	46	0.61 (0.45 to 0.81)
Gallbladder and biliary tract cancer
	0.5–2.5 years	137	2.36 (1.98 to 2.79)	0.5–0.7 years	42	1.17 (0.85 to 1.58)
	2.5–5.5 years	143	1.52 (1.28 to 1.79)	0.7–1.1 years	51	1.34 (1.00 to 1.76)
	5.5–7.7 years	61	0.39 (0.30-.50)	1.1–2.1 years	52	1.06 (0.79 to 1.39)
	>7.7 years	44	0.40 (0.29 to 0.53)	>2.1 years	45	0.73 (0.53 to 0.98)
Pancreatic cancer
	0.5–2.5 years	424	2.79 (2.53 to 3.06)	0.5–0.7 years	107	1.09 (0.90 to 1.32)
	2.5–5.5 years	467	1.87 (1.71 to 2.06)	0.7–1.1 years	123	1.20 (1.00 to 1.43)
	5.5–7.7 years	149	0.36 (0.30 to 0.42)	1.1–2.1 years	136	1.05 (0.88 to 1.25)
	>7.7 years	74	0.28 (0.23 to 0.37)	>2.1 years	124	0.78 (0.65 to 0.93)

Non-selective NSAIDs

Table 1 shows that 3428 (0.61%) of the non-selective NSAID users developed cancer, mainly colorectal (n=2017; 0.36%), pancreatic (n=490; 0.09%) and gastric cancers (n=260; 0.05%). Overall, there was a decreased risk of gastrointestinal cancer (SIR=0.79, 95% CI 0.77 to 0.82), and also for gastric (SIR=0.70, 95% CI 0.62 to 0.80), colorectal (SIR=0.74, 95% CI 0.71 to 0.77) and oesophageal (SIR=0.75, 95% CI 0.63 to 0.89) cancers analysed separately (and their main histological subtypes) (table 2). There was no evidence of decreased SIRs for the other types of gastrointestinal cancer types, although the effect sizes indicated a decreased SIR of small bowel and liver cancer. Longer duration of use of non-selective NSAIDs was associated with a decreased gastrointestinal cancer risk for all anatomical locations (table 3).

Selective COX-2 inhibitors

Overall, 100 (0.56%) COX-2 users developed some type of gastrointestinal cancer, predominantly colorectal (n=60; 0.33%), pancreatic (n=13; 0.07%), gastric (n=7; 0.04%) and oesophageal cancers (n=7; 0.04%). There was some evidence for a decreased risk of gastrointestinal cancer overall (SIR=0.89, 95% CI 0.73 to 1.09), although not statistically significant. None of the subanalyses showed strong evidence for an association (see online supplement 3).

Relatively selective COX-2 inhibitors

Among the non-selective NSAIDs users, 7609 individuals used relatively selective COX-2 inhibitors, of whom 74 (0.01%) developed cancer. There was no evidence for an association with any of the gastrointestinal cancer locations (see online supplement 3).

Aspirin with PPIs or statins

Users of aspirin with concomitant use of PPIs had higher SIRs for all gastrointestinal cancers compared with those not using PPIs, with all SIRs indicating an increased risk except for gallbladder cancer (table 4). The SIRs were especially increased for gastric cancer (SIR=1.89; 95% CI 1.73 to 2.06) and oesophageal cancer (SIR=1.94; 95% CI 1.71 to 2.20). When using Poisson regression to compare aspirin users using PPIs directly with aspirin users not using PPIs (instead of using the background population as reference), the risk was increased for all gastrointestinal cancers (IRR=1.19, 95% CI 1.11 to 1.26), with significantly increased risks for oesophageal, gastric, small bowel, liver and pancreatic cancer (see online supplement 4).

Table 4

The risk of gastrointestinal cancer among aspirin and non-selective non-aspirin non-steroidal anti-inflammatory drug (NSAID) users, stratified by additional use of proton pump inhibitors (PPIs) or statins compared with the total Swedish background population, expressed as standardised incidence ratios (SIRs) and 95% CIs.

