Use of 5‐alpha reductase inhibitors and risk of gastrointestinal cancers in men with benign prostatic hyperplasia: A population‐based cohort study

Pre‐clinical evidence suggests that 5‐alpha reductase inhibitors (5ARi's), prescribed in the treatment of benign prostatic hyperplasia, reduce colorectal and gastro‐oesophageal cancer incidence via action on the male hormonal pathway. However, few studies to date have investigated this association at the population level. Our study aimed to investigate the risk of colorectal and gastro‐oesophageal cancers with the use of 5ARi's. We conducted a retrospective cohort study of new users of 5ARi's and alpha‐blockers among patients with benign prostatic hyperplasia in the UK Clinical Practice Research Datalink. Patients were followed until a first ever diagnosis of colorectal or gastro‐oesophageal cancer, death from any cause or end of registration with the general practice or 31st of December 2017. Cox proportional hazards models with inverse probability of treatment weights were used to calculate weighted hazard ratios (HR) and 95% confidence intervals (CIs) of incident colorectal cancer or gastro‐oesophageal cancer associated with the use of 5ARi's compared to alpha‐blockers. During a mean follow‐up of 6.6 years, we found no association between the use of 5ARi's and colorectal (HR: 1.13, 95% CI 0.91–1.41) or gastro‐oesophageal (HR 1.14, 95% CI 0.76–1.63) cancer risk compared to alpha‐blockers. Sensitivity analysis showed largely consistent results when varying lag periods, using multiple imputations, and accounting for competing risk of death. Our study found no association between the use of 5ARi's and risk of colorectal or gastro‐oesophageal cancer in men with benign prostatic hyperplasia.

population-based retrospective cohort study found no evidence of an association between the use of 5-alpha reductase inhibitors and colorectal or gastro-oesophageal cancer risk, when compared to alpha-blockers.These results do not support the hypothesis that 5-alpha reductase inhibitors reduce cancer risk.

| INTRODUCTION
Cancers of the gastro-intestinal tract, including colorectal and gastrooesophageal are among the most common malignancies in men in the United Kingdom, many of which exhibit a male predominance. 1Thus, it has been hypothesised that the male sex hormones testosterone and dihydrotestosterone (DHT) are important in the initiation of both colorectal and gastro-oesophageal cancers.Testosterone and DHT exhibit their effects via action on the androgen receptor, subtypes of which have been shown to be differentially expressed in benign and malignant colonic tissue. 2 Moreover, in-vitro evidence suggests that the 5-alpha reductase enzyme is a factor in colorectal cancer progression and is associated with worse prognosis. 3Circulating sex hormones have been associated with oesophageal cancer incidence and the expression of the androgen receptor has been highlighted as a prognostic factor in gastric cancer. 4,5Alpha reductase inhibitors (5ARi's) are commonly prescribed for symptom control in benign prostatic hyperplasia (BPH).6 By inhibiting the action of 5-alpha reductase enzymes 5ARi's prevent conversion of testosterone to the more potent DHT.7,8 A previous case-control study including 361 users of a 5ARi found a reduced risk of gastro-oesophageal cancer with 5ARi use compared to non-users, 9 while another reported a reduction in oesophageal squamous cell carcinoma risk but not oesophageal or gastric adenocarcinoma.10 The results of these studies were limited by the use of a non-user comparator group from the general population, which can introduce confounding by indication and confounding by unmeasured factors such as frailty or healthcare utilisation.
Of note, Patients with BPH have been shown to have higher circulating testosterone levels, hypothesised to promote the growth in colorectal and gastro-oesophageal cancers. 5Therefore, observed differences in cancer risk in BPH patients may be confounded by baseline hormone levels rather than 5ARi treatment.To date no population-based studies have investigated the association of 5ARi's and colorectal cancer incidence.
Therefore, our study aims to investigate whether the use of 5ARi's reduces the risk of either colorectal or gastro-oesophageal cancers in men with BPH using a new-user active comparator design.
Alpha-blockers were chosen as a clinically relevant comparator group to minimise confounding by indication, as alpha-blockers are used for a similar indication as 5ARi.

