Angiotensin receptor blocker (ARB) use was reported to increase the risk for solid-cancers.¹ For prostate cancer (PrCA) the ARB-treated cohort had a meta-analytic risk ratio of 1.15 (C.I., 0.99,1.34, p = 0.076), that barely missed statistical significance.¹ The increased risk was not confirmed in subsequent studies or by the Food and Drug Administration (FDA)^2–6. In fact Uemura et.al., have found evidence that ARBs may have anti-proliferative effect of PrCA cells.^7–11 Thus, the residual concerns for increased risk of PrCA among ARB users is minimal, but since the majority of the follow-up studies were not done on individual level data, i.e. were meta-analysis, and when done with individual level data were based mainly on claims data from outside the United States, we decided to evaluate the safety signal in the Department of Veterans Affairs (VA).

The VA has provided medical care to >8 million male individuals. The VA researchers have access to claims data linked with individual patient level data from electronic medical records that includes pharmacy dispensation, laboratory results, cancer registry. This data when used with appropriate epidemiological research methods is able to provide valuable insight into the real-world relationship between ARB use and risk of prostate cancer.

We conducted an intention-to-treat (ITT) inverse-probability-of-treatment-weighted (IPTW)¹² retrospective cohort study to evaluate the impact of ARB prescribed for clinically indicated reasons on the incidence and histopathological grade of clinically detectable PrCA. The IPTW method is a propensity-score based method¹² that is able to correct mathematically for baseline differences (measured covariates) between comparison groups: in this retrospective study, IPTW is expected to balance differences in measured baseline covariates between the group that was assigned to receive ARB and the group that was assigned to not receive ARB. Regulatory approvals were obtained from the Institutional Review Board (IRB) of the William J.B. Dorn VAMC, the VA National Data Systems, the VA Patient Care Services and the Veterans Affairs Informatics and Computing Infrastructure (VINCI). We obtained linked individual-level data on all eligible Veteran patients from 1999 to 2011 from VA’s Central Cancer Registry (CCR), MedSAS, Decision Support System (DSS),¹³ Vital status file, health factors (for tobacco exposure),¹⁴ and the Corporate Data Warehouse (CDW).¹⁵

We based our cohort selection method on the methods developed by Hernan et.al.¹⁶ –where the recruitment process of a clinical trial was simulated using observational data. We simulated a randomly allocated intention to treat experiment that randomly assigns recruited patients to either ARB (treated) or non-ARB (not-treated) groups in 1:15 ratio continuously across calendar years 2003–2009 with a-priori end-point of December 31^st 2010. The cohort selection was performed in blocks of calendar years and then all individuals were pooled to form the final definitive cohort. As a first step, we identified new users of ARB between 2003 and 2009, who did not have an ARB dispensing in the previous years starting from 1999.

We first created the 2003 cohort. For patients receiving their first ARB dispensation in 2003, their start date of follow-up was defined as the closest date of outpatient VA clinician encounter ≤2 weeks before start date of ARB dispensing, ‘assigned to receive treatment’ (treated). If a patient was found to have started ARB without a corresponding VA clinical encounter, that patient was excluded. The rationale for this exclusion was that it is unlikely for a patient to start a new ARB treatment without a corresponding encounter with a clinician, except in such instances as a Veteran filling a non-VA prescription at a VA pharmacy. Including such individuals may introduce bias from missing information on comorbidity as the predominant care may be at a non-VA setting and thus not captured by VA electronic medical record system. The comparator groups, i.e. ‘assigned to receive no-treatment’ (not-treated) were 15-randomly selected individuals who had a VA clinician encounter in the year 2003, but did not receive ARB in the year of 2003 (from Jan 2003– Dec 2003). The start date of follow-up for not-treated group was a randomly selected date of their many respective actual clinical encounters in the year 2003. We achieved this by selecting the date corresponding to the lowest seed per person for that year, generated using SAS 9.2 Cary, NC function Proc RANUNI).

We then repeated this cohort selection method for the years 2004–2009, except that all patients who were already selected into prior year cohorts were not eligible to be selected into subsequent year cohorts. Finally, we pooled all 7-cohorts to form a single definitive cohort that was analyzed. Thus, a single patient was eligible to be represented in the definitive cohort only once. Baseline covariates were identified based on the date of start of follow-up.

We excluded from the staged selection process patients who were documented to have cancer in VA Central Cancer Registry (excluding non-melanoma skin cancer); had not established VA clinical, pharmacy and laboratory care at least 6-months prior to start date; those without information in VA health factor file for tobacco use; those with age < 55 or > 74 years (to ensure homogeneity between the two group, as both PrCA and ARBs are related to age); and those in the not-treated group who had propensity scores of either less than the 5^th percentile or greater than 95^th percentile of the treatment group (to reduce instability of IPTW).^17,18 All selected patients were then followed till the first of either the last date of VA healthcare benefit, death, date of diagnosis of prostate cancer, or December 31st 2010 (a-priori determined end-point), whichever came first.

