The objective of this methods project was to address whether PFS is an outcome related to psychological well-being or QOL. As described in the Background, using PFS as an outcome can be challenging, especially in terms of potential bias (assessment, evaluation, performance, attrition), detection error, and its utility as a surrogate for other outcomes. There is a growing literature describing the best practices to address sources of bias, including careful prospective definition of progression,39 blinding of patients and physicians to treatment,3, 29 use of blinded external radiological review,3, 24, 29, 30 and attention to patient attrition and data censoring.3, 25, 28 However, accurate measurement of PFS can still be problematic. In addition, the implications of PFS are dependent on the context defined by specific disease, disease stage, the types of drugs and their toxicities, and patient response to the risks and benefits of treatment.6 Regardless of these issues, PFS has been an attractive endpoint because, compared with OS, studies may be conducted more quickly using fewer subjects and at lower costs.

There has been growing evidence that PRO measures can provide important information for assessing the burden of cancer and effectiveness of treatments.67 A variety of relevant outcomes have been identified, including symptom status, functional status, measures of overall well-being, satisfaction with care, treatment adherence, and measures of QOL.

Symptom status has recently been of particular interest as a PRO measurement with potential value in the study of cancer treatments. The ASCPRO Multisystem Task Force has recently proposed that the measurement of symptomatic changes, as a subset of QOL, may be a sufficient outcome in clinical trials to make decisions about whether to introduce a new treatment.38 Evaluation of symptoms is complicated by the fact that they often occur in clusters and can be modified by comorbid conditions or the effects of supportive care, previous treatments, and other factors.38 A number of trial scenarios assessing the impact of symptoms are plausible, including treatment leading to a reduction in disease-related symptoms, treatment leading to a delay in the onset of disease-related symptoms, treatment itself producing symptoms, or two therapies resulting in equally effective results, but with differing toxicity profiles. Unfortunately, as Cleeland38 has noted, while it is well understood that symptoms in a patient can be produced by disease, treatment, or both or neither, attempts at symptom attribution are notoriously unreliable.

It has been noted that approaches for assessing specific symptoms by patient report in clinical trials have been lacking.68 A simple patient report on the presence or absence of a symptom, such as nausea or pain, can provide information about how the patient feels. However, this information is subjective and conveys little or no information about the impact of symptoms on functioning or other aspects of well-being. The benefit of adding a QOL measurement to the assessment of symptoms is the ability to determine symptom impact on functional status and/or global well-being.67 In measuring QOL there are a number of options, including the use of: (a) unidimensional domains or multidimensional domains (tools with one definable aspect or domain versus tools with many); (b) psychometric or preference-based measurements (tools measuring response along a subjective scale of well-being versus anchored scales where absolute conditions of well-being are designated); and (c) generic (tools applicable to a range of diseases) versus general cancer or site- or problem-specific cancer measurements.

We focused specifically on the association between an outcome, PFS, and other outcomes of importance to patients, such as QOL or disease symptoms. In contrast to a traditional comparison of interventions, ascertaining causal associations between PFS and other outcomes can be problematic, because progression can only be observed and not manipulated. This has implications for study design, interpreting results, quality evaluation, and even the feasibility of designing a study that might address the issues posed here. The study of PFS and how it relates to other outcomes of importance to patients must be addressed from the vantage of observational data and their inherent potential for bias.

We first performed a broad-based review of the literature to identify studies examining the direct impact of PFS status on patient anxiety, depression, or psychological well-being. The search revealed no studies that addressed this question. This is a notable gap in the evidence, because one can speculate that knowing they are progression-free might make patients feel relief,23 while news about disease progression might make them feel worried, but this has not been systematically studied. There is literature on the delivery of bad news to patients, including reports specifically addressing the direct psychological impact of telling patients they have been diagnosed with cancer.6971 How applicable this is to the setting of PFS remains unexamined. The psychological response to information on PFS may depend on the way it is delivered by a physician and on the patient’s understanding of what PFS actually means. It is not known whether patients would view PFS differently if news of measurable progression or lack thereof was delivered with an explanation that PFS may not predict OS.

The lack of studies addressing KQ1 is not surprising given the difficulty in obtaining reliable and objective measurements of both PFS and psychological well-being/QOL in a coordinated manner that extends data collection of QOL into the period beyond progression. More research is clearly needed on how patients view the risks and benefits of even small potential changes in treatment outcomes, on how knowledge about PFS status might affect psychological well-being/QOL, and on how this information might be used to inform treatment decisions.

We also conducted a focused review of trials administering drugs recently approved by one or more regulatory authorities, for which PFS was the primary endpoint. In this way we explored the available evidence regarding the association of PFS with QOL, disease symptoms, or other QOL-related outcomes. Four studies6164 were identified, each of which reported better QOL or disease symptoms among patients who remained progression-free compared with those who had progressed. However, interpretation of this association is compromised by the poor quality of all four studies.

