Nuthalapaty.net

Users' Guides to the Medical Literature
XIX. Applying Clinical Trial Results
A. How to Use an Article Measuring the Effect
of an Intervention on Surrogate End Points
Heiner C. Bucher, MD, MPH THE SEARCH
dial infarction, and CD4 cell count for Gordon H. Guyatt, MD, MSc Using MEDLINE you identify a study acquired immunodeficiency syn- of raloxifene for the treatment of os- drome [AIDS] and AIDS-related mor- Deborah J. Cook, MD, MSc teoporosis demonstrating an effect on tality) or measures of subclinical dis- Anne Holbrook, MD, MSc bone mineral density.1 You are won- ease (such as degree of atherosclerosis Finlay A. McAlister, MD dering whether this warrants adminis- on coronary angiography).
tration to lower your patient's risk of The use of surrogate end points is in- for the Evidence-Based Medicine dispensable for drug evaluation in phase 2 and early phase 3 trials geared to es- tablishing a drug's promise of benefit.
You are a physician seeing a 62-year- Ideally, clinicians making treatment de- In many countries, companies may ob- old woman with postmenopausal osteo- cisions should refer to methodologi- tain drug approval by demonstrating a porosis. Her bone mineral density, as cally strong clinical trials examining the positive impact on surrogate end points.
measured by dual-energy x-ray absorp- impact of therapy on clinically impor- The use of surrogate end points for tiometry, is 2.5 SDs below the mean value tant outcomes. By clinically important regulatory purposes reflects drug ap- in premenopausal women. Although she outcomes we mean outcomes that are proval decisions that regulators must does not have back pain, a spinal radio- important to patients: health-related make in the face of public health exi- graph shows an old vertebral fracture.
quality of life, morbid end points such The patient has not yet experienced prob- as stroke or myocardial infarction, or Reliance on surrogate end points may lems as a result of her vertebral frac- death. Often, however, conducting be beneficial or harmful. On the one ture, but she is disturbed by the pros- these trials requires such a large sample hand, use of the surrogate end point pect that she may end up like her mother size, or long-term patient follow-up, whose osteoporotic fractures have re- that researchers or drug companies look Author Affiliations: Medizinische Universita¨ts-
sulted in severe, long-term back pain.
for alternatives. Substituting surro- Poliklinik, Kantonsspital Basel, Basel, Switzerland The patient has reflux esophagitis (Dr Bucher); Department of Clinical Epidemiology and gate end points for the target event al- Biostatistics, McMaster University, Hamilton, On- and a past endoscopy revealed nonspe- lows conduct of shorter and smaller tri- tario (Drs Guyatt, Cook, and Holbrook); and Division cific gastritis. A specialist had pre- of General Internal Medicine, University of Alberta als, thus offering an apparent solution Hospital, Edmonton (Dr McAlister).
scribed alendronate, which the pa- to the dilemma.
The original list of members (with affiliations) ap- tient had to stop taking after several A surrogate end point may be de- pears in the first article of the series (JAMA. 1993; weeks because of dyspepsia. She 270:2093-2095). A list of new members appears in fined as "a laboratory measurement or the 10th article of the series (JAMA.1996;275:1435- searched the Web and discovered a new a physical sign used as a substitute for 1439). The following members of the Evidence- drug, raloxifene, and wonders whether Based Medicine Working Group contributed to this a clinically meaningful end point that article: Antonio Dans, MD, Leonilla Dans, MD, Pat this drug might be an alternative. You measures directly how a patient feels, Brill-Edwards, MD, Daren Heyland, MD, Les Irwig, know that this drug has been licensed MBBCh, PhD, FFPHM, Roman Jaeschke, MD, MSc, functions or survives."2 Surrogate end Hui Lee, MD, MSc, Mitchell Levine, MD, MSc, Vir- for the prevention of postmenopausal points include physiologic variables ginia Moyer, MD, MPH, and David Naylor, MD, DPhil.
osteoporosis. You promise to examine Corresponding Author and Reprints: Gordon H.
(such as bone mineral density as a sur- the literature and to get back to her.
Guyatt, MD, MSc, McMaster University Health Sci- rogate for long-bone fractures, blood ences Centre, 1200 Main St W, Room 2C12, Hamil- pressure for stroke, low-density lipo- ton, Ontario, Canada L8N 3Z5.
Users' Guides to the Medical Literature Section Editor:
See also pp 786 and 790.
protein cholesterol levels for myocar- Drummond Rennie, MD, Deputy Editor (West), JAMA.
1999 American Medical Association. All rights reserved.
JAMA, August 25, 1999—Vol 282, No. 8 771
USERS' GUIDES TO THE MEDICAL LITERATURE if there is a causal connection be- Are the Results Valid? Is There a
Table 1. Users' Guide for a Surrogate End
tween change in surrogate and change Strong, Independent, Consistent
in the clinically important outcome.
Association Between the
Are the results valid?• Necessary, but not sufficient: is there a Thus, the surrogate must be in the Surrogate End Point and
strong, independent, consistent association causal pathway of the disease process the Clinical End Point?
between the surrogate end point and theclinical end point? and an intervention's entire effect on the To provide a valid substitute for an im- • Is there evidence from randomized trials in clinical outcome of interest should be portant target outcome, the surrogate other drug classes that improvement in thesurrogate end point has consistently led to fully captured by a change in the sur- must be associated or correlated with that improvement in the target outcome?* rogate. This Users' Guide builds on pre- target. In general, researchers choose sur- • Is there evidence from randomized trials in the vious discussions of how one can es- rogate end points because they have same drug class that improvement in thesurrogate end point has consistently led to tablish a causal relationship9 and found a correlation between a surro- improvement in the target outcome?* presents an approach to critical ap- gate and a target outcome in observa- What were the results?• How large, precise, and lasting was the praisal of studies using surrogate end tional studies, and their understanding treatment effect? Effect should be large, points and application of their results of the biology makes it plausible that precise, and lasting to consider a surrogatetrial as possible basis for offering patients the to manage individual patients.
changes in the surrogate will invariably As our discussion will make evi- lead to changes in the important out- Will the results help me in caring for my dent, the clinician needs to assess far come. The stronger the association, the patients?• Are the likely treatment benefits worth the more than a single study to make the more likely the causal link between the potential harms and costs? Offer intervention decision about the adequacy of a sur- surrogate and the target. The strength of on basis of surrogate data only if patient's riskof the target outcome is high, patient places a rogate. Evaluation may require a com- an association is reflected in statistics high value on avoiding the target outcome, prehensive review of observational such as relative risk (RR) or odds ratio.
