[92] The limitations and potential hazards of using surrogate markers

Clinicians are often advised to use surrogate markers to help guide drug therapy. This typically occurs with guideline-based therapy for chronic conditions, such as type II diabetes mellitus¹ and cardiovascular disease.² The US Food and Drug Administration defines a surrogate as “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 that is expected to predict the effect of therapy”.³ Common surrogate markers include glycated hemoglobin (HbA1c) for diabetes mellitus, low-density lipoprotein cholesterol (LDL-C) and blood pressure for cardiovascular disease, and bone mineral density (BMD) for osteoporosis.

Surrogate markers can lead to misuse of scarce healthcare resources if a decision to initiate treatment is based on a “threshold” level of the surrogate marker without clinical relevance. Once drug therapy is started, surrogate markers are often used to monitor drug effectiveness and modify pharmacotherapy. We assume that the surrogate marker tells us whether or not drug therapy is reducing risks of morbidity or mortality.

This approach can be hazardous as shown in the examples below.

This Letter suggests an approach to the use of surrogate markers. It also considers the utility of treating to surrogate marker “targets” and the use of surrogates in risk prediction scores. When using surrogates to make recommendations for individual patients, clinicians should ask FOUR important questions:

1. Does “worsening” of a surrogate marker reliably indicate an increased risk of morbidity and mortality?

Using total cholesterol (TC) as an example, despite the widespread view that elevated TC is a risk factor for death, a number of studies have found that elevated TC is either associated with decreased mortality or not associated with mortality.^4-7 Thus TC is not a reliable marker for increased risk.

Lesson: A worsening surrogate marker does not always indicate increasing risk of morbidity and mortality.

2. Does “improving” a specific surrogate marker lead to less morbidity and mortality?

There are numerous examples where improving a surrogate marker is associated with harm as opposed to benefit. A few of these are as follows.

Torcetrapib substantially reduced LDL-C and substantially increased high-density lipoprotein cholesterol (HDL-C), yet it increased mortality and cardiovascular morbidity.⁸ Rosiglitazone lowered HbA1c in Type II diabetes but increased the incidence of myocardial infarction.^9,10 Fluoride increased BMD, but increased the incidence of fractures.¹¹ Erythropoiesis-stimulating agents (e.g. darbepoetin) increased hemoglobin levels in chronic kidney disease patients but increased stroke and vascular access thrombosis.¹² Furthermore, the magnitude of change to a surrogate marker does not necessarily predict the magnitude of clinical benefit. In the largest statin trial, the patient group achieving the most reduction in LDL-C had the same relative benefit (RR 0.79) as the group with the least reduction (RR 0.78).¹³

Lesson: “Improvement” of popular surrogate markers due to drug therapy failed to predict better outcomes in these examples. Using drug therapy to “improve” surrogate markers may be harmful, not helpful. Once such “contrary” evidence exists for a particular surrogate, clinicians should question its application in other settings. Understanding this can free both the clinician and patient from chasing numbers at the expense of the patient’s health.

3. Do benefits of using surrogate markers as targets for drug therapy outweigh the harms?

Clinicians are frequently advised in guidelines to adjust drug treatment to achieve surrogate outcome targets (e.g. blood pressure, LDL-C, or HbA1c). The scientific way to test whether “better” surrogate targets are indeed better for patients is to randomize patients to groups with the intent to achieve different surrogate marker targets and measure clinical outcomes. For LDL-C targets, this approach has never been tested in a randomized controlled trial (RCT).¹⁴ RCTs comparing the benefits of lower blood pressure targets (<135/85 vs. standard <140-160/90) have not shown that the benefits of attempting to achieve lower blood pressure targets outweigh the harms.^15,16 A new US blood pressure guideline developed according to the Institute of Medicine’s recommendations to avoid conflicts of interest recognized this evidence by revising “target” systolic BP upwards in 2014.¹⁷ In type 2 diabetes, the ACCORD trial demonstrated that attempting to achieve a lower HbA1C target (<6.0%), as compared to a standard target (7.0-7.9%) increased mortality and did not significantly decrease cardiovascular adverse events.¹⁸ The ADVANCE¹⁹ and VADT²⁰ trials also demonstrated disadvantages to the lower surrogate targets, also leading to changes in the relevant guidelines.

Lesson: Aiming for surrogate targets that are either unproven, or have been proven harmful, is not consistent with evidence-based principles.

4. Is it useful to employ surrogate marker values in risk prediction tools?

Some clinicians regularly employ risk prediction tools (e.g. Reynolds Risk Score²¹) to estimate or predict the risk of morbidity and mortality in an individual patient. Risk prediction tools include surrogate markers to calculate risk estimates in individuals. There is increasing concern that such prediction tools cannot even predict risk accurately for a population, let alone for an individual.^{22, 23} When a surrogate marker is not a reliable predictor of risk, this problem should not be surprising.

Lesson: When there is evidence that a “worsening” surrogate marker does not reliably predict risk, using risk prediction tools that incorporate the surrogate marker is potentially misleading.

Conclusion

• Most commonly used surrogate markers have not been proven to be consistently predictive of morbidity or mortality risk thus their use in risk calculators is questionable.
• Relying on surrogate markers to assess effectiveness of drug therapy has not been proven to yield clinically meaningful benefits and there are important examples where that strategy was harmful.
• Avoid chasing surrogate targets (e.g. LDL-C targets) that have not been proven to have a net health benefit.