We computed propensity scores using all variables listed in Table 1 and weighted the cohort using stabilized IPTW. The weighted cohort may now be expected to be similar to a cohort obtained from a random allocation experiment.¹⁹ Incidence curves were drawn for both types of exposures and for the absolute difference between exposures (Figure 1). Double-robust regression with IPTW after checking for Cox-Proportionality assumption was used to derive weighted hazard ratios (HR) with 95% confidence intervals. To evaluate if there was a difference in Gleason scores for the PrCAs diagnosed in the two groups, we conducted weighted CochranArmitage test for trend, as Gleason score is ordinal. All reported p-values are two-sided and all analyses on categorical data used exact methods when possible.

For the years 2003 to 2009 the un-weighted cohorts had respectively: 95,568; 99,664; 81,600; 70,944; 67,616; 66,080 and 62,352 individual patients. This formed a pooled unweighted cohort of 543,824 individuals. Weighting with IPTW resulted in excellent balance for all 54 variables that was used to compute propensity to receive treatment (Table 1). Propensity to receive treatment was most impacted by diabetes mellitus, serum creatinine, current use of insulin, use of Angiotensin Converting Enzyme inhibitors (ACEi), body mass index [BMI=weight(kg)/height(m)²], chronic renal failure, congestive heart failure, hypertension and low density lipoprotein; and least impacted by race, ethnicity, income, insurance status and religion. As expected, baseline PSA levels, prior utilization of PSA-based testing, prior prostate biopsy or use of 5-α-reductase inhibitor were not found to significantly impact the decision to assign ARB.

The weighted definitive cohort had, in treated and not-treated arms respectively 34,275 and 509,922; with PrCA rates of 506 (1.5%) and 8,269 (1.6%). The weighted hazard ratio (HR) for ARB was 0.91 (95% C.I. 0.84 to 0.99, p-value = 0.049). All independent HRs are reported in Table 2. Current smokers had a HR of 1.69 (95% C.I., 1.38, 2.06, p < 0.001) compared to never smokers, while patients with extremes of BMI where less likely to be diagnosed with PrCA compared to patients with normal BMI. We classified Gleason score based on aggressiveness into either <7, =7 or >7. The distribution of Gleason scores in the treated group was 225 (46.9%), 178 (37.1%) and 77 (16%); this was similar to the not-treated group 3,580 (46.3%), 3,006 (38.9%) and 1,142 (14.8%), with no statistical difference.

In this national cohort of veteran patients we observed a slight statistically significant reduction in the rate of clinically detected PrCA among patients assigned to receive ARB. An interesting incidental finding is that, this cancer reduction effect was not associated with PrCA grade migration as is expected in the 5-α reductase inhibitor clinical trials²⁰. The independent pathways by which ARBs are proposed to exert anti-cancer effect may be the reason for the lack of grade progression.^7,8,10

In our review of literature, we found that the biological evidence supporting the increased risk of PrCA from use of ARB is limited. Some explanations are related to imbalance in the local tissue level effect of Angiotensin II on inflammation and carcinogenesis. Angiotensin II influences the regulation of cell proliferation, angiogenesis, tissue repair, healing and development, and an imbalance may alter the risk of proliferation of cancer cells.^21–23 Unlike angiotensin converting enzyme inhibitors, ARBs do not suppress the production of Angiotensin II: thus in the presence ARB induced Angiotensin type-I receptor blockade, it maybe speculated that circulating Angiotensin II may have enhanced Angiogenesis or pro-inflammatory activity; either directly or through Angiotensin type-II receptors.^24,25 However, many researchers suggest a cancer protective effect.

Cancer protective effects of ARBs reported by other researchers include: reduction of basal and squamous cell carcinomas,²⁶ lung cancer metastatic burden,²⁷ and to improve overall lung cancer survival.²⁸ Specifically for PrCA, Candesartan has been shown to decrease PSA levels, improve performance status and decrease the need for analgesics among castration-resistant prostate cancer (CRPC) patients.¹⁰ Activation of AT1R has been shown to stimulate the proliferation of prostatic adenocarcinoma,⁹ while blockade of locally expressed AT₁ receptors by ARBs are reported to modulate local growth factors and cytokine expression in tumor tissues (local RAS).²⁹ In addition to the local RAS for cancer prevention, a prostate tissue-specific pathway may exist that would down-regulate the expression of androgen receptors.¹¹

The study limitations are many, but have been addressed to a large extent by careful study design and analysis. While computing propensity scores we have include many measured confounders and many instrumental variables. By balancing instrumental variables, we hope that we will be able to balance many unmeasured variables as well. To avoid errors due to missing information on Veterans who only receive part of their care in the VA, we required all individuals to have already established care in the VA at least 6-months prior to start of follow-up; still there may be some individuals who might receive their ARB dispensation from a non-VA pharmacy. A research study using Veteran only data may not be easily generalizable to non-Veteran population, and this concern is common to all research involving data from the VA. We conducted analysis by strictly adhering to the ITT paradigm, i.e. although we had information on treatment patterns during interval follow-up these data were not analyzed and switches in treatment/compliance was not taken into account. Also not analyzed were cumulative exposures, as these may be affected by the violation of the ITT assumption. Because our purpose was to evaluate the class effect of ARB on PrCA, we did not conduct sub-analyses stratified by ARB-subtype.

The findings of our study are insufficient to recommend the use of ARB as a PrCA chemoprevention modality. However, our finding of an ARB-related weak PrCA protective effect helps assuage any residual concerns of increased PrCA risk¹ and supports the conclusions of the FDA.³⁰