Are Surrogate Markers Adequate to Assess Cardiovascular Disease Drugs?

The use of surrogate end points as a basis for reaching conclusions about the benefit of therapy has been met with both rising enthusiasm, reflected in recent changes in the Food, Drug and Cosmetic Act¹ and some recent US Food and Drug Administration (FDA) actions, and rising concern, reflected in several strongly stated warnings.^{2 - 4} The mixed response is not surprising: reliance on surrogates, when the surrogate proves to predict clinical benefit, can bring treatment benefits to patients years before information on clinical outcomes could be available and at relatively low cost. But reliance on surrogates, when the effect on the surrogate does not lead to clinical benefit, can lead to the adoption of useless or even harmful therapies. The obvious community goal is to make decisions most likely to yield the former of these outcomes.

As defined in the preamble to the FDA's proposed accelerated approval rule, "A surrogate end point, or ‘marker,' is a laboratory measurement or physical sign that is used in therapeutic trials as a substitute for a clinically meaningful end point that is a direct measure of how a patient feels, functions, or survives and is expected to predict the effect of the therapy."⁵ An effect on the surrogate end point is thus not per se of any value to the patient. It is a benefit only to the extent that it causes or predicts an improved outcome (fewer myocardial infarctions, strokes, or deaths). Most surrogates are chosen because they are thought to be on the causal chain leading to the clinical outcome. Surrogates can be early or late in the presumed causal chain: cholesterol (a biochemical variable), blood pressure (a pathophysiologic variable), and coronary vessel diameter and left ventricular hypertrophy (morphological variables) are all surrogates,⁴ but the last 2 are closer to certain clinical events (myocardial infarction and heart failure). Some surrogates are not etiologic but are thought to reflect the activity of the underlying process that leads to adverse outcomes. Some surrogates can be thought of qualitatively (the surrogate and real end points generally move in the same direction) or more rigorously quantitatively, ie, the surrogate fully captures the effect of treatment on the outcome.⁶ Many of the theoretically possible surrogate-outcome relationships, and the problems they can pose when used as end points, are discussed by Fleming and DeMets.²

An intermediate (nonultimate) end point, which is sometimes confused with a surrogate, is a true clinical end point (a symptom or measure of function, such as symptoms of hyperglycemia, angina frequency, or exercise tolerance) but is not the ultimate end point of the disease, such as survival or the rate of other serious and irreversible morbid events. Improvement in an intermediate end point is of value to patients even if this does not lead to reduced mortality or morbidity and would ordinarily be a basis for marketing approval by the FDA. At the same time, an effect on an intermediate end point may also be taken as reason to expect a favorable ultimate outcome; in that sense, the intermediate end point plays the role of a surrogate. It is also important to gain the observed clinical benefit without an adverse effect on survival. A widely cited example of "surrogate misadventure," the adverse effect on survival of several kinds of inotropic agents used for heart failure,^{2 - 4 ,7 - 9} does not actually involve a surrogate end point. Rather, it is a case in which a short-term beneficial effect on heart failure symptoms was overwhelmed by long-term toxicity. Fortunately, this was learned before the drugs were marketed because the FDA condition for the approval of inotropic drugs requires that the manufacturer show that there is at least no adverse effect on survival.¹⁰

The Food, Drug and Cosmetic Act does not give direct guidance on what end points can provide evidence of effectiveness. It states only that the FDA should approve a new drug unless it finds a "lack of substantial evidence [evidence consisting of adequate and well-controlled clinical investigations] that the drug will have the effect it is represented to have under the conditions of use prescribed, recommended, or suggested in the proposed labeling."¹¹ This wording plainly does not suggest that a surrogate end point cannot be a legal basis for drug approval; indeed, it could be read to imply that any truthfully described finding could be the basis for approval.

However, the FDA requires that, in addition to being shown to be effective, drugs must be shown to be "safe" for their intended use. Safety means showing a favorable risk-benefit relationship (drugs are almost never safe in the sense of being risk-free). A drug with no effect of value (eg, with an effect on a surrogate end point not plausibly linked to clinical benefit) could not be considered safe, as its "risk-benefit ratio" would be infinite. A critical legal case (Warner-Lambert v Heckler)¹² in 1986 clearly established the FDA's ability to refuse to approve a drug because its demonstrated effect on a surrogate was of no known clinical value.

