Biomarkers and surrogate endpoints

When David Beckham leaves the field towards the end of a match, the man who replaces him is a surrogate. and although I suspect that many footballers, if asked, would say that Surrogate is a town in Yorkshire, the word actually comes from the Latin word subrogare, to substitute. The use of the term ‘surrogate marker’ in medicine dates from the late 1980s 1, but it had been preceded by some years by the term ‘biomarker’2 and was succeeded and replaced by yet another term, ‘surrogate endpoint’3. To see where these terms stand in relation to each other we need to define them from the bottom up. A surrogate endpoint has been defined as ‘a biomarker intended to substitute for a clinical endpoint’, the latter being ‘a characteristic or variable that reflects how a patient feels, functions, or survives’4. So, what is a biomarker? Well, that has been defined as ‘a characteristic that is objectively measured and evaluated as an indication of normal biologic processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention’4. Although what the marker marks is clearly defined here as being intrinsic, that is not true of the marker itself, which can be intrinsic or extrinsic. This leads me to propose a nosography of biomarkers, as listed in Table 1. The categories in the table could be further subdivided according to whether the markers are being used for diagnosis, staging, or monitoring of disease or for determining its response to therapy. Biomarkers are often cheaper and easier to measure than ‘true’ endpoints. For example, it is easier to measure a patient's blood pressure than to use echocardiography to measure left ventricular function, and it is much easier to do echocardiography than to measure morbidity and mortality from hypertension in the long term. Biomarkers can also be measured more quickly and earlier. Blood pressure can be measured today, whereas it takes several years to collect mortality data. In clinical trials the use of biomarkers leads to smaller sample sizes. For example, to determine the effect of a new drug on blood pressure a relatively small sample size of say 100–200 patients would be needed and the trial would be relatively quick (1–2 years). To study the prevention of deaths from strokes a much larger study group would be needed and the trial would take many years. There may also be ethical problems associated with measuring true endpoints. For example, in paracetamol overdose it is unethical to wait for evidence of liver damage before deciding whether or not to treat a patient; instead a pharmacological biomarker, the plasma paracetamol concentration, is used to predict whether treatment is required. There are many links in the chain of events that leads from the pathogenesis of a disease to its clinical manifestations; biomarkers can be used at any point in the chain, at the molecular, cellular, or organ levels. Likewise, a therapy might be developed to attack any one of these links, in order to try to manipulate the disease, symptomatically or therapeutically. Any measurement short of the actual outcome could be regarded as a surrogate endpoint biomarker. However, although all surrogate endpoints are biomarkers, not all biomarkers are useful surrogate endpoints. The ideal biomarker is one through which the disease comes about or through which an intervention alters the disease. For example, the serum cholesterol concentration should be an excellent diagnostic marker for cardiovascular disease; however, there is no clear cut-off point, and only about 10% of those who are going to have a stroke or heart attack have a serum cholesterol concentration above the reference range. But even if cholesterol is not a good diagnostic marker, it can still be used as a marker of therapeutic response to cholesterol lowering drugs. Other useful biomarkers are not directly related to the clinical endpoint, but are affected in parallel with the disease. In some cases they are good diagnostic markers but not good markers of progress (for example, prostate specific antigen in prostatic cancer), or conversely they may be good markers of progress but not helpful diagnostically (for example carcinoembryonic antigen in ovarian carcinoma). In looking for criteria for deciding which biomarkers are good candidates for surrogate endpoints we can turn to the guidelines that Austin Bradford Hill propounded for helping to analyse association in determining causation (Table 2) 5, 6. He propounded these guidelines in the context of environmental causes of disease, but they can be used in other spheres 7. Whenever a biomarker conforms to these guidelines, it is more likely to be useful. Note that simply because a biomarker fulfills the guidelines it will not necessarily be useful; it merely makes it more likely to be useful. Surrogate endpoints are most likely to be useful when the pathophysiology of the disease and the mechanism of action of the intervention are thoroughly understood. Otherwise, pitfalls await. For instance, smoking causes lung cancer, and a trial of the benefit of education in preventing lung cancer might use smoking as a surrogate endpoint rather than the occurrence of the cancer itself. On the other hand, if chemotherapy is used as a measure for treating lung cancer, smoking could not be used as a surrogate endpoint. This is obvious, but alerts us to the possibility of similar but less obvious examples, in which the mechanisms are not understood. Ventricular arrhythmias cause sudden death, and antiarrhythmic drugs prevent ventricular arrhythmias. It was therefore expected that antiarrhythmic drugs would prevent sudden death. In fact, in the Cardiac Arrhythmia Suppression Trial 8, Class I antiarrhythmic drugs increased sudden death significantly in patients with asymptomatic ventricular arrhythmias after a myocardial infarction, and the trial was stopped prematurely. The hypothesis was