and if there are no satisfactory alternativetherapies.
studies of the relationship between the We have presented a full discussion of *Answers to one or both of these questions should be "yes" surrogate and the target, and of some statistics reflecting the strength of asso- for surrogate trial to be an adequate guide for clinical or all of the randomized trials that have ciation in another article.11 Many bio- evaluated treatment impact on both the logically plausible surrogates are only surrogate and the target. While most cli- weakly associated with clinically impor- may lead to the rapid and appropriate nicians would hesitate to conduct such tant outcomes. For example, measures dissemination of new treatments. For an investigation, our guidelines will al- of respiratory function in patients with example, the Food and Drug Admin- low them to evaluate the arguments chronic lung disease, or conventional ex- istration's decision to approve new an- made by experts or the pharmaceuti- ercise tests in patients with heart and lung tiretroviral drugs based on informa- cal industry for prescribing treat- disease, are only weakly correlated with tion from trials using surrogate end ments on the basis of their effect on sur- capacity to undertake activities of daily points recognized the enormous need rogate end points.
living.12,13 When correlations are low, the for effective therapies for patients with surrogate is likely to be a poor substi- human immunodeficiency virus (HIV) tute for the target outcome.
infection. Subsequently, several of these THE GUIDES
In addition to the strength of the as- drugs have proved effective in random- In this guide, we follow the frame- sociation, one's confidence in the va- ized trials focusing on clinically impor- work of previous articles in the se- lidity of the association depends on tant outcomes.3-6 ries10 and ask 3 sorts of questions: are whether it is consistent across differ- On the other hand, reliance on sur- the results valid; what were the re- ent studies and after adjustment for rogate end points may lead to excess sults; and will the results help me in car- known confounders. For example, eco- morbidity and mortality. For example, ing for my patients? (TABLE 1). When
logic studies such as the Seven Coun- while cardiac inotropes may improve we consider the validity of a surro- tries Study14 suggested a strong corre- short-term cardiac hemodynamic func- gate, we must address 2 issues. First, lation between serum cholesterol levels tion in patients with heart failure, ran- to be consistently reliable, the surro- and coronary heart disease mortality domized clinical trials have demon- gate must be in the causal pathway from even after adjusting for other predic- strated excess mortality with a number the intervention to the outcome. Sec- tors such as age, smoking, and sys- of these agents.7 In particular, flose- ond, in considering a particular inter- tolic blood pressure. Subsequent co- quinan was widely prescribed after its vention, we must be confident that there hort studies confirmed this association release, but had to be withdrawn after are no important effects of that inter- and suggested that long-term reduc- a trial revealed its deleterious effects on vention on the outcome of interest that tions in serum cholesterol levels of are not mediated through, or captured 0.6 mmol/L (23 mg/dL) would lower How are clinicians to distinguish be- by, the surrogate. Our guides for va- the risk of coronary heart disease by ap- tween these 2 situations? Surrogate out- lidity (Table 1) bear directly on these proximately 30%. When a surrogate is come will be consistently reliable only associated with an outcome after ad- 772 JAMA, August 25, 1999—Vol 282, No. 8
1999 American Medical Association. All rights reserved.
USERS' GUIDES TO THE MEDICAL LITERATURE justing for multiple other potential precise, and lasting, and the benefit- suggested surrogate end points in heart prognostic factors we call the associa- risk trade-off must be clear.
failure have included ejection fraction, heart-rate variability, and markers of au- Similarly, cohort studies have consis- Is There Evidence From
tonomic function.37 The dopaminergic tently revealed that a single measure- Randomized Trials in Other
agent ibopamine positively influences all ment of plasma viral load predicts the Drug Classes That Improvement
3 surrogate end points, and yet a ran- subsequent risk of AIDS or death in pa- in the Surrogate End Point Has
domized trial demonstrated that the drug tients infected with HIV.15-20 For ex- Consistently Led to Improvement
increases mortality in heart failure.38 ample, in 1 study the proportion of pa- in the Target Outcome?
An example of a surrogate end point tients that progressed to AIDS after 5 Given the possibility of effects unre- is CD4 cell count, which has been vali- years in the lowest through the highest lated to the surrogate end point, patho- dated in randomized trials. A number of quartiles of viral load was 8%, 26%, 49%, physiologic studies, ecological stud- trials comparing different classes of anti- and 62%, respectively.20 Moreover, this ies, and cohort studies are insufficient retroviral therapies have demonstrated association retained its predictive power to establish that the link between sur- that patients randomized to more potent after adjustment for other potential pre- rogate and clinically important out- drug regimens had higher CD4 cell dictors such as CD4 cell count.15-19 comes is ironclad. We can confidently counts and were less likely to progress Returning to the scenario, you are rely on surrogate end points only when to AIDS or death.6,39 While there is no wondering if you can substitute bone long-term randomized trials have con- guarantee that the next trial using a dif- mineral density for fractures or health- sistently demonstrated that modifica- ferent class of drugs will show the same related quality of life in considering tion of the surrogate is associated with pattern, these results greatly strengthen whether to recommend raloxifene. A concomitant modifications in the tar- our inference that if therapy for HIV infec- large cohort study investigated risk fac- get outcome of interest. For example, tion increases the CD4 count, a reduc- tors for hip fracture.21 Postmeno- although ventricular ectopic beats are tion in AIDS-related mortality will result.
pausal women with a calcaneal bone associated with adverse prognosis in pa- Returning to our scenario, trials of eti- density in the highest third had a hip tients with myocardial infarction24 and dronate40,41 and alendronate42 for the pre- fracture rate of 9.4/1000 woman-years class 1 antiarrhythmic agents effec- vention of osteoporotic fractures in post- while women in the middle and lowest tively suppress ventricular arrhyth- menopausal women have shown parallel third had a fracture rate per 1000 wom- mias in animals and humans,25 these increases in bone mineral density and an-years of 14.7 and 27.3, respec- drugs have proved to increase mortal- reduced incidences of new vertebral frac- tively. Furthermore, after considering ity when evaluated in randomized tri- tures. This would suggest that clini- other risk factors for osteoporotic hip als.26 In this case, reliance on the sur- cians might rely on bone density to fractures including maternal history of rogate end point of suppression of evaluate new drugs in osteoporosis in hip fracture, previous fractures from any nonlethal arrhythmias led to the deaths making the assumption that if they saw site, poor self-rated health, use of long- of tens of thousands of patients.27 increases in bone density, decreases in acting benzodiazepines, impaired visual The treatment of heart failure pro- fractures would follow.