| Data source
The UK Clinical Practice Research Datalink (CPRD) GOLD is a database of electronic health records of over 17 million patients from a network of general practices. 11The CPRD GOLD has been shown to be a representative of the UK population in terms of age, sex, and ethnicity. 11Information recorded in the CPRD includes patient demographics, diagnoses (coded using read codes 12 ), referrals, and prescriptions, 11 which have been shown to be valid and of high quality. 13,14Moreover, the validity of colorectal and gastro-oesophageal cancer diagnosis recorded in the CPRD has been demonstrated in comparison to the UK National Cancer Data Repository with positive predictive values estimated at above 90% in some studies. 15,16

| Study population
We identified a cohort of the first 100,000 men diagnosed with BPH We excluded patients who were prescribed a 5ARi or alphablocker prior to BPH diagnosis (i.e., prevalent users).Patients with concomitant use of both 5ARi's and alpha-blockers at cohort entry were also excluded.Additionally, we excluded those with a previous diagnosis of any cancer (excluding non-melanoma skin cancer).In the analysis of colorectal cancer, we excluded a history of cystic fibrosis (which is associated with an increased risk of colorectal cancer), or rare cancer syndromes (familial adenomatous polyposis, lynch syndrome, Peutz-Jeghers syndrome) diagnosed at any time on or before cohort entry.We required patients to have at least 1 year of followup, that is, a 1-year lag period, to allow for a biologically plausible latency period and to minimise detection bias and reverse causation. 17hort exclusions and analysis were conducted separately for colorectal and gastro-oesophageal cancer.

| Exposure definition
5ARi and alpha-blockers users were considered exposed from 1 year after cohort entry and were followed using an intention-to-treat approach whereby patients were considered continually exposed to the cohort entry drug until end of follow up irrespective of switches, add-ons, or discontinuation.This approach assumes that the effect of exposure on outcome remains after treatment discontinuation.Thus, we followed up all patients who met the study criteria, from 1 year after cohort entry (i.e., beginning of person time at risk) until a diagnosis of colorectal cancer or gastro-oesophageal, censoring on death from any cause, end of registration with the general practice, or the end of study period (31 December 2017), which ever occurred first.

| Covariates
All models were adjusted for the following covariates measured at cohort entry: age, duration of BPH (modelled as continuous variables), year of cohort entry, body mass index (BMI, <18.5, 18.5-24.9,25-29.9,>30, unknown), smoking status (current, former, never, unknown), alcohol consumption (yes, no, unknown), and influenza vaccination (as measure of health-seeking behaviour).We also included the use of the following drugs in the year before cohort entry: nonsteroidal anti-inflammatory drugs, low-dose aspirin, and statins.For both colorectal cancer and gastro-oesophageal cancer models, individual comorbidities were also extracted based on the individual comorbidities included in the Charlson's Comorbidity Index including cerebrovascular disease, chronic pulmonary disease, congestive heart disease, dementia, diabetes, hemiplegia, liver disease, myocardial infarction, peripheral vascular disease, renal disease, and rheumatological disease identified any time before cohort entry. 18For colorectal cancer models we also included inflammatory bowel disease (including ulcerative colitis), cholecystectomy, and bisphosphonate use in the year prior cohort entry.Gastro-oesophageal cancer models included prior Heliobacter pylori infection, gastro-oesophageal reflux disorder, Barrett's oesophagus, and proton pump inhibitor use in the year prior to cohort entry.

| Statistical analysis
We used inverse probability of treatment weights (IPTW) to account for potential confounders.Logistic regression was used to calculate the propensity scores of 5ARi treatment versus alpha-blocker treatment conditional on the covariates listed above.Using the propensity score, weights were calculated based on IPTW in which 5ARi users are given a weight 1/propensity score and alpha-blockers are given a weight of 1/(1 À propesity score).Using this method the patients of each treatment group are weighted to represent the overall population to allow estimation of the average treatment effect of 5ARi use. 19We trimmed patients in the non-overlapping regions of the propensity score ensuring users of 5ARi and alpha-blockers for all values of the propensity score and reducing extreme weights. 20We used absolute standardised difference (ASD) to assess covariates before and after IPTW.Covariates with ASD below 0.10 were considered balanced. 21We calculated the crude incidence rate and 95% confidence intervals (CIs) based on Poisson distribution for 5ARi users compared to alpha-blocker users.Cox proportional hazards models with IPTW were used to estimate weighted hazard ratios and 95% CIs for colorectal cancer and gastro-oesophageal risk with 5ARi use using robust variance estimators.

| Secondary analysis
We conducted three secondary analyses.First, to determine whether there was a dose-response relationship between 5ARi use and colorectal or gastro-oesophageal cancer incidence we calculated cumulative defined daily doses (DDD).The DDD was calculated using the prescribed drug quantity (e.g., number of tablets) and strength (e.g., dose in mg) over the duration of follow-up for each participant according to the World Health Organisation (WHO) DDD methodology. 22As an example, for finasteride a 5 mg dose is considered one DDD.Therefore over 1 month a patient prescribed 28 tablets of 5 mg at 1 tablet per day would have 28 DDDs (drug usage in DDDs = [28 items Â 5 mg/per item/5 mg]).Using the calculated DDD we categorised cumulative use as <365 DDD, between 365 and 1095 DDD, and more than 1095 DDD corresponding to approximately <1 year, between 1 and 3 years, and more than 3 years, as a timevarying variable.Second, we categorised time since initiation beyond the 1-year lag period (i.e., time since beginning of follow-up) as 1 year or less, 1.1-3 years, and longer than 3 years, to determine if there was a duration-response relationship.Third, we also assessed the risk by type of 5ARi (finasteride and dutasteride) compared to alphablockers to determine if there were drug-specific effects.

