Analysis of Time-to-treatment Discontinuation of Targeted Therapy, Immunotherapy, and Chemotherapy in Clinical Trials of Patients With Non-Small-Cell Lung Cancer

G. M. Blumenthal; Y. Gong; K. Kehl; P. Mishra-Kalyani; K. B. Goldberg; S. Khozin; P. G. Kluetz; G. R. Oxnard; R. Pazdur


Ann Oncol. 2019;30(5):830-838. 

In This Article


Use of RWE to inform regulatory decision making are key priorities in the 21st Century Cures Act, the Prescription Drug User Fee Act VI reauthorization and several other major public initiatives.[13] RWE could be used in oncology to further characterize the safety and efficacy of drugs in the post-marketing setting. This may include confirmation of clinical benefit for an accelerated approval, expansion of labeling claims to new indications, testing of alternate doses and schedules of drugs, and assessment of drug activity in biomarker-defined subgroups and other special populations. The need for reliable end points for describing real-world treatment outcomes is not new—post hoc studies of treatment patterns can be a valuable approach to generate hypotheses for prospective study. However, such retrospective reports apply a range of methodologies, some attempting to calculate RECIST PFS outside of a trial, some studying 'clinical progression,' some studying TTNT, and others studying time-to-treatment failure (TTF) a closely related end point to TTD which attempts to discriminate between discontinuations due to adverse events and discontinuations for other reasons.[14]

Our data provides support for further exploration of TTD as a pragmatic RWE end point to assess the efficacy of anticancer therapies in advanced NSCLC, for several reasons. First, it is associated with a commonly used efficacy end point in oncology, PFS. Second, it has the potential to accurately capture the safety and efficacy of a cancer drug in the real world, because it may reflect the common practice of continued treatment beyond objective disease progression defined per RECIST. Third, in our analysis, TTD is censored less commonly compared to PFS and OS. Finally, TTD has the potential to be measured and abstracted from EHRs or claims databases. We selected TTD rather than the closely related end point TTF, due to the more complex censoring rules associated with TTF, rendering this end point potentially less pragmatic in the real world. In addition, we were not able to explore another potential end point, TTNT since this data was not routinely captured in the clinical trial datasets.

To our knowledge, this is the first study to indicate a patient-level association between TTD and PFS (r = 0.87) in NSCLC clinical trials across therapeutic classes. In addition, a moderate patient-level association was observed between TTD and OS (r = 0.68), and this TTD/OS correlation was higher for some modern targeted and immune therapies (ALK TKI, ICI). This study also described different patterns of TTD and PFS between the six therapeutic subgroups. In the oncogene-directed targeted therapy subgroups (e.g. EGFRm and ALK), median TTDs (13.4 and 14.1 months) reliably exceeded median PFS (11.4 and 11.3 months), and there was a meaningful minority of 'late TTD' cases (12.4% and 22.9%). The fact that some patients are able to safely continue treatment well beyond RECIST progression may be reflective of the underlying biology of oncogene-addicted lung cancer, where continued suppression of the driver mutation may be needed despite the emergence of alternate resistance mechanisms. In addition, the ability to continue treatment beyond progression may provide support for the tolerability of a therapy.

With ICI therapy, median PFS (4.2 months) was slightly longer than median TTD (3.5 months), with cases of both early (6.7%) and late (8.3%) TTD. This suggests that there were some cases of termination of ICI treatment due to immune-mediated adverse events where patients continued to have durable benefit, and other cases where patients were treated well beyond conventional progression (also termed pseudo-progression). This analysis included patients with mixed biomarker characteristics (PD-L1 high and low). Further investigation is needed to understand the behavior of TTD as an end point in enriched populations of patients with who are most likely to benefit from ICI therapy (e.g. PD-L1 expression ≥50%). With the chemotherapy subgroups, median PFS (5.6 and 4.1 months) exceeded median TTD (3.9 and 2.2 months). Early TTD occurred in a meaningful minority of chemotherapy-treated patients (7.7% and 15%), likely from patients stopping treatment due to adverse reactions.

The limitations of this analysis and the TTD end point must be acknowledged. This is a post hoc analysis of an end point that was not prespecified in these 18 trials. Additionally, in some cases treatment discontinuation was mandated by the clinical protocol upon RECIST progression, linking PFS and TTD by limiting the ability of an investigator to treat beyond progression. This likely improved the correlation between TTD and PFS and attenuated the occurrence of late TTD. This will not be the case in the routine clinical care setting and prospective validation of TTD as a meaningful real-world end point will be necessary. Future studies could consider adding TTD as a secondary end point to allow this data to be consistently captured for future analysis.

Another important limitation is the censoring of time-to-event end points such as PFS, TTD, and OS. The censoring was highest with OS (42%), followed by PFS (25%) and TTD (10%). A sensitivity analysis using a method to estimate the correlation of bivariate times under censoring was consistent with the overall results.[15] An additional sensitivity analysis calculating correlations of the paired end points using only those patients with an event in at least one of the two end points also was consistent with the overall results.

Another limitation is the potential bias inherent in the fact that TTD is a clinical (and thus subjective) end point, and highly dependent upon a patient's clinical circumstance. For example, if there is an opportunity to receive newer agents such as targeted therapy and immunotherapy, patients and physicians may be inclined to discontinue chemotherapy early, shortening the TTD. But if there are fewer alternate treatment options, patients and physicians may continue chemotherapy longer, lengthening the TTD. Finally, TTD would not be an appropriate end point for therapies that are given for a planned number of cycles, as discontinuation of treatment is prespecified. In these cases, other real-world end points should be considered.

Well-defined time intervals are critical in time-to-event analysis. PFS and TTD evaluations require subjective determination and therefore are subject to the potential for investigator-associated biases, such as selection bias. Using death as the terminus of time-to-event end points greatly improves measurement precision and minimizes bias, though capturing survival data may be challenging in real world settings. Another source of variability and potential bias in time-to-event analyses such as PFS and TTD is the initial point of measurement. Randomization ensures a nearly equal distribution of unknown covariables, including variation in a patient's condition at enrollment, thus reducing bias. To the extent possible, randomization should be conducted in RWE studies to minimize bias. While randomization in pragmatic trials is infrequent in oncology, it has been deployed in other diseases such as pulmonary and cardiovascular.[16,17]

In summary, this patient-level analysis of 8947 patients with advanced NSCLC indicates that TTD is associated with PFS in legacy clinical trials across therapeutic classes. Further research is needed to validate this end point for RWE studies, to see if TTD can be used in pragmatic randomized controlled trials at the point of care. Further exploration of additional RWE end points such as TTNT is warranted in situations where treatment-free interval may be beneficial.