function, and reduced physical activ- vides another instructive example. Tri- However, another secondary preven- ity, bone mineral density continued to als of angiotensin-converting enzyme in- tion trial in postmenopausal women us- predict the risk of hip fracture.21 These hibitors in heart failure treatment have ing sodium fluoride showed divergent re- findings are consistent across studies demonstrated parallel increases in exer- sults.43 Although sodium fluoride looking at the association between bone cise capacity28-31 and decreases in mor- increased bone mineral density at the density and fracture risk.22,23 Thus, bone tality,32 suggesting that clinicians may be lumbar spine by 35% over 5 years, more mineral density is a moderately strong, able to rely on exercise capacity as a valid vertebral and nonvertebral fractures oc- independent predictor of fracture, and surrogate. Milrinone33 and epopros- curred in the intervention group than in meets our first criterion for an accept- tenol34 have both demonstrated im- the placebo group (163 and 72 in 101 able surrogate end point.
proved exercise tolerance in patients with women with sodium fluoride vs 136 and While meeting this first criterion is symptomatic heart failure. However, 24 in 101 women with placebo). In an- necessary, it is not sufficient to support when these drugs were evaluated in ran- other randomized trial, fluoride again reliance on a surrogate outcome. As we domized controlled trials both showed showed a large increase in bone density will emphasize below (Table 1), before an increase in cardiovascular mortality without any change in fracture rate.44 In- offering an intervention on the basis of that in one instance was statistically sig- ferences on the basis of unchanged bone effects on a surrogate outcome, the cli- nificant,35 and in the second case led to density may also be problematic. A study nician should note a consistent relation- the early termination of the study.36 Thus, of calcium and vitamin D in the elderly ship between surrogate and target in exercise tolerance is inconsistent in pre- showed virtually no change in bone den- randomized trials; the effect of the in- dicting improved mortality and is there- sity, but a reduction in fracture risk of tervention on the surrogate must be large, fore an unsatisfactory substitute. Other approximately 50%.45 Thus, increase in 1999 American Medical Association. All rights reserved.
JAMA, August 25, 1999—Vol 282, No. 8 773
USERS' GUIDES TO THE MEDICAL LITERATURE bone mineral density as a surrogate end fibrates) have shown that these drugs Returning to the scenario, we have point has shown an inconsistent rela- reduce the incidence of myocardial in- established that because of the incon- tionship to osteoporotic fractures.
farction but increase the risk of mor- sistent relationship between increase in tality from other causes (with no im- bone mineral density and fracture re- Is There Evidence From
pact on overall mortality).53-55 duction we would be reluctant to offer Randomized Trials in the Same
These examples highlight the point patients a new antiosteoporotic agent Drug Class That Improvement
we made earlier: confidence in a sur- solely on the basis of evidence of its ef- in the Surrogate End Point Has
rogate outcome depends on the assump- fect on the surrogate end point. Ralox- Consistently Led to Improvement
tion that the treatment captures any rela- ifene, the drug we are considering for in the Target Outcome?
tionship between the treatment and the our patient, is a nonsteroidal benzo- Clinicians are in a stronger position to outcome.56,57 This assumption can be thiophene, a selective estrogen- rely on surrogate end points if the new violated in 2 ways. First, treatment may receptor modulator representing a new drug they are considering is from a class have a beneficial mechanism of effect class of drugs for the prevention of os- of drugs in which the relationship be- on the outcome independent of its effect teoporosis-related bone fractures. Thus, tween changes in the surrogate and on the surrogate. For instance, 1 expla- it is likely that the mechanisms of ac- changes in the target has been verified nation for the superior effect of angio- tion will be considerably different from in randomized trials. For instance, thia- tensin-converting enzyme inhibitors vs bisphosphonates and the conclusion zide diuretics and b-blockers have both calcium antagonists on clinically impor- that similar reductions in loss of bone been shown to reduce blood pressure tant outcomes is that angiotensin- density will lead to parallel reductions and clinically important outcomes such converting enzyme inhibition has bio- in clinical fractures is questionable. In as stroke in patients with hyperten- logical effects independent of lowering TABLE 2, we apply our validity criteria
sion. Thus, we would be much more blood pressure that reduce risk of stroke to a number of controversial examples comfortable relying on reduction in or death and that calcium antagonists of the use of surrogate end points.
blood pressure to justify administering do not share these effects.
a new b-blocker or thiazide diuretic than Second, treatment may have delete- What Were the Results? How
to justify offering a novel antihyperten- rious effects on the outcome that are not Large, Precise, and Lasting
sive agent from another class.46 mediated through the surrogate. Mor- Was the Treatment Effect?
For example, although 1 dihydropyri- tality-increasing effects of fibrates rather We are interested not only in whether dine calcium channel blocker has been than inability to lower morbidity and an intervention alters a surrogate end shown to reduce clinically important mortality through cholesterol reduc- point, but also in the magnitude, preci- outcomes in patients with hyperten- tion probably explain the lack of effect sion, and duration of the effect. If an in- sion,47 4 other trials have shown that of fibrates on clinically important out- tervention shows large reductions in the these agents are less efficacious than thia- comes. That such additional effects are surrogate end point, the 95% confi- zides or angiotensin-converting en- less likely across classes of drugs than dence intervals (CIs) around those large zyme inhibitors in preventing hard clini- within classes is what makes us more in- reductions are narrow, and the effect per- cal end points despite exerting similar clined to rely on within-class evidence sists over a sufficiently long period, our degrees of blood pressure lowering.48-51 from surrogate outcomes.
confidence that the target outcome will We will consider the example of cho- This criterion is complicated by the be favorably affected increases. Posi- lesterol reduction as a surrogate for car- variable definitions of drug class. A tive effects that are smaller, with wider diovascular outcomes such as myocar- manufacturer of a drug related to a class CIs, and shorter duration of follow-up dial infarction and death in part B of this of agents with a consistently positive as- leave us less confident.
