Disease-Free Survival

Disease-Free Survival in Adjuvant Cancer Trials and Economic Models

After a cancer has been removed by surgery or treated with radiotherapy, the question for an adjuvant treatment is whether it stops the disease coming back, and disease-free survival (DFS) is the endpoint many such trials use to answer it. A health economist meets DFS in two places. It is often the main efficacy evidence in appraisals of adjuvant treatments, reported years before overall survival is mature, and its curve sets how long a modelled cohort stays in the disease-free health state, where quality of life is usually highest and costs lowest. The value of the treatment then depends on whether a gain in DFS turns into longer life, a link that has been tested with individual patient data in adjuvant colon cancer but has to be argued afresh in other cancers and for other kinds of drug.

How the endpoint is defined and where trials differ

The FDA guidance on clinical trial endpoints for cancer drugs, issued in December 2018, says that DFS is generally defined as the time from randomisation until disease recurrence or death from any cause. The guidance names the adjuvant setting, after definitive surgery or radiotherapy, as the most frequent use of the endpoint, and says that "DFS can be an important endpoint in situations where survival may be prolonged, making a survival endpoint impractical". DFS is a composite endpoint, because a patient's time ends at whichever qualifying event comes first.

$$T_{DFS} = \min(T_{R}, T_{D}), \qquad S_{DFS}(t) = P(T_{DFS} > t)$$

where $T_{DFS}$ is the time from the origin (usually randomisation) to the first DFS event, $T_{R}$ is the time to recurrence, $T_{D}$ is the time to death from any cause, $S_{DFS}(t)$ is the DFS survival function and $t$ is time since the origin. Patients with neither event by the end of follow-up are censored, as in any time-to-event analysis.

The label is less fixed than the formula suggests. Punt and colleagues reported in 2007 a review of 52 studies of adjuvant treatment for colon cancer published between 1997 and 2006, which used eight endpoints in addition to overall survival, and found that the definitions of those endpoints, and the starting point for measuring time to the events, varied widely. Their expert panel then agreed a definition for each endpoint.

The DATECAN initiative returned to the problem in 2020 with consensus definitions for adjuvant colon cancer trials. Its panel recommended DFS, rather than recurrence-free survival (RFS, the abbreviation also used for relapse-free survival) or time to recurrence, as the primary endpoint when overall survival is not primary, and defined RFS with the same events as DFS except second primary colorectal cancers. The consensus DFS definition counts death from any cause, a second primary colorectal cancer, anastomotic relapse and metastatic relapse as events, but not a second primary cancer outside the colon or rectum. Whether second primary cancers count is therefore part of what a DFS result means.

Breast cancer has its own scheme. Hudis and colleagues proposed the STEEP system in 2007 so that a given endpoint name would always cover the same set of events, with each event defined the same way across endpoints and studies. Invasive disease-free survival is a separately named endpoint under STEEP, so a DFS result from a breast cancer trial and one from a colon cancer trial need not count the same events.

DFS as a surrogate for overall survival in adjuvant colon cancer

Whether DFS can stand in for overall survival is the question that matters most for economic evaluation, because an adjuvant treatment that delays recurrence would be expected to add QALYs mainly through longer life. A large test used pooled individual patient data from adjuvant colon cancer trials.

Sargent and colleagues pooled individual patient data from 18 randomised phase III trials, with 43 arms and 20,898 patients, to test whether DFS with 3 years of follow-up could replace overall survival with 5 years of follow-up as the primary endpoint. They reported in 2005 a correlation of 0.89 between 3-year DFS and 5-year overall survival.

Yin and colleagues revisited the question in 2022 with individual patient data from 9 randomised trials conducted between 1998 and 2009. They found that the trial-level association between treatment effects on 3-year DFS and on 5-year overall survival remained strong, with an $R^{2}$ of 0.82 by weighted least squares and 0.92 by a copula model, and concluded that 3-year DFS remains a validated surrogate endpoint for 5-year overall survival in adjuvant colon cancer trials.

That evidence belongs to one disease and mainly to chemotherapy. It does not show that a DFS gain from an immunotherapy in melanoma, or from a targeted drug in lung cancer, will be matched by a survival gain of the size the colon cancer relationship would predict.

How regulators and NICE treat DFS evidence

Regulators and HTA bodies ask different questions of the same DFS result. A regulator asks whether the effect is clinical benefit or reasonably likely to predict it, while an HTA body has to put a number on the survival and QALY gain that follows.