The law and regulations gave little guidance as to how to determine that an end point has sufficient clinical value to be a basis for approval until a 1992 FDA regulation on "accelerated approval,"¹³ incorporated into law in 1997,¹ provided indirect guidance. Recognizing that, in the past, approval of drugs had been based on effects on "well-established" surrogate end points (not defined), the regulation sought to facilitate approval of drugs to treat serious diseases without available therapy by allowing approval based on "an effect on a surrogate end point that is reasonably likely, based on epidemiologic, therapeutic, pathophysiologic, or other evidence, to predict clinical benefit."¹³ The "reasonably likely" standard was an explicitly lower standard than would have been used to support ordinary approval and allowed for "some uncertainty as to the relation of that end point to clinical benefit."¹³ The "well-established" surrogates, therefore, had to be more than "reasonably likely" to predict benefit, leaving little uncertainty that they did so. Approval under the accelerated approval rule is conditioned on a requirement that the drug be studied further "to verify and describe its clinical benefits,"¹³ ie, that the validity of the surrogate be confirmed.

The risk of reliance on a surrogate is that, for 2 distinct reasons, the risk-benefit calculation based on the available data may be wrong.

First, the surrogate may not be valid. A candidate surrogate end point may not be causally related to the clinical event as supposed. A simple illustration is that an elevated white blood cell count accompanies pneumonia, but treating it with cytotoxic agents would not improve the condition of the patient with pneumonia. The example seems absurd only because we understand infection and the body's response to it. It is now accepted that elevated blood pressure is a direct cause of stroke, heart failure, renal failure, and accelerated coronary artery disease and that lowering blood pressure reduces the morbidity and mortality rates of those outcomes. But in the 1960s, before the controlled outcome studies of antihypertensive drugs, this was actively debated; there was a strong view that elevated blood pressure was an "adaptive" response to underlying vascular disease and that lowering it would be harmful.¹⁴ Recent data on the benefits of 3-hydroxy-3-methylglutaryl coenzyme A (HMG-CoA) reductase inhibitors may have partly settled the surrogate question for those cholesterol-lowering drugs, but the real value of other cholesterol-lowering agents is not as clear. The current status of most other potential cardiovascular surrogate end points is highly uncertain.

In general, the persuasiveness of a surrogate end point in supporting effectiveness of a drug is based on biologic plausibility (evidence that the surrogate biochemical, anatomic and/or morphological, or pathophysiologic end point is on the causal pathway to the adverse outcome or is a regular finding associated with that outcome and is plausibly related to a common causal factor) and a history of successful intervention with a pharmacologically related agent or (even better) a range of pharmacologically unrelated agents. Table 1 lists some of the kinds of evidence and experience that would strengthen or weaken support for a surrogate as well as practical and public health considerations that could also influence the decision to rely on a surrogate.

The second risk of using a surrogate is that unexpected unfavorable effects of the drug may lead to a net unfavorable outcome. Although this is often identified as a surrogate problem, it is actually a general safety issue.^{2 ,4 ,15} For any treatment, whether it affects a surrogate or a clinical end point, unexpected adverse effects will be detected only to the extent that an adequate patient population is observed and studies are designed to detect these effects. For idiosyncratic effects (such as hematologic or hepatic toxicity) that are extremely rare in the untreated population and where 1 or 2 events constitute an interpretable signal, this is simply a matter of adequate total exposure. More difficult to discover are increases in the rates of events already relatively common in the population receiving the drug (eg, sudden death in postinfarction patients or patients with heart failure). These increases can be detected reliably (unless the effect is very large) only in large randomized trials. This is sometimes expressed by the phrase "there is no surrogate for safety."