| Sensitivity analysis
We conducted four sensitivity analyses to assess the robustness of our findings.First, we varied the length of the lag period to 2 and 3 years to account for the long latency of cancer onset.Second, we used multiple imputations for variables with missing data (smoking, alcohol, and BMI). 23,24An ordinal logistic regression model was used to impute variables with missing information with explanatory variables, outcome, cumulative hazard (as recommended 25 ), exposure, and covariates.Ten imputations were conducted, and the results combined using Rubin's rules. 26Third, to minimise exposure misclassification we censored on switching of the study drugs and used stabilised inverse probability of censoring weights (IPCW) to account for informative censoring.As before, a 1 year lag period was applied thus, all patients were followed from 1 year after cohort entry (i.e., after the lag period) until an incident diagnosis of colorectal or gastrooesophageal cancer, 1 year after switching to one of the study drugs, death from any cause, end of registration with the general practice, or the end of the study period.Under this exposure definition if a patient switched between the study drugs, for example, from a 5ARi's to alpha-blocker, any event occurring within the 1-year period would be attributed to 5ARi exposure time.Logistic regression was used to calculate the probability of remaining uncensored conditional on the covariates mentioned previously and the resulting IPCW was multiplied with the IPTW for analysis.Fourth, IPCW was used to account for the competing risk of mortality. 27l analysis was performed using STATA Release 16 (StataCorp, College Station, TX, USA).

| Study population
The study population consisted of 39,930 and 39,965 patients for colorectal and gastro-oesophageal cohorts, respectively (Figure 1).Characteristics of 5ARi and alpha-blocker users for the colorectal cancer cohort before and after weighting are presented in Table 1.Patient characteristics were similar for the colorectal cancer and gastro-oesophageal cancer cohorts.Baseline characteristics for the gastro-oesophageal cancer cohort are given in Table S1.
Overall, 5ARi users were older (mean age, 71.3 vs. 68.1 years) in comparison with alpha-blocker users.5ARi users were also more likely to have comorbidities and use of low-dose aspirin.The mean followup time for each cohort was 6.6 years (standard deviation, 4.2 years).
After IPTW, covariates were balanced between the exposure groups, with all ASDs below 0.10.
There was no observed association between 5ARi use and colorectal cancer risk (hazard ratios [HR]: 1.13, 95% CI: 0.91-1.41).Secondary analysis of dose-response using DDD categories showed null Weighting was conducted using inverse probability of treatment weighting.
In sensitivity analysis, the effect estimates remained highly consistent with the primary analysis as summarised in Figure 2A (given in detail in Table S3-S6).The HR ranged from 1.13 for the multiple imputation analysis to 1.23 for the 3-year lagged analysis with largely overlapping CI's.

| 5ARi's and gastro-oesophageal cancer
From our gastro-oesophageal cancer cohort, we identified 5024 5ARi and 34,941 alpha-blocker new users for our analysis (Figure 1).
In sensitivity analysis, the effect estimates remained largely consistent with the primary analysis as summarised in Figure 2B (given in detail in Table S3-S6).The HR ranged from 1.07 for the 2-year lagged analysis to 1.30 for the censoring on switching or add-on of study drugs analysis with overlapping CI's observed across the analyses.