Users' Guide.52 Briefly, several large tri- sociation between modification of a sur- We have already cited evidence sug- als of primary and secondary preven- rogate end point and modification of the gesting that CD4 cell counts may be an tion of coronary heart disease with target (such as a b-blocker) will natu- acceptable surrogate for mortality in pa- statins have consistently shown that rally argue for a broad definition of tients with HIV infection. A random- these drugs reduce cardiovascular out- class. Manufacturers of agents that are ized controlled trial of immediate vs de- related to drugs with known or sus- layed zidovudine therapy in HIV- We could therefore make the as- pected adverse effects on target events infected asymptomatic individuals sumption that a new statin with a simi- (clofibrate, or some calcium antago- declared a positive result for immediate lar low-density lipoprotein cholesterol– nists) are likely to argue, on the other therapy, largely on the basis of a greater lowering potency may also reduce hand, that the chemical or physiologi- proportion of treated patients with CD4 clinically important outcomes. How- cal connection is not sufficiently close cell counts above 435 3 106/L at a me- ever, we would be reluctant to gener- to consider the new drug to be in the dian follow-up of 1.7 years.58 Subse- alize to another class of lipid-lowering same class as the harmful agent. Part B quently, the Concorde study addressed agents since trials of 1 such class (the will address these issues more fully.52 the same question in a randomized trial 774 JAMA, August 25, 1999—Vol 282, No. 8
1999 American Medical Association. All rights reserved.
USERS' GUIDES TO THE MEDICAL LITERATURE with a median follow-up of 3.3 years.59 Returning to our scenario, the trial of same ones we have suggested for any is- The Concorde investigators found a con- raloxifene in women with osteoporosis sue of therapy or prevention 60 and elabo- tinuous decline in CD4 cell counts in demonstrated that after 2 years of treat- rated on in our Users' Guide regarding both treated and control groups, but the ment, raloxifene-treated patients in the applicability.61 These 3 questions have to median difference of 30 3 106/L in fa- group receiving the highest dosage do with whether the results can be ap- vor of treated patients at study termina- showed an increase in bone mineral den- plied to your patient's care, whether all tion was statistically significant. How- sity at the lumbar spine of 2.2% (SE, important outcomes were considered, ever, the study showed no effect of 0.3%) compared with a slight decrease and whether the likely benefits are worth zidovudine in terms of reduced progres- in the control group 0.8% (SE, 0.3%).
the down sides of treatment.
sion to AIDS or death. The median CD4 This difference in change over time was "Can the results be applied to my pa- cell count difference was insufficient to statistically significant (P,.03). Ide- tient's care" refers to the extent to which have an impact on clinically important ally, the investigators would have pro- the patient before you is similar to those outcomes. The Concorde authors made vided us with a CI around the 3% dif- who participated in the published stud- the following conclusion: the small, but ference in percentage change in bone ies under consideration, and the ex- highly significant persistent difference in mineral density in the treatment and con- tent to which the therapy, and the as- CD4 cell counts between the groups was trol groups. As we will illustrate when sociated technologies for monitoring not translated into a significant clinical we consider weighing benefits and and responding to complications, are benefit and "called into question the un- harms, the magnitude of the effect on the available in your setting. "Were all im- critical use of CD4 cell counts as a sur- surrogate may (or may not) help us es- portant outcomes considered" relates rogate endpoint." Had the Concorde timate the size of a possible affect on the to the focus of this Users' Guide, and analysis showed significantly shorter target outcome.
all the issues we have raised thus far: times to reach a CD4 cell count of was the primary outcome really the one 350 3 106/L in the control group and Will the Results Help in Caring
in which patients will be interested? been regarded as fundamental, the trial for My Patients?
This second criterion also draws is- might have been stopped early with a The questions clinicians should ask sues of adverse intervention effects to themselves in applying the results are the our attention. Applying the third cri- Table 2. Selected Examples of Applied Validity Criteria for the Critical Evaluation of Studies Using Surrogate End Points
Is There a Strong,
Is There Evidence
Is There Evidence
Trials in Other Drug
Trials in the Same
Classes That
Drug Class That
Between the
Improvement in the
Improvement in the
Surrogate End Point
Surrogate End Point
Point and the
Has Consistently Led
Has Consistently Led
Clinical End
to Improvement in
to Improvement in
the Target Outcome?
the Target Outcome?
End Point
End Point
Bone mineral density Osteoporotic fractures Protease inhibitor* Human immunodeficiency virus plasma load immunodeficiencysyndrome or death Reverse transcriptase inhibitor Human immunodeficiency virus viral plasma load immunodeficiencysyndrome or death Protease inhibitor* immunodeficiencysyndrome or death Reverse transcriptase inhibitor immunodeficiencysyndrome or death Antihypertensive drugs Stroke, myocardial Dihydropyridine calcium New thiazide diuretic Antilipidemic drugs Cholesterol or low-density Myocardial infarction, lipoprotein cholesterol *In combination therapy with 2 reverse transcriptase inhibitors.
1999 American Medical Association. All rights reserved.
JAMA, August 25, 1999—Vol 282, No. 8 775
USERS' GUIDES TO THE MEDICAL LITERATURE terion, judging whether the benefits are the limitations of this approach pointed deadly and usually relentless progres- worth the down sides of treatment, pre- out above) examine the results of ran- sion of HIV infection, and the paucity sents particular challenges when inves- domized controlled trials of alendro- of alternative therapies, has contrib- tigators have focused on surrogate end nate (a drug from a different class for uted to the readiness of patients, clini- points, and we will discuss this crite- which we have data on the same surro- cians, and regulatory agencies to ac- rion in some detail.
gate end point as well as clinical end cept evidence from surrogate end points points such as fracture reduction). While in instituting novel therapies in pa- Are the Likely Treatment Benefits
alendronate appears to improve verte- tients infected with HIV. In osteopo- Worth the Potential Harms
bral bone density by 7.5% over 2 years rosis, in which the consequences of the and Costs?
