Document resource
In the past three decades, the US Food and Drug Administration (FDA) has increasingly relied on surrogate endpoints (eg, disease-free and progression-free survival) to approve cancer drugs.1 2 In 2017, the percentage of approved cancer drugs demonstrating improvement in overall survival (OS) fell to 7% (n=2), compared with the 93% (n=25) demonstrating improvements in surrogate endpoints only.2 Health economists are thus faced with the challenge of approximating OS in model-based economic evaluations, as part of the health technology assessment process to help guide reimbursement decisions for both private and public insurers. Here are three characteristics of contemporary cancer survival that health economists and clinicians need to consider for more robust economic evaluations.The correlation between OS and surrogate endpoints may be low In the absence of OS data, surrogate endpoints must be used to predict OS in economic modelling. The ideal surrogate endpoints for OS are those that have (1) a strong correlation with OS (lower limit of the 95% confidence interval [CI] of the correlation coefficient is ≥0.85 3) and (2) between-group differences (in intervention effect) that may be used to predict between-group differences in OS.4 However, these two conditions are often not assessed. Between 1992 and 2019, the FDA approved 64 cancer drugs based on surrogate endpoints. Of these, only 3 reported high correlation between the assessed surrogate endpoint and OS; 1 reported medium correlation, 10 reported poor correlation and 39 did not document any correlation.1 Characteristics of the trial design (eg, follow-up duration, use of a crossover or parallel group, censoring methods) may further impact the degree of discrepancy between the observed surrogates and OS.4–6Some low-risk cancers have OS prognosis comparable to the general population According to the Surveillance, Epidemiology and End Results (SEER) programme, the 5-year relative survival rate (RSR) of all cancers in the USA was 69% between 2014 and 2020. 7 RSR is the ratio of observed survival of people with cancer to the expected survival of people with similar characteristics without cancer (in practice, the general population). Meanwhile, the 5- year RSR of localised breast (female), melanoma, prostate and thyroid cancers all exceeded 99%.7 Similar statistics are observed in Canada.8 Because the OS of people with cancer cannot, in theory, exceed the age-specific and sex-specific OS of the general population, these high RSRs indicate that patients with low-risk cancers have nearly the same survival rate as people without cancer. It can therefore be reasonably expected that new cancer interventions for low-risk cancers will have minimal effect on OS, even if they demonstrate improvements in surrogate endpoints.9The conditional relative survival of most cancers improves over time Cancer-specific mortality is modelled to increase over time in most state-transition models. As a simple example, consider a Markov model with a monthly cycle that includes four health states: (1) progression-free, (2) progression, (3) metastasis and (4) death. People in health states (1) to (3) may transition to the death state each month based on the monthly probability of death (mainly due to cancer, but may also be due to other competing diseases and conditions 10) estimated using the cancer stage and demographic prognostic variables. This means at each cycle, people in the metastatic health state are assigned a higher probability of death than those in other states. Over the model time horizon, people will transition from earlier health states to reflect cancer progression, leading to an increasing probability of death over time.While this model structure appears logical and is widely used in the literature, it may not align with observed survival statistics from cancer registries. In the USA and Canada, the rate of cancer-specific mortality is often higher in the first few years after diagnosis, followed by declining excess deaths thereafter. This phenomenon can be depicted by the conditional relative survival (CRS), which represents the probability of a person with cancer surviving a specific time into the future, after having survived a given number of years following diagnosis. SEER data from 2000 to 2020 show that the 5-year CRS among all people with cancer who have survived 1, 5 and 10 years since their diagnosis is 81.8%, 92.6% and 95.0%, respectively.7 Similar trends have been reported in Canada.8 This data clearly indicates that over time, the probability of surviving cancer improves with each subsequent year people remain alive since diagnosis, which is in contrast to the Markov model structure and results we demonstrated earlier.In conclusion, health economists and clinicians need to consider these three characteristics of contemporary cancer survival, which are critical for robust economic evaluations of new cancer interventions.