RECIST 1.1 for Response Evaluation Apply Not Only to Chemotherapy-Treated Patients But Also to Targeted Cancer Agents: A Pooled Database Analysis

Abstract

PURPOSE

The mode of action of targeted cancer agents (TCAs) differs from classic chemotherapy, which leads to concerns about the role of RECIST in evaluating tumor response in trials with TCAs. We investigated the performance of RECIST using a pooled database from 50 clinical trials with at least one TCA.

METHODS

We examined the impact of the number of target lesions (TLs) on within-patient variability of tumor response. The prognostic effect of TL response (at 12 weeks or on study on the basis of a maximum five TLs) on survival was studied through landmark and time-dependent Cox models adjusted for baseline tumor load, occurrence of new lesions, or unequivocal progression of nontarget disease.

RESULTS

Data were obtained from 23,259 patients with cancer (36% lung, 28% colorectal, 11% breast, and 25% other); 15,620 received TCAs, predominantly transduction or angiogenesis inhibitors, as a single agent (37%), combined with other TCAs (7%), or as chemotherapy (56%); 28% received chemotherapy only; and 5% received best supportive care or placebo. A total of 17,222 patients contributed to the analyses. Within-patient variability decreased with increasing number of TLs, similarly for TCAs (with/without chemotherapy) and chemotherapy only. Mixed responses occurred proportionally in all treatment classes. Landmark analyses showed an ordinal relationship between percentage change from baseline to 12 weeks and overall survival, and demonstrated a clear distinction between tumor shrinkage and progressive disease according to RECIST. Time-dependent analysis showed no marked improvement in the ability to predict survival on the basis of TL tumor growth compared with nontarget progression or new lesion occurrence, regardless of treatment. Similar results were seen for major tumor types and different classes of TCAs.

CONCLUSION

This work reinforces that RECIST version 1.1 perform well for response assessment of TCAs.

INTRODUCTION

The assessment of change in tumor burden, which is a mainstay of the evaluation of cancer therapeutics, defines objective response and disease progression. Both are increasingly important end points in cancer clinical trials, especially for progression and disease-free survival, which are used frequently for drug registration. Because RECIST was published in 2000 they have been widely adopted to assess response in clinical trials.^1,2 The RECIST Working Group further standardized and clarified these response criteria for version 1.1 after validation on a large warehouse containing more than 6,500 patients treated with chemotherapy.

Over the past decades, numerous targeted cancer and immunotherapeutic drugs have been and are being developed, with many already used in routine clinical care. Targeted cancer agents (TCAs) block the growth and spread of cancer by interfering with specific molecules that are involved in the growth, progression, and spread of cancer. Therefore, their mode of action differs from that of chemotherapy for which RECIST was initially developed and validated. Although chemotherapy causes the tumor to shrink, TCAs may not lead to obvious tumor shrinkage or might induce heterogeneous effects on different sites of metastases, which has raised questions about whether variations in response criteria may be required to evaluate the activity of these TCAs. Moreover, it has been questioned whether, for example, signal transduction inhibitors versus angiogenesis inhibitors with different modes of action affect the tumor response differently. To address these questions, the RECIST Working Group compiled a large warehouse that comprises studies performed by pharmaceutical companies and academia, including TCA studies.

Immunotherapeutics were not included in the warehouse because not enough data were available at the time and tumors seem to respond differently compared with chemotherapeutic and targeted drugs. A consensus guideline, iRECIST, was recently developed to ensure consistent design and data collection to allow validation in a separate database.³ Here, we provide a summary of the various analyses that were performed on the TCA warehouse and address the value of RECIST 1.1 in TCAs.

METHODS

The Data

In 2011, the RECIST Working Group launched the first calls for a data warehouse, and the final database was successfully compiled at the European Organisation for Research and Treatment of Cancer (EORTC) Headquarters on the basis of 50 phase II and phase III trials with clinical data from patients treated with TCAs or TCAs in combination with chemotherapy compounds (Appendix Table A1, online only). Data on 23,259 patients were shared by partners from industry (66%) and academia (34%), including general patient information (eg, the start of treatment, survival information, tumor type) and detailed longitudinal tumor measurements (measurement/evaluation date, site of the lesion, method of measurement, size of measured lesions, information on nontarget lesions, and occurrence of new lesions), as available in the study case report forms.

Although the majority of studies were based on RECIST 1.0, some also used modified WHO criteria or RECIST 1.1 (Appendix Table A1), which resulted in heterogeneity in the type of measurable lesions reported. To homogenize this, the definition of a measurable lesion according to RECIST 1.1 was adopted throughout (ie, at least 10 mm in longest diameter [if non-nodal] at baseline, as assessed by computed tomography, spiral computed tomography, or magnetic resonance imaging consistently throughout all assessment times).