wrong. Another good example is enalapril and vasodilators, such as hydralazine and isosorbide, whose haemo-dynamic effects and effects on mortality associated with heart failure are dissociated. Vasodilators improved exercise capacity and improved left ventricular function to a greater extent than enalapril. However, enalapril reduced mortality significantly more than vasodilators 9. So in this case haemodynamic effects are not a good surrogate. Patients with asthma feel breathless if they have a low peak expiratory flow rate (PEFR). However, in one study different drugs produced different relationships between PEFR and breathlessness 10. Patients taking beclomethasone did not feel as breathless as those taking theophylline for a given PEFR. So what should the surrogate marker be – the ‘hard’ endpoint of peak flow or the ‘soft’ marker of how the patients felt? This also raises the question of whether more than one surrogate endpoint should be used in clinical trials. Confounding factors can nullify the value of surrogate endpoints. For example, serum T3 is used as a marker of the tissue damage that thyroid hormone causes in patients with hyperthyroidism. However, its usefulness is blunted in patients taking amiodarone, which interferes with the conversion of T4 to T3 without necessarily altering thyroid function. A surrogate endpoint has been defined statistically as ‘a response variable for which a test of the null hypothesis of no relationship to the treatment groups under comparison is also a valid test of the corresponding null hypothesis based on the true endpoint’11. Often the surrogate endpoint is used as an entry criterion in clinical trials, and it is important to be aware that this can lead to statistical problems 12. It introduces heterogeneous variance and the problem of regression to the mean. If someone is entered into a trial on the basis of an abnormal surrogate marker and then receives no treatment, the surrogate endpoint will still improve, simply because of the statistical variation in the measurement of variables. This reduces the power of a study. There is also a high likelihood of missing data when surrogate endpoints are used. Using a small sample size when using a surrogate endpoint may also mean that a study is not big enough to detect adverse effects of drugs. We use biomarkers all the time. We measure the time to relapse in a patient with cancer as a surrogate endpoint for survival time. We measure ocular pressure instead of loss of vision in patients with glaucoma. We use biomarkers to stage disease (for example the number of lymph nodes affected by cancer), in diagnosis (for example serum T3, electrocardiography, and autoantibodies), and to monitor the progress of a disease or its treatment (for example, blood glucose concentration, blood pressure, and FEV1). The search for useful biomarkers is a constant one. For example, in this month's issue of the Journal Dumont et al. have investigated neuropsychological, neurophysiological, and neuroendocrine tests and motor skills as biomarkers of the effects of SSRIs in healthy subjects 13. They applied three of the Bradford Hill guidelines: consistency, dose-responsiveness in the therapeutic range, and biological plausibility. They concluded that there is no single marker of value, but that a combination of markers is best. Their search was impressively thorough – 171 different tests in 56 studies – but it is a little surprising that among the 79 references that they cited they did not include the citation classic by Hindmarch 14, which featured in our thirtieth anniversary special issue last year, and in which the same conclusion was reached about psychoactive drugs in general 15,16. Another type of biomarker appears elsewhere in this issue of the Journal. If a drug is adsorbed by charcoal and is excreted into the gut via the liver or secreted via enterocytes, activated charcoal will prevent its reabsorption. This property of activated charcoal has been put to good use in the treatment of self-poisoning. For example, in a large, randomized, placebo-controlled trial in cardiac glycoside poisoning in Sri Lanka, due to ingestion of yellow oleander seeds taken with suicidal intent, multiple-dose activated charcoal reduced mortality from 8% to 2.5%, a striking effect 17. Stass et al.18 have used this action of activated charcoal for a different purpose – as an exogenous pharmacological biomarker of the extent to which enterohepatic or enteroenteric recycling contributes to the systemic availability of moxifloxacin. Oral charcoal increased the clearance of a single intravenous dose of moxifloxacin by about 24%. Other papers in this issue of the Journal imply the use of biomarkers, whether you realize it or not: a study of adherence of middle-aged patients to statins, given the premise that age and pre-existing cardiovascular disease are the best markers of the risk of stroke and heart attack 19; the potential use of the prevalence of khat chewers in populations in which the habit is common as a marker of the effectiveness of a heart attack prevention programme, since chewing khat is associated with a greatly increased risk of myocardial infarction 20; the use, in a Chinese population taking warfarin, of the INR as a marker of the risks of major bleeding or thromboembolism 21; the potential use of measurement of the N-terminal propeptide of type III procollagen as a marker for the haemodynamic effects of spironolactone 22. There are clear potential benefits in using biomarkers. Information can be obtained earlier, more quickly, and more cheaply. However, the chain of events in a disease process linking pathogenesis to outcome is fragile and the better we understand the nature of the path a disease takes and the pharmacology of a drug that affects it the better biomarkers we will be able to develop in diagnosing, staging, and monitoring disease and its response to therapy.