Such concerns are equally applicable, however, to drugs approved for purely symptomatic treatments (eg, drugs for penile erectile dysfunction, obesity, depression, or seasonal allergies) or drugs approved for effects on intermediate end points (eg, angina rate or exercise tolerance in heart failure)² and raise general questions of how much and what kind of safety data should be available to support marketing of a new drug. Concerns about unexpected adverse effects are especially prominent in the case of surrogate-based approvals, probably for 2 reasons. First, reliance on a surrogate is usually an alternative to a large outcome trial. A large outcome trial provides a large, controlled safety database, and the loss of those data in a smaller surrogate study is conspicuous. Second, and notably in the cardiovascular area, the pharmacologic effect thought to be beneficial, or a closely related effect, has been found in several important cases to have adverse effects on the very outcomes the drugs were intended to improve.

High-dose diuretics, for example, lower blood pressure and decrease stroke rates but have little effect on cardiovascular events in hypertensive patients,^{16 - 17} perhaps because diuretics also cause hypokalemia, which increases the risk of sudden death.¹⁸ Type 1C antiarrhythmics (encainide hydrochloride, flecainide acetate, and moricizine hydrochloride) provoke new serious arrhythmias in postinfarction patients^{19 - 20} ; that is, they cause the very events that their effective ventricular premature beat (VPB) reduction was intended to prevent. The adverse effect of β-agonist and phosphodiesterase-inhibitor inotropes on survival in severe heart failure^{2 - 4 ,7 ,9} appears to result from a proarrhythmic effect or an adverse effect on the progression of underlying myocardial damage that may be inseparable from the mechanism of inotropy.

Distinguishing concern about the validity of the surrogate from the more general question of safety is important because it affects the kind of data that can be used to assess the benefits and risks of treatment. If there is doubt about the surrogate itself, only an outcome study in the specific disease can determine the value of the drug. But if the validity of the surrogate is accepted, studies in a variety of settings may be pertinent to assessment of safety. For example, the effectiveness of lowering blood pressure in decreasing the consequences of elevated blood pressure is well established by numerous large trials using a wide range of drugs.^{16 - 17 ,21} A drug that lowers blood pressure may therefore be about as certain to provide a clinical benefit as would be an antianginal drug. Nonetheless, a particular antihypertensive or antianginal drug could have an adverse effect that undermines its benefits. Establishing safety of the drug in either of these cases probably would not necessarily require large studies of the drug in hypertension or angina. Data from another population, preferably one more vulnerable to cardiovascular toxicity, would also be pertinent, as would data from studies of pharmacologically related agents in either population. These latter studies would indicate whether the drug's basic pharmacologic properties were dangerous. The absence of outcome data for newer classes of antihypertensives (calcium channel blockers and angiotensin-converting enzyme inhibitors [ACEIs]) has been cited by critics of the use of surrogates^{2 - 3} ; they contrast this lack of outcome data with the extensive outcome data in hypertension on diuretics and β-blockers and with the clear guidance provided by large outcome studies of ACEIs in ischemic heart disease.³

There are several problems with these contrasts. First, data from placebo-controlled trials of older hypertension treatments, except for high-dose diuretics (no longer recommended) and reserpine,^{22 - 23} are relatively sparse; moreover, most trials used multiple drugs and added treatments until a goal was reached, which made the role of any specific agent difficult to discern. The evidence supporting the value of lowering blood pressure comes most persuasively from the totality of all outcome trials¹⁷ and the uniform benefit seen with all therapies, not from the data on any particular intervention. Second, the 1 published placebo-controlled outcome study with a dihydropyridine calcium channel blocker (nitrendipine)²¹ gave results approximating the best attained with other interventions. Third, outcome studies were necessary to assess the effects of ACEIs in ischemic heart disease because no persuasive surrogates were available. Finally, ACEIs and calcium channel blockers have been extensively studied in large clinical trials for indications other than hypertension and do not have unexpected toxicity. They would therefore be expected to provide the benefits common to other drugs that lower blood pressure.