	Aspirin users (n=783 870)		Aspirin with PPI (n=200 828)		Aspirin with statins (n=432 996)
	Number of cases	SIRs (95% CI)	Number of cases	SIRs (95% CI)	Number of cases	SIRs (95% CI)
All gastrointestinal cancer	10 969	1.02 (1.00 to 1.04)	3617	1.25 (1.21 to 2.29)	6210	0.99 (0.96 to 1.01)
Oesophageal cancer	539	1.10 (1.01 to 1.19)	247	1.94 (1.71 to 2.20)	299	0.98 (0.87 to 1.09)
Gastric cancer	1079	1.08 (1.01 to 1.14)	509	1.89 (1.73 to 2.06)	619	1.05 (0.98 to 1.13)
Small bowel cancer	253	1.05 (0.93 to 1.19)	107	1.67 (1.37 to 2.01)	139	0.97 (0.81 to 1.14)
Colorectal cancer	6919	1.00 (0.98 to 1.03)	2004	1.07 (1.02 to 1.12)	3893	0.97 (0.94 to 1.00)
Liver cancer	645	1.11 (1.03 to 1.20)	231	1.52 (1.33 to 1.73)	351	0.99 (0.89 to 1.10)
Gallbladder and biliary tract cancer	385	0.92 (0.83 to 1.01)	117	1.00 (0.82 to 1.19)	212	0.90 (0.78 to 1.03)
Pancreatic cancer	1114	1.04 (0.98 to 1.11)	391	1.36 (1.23 to 1.50)	678	1.07 (0.99 to 2.15)

Among aspirin users exposed to statins, the SIRs were close to unity for each anatomical location (table 4). When aspirin users using statins were directly compared with aspirin users not using statins, risks were decreased for all gastrointestinal cancers (IRR=0.81, 95% CI 0.77 to 0.85), with significant decreases for all cancer locations except for colorectal and pancreatic cancer (see online supplement 4).

Non-selective NSAIDs with PPIs or statins

In users of non-selective NSAIDs on therapy with PPIs, the SIRs were increased for all gastrointestinal cancer types (and again higher than among those not using PPIs), except for colorectal cancer (table 4). When users of non-selective NSAIDs using PPIs were directly compared with those not using PPIs, risks were increased for all gastrointestinal cancers (IRR=1.61, 95% CI 1.49 to 1.74), and each individual cancer location except for gallbladder cancer (see online supplement 4). Among non-selective NSAIDs users using statins, the SIRs were lower than among all users of non-selective NSAIDs, and significantly reduced for oesophageal, gastric and colorectal cancers. When users of non-selective NSAIDs using statins were directly compared with those not using statins, risks were decreased for all gastrointestinal cancers (IRR=0.86, 95% CI 0.76 to 0.96), yet not significant for the individual cancer locations (see online supplement 4).

Discussion

This study on contemporary use of NSAIDs showed a decreased risk of all types of gastrointestinal cancer among long-term users of aspirin (>5.5 years) and non-selective NSAIDs users even for shorter duration of use (>0.7 years). Long-term users of non-selective NSAIDs were at a particularly decreased risk for gastric, oesophageal and colorectal cancers. These seemingly protective associations might be counteracted by concomitant PPI therapy, and enhanced by concomitant statin use.

The main strengths of this study are the population-based design and large sample size, including all adults residing in Sweden during the study period, which enabled separate analyses for contemporary use of different types of NSAIDs, and evaluation of less common types of gastrointestinal cancer which could not be assessed previously because of insufficient power, in particular for non-aspirin NSAIDs. Other advantages include the complete follow-up and accurate censoring for mortality. The data on the exposures (medications) and outcomes (gastrointestinal cancers) were highly accurate due to the validity and completeness of the Swedish registries, eliminating recall bias.

Although our findings for aspirin are largely consistent with the literature, that the protective effect is only seen after 5 years, reverse causation, confounding and/or bias appear to influence the aspirin analyses because of the apparent initial increased risk of cancer among short-term users. By excluding all individuals diagnosed with cancer within a year after enrolment, and only including those with a minimal accumulated duration of use of 6 months, the risk of reverse causation should be reduced. Yet, our results indicate that those with an estimated duration shorter than 5 years have an apparent increased risk, which might be because they take aspirin because of yet undiagnosed cancer-related pain or thrombotic events—indicating confounding by indication and reverse causation among the group with the shortest exposure time, an effect which could not have been detected in intervention trials or in case-control studies with a study-design-inherent more restrictive selection of study participants.15 35 As previous studies reported, 15%–20% of patients with cancer have thrombotic complications during the course of the disease (often as early manifestation of an occult malignancy),36 yet these complications (eg, deep venous thrombosis) are more likely to be treated with anticoagulants than aspirin. However, when only looking at those exclusively using low-dose aspirin for ≥180 days, that is, the platelet aggregation inhibitors, the protective effects were also visible in the overall analyses not taking into account duration, with SIR=0.86 (95% CI 0.85 to 0.88). This indicates that the apparent increased risks are mainly because of the small group using aspirin as analgesic (high dose), which shows it is important to distinguish between both groups of aspirin use. Reverse causality seems to be less of a problem for other NSAIDs users, although these may be used as analgesics.37 Yet, individuals using NSAIDs may be at a lower a priori risk of developing gastrointestinal cancer, because individuals with upper gastrointestinal symptoms are less likely to be chronic NSAIDs users due to the risk of gastrointestinal side effects.

