It is intuitive and obvious that no person desires a myocardial infarction, stroke, or recurrence or metastasis of a previously localised cancer. Reducing the risk of these endpoints is therefore used as the evidentiary basis of a wide range of drugs, devices, and surgical procedures. However, recent trials in cardiovascular and cancer medicine force us to revisit the assumption of their value as endpoints. Several studies have produced discordant results, with interventions that reduce myocardial infarction, stroke, or metastasis having no effect on cardiovascular death, health related quality of life, cancer related death, and overall mortality—four harder clinical endpoints.
Is the avoidance of myocardial infarction, stroke, and cancer recurrence or metastasis less meaningful than it used to be? Or has the severity of these events lessened, partly because of advances in imaging and assay sensitivity? As technology advances and diagnostic categories broaden we need to consider whether there comes a point when endpoints that are widely believed to be measures of patient centred outcomes become test based, non-clinical endpoints. We have selected four recent randomised trials that highlight how benefits hinge on the nature of events averted.
Trials with discordant results
The FOURIER trial compared the effect of evolocumab, a monoclonal antibody that inhibits proprotein convertase subtilisin-kexin type 9 (PCSK9), with placebo in patients with cardiovascular disease.1 The trial used a primary composite endpoint of cardiovascular death, myocardial infarction, stroke, hospital admission for unstable angina, or coronary revascularisation and found a benefit for the drug (1344/13 784 (9.8%) had reductions in the composite endpoint v 1563/13 780 (11.3%) with placebo; hazard ratio=0.85, 95% confidence interval 0.79 to 0.92; P<0.001). However, the drug had no favourable effect on cardiovascular death (251/13 784 (1.8%) v 240/13 780 (1.7%); HR=1.05, 95% CI 0.88 to 1.25; P=0.62) or overall mortality (444/13 784 (3.2%) v 426/13 780 (3.1%); HR=1.04, 95% CI 0.91 to 1.19; P=0.54).
The findings contrast with those of older secondary prevention trials in which cardiovascular endpoints and mortality improved in parallel. A decade earlier, for example, the PROVE IT–TIMI 22 trial randomised patients admitted to hospital with an acute coronary syndrome to standard dose pravastatin or high intensity atorvastatin.2 At two years follow-up, the trial found significant improvement in the composite primary endpoint of stroke, death from any cause, myocardial infarction, unstable angina requiring hospital admission, or, the most common event contributing to the composite, revascularisation. This occurred alongside non-significant improvements in death from all cause (3.2% standard dose statin v 2.2% with high intensity statin; absolute numbers and P value were not reported), and death from coronary heart disease (1.4% v 1.1%).
Patients in FOURIER and PROVE IT-TIMI 22 were evaluated at comparable time points (average follow up 26.4 months and 24 months, respectively). In other words, the 15% relative risk reduction in the primary composite endpoint in FOURIER was associated with no perceivable difference in harder endpoints, whereas the 16% reduction in the primary endpoint of PROVE-IT-TIMI-22 occurred alongside suggestion of improvement in harder outcomes.
Cancer medicine provides three further examples, with the endpoint of recurrence or metastasis after potentially curative therapy not translating into survival benefits. For prostate cancer, the ProtecT investigators found no difference between active surveillance, radical prostatectomy, and radical radiotherapy with hormones on cancer specific mortality among men with localised prostate cancer.3 Among the arms, prostatectomy had the greatest negative effect on sexual function and urinary continence, while radiotherapy had significant negative effects on sexual function and bowel function. Higher rates of metastases were observed in the active monitoring group.
There was no difference in health related quality of life between the arms, but, because of the increased rate of metastases, some experts argue that active surveillance is inferior.4 In a defence of this view, D’Amico writes, “Developing metastatic PC is a life-changing event … which brings the fear of dying of PC into” patients’ lives,4 and, “Is it really considered a success if a man suffers recurrence with metastatic disease, which often means a painful bone fracture along with castration and all of its toxicities”?4
Another example is sunitinib, which the US Food and Drug Administration approved for the adjuvant treatment of resected kidney cancer in 2017.5 This approval was granted on the basis of one randomised trial (S-TRAC) that showed sunitinib, when compared against observation, delayed the time until the composite endpoint of recurrence or death (hazard ratio=0.76, 95% CI 0.59 to 0.98; P=0.03),6 though it did so without any improvement in overall survival (1.01, 0.72 to 1.44; P=0.94). Overall survival curves, presented in the paper’s supplement, are superimposable over the first eight years of follow-up, suggesting that the lowered rate of recurrence has shown no evidence of translating into improved survival.
Finally, consider gefitinib, a small molecular inhibitor of the epidermal growth factor receptor. In a randomised trial of patients with lung cancer who had their tumours resected (ADJUVANT), the drug improved the time till disease recurrence or death (hazard ratio=0.60; 95% CI 0.42 to 0.87; P=0.005).7 Overall survival was virtually identical between the arms, with 34.2% of patients in the trial dying, whereas in previous studies of non-small cell lung cancer, improvements in disease recurrence strongly predicted improvements in overall survival.8
Weighing the benefits and harms
Together these examples show a new challenge facing patients and providers. If the myocardial infarctions, strokes, and metastases averted in these studies are severe, then the interventions probably offer net benefit. At the same time, if the events averted are mild— for instance, if most metastases are asymptomatic or indolent—then net benefit becomes questionable, particularly when weighed against side effects and therapeutic and financial burden.
At some point, if observed endpoints do not directly affect patients’ perceived health or health related quality of life, they may be thought of as a risk factor for a future clinical event, and not an event themselves. An easy way to adjudicate these concerns is appealing to health related quality of life—did it improve? However, in the trials described above this was not reported for evolocumab, showed no difference in ProtecT, was worse in S-TRAC, and modestly improved when comparing gefitinib with chemotherapy in the ADJUVANT trial.