By 1995, ACEIs had been studied in placebo-controlled trials in more than 7000 patients with symptomatic heart failure²⁴ ; 4200 additional patients with asymptomatic ventricular dysfunction were observed in the Studies of Left Ventricular Dysfunction (SOLVD) prevention trial.²⁵ Still larger numbers of patients (95,000) with recent acute myocardial infarction have been studied in reported controlled trials.⁷ These studies, in patients with structural heart disease who are highly vulnerable to most cardiac adverse effects, showed favorable effects of ACEIs on survival and/or progression of heart failure. Although these data are not a perfect substitute for data in hypertensive patients, it seems reasonable to conclude that if ACEIs do not produce adverse effects in these patients, they will not cause them in hypertensive patients.

Similarly, controlled trials of verapamil hydrochloride and diltiazem hydrochloride in many thousands of postinfarction patients showing no overall effect on survival and a suggestion of reduced rates of myocardial infarction are pertinent to the safety of calcium channel blocker use in hypertension and to concerns raised by observational studies reporting that these drugs caused an excess of myocardial infarction²⁶ or death²⁷ in treated hypertensive patients. An international panel²⁸ concluded that the existing evidence does not suggest such adverse effects.

Although studies in a more vulnerable population seem relatively persuasive evidence of safety in a healthier population, evidence of drug safety in relatively healthy patients (eg, hypertensive) might not be applicable to safety in a postinfarction population or population with heart failure. Encainide and flecainide, for example, appeared to be relatively well tolerated by people with frequent premature beats and no acute or severe coronary artery disease but were lethal in a postinfarction population.^{19 - 20}

Whether a surrogate or intermediate end point is appropriate to use as a basis for marketing continues to be a matter of scientific judgment. Table 1 lists some of the scientific factors and public health or risk/benefit considerations, that would support, or argue against, use of a surrogate end point. The results of new outcome studies, the availability of other therapies, and evolving public attitudes toward the urgency of drug availability and drug safety all can affect decisions about use of a surrogate end point in a particular case.

Some factors given in Table 1 could be used to argue for or against use of a surrogate end point. For example, a history of successful clinical outcomes with a drug that affects a surrogate argues for the credibility of the surrogate, but it also raises the question of why we should accept any risk that a different drug, without outcome data, might not have that effect. Similarly, evaluating treatment for a fatal disease that occurs infrequently and after prolonged delay increases both the desire to rely on a surrogate (to make therapy available in a reasonable time) and concern that the drug might cause an undesirable adverse effect during its long use. Table 2 gives the current status of the use of surrogate and intermediate end points in a wide range of cardiovascular conditions. The only surrogate end points currently used as a basis for approval of cardiovascular drugs are blood pressure and serum cholesterol level.

Effect on blood pressure is the basis of approval of new antihypertensive drugs. The most persuasive support for the surrogate end point of blood pressure is experience from numerous long-term outcome studies of diverse antihypertensive drugs showing a clear effect on stroke and at least favorable trends on cardiovascular events and survival rates.¹⁷ In addition, substantial epidemiologic evidence indicates that blood pressure is continuously related to the risk of stroke and coronary heart disease.²⁹ Few active drugs have been compared directly to determine whether factors other than the effect of blood pressure modulate hypertensive benefit, but direct comparisons of high-dose diuretics and β-blockers showed no real difference,¹⁷ even for cardiovascular events, for which β-blockers might have been expected to be superior because of their benefit postinfarction³⁰ and lack of hypokalemic effect. As Collins and colleagues¹⁷ point out, a study to show such a difference would have to be very large, such as the ongoing Antihypertensive and Lipid Lowering Treatment to Prevent Heart Attack Trial (ALLHAT).³¹ Although the importance of such comparisons was emphasized more than a decade ago by the FDA,³² comparative studies have not been required of individual manufacturers. If ongoing large trials demonstrate differences in outcomes with drugs approved using the same surrogate procedures, that policy could change.

Currently, platelet aggregation inhibitors or anticoagulants in various settings (postinfarction or stroke, peripheral vascular disease, acute coronary syndrome, postangioplasty or postbypass) are studied using clinical end points (death, new infarction, urgent procedural intervention). As yet, although various antiplatelet treatments have a long and growing record of success in preventing adverse outcomes, there is no effect on a platelet aggregation or coagulation surrogate end point that has been convincingly shown to correspond to a clinical benefit and to define the risk of bleeding.