(vs control),42 raloxifene is associated condition are less immediately devas- To know whether to offer a treatment with only a 3.0% improvement over the tating, and a variety of agents are avail- to their patients, clinicians must be able same time frame. A systematic over- able, the case for relying on surrogate to estimate the magnitude of the likely view of the alendronate trials80 reported end points is far less compelling.
benefit. When the data available are lim- a 29% reduction in RR of nonvertebral ited to the effect on a surrogate end fracture over 2 years. Only 1 trial looked point, estimating the extent to which at symptomatic vertebral fractures in OF THE SCENARIO
treatment will reduce clinically impor- women with decreased bone density and We have found a strong, consistent, in- tant outcomes becomes a challenge.
an existing vertebral fracture.81 This dependent, and biologically plausible One approach is to extrapolate from study demonstrated an RR reduction of association between bone mineral den- 1 or more randomized trials assessing a 55% with alendronate and suggested that sity and vertebral and nonvertebral frac- related intervention in a similar patient our patient's risk over 3 years of a non- tures. Randomized trials, however, have population that provides both surro- vertebral fracture would be approxi- failed to show a consistent association gate end point and clinical outcome data.
mately 15%; symptomatic vertebral frac- between increased bone density and re- For example, until recently there were ture would be about 5%. Given the RR duction in fracture across all drug little long-term data on the efficacy of lo- reductions with alendronate, one would vastatin in reducing clinically impor- need to treat approximately 25 women Because our patient is at substantial tant outcomes. However, one could ex- to prevent a nonvertebral fracture and risk of fracture over the short term, the trapolate from short-term dose efficacy 40 women to prevent a symptomatic ver- number needed to treat to prevent both studies assessing the surrogate end point tebral fracture over a 3-year period.
nonvertebral and vertebral fractures is of cholesterol lowering. Thus, since 40 Since the improvement in bone min- moderate, as is the absolute benefit she mg of lovastatin produced a similar de- eral density with raloxifene is at best might expect. Moreover, she is inter- gree of lowering of low-density lipopro- 50% of the effect of alendronate, we ested in longer-term fracture preven- tein cholesterol as 40 mg of pravastatin would anticipate a considerably lower tion, and her risk will grow over time.
(31% vs 34% reduction) in the CURVES reduction in fracture risk with raloxi- One might offer her alternative inter- Study,77 one could theorize that lovas- fene. However, interim analysis of an ventions, including hormone replace- tatin would have similar long-term ben- ongoing raloxifene trial62 reported a ment therapy, calcium and vitamin D, efits to pravastatin. Subsequently, the AF- 46% RR reduction with this therapy bisphosphonates, or calcitonin.
CAPS/TexCAPS Trial (a 5-year trial (despite less of an increase in bone min- While there is strong evidence from assessing the efficacy of lovastatin in the eral density than seen with the alen- randomized trials supporting the use of primary prevention of ischemic heart dis- dronate trials). This serves to empha- bisphosphonates to decrease osteopo- ease)78 confirmed that this agent had a size the dangers of extrapolating results rotic fractures, randomized trial data beneficial profile similar to pravastatin across classes when it is uncertain that showing fracture reduction in popula- (as determined by the 5-year, primary the effects on clinically important out- tions similar to our patient with the other prevention WOSCOPS Trial)79: the RR comes are mediated in the same fash- agents is limited. Our patient is con- reductions (and 95% CIs) for myocar- ion by the 2 comparison drugs.
cerned about her long-term risk. Ralox- dial infarction were 40% (17%-57%) and In deciding whether the likely mag- ifene was well tolerated during this 2-year 31% (17%-43%), respectively. How- nitude of the treatment effect war- trial but no information is available about ever, this approach is likely to be seri- rants offering patients the interven- long-term adverse effects including car- ously flawed when one is extrapolating tion, clinicians must consider not only diovascular disease, venous thrombo- from trials of another class of drugs.
the uncertainty associated with that es- embolism, breast and endometrial can- Returning to our scenario, to esti- timate, but the trade-off with poten- cer, and menopausal symptoms. While mate the magnitude of the fracture tial toxic effects and costs of therapy.
a number of options (including a trial of reduction we might expect with ralox- In addition, clinicians must ponder the etidronate, offering hormone replace- ifene (in which we have only surrogate consequences of not treating, and the ment therapy, calcium and vitamin D, end point data), we could (recognizing available management alternatives. The calcitonin, or suggesting only a bal- 776 JAMA, August 25, 1999—Vol 282, No. 8
1999 American Medical Association. All rights reserved.
USERS' GUIDES TO THE MEDICAL LITERATURE anced diet and exercise) might be rea- comes of unequivocal importance to pa- 5. Saravolatz LD, Winslow DL, Collins G, et al. Zid-
ovudine alone or in combination with didanosine or
sonable, ideally the clinician would sub- tients is the only ironclad solution to the zalcitabine in HIV-infected patients with the ac- ject these options to the same scrutiny surrogate outcome dilemma. When cli- quired immunodeficiency syndrome or fewer than 200 applied to raloxifene.
nicians must choose between alterna- CD4 cells per cubic millimeter. N Engl J Med. 1996;335:1099-1106.
Data indicating a reduction in frac- tive interventions, trials should make 6. Hammer SM, Squires KE, Hughes MD, et al. A con-
ture rate would greatly strengthen the head-to-head comparisons between com- trolled trial of two nucleoside analogues plus indina-vir in persons with human immunodeficiency virus in- case for including raloxifene as the pre- peting treatments rather than restrict- fection and CD4 cell counts of 200 per cubic millimeter ferred option. Just as you are about to see ing comparisons of treatment to con- or less. N Engl J Med. 1997;337:725-733.
the patient (and, for us, just before this trol or placebo. We expand on this issue 7. Niebauer J, Coats AJ. Treating chronic heart fail-
ure: time to take stock. Lancet. 1997;349:966-967.
article went to press) you pick up a few in Part B of this Users' Guide. However, 8. Massie BM, Berk MR, Brozena SC, et al. Can fur-
of your latest editions of JAMA from the when patients' risk of serious morbidity ther benefit be achieved by adding flosequinan to pa-tients with congestive heart failure who remain symp- pile in the corner of your office, and find or mortality are high, this "wait-and- tomatic on diuretic, digoxin, and an angiotensin 2 highly relevant randomized trials.82,83 see" strategy may pose problems for converting enzyme inhibitor? results of the flosequinan-ACE inhibitor trial (FACET). Circulation. 1993;88:492- The results show that, in women like many patients and their physicians.
your patient with a prevalent vertebral We encourage clinicians to criti- 9. How to read clinical journals, IV: to determine eti-
fracture, raloxifene decreased radiologi- cally question therapeutic interven- ology or causation. CMAJ. 1981;124:985-990.
