Nonsteroidal Anti-inflammatory Drugs But Not Aspirin Are Associated With a Lower Risk of Post-colonoscopy Colorectal Cancer

Ka Shing Cheung; Lijia Chen; Esther W. Chan; Wai Kay Seto; Ian C.K. Wong; Wai K. Leung


Aliment Pharmacol Ther. 2020;51(9):899-908. 

In This Article


Study Design and Data Source

This was a retrospective cohort study with data retrieved from a territory-wide electronic healthcare database, the Clinical Data Analysis and Reporting System (CDARS), which is managed by the Hong Kong Hospital Authority. The Hong Kong Hospital Authority is the only statutory public healthcare provider offering 90% of all primary, secondary and tertiary care services of Hong Kong with a population of 7.3 million. The CDARS records all patient's demographics and clinical data including hospitalisation, visits to outpatient clinics, investigation procedures and results, endoscopic and surgical procedures, as well as drug prescription and dispensing history. A number of territory-wide studies have been conducted using the CDARS, with a high degree of coding accuracy (>90%) of the International Classification of Diseases, Ninth Revision (ICD-9) codings.[27–33]

Outcome Definition and Study Subjects

Individuals aged at least 40 years and had undergone colonoscopy between 2005 and 2013 in all public hospitals in Hong Kong were identified. We excluded patients with prior CRC, inflammatory bowel disease, prior colectomy and detected CRC (defined as cancer found within 6 months of the index colonoscopy). The patient selection process is depicted in Figure 1. The primary outcome of interest was PCCRC between 6 and 36 months (PCCRC-3y), which is the definition of the World Endoscopy Organization (WEO) consensus on 'PCCRC rate for an interval of 3 year'.[7] This definition was also adopted by other previous studies.[34–39] In contrast, detected CRC was defined as CRC diagnosed within 6 months of index colonoscopy, presuming that CRC suspected at index colonoscopy would be confirmed within 6 months.[34] The secondary outcomes of interest were (a) PCCRC-all (ie all PCCRC cases developing >6 months after index colonoscopy) and (b) PCCRC>3y (ie PCCRC cases developing >36 months after index colonoscopy) (Figure 2). Cancer site was subcategorised into proximal (from caecum to transverse colon [ICD-9 codes 153.4, 153.6, 153.0, 153.1]) and distal colon (from splenic flexure to rectum [ICD-9 codes 153.2, 153.3, 153.7, 154.0, 154.1]).

Figure 1.

Patient selection flow diagram. CRC, colorectal cancer; CLN, colonoscopy

Figure 2.

Study time frame. Abbreviations: NSAID, nonsteroidal anti-inflammatory drug; CRC, colorectal cancer; CLN, colonoscopy. Detected CRC: CRC diagnosed within 6 months after index colonoscopy. PCCRC-3y: CRC diagnosed between 6 and 36 months after index colonoscopy. PCCRC-all: CRC diagnosed >6 months after index colonoscopy. PCCRC>3y: CRC diagnosed >36 months after index colonoscopy

To investigate the primary outcome, we observed patients from 6 months after index colonoscopy (ie index date) and censored them at PCCRC-3y diagnosis, death or end of 36 months. For the secondary outcomes, we observed patients from 6 months after index colonoscopy and censored them at CRC diagnosis, death or end of study (31 December 2017).

Data Validation

Due to anonymisation of patient's identity in the electronic database, only data on the outcome of PCCRC-3y (n = 137) from our own centre, Queen Mary Hospital, could be retrieved for validation. The coding accuracy was 97.1%.

Study Variables

The primary exposure of interest was NSAID use before index colonoscopy. Aspirin use was considered as a secondary exposure of interest. Covariates taken into analysis for PCCRC-3y risk included patient's factors and endoscopy centres' performance (annual endoscopy volume and polypectomy rate).[34,36,37,40] Specifically, patient's factors included age at index colonoscopy, sex, history of colonic polyps, polypectomy at index colonoscopy, smoking status, heavy alcohol consumption, comorbidities (cardiovascular, metabolic, neurological, renal and liver diseases) (Table 1) and concurrent usage of medications (aspirin,[41] cyclooxygenase [COX]-2 inhibitors[15] and statins[39,42]). Table S1 provides details of ICD-9 codes of each disease. Smoking was identified by ICD-9 code of V15.82 and by proxy of chronic obstructive pulmonary disease (COPD). Heavy alcohol consumption was inferred from the presence of alcohol-related disorders, including hepatic, gastrointestinal, neurological and psychiatric diseases.

Medication prescription and dispensing data were traced up to 5 years before index colonoscopy. All medication use, including NSAIDs, was defined as usage for ≥90 days as in our previous study.[39] The treatment duration of individual prescription between prescription start date and end date was calculated for a particular drug, and was then summed up as total treatment duration. Effects of individual NSAID (including diclofenac, naproxen, ibuprofen, mefenamic acid, indomethacin, sulindac, piroxicam and ketoprofen) on PCCRC-3y were also analysed.

To study dose-response relationship, duration of NSAID use was categorised into three groups: (a) never use, (b) ≤1 year and (c) >1 year. Frequency of NSAID use was also categorised into three groups: (a) never use, (b) <weekly use and (c) ≥weekly use. The frequency of use was calculated by dividing the number of days of NSAID use by 5 years.

We further explored the association between the timing of NSAID uses before index colonoscopy and PCCRC. Current NSAID users were defined when the last prescription ended ≤6 months before the index colonoscopy, while past users were defined when the last prescription ended >6 months before the index colonoscopy. NSAID nonusers were defined when there was no recorded prescription both before and after index colonoscopy.

In addition, we determined the association of post-colonoscopy NSAID use (defined as ≥90-day use after index colonoscopy) on risks of PCCRC.

Statistical Analyses

All statistical analyses were performed using R version 3.2.3 (R Foundation for Statistical Computing) statistical software. Continuous variables were expressed as median and interquartile range (IQR). Mann-Whitney U test was used to compare continuous variables of two groups. Chi-squared test or Fisher's exact test was applied for categorical variables. Propensity score (PS) regression adjustment was used as the primary analysis method to determine the effect of NSAIDs on PCCRC-3y risk.[43,44] PS represented the probability of NSAID use predicted by the 22 aforementioned covariates in a logistic regression model. Cox proportional hazards model with PS regression adjustment was used to calculate the adjusted hazard ratio (aHR) of PCCRC-3y with NSAID use.

Stratified analysis was performed according to cancer location (proximal or distal colon). Subgroup analysis was performed according to age, sex, history of diabetes mellitus and colonic polyps. To determine the effect of NSAIDs on secondary outcomes, the aHR was derived by Cox proportional hazards model with PS regression adjustment.

PS matching was also performed to achieve balance in covariates between the two groups.[43–45] NSAID users were matched to NSAID nonusers in a 1:2 ratio without replacement using a greedy distance-based matching algorithm with the logit of the PS within 0.1 standard deviation. Absolute standardised difference (ASD) allows an objective assessment of the matching result. It was defined as absolute difference in means or proportions divided by pooled standard deviation. Balance of covariates between two groups was achieved if an ASD was less than 0.20.[46] A two-sided p-value of < 0.05 was used to define statistical significance.