| DISCUSSION
This large population-based cohort study found no evidence of a reduction in the risk of colorectal or gastro-oesophageal cancers among users of 5ARi's compared to alpha-blockers.This is the first study to evaluate the association between the use of 5ARi and colorectal cancer risk.In a prior case-control study T A B L E 2 Crude and weighted hazard ratios of colorectal cancer and gastro-oesophageal cancer associated with the use of 5-alpha reductase inhibitors vs alpha-blockers in men with benign prostatic hyperplasia.suggested a potential reduced risk of gastro-oesophageal cancer with use of finasteride using a 2-year lag (OR: 0.68, 95% CI: 0.48-0.97). 9In addition, a population-based cohort study found a reduced risk of oesophageal squamous cell carcinoma was noted (HR: 0.49, 95% CI: 0.37-0.65)but not with overall gastro-oesophageal cancer (HR: 0.92, 95% CI: 0.82-1.02). 10Important limitations of our study were the possible inclusion of prevalent users in analysis, no lag period to account for cancer latency, and they were unable to control for important lifestyle factors such as BMI and alcohol use. 10 Importantly, both of these studies compared to non-users from the general population and did not restrict to men with a diagnosis of BPH.A notable limitation as patients with BPH may have lower circulating levels of sex hormones, an effect of treatment with 5ARi's, 28,29 therefore patients with BPH may respond differently to changes in hormonal levels. 30,31 advantage of our study is the use of an active comparator new user design and limiting the analysis to BPH patients only to mitigate potential confounding by indication, selection bias, and unmeasured confounding. 32-vitro evidence regarding colorectal cancer has highlighted the potential role of the type 1 5-alpha reductase enzymes in cancer development. 2,3Furthermore, treatment of cancer cells with the 5ARi dutasteride, a dual inhibitor of type 1 and 2 enzymes, was capable of inhibiting cancer processes in vitro. 3This evidence suggests a role of the type 1 enzyme targeted by dutasteride and not There was no evidence of an association between dutasteride and colorectal cancer risk in our study however the number of dutasteride users were limited.Levels of circulating sex hormones have also been highlighted in colorectal cancer epidemiology in which men with higher levels of sex hormone binding globulin (SHBG), an important hormone in the male hormonal pathway, were shown to have a lower incidence of colorectal cancer but no association with testosterone was shown. 33is is reflected in our study with no effect of 5ARi treatment on colorectal cancer risk.5][36] Moreover, a study found circulating androgens and SHBG were associated with the incidence of gastrooesophageal cancers. 4,33This may support a protective role of 5ARi's, via the reduced effect of androgens, in gastro-oesophageal cancer however this was not observed in our study.On the other hand, it has previously been highlighted that female hormones may also play a protective role in gastro-oesophageal cancers therefore male hormones may play a smaller role in cancer development. 37,38r study had several strengths.First, this was a large populationbased cohort study using a new-user active comparator design.By restricting analysis to patients with a previous diagnosis of BPH and those newly treated with 5ARi's or alpha-blockers, we minimised

(B) (A)
F I G U R E 2 Forest plots of primary and sensitivity analysis of the risk of (A) colorectal cancer and (B) gastrooesophageal cancer with the use of 5-alpha reductase inhibitors.All analyses were weighted using IPTW.Treatment switch and Competing risk of death analysis combined IPTW and IPCW.CI, confidence interval; HR, hazard ratio; IPCW, inverse probability of censoring weights; IPTW, inverse probability of treatment weights.
prevalent user bias, confounding by indication and detection bias following the increased surveillance after a BPH diagnosis.Utilization of the UK CPRD allowed for adjustment of important confounders including lifestyle factors such as smoking, alcohol use, and BMI.Our study also included a lag period to account for the long latency between the use of 5ARi's and alpha-blockers and the onset of cancer.
Our study also had several limitations.In particular, low numbers of events in some analysis limited statistical power and we were unable to look at tumour type or location for each cancer.The UK CPRD provided information on many confounders however as with any observational study residual confounding may still be present from unknown or unmeasured confounders such as family history, diet, and ethnicity.The prescription information from the CPRD includes only those written in general practice, thus misclassification of exposure is possible for example as a result of non-adherence or prescriptions received in other specialised settings.However, in general the management of BPH is controlled by general practitioners in primary care and exposure misclassification would be expected to be non-differential between 5ARi and alpha-blocker groups.Finally, BPH diagnosis in primary care is often only confirmed with medical history and physical examination, with no imaging required, so a small level of misclassification of BPH may also be present. 6

| CONCLUSION
In summary, this population-based cohort study indicates that 5ARi use in men with BPH is not associated with a reduced risk of colorectal or gastro-oesophageal cancers, when compared to alpha-blockers.
Our study does not support the hypothesis that 5ARi action may reduce colorectal or gastro-oesophageal cancer risk by reducing androgen activity.Further studies with increased event numbers would help corroborate these findings in men with BPH.

between 1
February 1998 and 31 December 2014.Patients were required to be at least 40 years of age at BPH diagnosis.Patients with <1 year of medical history in the CPRD prior to BPH diagnosis date were excluded to allow a minimum look-back period to ensure identification of new users of 5ARi's or alpha-blockers.From the base cohort of men with a BPH diagnosis, we identified all first prescriptions of a 5ARi (finasteride or dutasteride) or an alpha-blocker (tamsulosin, alfuzosin, doxazosin, prazosin, or terazosin) at or post-BPH diagnosis.The date of this first prescription was considered the date of cohort entry.
Baseline characteristics of the study population before and after weighting for colorectal cancer cohort.
F I G U R E 1 Flow chart outlining the selection of the colorectal and gastro-oesophageal cancer cohorts of new users of 5-alpha reductase inhibitors (5ARi's) and alpha-blockers (AB).BPH, benign prostatic hyperplasia; CPRD, Clinical Practice Research Datalink.T A B L E 1Abbreviations: 5ARi, 5-alpha reductase inhibitor; ASD, absolute standardised difference; NSAID, non-steroidal anti-inflammatory drug.a