Evidence for a causal relationship in these studies remains unclear. It could be that a common underlying mechanism improves both PFS and QOL. For example: tumor shrinkage leads to improved pain that is detected via a QOL questionnaire and also is seen as disease regression on imaging. Alternatively, it could be that patient knowledge of progression status itself affects subjective impressions of overall QOL (or affects a specific symptom that impacts overall QOL measure scores, such as anxiety). The issue of causality is important in terms of considering future use of the PFS endpoint in clinical research, particularly when PFS is not serving as a surrogate for OS, but rather as a marker of QOL. If the underlying mechanism affecting PFS and QOL are the same, then the clinically meaningful endpoint is actually QOL, with PFS as an intermediate or surrogate outcome. In this case, a logical approach would be to design studies making QOL the primary focus.

Several sources of potential bias associated with PFS ascertainment were identified and discussed in the Methods section; all four sources of bias were evident in the included studies, weakening confidence in the reported PFS/QOL association. All four studies were limited by a failure to develop an a priori hypothesis regarding the association of PFS and QOL, by censoring at the time of progression, and by substantial rates of missing data not at random. While all four studies did describe analytical approaches to address missing data not at random, the ability to account for the large dropout observed is doubtful. In addition, two61, 64 of the four studies failed to provide patient-investigator blinding to treatment.

All four of the studies adjusted for QOL baseline scores in their analyses. This is particularly important because patients with better baseline scores might have longer PFS. Although the use of QOL as a prognostic factor for PFS is not the focus of this methods project, there is evidence that QOL itself is a prognostic indicator of survival.7275 The basic mechanism involved appears unrelated to the association of PFS with QOL; however, if QOL is prognostic for survival, there is the potential to confound studies evaluating the impact of PFS on QOL.

Duration of PFS and OS was relatively short in these studies (i.e., weeks to months). If treatments are making only small improvements in survival outcomes, it becomes that much more important to develop reliable estimates of the impact on disease symptoms and QOL, so that patients and physicians can make informed choices. It is expected that patients’ interpretations of what might be meaningful changes in outcomes may differ from those of other stakeholders; in particular, best-case scenarios regarding their own outcomes may drive patient decisions. Studies determining how delivery of news may impact a patient’s understanding, QOL, and use of information are needed.

Heterogeneous patient populations make interpreting the included studies challenging. More studies with a clear description of sub-populations of interest are needed to help address these issues. Future studies using patient level data and analysis would also be useful.

The studies identified here were accompanied by important sources of bias. For that reason they provided limited evidence to address the questions posed. Ideally, studies would be free from bias. Describing a study design eliminating all biases76 facilitates understanding the challenges ascertaining how treatment, disease progression (and knowledge of it), and other outcomes are associated. The purpose of such an exercise is to better understand issues that need to be addressed in future studies, not to outline a perfect study. In order to address known sources of bias an ideal study would include the following:

  • Two treatments accompanied by similar overall survival rates and toxicities (QOL outcomes are not confounded by differential toxicities).
  • Patients failing either treatment due to disease progression do not receive further therapy (eliminates time-dependent confounding).
  • One treatment results in longer PFS so that the two study arms are similar in all respects except PFS.
    (Under these conditions, an unbiased comparison of other clinically important outcomes and their association with disease progression could be assessed.)
  • A second randomization to being informed about or blinded to PFS status would determine if knowledge about PFS impacts QOL outcomes. Patients would also be informed that PFS may not be associated with overall survival.

These characteristics could not accompany any real study, and it is difficult to consider blinding patients to knowledge of progression. Accordingly, the biases present in studies examining the questions posed here must be addressed. At a minimum, disease progression must be sufficiently measurable; no design or analysis can obviate large measurement error. Second, a therapeutic failure evidenced by disease progression would be typically followed by subsequent treatment (we assume avoiding second therapy is a potential benefit). Finally, because one is observing, not manipulating PFS, it is useful to posit a basic causal model to place these abstract notions in context (Figure 2).

Figure 2 depicts relationships or causal pathways between the variables of interest in a basic causal model for treatments, progression, and outcomes for advanced tumors. In the figure, initial treatment may impact disease progression and influence quality of life and/or survival either through progression or other pathways. Initial treatment may lead (1) directly or through subsequent treatment to survival or death, (2) directly to progression and with or without knowledge of progression to an endpoint of quality of life, (3) directly or indirectly to an endpoint of quality of life or (4) directly with or without subsequent progression to an endpoint of dead/alive. The path from initial to subsequent treatment reflects that a second line agent, if used, will not be chosen independently. The potential second treatment is also a time-dependent confounder. The bolded arrows or paths are of primary interest here; that is, the relationship between progression and QOL (e.g., patient-reported outcomes).

Figure 2

Basic causal model for treatments, progression, and outcomes. PRO = patient-reported outcomes; QoL = quality of life

The graph depicts relationships among the variables of interest, both causal and confounding. Beginning on the left lower corner, initial treatment may impact disease progression and influence survival either through progression or other mechanisms (arrow from treatment to dead/alive). Treatment can influence QOL or other PROs through toxicities. A second treatment requires considering time-dependent confounding. The remainder of the graph can be interpreted in a likewise manner.