The 2018 FDA guidance states that a treatment effect measured by DFS or event-free survival can be a surrogate endpoint to support accelerated approval, a surrogate endpoint to support traditional approval, or it can represent direct clinical benefit.

The NICE technology appraisal and highly specialised technologies manual (PMG36) deals with surrogates in sections 4.6.5 to 4.6.7. Section 4.6.5 says that clinical end points reflecting how a patient feels, functions or how long a patient lives are more informative than surrogate outcomes. Section 4.6.6 sets out three levels of evidence for a surrogate relationship, citing Ciani and colleagues, with level 1 being randomised controlled trial evidence that the technology's effect on the surrogate corresponds to a commensurate effect on the final outcome. Section 4.6.7 asks for good evidence that the relative effect of a technology on the surrogate end point is predictive of its relative effect on the final outcome, preferably from a meta-analysis of level 1 evidence using the bivariate methods in the Decision Support Unit's Technical Support Document 20. A DFS hazard ratio from an adjuvant trial with immature survival data is therefore a surrogate argument in NICE terms, and the company has to justify how it is carried through to overall survival in the model.

DFS states in adjuvant economic models

Adjuvant models usually start everyone in a disease-free state and move people out of it on recurrence. Most add separate states for locoregional recurrence and for distant metastatic disease, because the treatments, costs and survival prospects after each kind of recurrence are different.

NICE technology appraisal TA1043, published in February 2025 and covering osimertinib after complete resection of EGFR mutation-positive non-small-cell lung cancer, is a recent example. The company used a semi-Markov model with five states (disease-free, locoregional recurrence, first-line distant metastases, second-line distant metastases and death) and a 37-year time horizon, and DFS data from the ADAURA trial informed the choice of distribution for moving out of the disease-free state into the locoregional recurrence or first-line distant metastases states. The model also included a cure assumption, and the committee was concerned about uncertainty in the modelling of long-term outcomes.

This structure is a state transition model rather than a partitioned survival model. A partitioned model would read the disease-free share directly from the DFS curve and the share alive off the overall survival curve, but in adjuvant trials overall survival is usually immature, so the post-recurrence states are commonly built from other evidence on metastatic disease. Survival then comes out of the model as the result of the recurrence pattern and the treatment that follows each recurrence, and the implied overall survival curve can be checked against any trial survival data that exist.

Three modelling choices often decide adjuvant results. The first is the extrapolation of the DFS curve beyond trial follow-up, including whether its hazard flattens. The second is whether people still disease-free after some years are treated as cured, with only background mortality from then on. The third is what the comparator arm receives at recurrence, since an effective treatment for metastatic disease narrows the survival gap that a DFS gain would otherwise produce.

Worked example with an illustrative DFS gain

All numbers in this example are illustrative and simplified. The example ignores costs and QALY differences during the first 3 years and treats the remaining QALY figures as already discounted. An adjuvant drug raises 3-year DFS from 0.70 to 0.78, and people still disease-free at 3 years are assumed cured.

Step 1 (DFS gain). The absolute gain is $\Delta DFS = 0.78-0.70 = 0.08$, so in a cohort of 1,000 patients $N_{DF} = 1{,}000 \times 0.08 = 80$ more people are disease-free at 3 years.

Step 2 (QALYs). For simplicity every DFS event is treated as a recurrence, although in practice DFS also counts deaths without recurrence. A person disease-free at 3 years has 12.5 remaining QALYs and a person who has had a recurrence has 2.5. The QALY gain per patient is $\Delta Q = 0.08 \times (12.5-2.5) = 0.08 \times 10 = 0.8$.

Step 3 (costs). Under the same simplification, every DFS event is a recurrence that needs treatment. The adjuvant drug costs GBP 28,000 per patient, and each recurrence avoided saves GBP 50,000 of treatment for metastatic disease. The saving per patient is $S = 0.08 \times 50{,}000 = 4{,}000$, so the net cost is $\Delta C = 28{,}000-4{,}000 = 24{,}000$.

Step 4 (ICER). The ICER is $ICER = 24{,}000 / 0.8 = 30{,}000$, or GBP 30,000 per QALY.

