Incremental effects of antihypertensive drugs: instrumental variable analysisBMJ 2017; 359 doi: https://doi.org/10.1136/bmj.j5542 (Published 22 December 2017) Cite this as: BMJ 2017;359:j5542
All rapid responses
Understanding the benefits and complications of using multiple blood pressure treatment classes is an important research question with potentially meaningful clinical implications. When addressing this research question, one must carefully scrutinize the assumptions made to understand the validity of the results. The present study  argues that being randomized to the intensive treatment arm of the SPRINT trial serves as an instrumental variable (IV) for the number of blood pressure treatment classes prescribed, allowing for an analysis that corrects for confounding by indication when evaluating how the number of blood pressure treatment classes prescribed affects systolic blood pressure and cardiovascular events. Critically, a valid IV analysis must have an IV that is independent of other unobserved variables that affect the predictor variable (number of drug classes) and the outcomes (systolic blood pressure and cardiovascular events) [2,3]. Unfortunately, given the treatment strategies employed in the SPRINT trial , we are concerned that the IV assumption is untenable; indeed, as the authors allude to in their discussion, choices such as the dosage of drugs, adherence, and/or the total number of blood pressure drugs prescribed were not available for consideration in the analysis, and these unobserved factors can affect the number of drug classes prescribed, blood pressure levels, and cardiovascular outcomes. A person in the intensive treatment arm may not just be treated with more classes of medication, but with fewer or the same number of classes but at higher doses of those classes; this dosage issue violates the central assumption behind the IV analysis. As we show below, this violation of the central IV assumption is not merely a caveat, but can unfortunately invalidate the principal conclusions of the analysis.
First, we find that reasonable heuristics for the critical IV assumption are violated. In particular, if the only effect of randomization on systolic blood pressure was through the number of blood pressure treatment classes, except for due to confounding by indication, one would expect the blood pressure levels to be similar among patients taking an equal number of blood pressure treatment classes in each study arm. Yet the trial data show that participants with just one blood pressure treatment class had a 13.0 mmHg difference (95% CI 11.0 to 15.0 mmHg) in final systolic pressure between the treatment arms, participants with two classes had a 15.8 mmHg difference (95% CI: 14.4 to 17.2 mmHg), participants with three drug classes had a 16.5 mmHg difference (95% CI: 14.8 to 18.1 mmHg), and participants with four classes had a 16.5 mmHg difference (95% CI: 13.7 to 19.4 mmHg), suggesting that drug dosage or adherence--not just number of drug classes--may have affected results. Further, consider the change in number of blood pressure classes and the change in systolic blood pressure from the randomization visit to the final visit in the study. If the IV were valid, among individuals for whom there was no change in number of blood pressure classes, there should be very little change in average systolic blood pressure; however, the trial data show a change of -12.75 mmHg (95% CI -14.40 to -11.10 mmHg), again suggesting drug dosage or adherence are likely violating the central IV assumption. Intensive treatment assignment could therefore exacerbate correlations between drug classes and unobserved factors like drug dose or adherence, so the effect sizes can become substantially biased when attributed to drug classes.
Second, we find that the violations of the major IV assumption can unfortunately lead to effect sizes of the same magnitude and reported statistical significance as noted in the study, even if there is no real independent effect of blood pressure treatment classes on cardiovascular events. Suppose we fit an Aalen additive hazards model  to the individual participant data from SPRINT using other covariates in the trial but not drug classes, to generate a model for the null hypothesis of no effect of drug classes on cardiovascular events. The IV analysis employed in the present study  produces a nonzero (false) estimate of 5.22 lower cardiovascular events per 1000 person-years for each additional drug class (95% CI: 1.53 to 9.29) on data simulated from the null effect model (which is just one of numerous models that illustrate the problem; see https://syadlowsky.github.io/hypertension-classes-iv/sprint_invalid_iv.html for statistical code). Hence, even if the number of drug classes prescribed has no effect on cardiovascular outcomes beyond the systolic blood pressure reduction from an additional medication of any class, the IV analysis would still incorrectly show a ‘statistically significant’ reduction in cardiovascular events per additional drug class due to violations of the central IV assumption.
When the central IV assumption is violated, an IV analysis--far from properly adjusting for confounding by indication--can produce extremely biased effect size estimates, even in the wrong causal direction, and produce false assurances of a statistically significant result where there is none . Our comment does not imply that there is no real relationship between number of drug classes and the outcomes of interest, only that the IV analysis conducted in the present study  does not reliably support such a relationship.
Steve Yadlowsky, M.S.
Departments of Statistics and of Electrical Engineering, Stanford University
Sanjay Basu, M.D., Ph.D.
Centers for Primary Care and Outcomes Research, Departments of
Medicine and of Health Research and Policy, Stanford University
1 Markovitz AA, Mack JA, Nallamothu BK, et al. Incremental effects of antihypertensive drugs: instrumental variable analysis. BMJ 2017;359:j5542.