The association of primary interest here is between progression (and subsequent progression) and QOL (PROs or psychological well-being). To define that association, confounding must somehow be accounted for—a potentially difficult task. It is necessary to take into account initial and subsequent treatments, whether patients are aware of progression, treatment efficacy, and toxicities. Even if the many shortcomings identified in the studies reviewed here were absent, the data would be limited to inform the relevant associations.

In summary, there is evidence that PFS provides information on the direct effect of treatment on tumor burden,23, 77 although there is a lack of consensus regarding its use. This is not surprising given the fact that there is a significant gap in information on the direct and indirect relationship between PFS and QOL or other related outcomes. Our assessment of the state of the evidence on PFS and QOL or related outcomes led to some important observations not appreciated a priori: PFS is observed, not randomized; studies examining the associations of interest are generally of poor quality; standard quality assessment tools have shortcomings in this setting requiring novel quality measures; and positing a causal model facilitates understanding the limitations of current evidence and can inform future research.

Given these underpinnings, we suggest future research adopt the following approach to assess the relationship between PFS and QOL. While it is not possible to randomize PFS status, it may be informative to identify patients with or without progression and to observe whether the state of progression can be linked to QOL or other PRO. In a best case scenario, it might be possible to consider PFS a surrogate for QOL using statistical techniques described in the PFS/OS setting (hypothesis testing, estimation, prediction).6 It is unlikely blinding would be feasible, but alternative designs could be explored. For example, PFS information could be conveyed to one group of randomized patients using standard techniques (most commonly verbal interaction between an oncologist and patient). The second randomized group would receive information through the use of decisional aides that are constructed to clearly and correctly communicate the meaning of PFS as an outcome. A comparison could then be performed between the two groups to determine how delivery of information affects QOL outcomes. Whether PFS might be an indirect method of predicting QOL, or whether direct measurement of QOL in parallel with PFS represents a more practical approach, remains an open issue.

Evidence Gaps

There is a need for prospective research to evaluate: 1) the causal relationship between patient knowledge of progression status and other outcomes of importance to patients, including specific symptoms and overall QOL; 2) patient impressions of the meaningfulness of progression as a health outcome in the absence of association with overall survival; and 3) the extent to which improvement in patient-reported outcomes associated with improvement in PFS is related to a common underlying mechanism, such as tumor shrinkage positively impacting symptomatology and therefore QOL. Because PFS is a variable which cannot be manipulated, but only observed, there is also a significant need to explore what models are available to better understand this endpoint. In general, better tools are needed to measure QOL, and more robust studies of QOL with clear a priori hypotheses and endpoints are needed.

Pending further study of the association between PFS and outcomes of importance to patients, there is insufficient evidence to support the use of PFS alone to demonstrate QOL or related outcomes. It is acknowledged that in some situations, measurement of OS is impractical or unfeasible.10 In such cases, the direct measurement of PFS and PROs, such as QOL or patient-reported symptoms, may be a practical alternative. Further study of these drugs and confirmation and expansion of the findings reported here would advance our understanding of these complex relationships. In particular, future studies should address and remedy the biases and quality issues described in this report.


In this methods project, what was being studied was not a comparison of treatment interventions, but rather the relationship between two outcomes of treatment; therefore the standard approaches used for randomized clinical trials were not applicable. We posited a causal model for addressing this issue, but recognize that this is a hypothetical construct that requires further study and validation.

Existing validated quality assessment tools were not entirely applicable for our purposes. This resulted in the adaptation of the quality assessment tool to include items addressing potential biases unique to the study of the association between PFS and QOL. However, the modified quality assessment is unvalidated.

It would be interesting to know whether improvements in PFS in advanced disease would be predictive of a survival benefit from the use of the same treatment in the adjuvant setting. However, we did not address this issue as part of our key questions or data inquiry and are unable to address whether such a relationship exists.

Finally, in order to provide a focused review commensurate with the resources available to this methods project, the search for articles on outcomes of interest to patients used terms for psychological outcomes (depression, depressive disorder, anxiety, psychological), the term Quality of Life, and the term PRO. Had this search been broadened by use of separate search terms for symptoms or toxicities, additional articles may have been identified, but this was beyond the scope of the project. In selecting articles, we chose only reports that included a direct quantitative comparison of PFS status to the QOL-related outcomes of interest. It is unlikely that an expanded review of studies would have overcome the challenge of addressing the issues of heterogeneity and quality that prevented pooling of data in our analysis.


The objective of this methods project was to address whether PFS is an outcome related to psychological well-being or QOL. There were no studies that directly addressed the question of a causal relationship between knowing PFS status and patient anxiety, depression, or psychological well-being. Due to limitations in their design, the four studies demonstrating an association between PFS and better QOL or disease symptoms were all of poor quality. Hence, there is insufficient evidence to make any conclusions about the association between PFS and QOL or related outcomes. The direct measurement of both PFS and QOL may be a practical and informative alternative when measurement of OS is unfeasible.