Especially for aspirin users (with a high cardiovascular mortality, higher average age than NSAIDs users and more chronic comorbidities), death is a competing risk for the development of cancer, reducing the number at risk to develop cancer. Therefore, we censored follow-up time at time of death. In this cohort, the standardised mortality risks were 9.64 (95% CI 9.60 to 9.69) for aspirin users and 2.08 (95% CI 2.05 to 2.11) for non-selective non-aspirin NSAIDs users indeed showing a higher risk of death competing with the risk of cancer, leading to an overestimation of the protective effect in particular among aspirin users

Another limitation is potential confounding, for example, by socioeconomic status, dietary factors, obesity, tobacco smoking and alcohol consumption, which could not be taken into account since such information was not available for the total background population. However, we adjusted for age, sex and calendar period. Cancer-type specific confounders and their treatment such as Helicobacter pylori for gastric cancer, hepatitis B/C infection for liver cancer and chronic inflammatory diseases such as inflammatory bowel disease for bowel cancer and pancreatitis for pancreatic cancer, may also have contributed to the risk of cancer and timing of diagnosis. We may have incomplete exposure ascertainment (and underestimation of duration of use) for part of our cohort since no information was available on prescriptions before July 2005 or over-the-counter use. Yet, potential long-term (protective) effects may be expected to decrease gradually yet significantly after treatment cessation, reducing the potential effect of misclassification on our results due to exposure before 2005. We used the minimal exposure criterion of 180 days to exclude occasional users who are more likely to obtain their NSAIDs over the counter, so at a higher price. We did not have data on used daily dosage or duration of use, and used a proxy variable for duration based on accumulation of the average DDDs per package. This explains why some aspirin users had an estimated exposure time longer than the duration of follow-up, indicating a high daily dose. The high variability in actual and estimated administered dosage also hindered assessment of recency of use. Some previous studies subdivided aspirin use into ‘low dose’ and ‘high dose’ based on prescribed dosages (eg, <75 mg/day38 or <100 mg/day14), but since we did not have information on the number of prescribed pills per day, and different dosages could have been prescribed during the study period, we used the definition based on ATC coding and assessed the estimated duration of use, with the additional advantage that the low-dose aspirin was only available on prescription. This should also be a more accurate reflection of duration of use than the number of prescriptions.9 In our study, the DDD per package could range from <5 to 500 (for other NSAIDs) or 1000 (for aspirin), which illustrates the variation between prescriptions. Since 1.4 million individuals were exposed to NSAIDs (≥180 days), that is, one-fifth of the adult population in Sweden, our results are likely to be diluted since we compared them with the total background population. Yet, despite this dilution the associations among long-term users were strongly decreased.

Compared with previous studies, this study was better powered to separately analyses different gastrointestinal cancers and types of NSAIDs.39 The above-mentioned meta-analysis15 identified only two cohort studies including over 100 000 individuals assessing colorectal cancer risk among aspirin users.40 41 Even our exposed groups for aspirin and non-selective NSAIDs alone were 5–7 times larger than earlier large studies. The decrease in gastrointestinal cancer risk became evident only after longer exposure, which has also been shown in previous research,15 42 and is biologically plausible given the expected time latency for (hindering) cancer progression.

The higher risk among aspirin and other NSAID users also using PPIs should be interpreted with some caution. PPIs are often prescribed for gastro-oesophageal reflux and peptic ulcers, which are risk factors for oesophageal and gastric cancer, respectively. Therefore, a higher cancer risk was expected for those locations, yet not for the other gastrointestinal cancer types. PPIs can also be used to prevent peptic ulcers in users of aspirin and other NSAIDs, usually in individuals without any gastrointestinal morbidity. Two other studies of our group based on the same source cohort,23 25 investigated the risk of gastric and oesophageal cancer among PPI maintenance users, which suggested an increased risk in all indication groups for PPI (including those without gastrointestinal symptoms, and aspirin/NSAID users), which also supports a potential independent role for PPI in carcinogenesis as also suggested recently by other groups.43 44 Together with a potential increased risk of mortality related to long-term PPI use,45 we believe a more careful approach should be considered when prescribing PPIs to prevent gastrointestinal complications in long-term NSAIDs users. Yet, the risk for gastrointestinal complications such as bleeding should be assessed on an individual basis based on other research investigating short-term effects.46 Before considering implementing aspirin or other NSAIDs as widespread intervention, safety, in particular considering long-term use, needs to be considered, with previous research tending towards a ‘favourable benefit harm-profile’ despite an excess risk of bleeding.47

To conclude, this large Swedish nationwide and population-based cohort study on contemporary and long-term use of NSAIDs indicates a strongly protective effect of long-term use of both (low-dose) aspirin and other non-selective NSAIDs on gastrointestinal cancer development.