Step 5 (better treatment after recurrence). Suppose a more effective and more expensive treatment becomes standard at recurrence, given to people who recur in either arm, so a person who recurs has 5.0 remaining QALYs and each recurrence costs GBP 80,000. The QALY gain becomes $\Delta Q = 0.08 \times (12.5-5.0) = 0.08 \times 7.5 = 0.6$, the saving becomes $S = 0.08 \times 80{,}000 = 6{,}400$, the net cost is $\Delta C = 28{,}000-6{,}400 = 21{,}600$ and the ICER is $ICER = 21{,}600 / 0.6 = 36{,}000$, or GBP 36,000 per QALY.

The DFS gain is the same in both runs. The ICER rises by GBP 6,000 per QALY because the value of keeping a person disease-free depends on what happens to people who do recur, which is exactly the part of the pathway a DFS result does not measure.

Boundaries with nearby survival endpoints

DFS sits among several time-to-event endpoints whose names overlap and whose events differ. The trial protocol, not the name, decides which events count.

Relapse-free survival is the closest neighbour. The difference can lie in how a trial counts second primary cancers, and the time origin can also differ between definitions (randomisation, surgery or the end of treatment), so the two names cannot be compared without the protocols. Event-free survival usually suits neoadjuvant or perioperative designs, where a patient can fail before ever reaching a disease-free state. Progression-free survival plays the same role in advanced disease that DFS plays after curative treatment, and the partitioned survival models used for metastatic cancer are built around it.

Limitations and common misreadings

DFS combines events that matter very differently to patients and to costs. A local recurrence that is cured by further surgery and a death both count as one DFS event, so the hazard ratio says nothing about the mix.

1. A DFS gain proves a survival gain. The correlations reported by Sargent and colleagues and by Yin and colleagues come from adjuvant colon cancer trials, and NICE PMG36 section 4.6.7 asks for good evidence that the relative effect of the technology on the surrogate predicts its relative effect on the final outcome.
2. DFS means the same thing in every trial. Second primary cancers, the time origin and the events counted all vary, which is why the Punt review and the DATECAN panel proposed standard definitions.
3. A flat tail in the DFS curve shows cure. A plateau seen in a few patients at the end of follow-up is weak evidence, and a cure assumption in an adjuvant model needs separate support.

Sources

• US Food and Drug Administration. Clinical Trial Endpoints for the Approval of Cancer Drugs and Biologics. Guidance for Industry. Silver Spring, MD: FDA; December 2018. https://www.fda.gov/media/71195/download
• Punt CJA, Buyse M, Köhne CH, et al. Endpoints in adjuvant treatment trials: a systematic review of the literature in colon cancer and proposed definitions for future trials. Journal of the National Cancer Institute. 2007;99(13):998-1003. https://doi.org/10.1093/jnci/djm024
• Cohen R, Vernerey D, Bellera C, et al. Guidelines for time-to-event end-point definitions in adjuvant randomised trials for patients with localised colon cancer: results of the DATECAN initiative. European Journal of Cancer. 2020;130:63-71. https://doi.org/10.1016/j.ejca.2020.02.009
• Hudis CA, Barlow WE, Costantino JP, et al. Proposal for standardized definitions for efficacy end points in adjuvant breast cancer trials: the STEEP system. Journal of Clinical Oncology. 2007;25(15):2127-2132. https://doi.org/10.1200/JCO.2006.10.3523
• Sargent DJ, Wieand HS, Haller DG, et al. Disease-free survival versus overall survival as a primary end point for adjuvant colon cancer studies: individual patient data from 20,898 patients on 18 randomized trials. Journal of Clinical Oncology. 2005;23(34):8664-8670. https://doi.org/10.1200/JCO.2005.01.6071
• Yin J, Salem ME, Dixon JG, et al. Reevaluating disease-free survival as an endpoint vs overall survival in stage III adjuvant colon cancer trials. Journal of the National Cancer Institute. 2022;114(1):60-67. https://doi.org/10.1093/jnci/djab187
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). Published 2022; last updated 13 October 2026. Sections 4.6.5 to 4.6.7. https://www.nice.org.uk/process/pmg36/chapter/economic-evaluation-2
• National Institute for Health and Care Excellence. Osimertinib for adjuvant treatment of EGFR mutation-positive non-small-cell lung cancer after complete tumour resection (TA1043). February 2025. Section 3, committee discussion. https://www.nice.org.uk/guidance/ta1043/chapter/3-Committee-discussion