2 Greenland S. An introduction To instrumental variables for epidemiologists. Int J Epidemiol 2000;29:1102.
3 Wooldridge JM. Introductory Econometrics: A Modern Approach. Cengage Learning 2015.
4 SPRINT Research Group, Wright JT Jr, Williamson JD, et al. A Randomized Trial of Intensive versus Standard Blood-Pressure Control. N Engl J Med 2015;373:2103–16.
5 Martinussen T, Vansteelandt S, Gerster M, et al. Estimation of direct effects for survival data by using the Aalen additive hazards model. J R Stat Soc Series B Stat Methodol 2011;73:773–88.
Competing interests: No competing interests
Markovitz AA, et al1 assessed the incremental effects of adding extra antihypertensive drugs from a new class to a patient's regimen. In instrumental variable models, the addition of an antihypertensive drug from a new class led to clinically important reductions in systolic blood pressure (-14.4 mm Hg) and fewer major cardiovascular events (absolute risk -6.2). Incremental reductions in systolic blood pressure remained large and similar in magnitude for patients already taking drugs from zero, one, two, or three or more drug classes. This finding was consistent across all subgroups of patients.
Patients with hypertension often require multiple antihypertensive drugs. Observational studies suggest that benefits diminish when additional antihypertensive drugs are prescribed.2,3 However, the current study challenges the view that the effects of antihypertensive drugs will diminish with each added drug class while increasing the adverse effects.
Physicians are now in chaos because of recent controversial clinical trials. Various strategies to reduce residual cardiovascular risk in hypertensive patients included treating BP to lower target goals and using different classes of antihypertensive medications, however still considerable residual risk remained. In contrast, controlling hypercholesterolemia in hypertensive patients by statins is very effective in reducing residual cardiovascular risk by 35% to 40%.4 Also, cross-talk between hypercholesterolemia and rennin angiotensin system (RAS) exists at multiple steps of insulin resistance and endothelial dysfunction. In this regard, combined therapy with statins and RAS blockers (one class of antihypertensive medications) demonstrate additive/synergistic effects on endothelial dysfunction and insulin resistance in addition to lowering cholesterol levels and blood pressure when compared with either monotherapy in patients.5,6 This is mediated by both distinct and interrelated mechanisms. Therefore, combined therapy with statins and RAS blockers may be important in developing optimal management strategies in patients with hypertension, hypercholesterolemia, diabetes, metabolic syndrome, or obesity to prevent vascular risk.7-9
The American Heart Association and American College of Cardiology recently lowered the national blood pressure target from 140/90 mm Hg to 130/80 mm Hg for the general population, in a guideline endorsed by nine additional groups. The new guidelines mean that more than 100 million Americans now qualify for hypertension diagnosis and suggest that already diagnosed individuals be managed with more antihypertensive medications to reach a new target.
Questions have been raised, however, about which strategy is better. To add which one? Antihypertensive or statins drugs?
Funding: None, Disclosures: None
1. Markovitz AA, Mack JA, Nallamothu BK, Ayanian JZ, Ryan AM. Incremental effects of antihypertensive drugs: instrumental variable analysis. BMJ 2017;359:j5542.
2. Rana BK, Dhamija A, Panizzon MS. Imputing observed blood pressure for antihypertensive treatment: impact on population and genetic analyses. Am J Hypertens 2014;27:828-37.
3. Timbie JW, Hayward RA, Vijan S. Variation in the net benefit of aggressive cardiovascular risk factor control across the US population of patients with diabetes mellitus. Arch Intern Med 2010;170:1037-44.
4. Egan BM, Li J, Qanungo S, Wolfman TE. Blood pressure and cholesterol control in hypertensive hypercholesterolemic patients: National health and nutrition examination surveys 1988-2010. Circulation 2013;128:29-41.
5. Koh KK, Quon MJ, Han SH, et al. Additive beneficial effects of losartan combined with simvastatin in the treatment of hypercholesterolemic, hypertensive patients. Circulation 2004;110:3687-92.
6. Koh KK, Quon MJ, Han SH, et al. Vascular and metabolic effects of combined therapy with ramipril and simvastatin in patients with type 2 diabetes. Hypertension 2005;45:1088-93.
7. Koh KK. Quon MJ. Targeting converging therapeutic pathways to overcome hypertension. Int J Cardiol 2009;132:297-9.
8. Lim S, Sakuma I, Quon MJ, Koh KK. Potentially important considerations in choosing specific statin treatments to reduce overall morbidity and mortality. Int J Cardiol 2013;167: 1696-1702.
9. Koh KK, Lim S, Choi H, et al. Combination pravastatin and valsartan treatment has additive beneficial effects to simultaneously improve both metabolic and cardiovascular phenotypes beyond that of monotherapy with either drug in patients with primary hypercholesterolemia. Diabetes 2013;62:3547-552.
Competing interests: No competing interests
ALLHAT concluded that thiazide-type diuretics are superior than calcium channel blockers or angiotensin-converting enzyme inhibitors in preventing 1 or more major forms of CVD. 
Additional classes of antihypertensive drugs should be added in prescribed therapeutic schemes only if maximum tollerated daily doses of diuretics do not manage to reduce systolic blood pressure below accepted targets.
Competing interests: No competing interests