10. Oxman AD, Sackett DL, Guyatt GH. Users' guides
cal vertebral fracture risk (for 60 mg: tions in which the only proof of effi- to the medical literature, I: how to get started. JAMA. number needed to treat = 16 [RR, 0.7; cacy is from surrogate end point data.
1993;270:2093-2095.
11. Guyatt G, Walter S, Shannon H, Cook D, Jae-
95% CI, 0.6-0.9]; and for 120 mg: num- When the surrogate end point meets all schke R, Heddle N. Basic statistics for clinicians, IV: cor- ber needed to treat = 10 [RR, 0.5; 95% CI, our validity criteria, the effect of the in- relation and regression. CMAJ. 1995;152:497-504.
0.4-0.7]), but did not decrease the inci- tervention on the surrogate end point 12. Guyatt GH, Thompson PJ, Berman LB, et al. How
should we measure function in patients with chronic heart
dence of nonvertebral fracture. In help- is large, the patient's risk of the target and lung disease? J Chronic Dis. 1985;38:517-524.
ing your patient to decide on the right outcome is high, the patient places a 13. Mahler DA, Weinberg DH, Wells CK, Feinstein
AR. The measurement of dyspnea: contents, interob-
course of action, you realize you will have high value on avoiding the target out- server agreement, and physiologic correlates of two to consider other effects of raloxifene: the come, and there are no satisfactory al- new clinical indexes. Chest. 1984;85:751-758.
14. Verschuren WM, Jacobs DR, Bloemberg BP, et al.
JAMA articles also show a 76% RR re- ternative therapies, clinicians can rec- Serum total cholesterol and long-term coronary heart duction of breast cancer as detected by ommend therapy on the basis of disease mortality in different cultures. JAMA. 1995; mammography (number needed to treat, randomized trials evaluating only sur- 274:131-136.
15. Mellors JW, Rinaldo CR Jr, Gupta P, et al. Prog-
126), a 3-fold increase in the risk of ve- rogate end points. In other situations, nosis in HIV-1 infection predicted by the quantity of nous thromboembolism, and an in- clinicians must carefully consider the virus in plasma. Science. 1996;272:1167-1170.
16. Mellors JW, Kingsley LA, Rinaldo CR, et al. Quan-
creased incidence of hot flashes, leg known adverse effects and cost of titation of HIV-1 RNA in plasma predicts outcome after cramps, influenzalike syndromes, and therapy, and the possibility of unan- seroconversion. Ann Intern Med. 1995;122:573-579.
17. Ruiz L, Romeu J, Clotet B, et al. Quantitative HIV-1
peripheral edema.
ticipated adverse effects, before recom- RNA as a marker of clinical stability and survival in a When we use surrogate end points mending an intervention solely on the cohort of 302 patients with a mean CD4 cell count of to make inferences about expected ben- basis of surrogate end point data.
300 3 10(6)/1. AIDS. 1996;10:F39-F44.
18. O'Brien TR, Blattner WA, Waters D, et al. Serum
efit, we are making assumptions re- HIV-1 RNA levels and time to development of AIDS garding the link between the surro- Acknowledgment: We are grateful to Cliff Rosen, MD,
in the Multicenter Hemophilia Cohort Study. JAMA. for his helpful comments concerning the scenario and gate end point and the target outcome.
the associated discussion. Deborah Maddock pro- 19. Yerly S, Perneger TV, Hirschel B, et al. A critical
We have outlined criteria clinicians can vided invaluable coordination for the EBM Working assessment of the prognostic value of HIV-1 RNA lev- Group in the development of this article.
use to decide when these assumptions els and CD4+ cell counts in HIV-infected patients. ArchIntern Med. 1998;158:247-252.
might be appropriate. Even if a surro- 20. Ho DD. Viral counts in HIV infection. Science.
gate end point meets all of these crite- 1. Delmas PD, Bjarnason NH, Mitlak BH, et al. Ef-
21. Cummings SR, Nevitt MC, Browner WS, et al. Risk
ria, inferences about a treatment ben- fects of raloxifene on bone mineral density, serum cho- factors for hip fracture in white women. N Engl J Med. efit may still prove misleading. Thus, lesterol concentrations, and uterine endometrium in 1995;332:767-773.
22. Marshall D, Johnell O, Wedel H. Meta-analysis
treatment recommendations based on postmenopausal women. N Engl J Med. 1997;337:1641-1647.
of how well measures of bone mineral density pre- surrogate outcome effects can never be 2. Temple RJ. A regulatory authority's opinion about
dict occurrence of osteoporotic fractures. BMJ. 1996; strong. Furthermore, difficulties in es- surrogate endpoints. In: Nimmo WS, Tucker GT, eds.
Clinical Measurement in Drug Evaluation. New York, 23. Huang C, Ross PD, Wasnich RD. Short-term and
timating the magnitude of effects on NY: John Wiley & Sons Inc; 1995:57.
long-term fracture prediction by bone mass measure- clinically important end points com- 3. Hammer SM, Katzenstein DA, Hughes MD, et al.
ments. J Bone Miner Res. 1998;13:107-113.
A trial comparing nucleoside monotherapy with com- 24. Bigger JT Jr, Fleiss JL, Kleiger R, et al. The relation-
promises economic analysis examin- bination therapy in HIV-infected adults with CD4 cell ships among ventricular arrhythmias, left ventricular dys- ing the cost-effectiveness of alterna- counts from 200 to 500 per cubic millimeter. N Engl J function, and mortality in the 2 years after myocardial tive management strategies.
4. Delta Coordinating Committee. Delta: a ran-
25. McAlister FA, Teo KK. Antiarrhythmic therapies
These considerations emphasize that domised double-blind controlled trial comparing com- for the prevention of sudden cardiac death. Drugs. waiting for randomized trials investigat- binations of zidovudine plus didanosine or zalcitab- ine with zidovudine alone in HIV-infected individuals.
26. Echt DS, Liebson PR, Mitchell LB, et al, and the
ing the effect of the intervention on out- Cardiac Arrhythmia Suppression Trial. Mortality and 1999 American Medical Association. All rights reserved.