References

1. Canadian Diabetes Association Clinical Practice Guidelines Expert Committee. Canadian Diabetes Association 2013 Clinical Practice Guidelines for the Prevention and Management of Diabetes in Canada. Can J Diabetes 2013;37(suppl 1):S1-S212.
2. Anderson TJ, Gregoire J, Hegele RA et al. 2012 Update of the Canadian Cardiovascular Society Guidelines for the Diagnosis and Treatment of Dyslipidemia for the Prevention of Cardiovascular Disease in the Adult. Canadian Journal of Cardiology 2013;29(2):151-167.
3. US FDA. New drugs, antibiotic and biological drug product regulations; accelerated approval. Proposed rule. 57 federal register 13234-13242 (1992).
4. Beaglehole R, Foulkes MA, Prior IA, Eyles EF. Cholesterol and Mortality in New Zealand Maoris. British Medical Journal 1980;280(6210):285-7.
5. Krumholz HM, Seeman TE, Merrill SS et al. Lack of Association Between Cholesterol and Coronary Heart Disease Mortality and Morbidity and All-Cause Mortality in Persons Older Than 70 Years. JAMA. 1994;272:1335-1340.
6. Petursson H, Sigurdsson JA, Bengtsson C et al. Is the use of cholesterol in mortality risk algorithms in clinical guidelines valid? Ten years prospective data from the Norwegian HUNT 2 study. J Eval Clin Pract. Feb 2012; 18(1): 159–168.
7. Iso H, Naito Y, Kitamura A et al. Serum total cholestrol and mortality in a Japanese population. J Clin Epidemiol. 1994 Sep;47(9):961-9.
8. Barter PJ, Caulfield M, Eriksson M et al. Effects of Torcetrapib in Patients at High Risk for Coronary Events. N Engl J Med 2007; 357:2109-2122.
9. Richter B, Bandeira-Echtler E, Bergerhoff K et al. Rosiglitazone for type 2 diabetes mellitus. Cochrane Database of Systematic Reviews 2007, Issue 3. Art. N o.: CD00 6063. DOI: 10.1002/14651858.CD006063.pub2
10. Nissen SE, Wolski K. Rosiglitazone revisited: an updated meta-analysis of risk for myocardial infarction and cardiovascular mortality. Arch Intern Med. 2010 Jul 26;170(14):1191-1201.
11. Haguenauer D, Shea B, Tugwell P et al. Fluoride for treating postmenopausal osteoporosis. Cochrane Database of Systematic Reviews 2000, Issue 4. Art. No.: CD002825. DOI: 10.1002/14651858.CD002825.
12. Palmer S, Navaneethan SD, Craig JC et al. Meta-analysis: Erythropoiesis-Stimulating Agents in Patients With Chronic Kidney Disease. Ann Intern Med. 2010;153:23-33.
13. Heart Protection Study Collaborative Group. MRC/BHF Heart Protection Study of cholesterol lowering with simvastatin in 20 536 high-risk individuals: a randomized placebo-controlled trial. Lancet 2002;360:7-22.
14. Hayward RA, Krumholz HM. Three Reasons to Abandon Low-Density Lipoprotein Targets An Open Letter to the Adult Treatment Panel IV of the National Institutes of Health. Circ Cardiovasc Qual Outcomes. 2012;5:2-5.
15. Arguedas JA, Perez MI, Wright JM. Treatment blood pressure targets for hypertension. Cochrane Database of Systematic Reviews 2009, Issue 3. Art. No.: CD004349. DOI: 10.1002/14651858.CD004349.pub2.
16. Arguedas JA, Leiva V, Wright JM. Blood pressure targets for hypertension in people with diabetes mellitus. Cochrane Database of Systematic Reviews 2013, Issue 10. Art. No.: CD008277. DOI: 10.1002/14651858.CD008277.pub2.
17. James PA, Oparil S, Carter BL et al. 2014 Evidence-Based Guideline for the Management of High Blood Pressure in Adults. Report From the Panel Members Appointed to the Eighth Joint National Committee (JNC 8). JAMA. 2014;311(5):507-520.
18. The Action to Control Cardiovascular Risk in Diabetes Study Group. Effects of intensive glucose lowering in type 2 diabetes. N Engl J Med 2008;358:2545-59.
19. The ADVANCE Collaborative Group. Intensive Blood Glucose Control and Vascular Outcomes in Patients with Type 2 Diabetes. N Engl J Med 2008; 358:2560-2572.
20. Duckworth W, Abraira C, Mortiz T et al. Glucose Control and Vascular Complications in Veterans with Type 2 Diabetes. N Engl J Med 2009; 360:129-139.
21. Reynolds Risk Score. (n.d.). Retrieved January 16, 2015, from http://www.reynoldsriskscore.org/
22. Ridker PM, Cook NR. Statins: new American guidelines for prevention of cardiovascular disease. Lancet 2013; 382 (9907):1762–1765.
23. Nissen SE. Prevention Guidelines: Bad Process, bad outcome. JAMA Internal Med 2014;174(12):1972-73.