Evidence exists of a progressive expansion of all these disease categories over time. Diagnoses of myocardial infarctions have changed substantially since the introduction of more sensitive cardiac markers, including creatinine phosphokinase (CK), myocardial band fraction of CK, and high sensitivity cardiac troponin.910 This expansion requires that we consider two new aspects when interpreting the biomarkers: a gradient risk in biomarker levels, as suggested by the prognostic value of different troponin levels,11 and increased sensitivity of the disease.12 Professional cardiology societies have cautioned about the importance of applying constant definitions when tracking time trends in the rate of myocardial infarction when diagnostic criteria are altered.13
In the case of stroke, improvements in the availability and quality of imaging techniques may have resulted in more silent infarcts being identified, leading to a lower diagnostic threshold.14 Asymptomatic ischaemic lesions occur in 7-28% of elderly people, more than fivefold higher than the prevalence of symptomatic stroke.1415 These infarcts have been reported to be associated with subtle deficits in physical and cognitive function and increased risk of subsequent stroke or dementia,14 but it is not clear whether these events compromise health related quality of life and, most importantly, whether their avoidance is worth the side effects and burden of intervention.
In the evolocumab study, the drug’s massive effect on low density lipoprotein cholesterol level—reducing it from a mean of 2.3 mmol/L in the control arm to 0.8 mmol/L in the intervention arm—is likely to have effectively unblinded the study.16 Knowledge of LDL may have influenced decision making and coding of subsequent events. Cardiologists may be more likely to recommend percutaneous coronary angioplasty in a patient with LDL over 2.6 mmol/L than below 1.3 mmol/L at similar levels of chest pain or discomfort. And, though the FOURIER authors report there was no increase in type 4 (or percutaneous angioplasty related) myocardial infarctions between the arms, they fail to report what percentage of infarctions were ST elevated, leaving open the question of the clinical severity of the events.17
Moreover, although the trial met its primary composite endpoint, cardiovascular death and overall mortality were not improved. Beyond type 4 myocardial infarctions, knowledge of LDL may influence the coding of other cardiac events that require provider discretion. For instance, in a case of suspected unstable angina in someone with an ambiguous clinical history, a clinician may rely on a low but detectable troponin level, and knowledge of a patient’s most recent LDL measurement may affect such decisions.1618
Interpretation of the FOURIER trial requires knowledge of the rates of ST elevated infarction (an echocardiography driven endpoint) and infarction leading to cardiogenic shock requiring pressors, another more objective endpoint. Additionally, further data, including the presence of new ST-T changes or Q waves, new left bundle branch block, imaging evidence of loss of viable myocardium, or new regional wall motion abnormality should be reported.19
In the case of ProtecT, the higher incidence in metastasis did not worsen a true patient centred endpoint: health related quality of life. The trial defined metastasis broadly as the spread of prostate cancer to bone, viscera, or lymph nodes or a PSA level >100 ng/mL, and it does not report how often metastasis led to fracture. Earlier trials used narrower definitions. For example, compared with the Scandinavian Prostate Cancer Group 4 study,2021 ProtecT added PSA rise over 100 ng/mL and regional lymph node disease to the definition of metastasis.
This disease expansion matters because metastatic disease becomes a composite outcome of several discrete events, not unlike composite outcomes in cardiology. And while there is clear evidence that skeletal related events do confer a poor prognosis,22 there is not comparable evidence for each of the other events included in the composite.
Making endpoints more meaningful
An important step towards untangling these endpoints is further details on trial outcomes. The evolocumab investigators could break down myocardial infarction into ST elevation, non-ST elevation, those that occurred immediately after revascularisation and those that occurred spontaneously, and those that resulted in shock or long term diminished ejection fraction and those that did not. Among infarctions without ST elevation, the investigators could provide breakdown by TIMI risk score to further gauge severity and, again, those leading to shock or systolic dysfunction. Authors should report events based on the third universal definition of myocardial infarction.19 The ultimate solution to interpreting endpoints would be the reporting of outcomes with individual endpoint biomarker or quantitative imaging data for independent review.
In the case of ProtecT, given concern about metastasis leading to fracture, this outcome could be explicitly reported. There seems little value in having experts speculate about the consequences of metastasis4 when these are known. Data sharing has potential to lead to greater clarity and transparency, as other investigators may further refine the breakdown and transparency of endpoints, particularly in trials over time.
Finally, validity of outcomes may also be improved by greater patient involvement at the outset of clinical trials, in their design and conduct. Patients can help to develop instruments that best capture their burden of symptoms or sequelae of disease. Promising efforts are already being made, including the collaborative group OMERACT in rheumatoid arthritis outcomes. We suggest that this process is iterative, performed before, during, and after trials. This way, patient experiences during the clinical trial process may improve capture of outcomes in subsequent investigations.
It is understandable that investigators would be tempted to declare reductions in cancer metastasis, myocardial infarction, or stroke as proof of therapeutic efficacy. However, because diagnostic drift now includes illness of lesser severity, it is no longer clear that any of these events implies loss in health related quality of life. Direct consideration of health related quality of life has the added advantage of balancing benefit against the harm of interventions, such as side effects, toxicity, treatment burden, time commitment, and financial costs. We believe greater detail in outcomes reporting and data sharing can overcome this challenge, which, to a large degree, represents our success in the technological advancement of diagnostic testing.
Improved diagnostic technology and expanding definitions may be reducing the relevance of accepted trial endpoints
Reductions in myocardial infarction, stroke, and cancer metastases may not always translate to improved quality of life or survival
Trialists must routinely measure and report health related quality of life
Data sharing and patient involvement in developing trial protocol may ameliorate these concerns