JAMA, August 25, 1999—Vol 282, No. 8 777
USERS' GUIDES TO THE MEDICAL LITERATURE morbidity in patients receiving encainide, flecainide, double-blind comparison of placebo and active treat- trial of addition of lamivudine or lamivudine plus lo- or placebo. N Engl J Med. 1991;324:781-788.
ment for older patients with isolated systolic hyperten- viride to zidovudine-containing regimens for pa- 27. Moore TJ. Deadly Medicine. New York, NY: Si-
sion. Lancet. 1997;350:757-764.
tients with HIV-1 infection: the CAESAR trial. Lancet. mon & Schuster; 1995.
48. Psaty BM, Siscovick DS, Weiss NS, et al. Hyperten-
28. Drexler H, Banhardt U, Meinertz T, et al. Con-
sion and outcomes research: from clinical trials to clini- 67. Fischl M, Greenberg S, Clumeck N, et al. Safety
trasting peripheral short-term and long-term effects cal epidemiology. Am J Hypertens. 1996;9:178-183.
and activity of abacavir (1592, ABC) with 3TC/ZDV of converting enzyme inhibition in patients with con- 49. Borhani NO, Mercuri M, Borhani PA, et al. Final
in antiretroviral naive subjects. In: Abstracts from the gestive heart failure: a double-blind, placebo- outcome results of the multicenter isradipine diuretic 12th World AIDS Conference: Geneva, Switzerland; controlled trial. Circulation. 1989;79:491-502.
atherosclerosis study (MIDAS): a randomized con- June 28-July 3, 1998.
29. Lewis GR. Comparison of lisinopril versus pla-
trolled trial. JAMA. 1996;276:785-791.
68. Katzenstein DA, Hammer SM, Hughes MD, et al,
cebo for congestive heart failure. Am J Cardiol. 1989; 50. Tatti P, Pahor M, Byington RP, et al. Outcome re-
and the AIDS Clinical Trials Group Study 175 Virology sults of the fosinopril versus amlodipine cardiovascular Study Team. The relation of virologic and immunologic 30. Giles TD, Fisher MB, Rush JE. Lisinopril and cap-
events randomized trial (FACET) in patients with hyper- markers to clinical outcomes after nucleoside therapy topril in the treatment of heart failure in older pa- tension and NIDDM. Diabetes Care. 1998;21:597-603.
in HIV-infected adults with 200 to 500 CD4 cells per tients. Am J Med. 1988;85:44-47.
51. Estacio RO, Jeffers BW, Hiatt WR, Biggerstaff SL,
cubic millimeter. N Engl J Med. 1996;335:1091-1098.
31. Riegger GA. Effects of quinapril on exercise tol-
Gifford N, Schrier RW. The effect of nisoldipine as com- 69. Collins R, Peto R, MacMahon S, et al. Blood pres-
erance in patients with mild to moderate heart fail- pared with enalapril on cardiovascular outcomes in pa- sure, stroke, and coronary heart disease, part 2: short- ure. Eur Heart J. 1991;12:705-711.
tients with non-insulin-dependent diabetes and hy- term reductions in blood pressure; overview of ran- 32. Garg R, Yusuf S. Overview of randomized trials
pertension. N Engl J Med. 1998;338:645-652.
domised drug trials in their epidemiological context.
of angiotensin-converting enzyme inhibitors on mor- 52. McAlister FA, Laupacis A, Wells GA, Sackett DL,
tality and morbidity in patients with heart failure. JAMA. for the Evidence-Based Medicine Working Group. Us- 70. MacMahon S, Peto R, Cutler J, et al. Blood pres-
ers' guide to the medical literature, XIX: applying clini- sure, stroke, and coronary heart disease, part 1: pro- 33. Di Bianco R, Shabetai R, Kostuk W, et al. A com-
cal trial results part B: guidelines for determining longed differences in blood pressure; prospective ob- parison of oral milrinone, digoxin, and their combi- whether a drug is exerting (more than) a class effect.
servational studies corrected for the regression dilution nation in the treatment of patients with chronic heart JAMA. In press.
bias. Lancet. 1990;335:765-774.
failure. N Engl J Med. 1989;320:677-683.
53. Bucher HC, Griffith LE, Guyatt GH. Systematic re-
71. Psaty BM, Smith NL, Siscovick DS, et al. Health
34. Sueta CA, Gheorghiade M, Adams KF, et al, and
view on risk and benefit of different cholesterol low- outcomes associated with antihypertensive therapies the Epoprostenol Multicenter Research Group. Safety ering interventions. Arterioscler Thromb Vasc Biol. used as first-line agents: a systematic review and meta- and efficacy of epoprostenol in patients with severe con- analysis. JAMA. 1997;277:739-745.
gestive heart failure. Am J Cardiol. 1995;75:34A-43A.
54. Muldoon MF, Manuck SB, Matthews KA. Low-
72. Heinonen TM, Stein E, Weiss SR, et al. The lipid-
35. Packer M, Carver JR, Rodeheffer RJ, et al, for the
ering cholesterol concentration and mortality. BMJ. lowering effects of atorvastatin, a new HMG-coA re- Promise Study Research Group. Effect of oral milri- ductase inhibitor: results of a randomized, double- none on mortality in severe chronic heart failure.
55. Smith GD, Song F, Sheldon TA. Cholesterol low-
masked study. Clin Ther. 1996;18:853-863.
N Engl J Med. 1991;325:1468-1475.
ering and mortality. BMJ. 1993;306:1367-1373.
73. Bakker-Arkema RG, Davidson MH, Goldstein RJ,
36. Califf RM, Adams KF, McKenna WJ, et al. A ran-
56. Prentice RL. Surrogate endpoints in clinical trials.
et al. Efficacy and safety of a new HMG-coA reduc- domized controlled trial of epoprostenol therapy for Stat Med. 1989;8:431-440.
tase inhibitor, atorvastatin, in patients with hypertri- severe congestive heart failure. Am Heart J. 1997; 57. Fleming TR. Surrogate markers in AIDS and can-
glyceridemia. JAMA. 1996;275:128-133.
cer trials. Stat Med. 1994;13:1423-1435.
74. Law MR, Wald NJ, Thompson SG. By how much
37. Yee KM, Struthers AD. Can drug effects on mor-
58. Cooper DA, Gatell JM, Kroon S, et al, and the Eu-
and how quickly does reduction in serum cholesterol tality in heart failure be predicted by any surrogate mea- ropean-Australian Collaborative Group. Zidovudine in concentration lower risk of ischaemic heart disease? sure? Eur Heart J. 1997;18:1860-1864.
persons with asymptomatic HIV infection and CD4+ 38. Hampton JR, van Veldhuisen DJ, Kleber FX, et al.
cell counts greater than 400 per cubic millimeter.