Although increasing the patency of an occluded coronary artery seems to predict the effectiveness of thrombolytic agents in a qualitative sense, a clear quantitative relation of patency to survival and rate of hemorrhagic stroke has not been established. Patency alone thus does not adequately define the risks and benefits of treatment, and patient outcomes must be studied for a thrombolytic agent to be approved.

Increased mortality with 2 classes of inotropic agents and an inotropic vasodilator drug clearly indicates that hemodynamic or symptomatic benefit in heart failure does not predict improved survival. Therefore, for a drug to be approved for heart failure symptom improvement, evidence of a symptomatic benefit needs to be supported by showing that there is at least no adverse effect on survival. The adverse outcome effects of studies of inotropes^{7 - 9} have been cited^{2 - 4} as evidence that surrogate and intermediate end points are treacherous, but these results were not as astonishing as is sometimes suggested. Well before the large inotrope outcome trials⁹ were conducted, the FDA concluded that there should "be reasonable assurance that survival in high-risk patients is not impaired; the controlled trials thus need to be of sufficient size to detect a substantial increase in mortality."¹⁰ This conclusion was based in part on early suggestions of rapid deterioration in open studies of inotropes and in part on the known adverse effects of digoxin. The inotrope trial results obviously were not expected, but the studies supported the requirement that any novel pharmacologic agent shown to improve symptoms would also need to be shown not to decrease survival.

Antianginal drugs are approved based on improvement in exercise tolerance or reduction in symptoms of angina; no current treatments have been shown to improve outcome. Safety of antianginal drugs is well supported by studies of calcium channel blockers,²⁸ and β-blockers³¹ in the postinfarction setting. Silent ischemia, like symptomatic ischemia, predicts an increased rate of death and myocardial infarction, and it has been proposed that a reduced rate of silent episodes should be a basis for approval. The FDA has not accepted this suggestion,³² concluding instead that drugs for silent ischemia needed to show an effect on a clinical end point, such as survival or rate of new infarction. It did not seem reasonable to conclude, without such data, that agents known only to affect ischemia would provide a benefit, when the same drugs used to treat symptomatic angina had not been shown to improve outcome. It also seemed at least possible that ischemia stimulated growth of collateral vessels,³³ which could improve outcome.

The most potent single example of an erroneous surrogate is the stunning result (mortality was increased 2.5-fold by encainide and flecainide compared with placebo) of the Cardiac Arrhythmia Suppression Trial (CAST)^{19 - 20} that definitively established that effective suppression of VPBs does not decrease mortality, despite the well-established association between elevated VPB rates and early arrhythmic death. But although the markedly adverse outcome was certainly unexpected, labeling for encainide and flecainide before the CAST study specifically pointed out the absence of known survival benefit from VPB suppression, the lack of any information on safety or effectiveness in the postinfarction state, and the drugs' ability to cause worsened arrhythmias. The indicated uses for both drugs were limited to patients with documented life-threatening arrhythmias and symptomatic patients with nonsustained ventricular and frequent VPBs. Since the CAST results were reported, approval of drugs for ventricular arrhythmias that are not immediately life-threatening has required either an improved survival outcome or, if a symptomatic claim was sought, evidence of no adverse effect on survival. However, no drugs have been approved under this standard.

In many cardiovascular diseases, surrogate and intermediate end points have not proved reliable predictors of outcome, the most striking examples being the failure of effects of antiarrhythmic drugs on VPB rates to improve survival and the similar failure of inotropic and vasodilator drugs with effects on exercise ability to improve heart failure outcome. On the other hand, many studies have shown that the ability of drugs to lower blood pressure predicts outcome and the large effects on cholesterol induced by HMG-CoA reductase inhibitors are also establishing a pattern of improved outcome. Surrogate end points are thus neither consistent successes nor consistent failures. The safety concerns left unanswered by reliance on a surrogate need to be satisfied in some other way: by reference to the drug's whole database, to past experience with related compounds, or to studies in other settings. The questions of how large the safety database for a widely used drug needs to be and how much assurance of the lack of an adverse effect on outcome there should be are not questions about surrogates, but rather more general questions concerning the level of safety assurance needed at the time any drug is marketed.