Are Large Sample Size Studies the Answer to Evaluate Effects of Drug Use in Non–Clinical Trial Settings?

Abstract

TO THE EDITOR: We read with interest the article by Chang et al 1 published in a recent issue of Journal of Clinical Oncology. The authors used a new-user case-control study design that was expected to minimize prevalent user bias and improve the reliability of any significant findings. 2 However, we found at least three major flaws in their study design and data analysis that raised great concerns about their conclusions regarding the risk of cancer associated with the use of angiotensin II receptor blockers (ARBs), as a class or individually.First, the data analysis did not adjust for indications of use of ARBs as a class or individually when appropriate. ARBs belong to a class of drugs that reduce blood pressure and provide cardiorenal protection and are often used in individuals who cannot tolerate angiotensin converting enzyme inhibitors. In the absence of randomized clinical trial data, the confounding effects of drug indication can be adjusted by using a propensity score as a matching criterion or by including the score or the probability as a covariate in the conditional logistic regression analysis. Failure to take these factors into consideration can significantly and falsely increase the hazard ratio (HR) of the use of ARBs with regard to cancer risk because of the high-risk nature of this drug group for cancer. Consequently, reduced cancer risk may represent a lack of association, neutral effect size may represent increased cancer risk, and small increased risk may represent a large increase in cancer risk.To ask the critical question of whether ARB users are at increased cancer risk compared with nonusers, we examined the HR of the propensity score of the use of ARBs for cancer in our Hong Kong Diabetes Registry. 3 We found that the probability of the use of ARBs as a class was associated with a markedly increased risk of cancer in univariable analysis (HR, 17.51; 95% CI, 3.94 to 77.92) and multivariable analysis (HR, 13.15; 95% CI, 1.47 to 117.99). (See model 2 of Yang et al's Table 2 for the list of covariates used in the adjustment). 3 The confounding effect of the high risk of cancer in diabetic patients who require blood pressure lowering drugs is also evident by the increased cancer risk associated with beta blockers, calcium channel blockers, and thiazides, with relative risks ranging from 1.15 for calcium channel blockers to 1.32 for thiazides. 1 The observation that increased cancer risk with the use of antihypertensive drugs other than angiotensin converting enzyme inhibitors and ARBs also shows that adjustment for comorbidities failed to remove the indication effect of drug usage. Thus, given the high risk of cancer associated the use of ARBs, the lack of association of ARBs with cancer reported by Chang et al 1 might in fact suggest reduced cancer risk with the use of ARBs.Second, the matching used in the study did not take into account follow-up time. To be qualified as a control for a patient, the follow-up time of the control must be no shorter than that of the patient. Failure to co...

Addressing different biases in analysing drug use on cancer risk in diabetes in non‐clinical trial settings—what, why and how?

“…Estimation of hazard ratios and 95% confidence intervals (CIs) of statin use during follow-up for cardiovascular disease (CVD) using a 1: 2 matched nested user-non-user cohort selected from the Hong Kong Diabetes Registry (refer to figure 1 for patient selection). users [19]. In response to a commentary on our report of reduced cancer risk in insulin users, we demonstrated the discrepant results using different analysis methods.…”

“…The latter has been shown to be associated with RAS activation known to be associated with cancer development [15–18]. In the same study [8], the authors also reported increased cancer risk with other classes of antihypertensive drugs such as beta blockers, calcium channel blockers and thiazides with relative risks ranging from 1.15 to 1.32 [8], suggesting that drug indication rather than drug effects might explain these drug‐cancer associations [19].…”

“…For example, in one study, failure to consider indications of ARB usage has led to confusing conclusions with ARB as a group not affecting cancer risk while an individual ARB, losartan, decreased cancer risk [odds ratio (OR) = 0.78; 95% confidence interval (CI), 0.63 to 0.97] and another ARB, candesartan, increased cancer risk (OR, 1.79; 95% CI, 1.05 to 3.06) [8]. In response to this study [8], we used data from the Hong Kong Diabetes Registry and demonstrated that the propensity score for ARB use as a group was independently associated with 13‐fold cancer risk compared to non‐ARB users [19]. In response to a commentary on our report of reduced cancer risk in insulin users, we demonstrated the discrepant results using different analysis methods.…”

Additive effects of blood glucose lowering drugs, statins and renin-angiotensin system blockers on all-site cancer risk in patients with type 2 diabetes

“…First, because the condition associated with drug use may contribute to the clinical outcome, for which the drug is prescribed to reduce risk, the propensity score for use of the drug is used to adjust for the drug indication [14]. Second, inclusion of existing users of drugs of interest will introduce prevalent user bias.…”

Diabetes and cancer: the mechanistic implications of epidemiological analyses from the Hong Kong Diabetes Registry

“…In a commentary, we pointed out that the Taiwan study had not adjusted for the propensity of use of ARBs either as a group or individually. We further used data from the HKDR to illustrate that the propensity score for ARB use as a group was independently associated with 13‐fold cancer risk compared with non‐ARB users . To this end, there is emerging evidence supporting the new roles of angiotensin II in inflammation, oxidative stress and ageing , all of which have been implicated in cancer development.…”