75. Winocour PH, Durrington PN, Bhatagnar D, et al.
Randomised study of effect of ibopamine on survival N Engl J Med. 1993;329:297-303.
The effect of bezafibrate on very low density lipopro- in patients with advanced severe heart failure. Lancet. 59. Concorde Coordinating Committee. Concorde:
tein (VLDL), intermediate density lipoprotein (IDL), and MRC/ANRS randomised double-blind controlled trial low density lipoprotein (LDL) composition in type 1 dia- 39. Cameron DW, Heath-Chiozzi M, Danner S, et al,
of immediate and deferred zidovudine in symptom- betes associated with hypercholesterolaemia or combined and the Advanced HIV Disease Ritonavir Study Group.
free HIV infection. Lancet. 1994;343:871-881.
Randomised placebo-controlled trial of ritonavir in ad- 60. Guyatt GH, Sackett DL, Cook DJ, for the Evidence-
76. Jones IR, Swai A, Taylor R, Miller M, Laker MF,
vanced HIV-1 disease. Lancet. 1997;351:543-549.
Based Medicine Working Group. Users' guides to the Alberti KG. Lowering of plasma glucose concentra- 40. Watts NB, Harris ST, Genant HK, et al. Intermit-
medical literature, II: how to use an article about tions with bezafibrate in patients with moderately con- tent cyclical etidronate treatment of postmenopausal therapy or prevention A: are the results of the study trolled NIDDM. Diabetes Care. 1990;13:855-863.
osteoporosis. N Engl J Med. 1990;323:73-79.
valid? JAMA. 1993;270:2598-2601.
77. Jones P, Kafonek S, Laurora I, Hunninghake D.
41. Storm T, Thamsborg G, Steiniche T, Genant HK,
61. Dans AL, Dans LF, Guyatt GH, Richardson S, for
Comparative dose efficacy study of atorvastatin ver- Sorensen OH. Effect of intermittent cyclical etidro- the Evidence-Based Medicine Working Group. Us- sus simvastatin, pravastatin, lovastatin, and fluvas- nate therapy on bone mass and fracture rate in women ers' guides to the medical literature, XIV: how to de- tatin in patients with hypercholesterolemia (the with postmenopausal osteoporosis. N Engl J Med. cide on the applicability of clinical trial results to your CURVES study). Am J Cardiol. 1998;81:582-587.
patient. JAMA. 1998;279:545-549.
78. Downs JR, Clearfield M, Weis S, et al. Primary pre-
42. Liberman UA, Weiss SR, Broll J, et al, and the Alen-
62. Ettinger B, Black D, Cummings S, et al. Raloxi-
vention of acute coronary events with lovastatin in men dronate Phase III Osteoporosis Treatment Study Group.
fene reduces the risk of incident vertebral fractures: and women with average cholesterol levels. JAMA. Effect of oral alendronate on bone mineral density and 24 month interim analyses. Osteoporos Int. 1998;8 the incidence of fractures in postmenopausal osteo- (suppl 3):11.
79. Shepherd J, Cobbe SM, Ford I, et al. Prevention of
porosis. N Engl J Med. 1995;333:1437-1443.
63. Clendeneninn N, Quart B, Anderson R, Knowles
coronary heart disease with pravastatin in men with hy- 43. Riggs BL, Hodgson SF, O'Fallon WM, et al. Ef-
M, Chang Y. Analysis of long-term virologic data from percholesterolemia. N Engl J Med. 1995;333:1301-1307.
fect of fluoride treatment on the fracture rate in post- the viracept (nelfinavir) 511 protocol using 3 HIV-RNA 80. Karpf DB, Shapiro DR, Seeman E, et al. Preven-
menopausal women with osteoporosis. N Engl J Med. assays. In: Abstracts from the 5th Conference on Ret- tion of nonvertebral fractures by alendronate: a meta- roviruses and Opportunistic Infections; Chicago, Ill; 1998.
analysis. JAMA. 1997;277:1159-1164.
44. Meunier PJ, Sebert J-L, Reginster J-Y, et al. Fluo-
64. Brun-Vezinet F, Boucher C, Loveday C, et al, and
81. Black DM, Cummings SR, Karpf DB, et al, and the
ride salts are no better at preventing new vertebral frac- the Delta Virology Working Group and Coordinating Fracture Intervention Trial Research Group. Ran- tures than calcium-vitamin D in postmenopausal os- Committee. HIV-1 viral load, phenotype, and resis- domised trial of effect of alendronate on risk of frac- teoporosis. Osteoporos Int. 1998;8:4-12.
tance in a subset of drug-naive participants from the ture in women with existing vertebral fractures. Lan- 45. Dawson-Hughes B, Harris SS, Krall EA, Dallal GE.
Delta trial: the national virology groups. Lancet. 1997; Effect of calcium and vitamin D supplementation on 82. Cummings SR, Eckert S, Krueger KA, et al. The
bone density in men and women 65 years of age and 65. Montaner JS, Reiss P, Cooper D, et al. A random-
effect of raloxifene on risk of breast cancer in post- older. N Engl J Med. 1977;337:670-676.
ized, double-blind trial comparing combinations of ne- menopausal women. JAMA. 1999;281:2189-2197.
46. McAlister FA, Straus S, Sackett DL. Randomized
virapine, didanosine, and zidovudine for HIV- 83. Ettinger B, Black DM, Mitlak BH, et al. Reduction
clinical trials of antihypertensive drugs: all that glit- infected patients: the Netherlands, Canada, and of vertebral fracture risk in postmenopausal women with ters is not gold. CMAJ. 1998;159:488-490.
Australia (INCAS) study. JAMA. 1998;279:930-937.
osteoporosis treated with raloxifene: results from a 3-year 47. Staessen JA, Fagard R, Thijs L, et al. Randomised
66. CEASAR Coordinating Committee. Randomised
randomized trial. JAMA. 1999;282:637-645.
778 JAMA, August 25, 1999—Vol 282, No. 8
1999 American Medical Association. All rights reserved.

Source: http://www.nuthalapaty.net/kb/ebm/docs/jama19a.pdf