Whenever a measurement was missing at an intermediate assessment, the last available one before that assessment for that lesion was imputed to still enable calculation of overall response at that time point. This occurred for at least one lesion in 1,962 patients of the full data set (out of 23,259 [8.4%]). Wherever the measurement method of a target lesion changed from the one used at baseline, the reported measurement was replaced by the last available one before the assessment recorded using the baseline method. This affected at least one lesion in 702 patients of the analysis data set (out of 17,222 [4.1%]). Nodal lesions were considered potential target lesions if they had a short axis (if available) of at least 15 mm. Pathologic lymph nodes between 10 and 15 mm at baseline were considered part of nontarget disease and were considered to represent unequivocal progression if they doubled in size. This changed the nontarget response assessment to progressive disease (PD) for 93 patients (out of 17,222); however, the RECIST 1.1 assessment changed from non-PD to PD for only 16 patients, and the remaining 77 patients already had PD on the basis of either target or new lesions. Furthermore, target lesions selected in the brain and osseous structures (only a few cases reported), or for which the site of metastasis could not be properly classified in one of the categories listed in Appendix Table A3 (online only) were not considered.

Target lesions were then selected from the measurable lesions according to size (ie, the largest first with a maximum of two per site). Nonselected but measurable lesions were demoted to the status of nontarget disease and, for the purpose of this analysis, considered to have unequivocally progressed if they doubled in size. RECIST 1.1 require that nontarget disease results in a 73% increase in volume in the total disease burden to call unequivocal progression of nontarget disease. For the current analysis, a more conservative rule was adopted to avoid the possibility that nontarget PD would be called on the basis of one of these demoted lesions alone. This changed the nontarget response assessment to PD for 541 patients and the RECIST 1.1 assessment from non-PD to PD for 20 patients (out of 17,222). When lesions were surgically removed (as far as this was possible to deduce from the data), the measurements were censored at the last assessment before surgery.

Statistical Methodology

Variability of within-patient lesions: impact of number of target lesions.

Because of their focused mechanism of action, mixed type of responses may be seen in patients treated with TCAs. We explored this by studying the variability in the activity of a TCA on different lesions within a patient. We investigated the impact of the number of target lesions selected on the variability in response assessment.⁴ For this purpose, all possible groupings of the available target lesions for each patient were considered. For instance, for a patient with two lesions, there are three possible combinations (lesion 1, lesion 2, and lesion 1 and 2); for a patient with 10 lesions, there are 1,023 possible combinations.

For 96% of patients, at least one follow-up assessment was available within 12 weeks after study initiation. Therefore, the percentage change from baseline of the sum of lesion diameters (longest diameter for non-nodal lesions, short axis subtracted by 10 mm [with 0 as lower bound] for nodal lesions [ie, a pragmatic approach to adjust for nodal lesions that returned to normal as they regress to < 10 mm in size]) was determined (see Appendix, online only, for more details). A positive percentage change corresponds to a decrease in the sum from baseline; a negative percentage change reflects an increase in the sum from baseline. For all possible combinations of target lesions, the percentage change from baseline was determined and categorized according to RECIST as either complete response (CR; 100%), partial response (PR; 100% to 30%), stable disease (SD; 30% to −20%), or PD (≤ −20%).

Association with survival.

In the absence of a gold standard for the immediate ascertainment of tumor response/progression, we used overall survival to validate the RECIST response/PD definitions. To avoid lead-time bias because the response is observed while on study treatment, a landmark approach was adopted.⁵ The percentage change from baseline to 12 weeks (as introduced in the previous section, but based on a RECIST 1.1 selection of lesions) was associated with survival, landmarked at the same time point, using a Cox proportional hazards regression model adjusted for baseline tumor load, occurrence of new lesions, or unequivocal progression of nontarget disease before the landmark.

The main disadvantage of a landmark analysis is the loss of information because only patients who survive beyond the landmark can be taken into account. Alternatively, Cox models can be used with time-varying covariables to capture the effect of the covariable over time. By following the approach of Litière et al,⁶ we explored whether the components of progression, which varied over time, can improve prediction of survival in our warehouse and whether this differs by treatment class. Overall survival was analyzed using a Cox proportional hazards regression model that used a multivariable approach to adjust for baseline tumor load, and at each assessment time, best target response as best percentage improvement from baseline, tumor growth of target lesions as worst percentage change from nadir or as worst increase from nadir (millimeters per week), presence of new lesions, and occurrence of progression in nontarget lesions (see Appendix for more details). These analyses were stratified by trial.

Role of the Funding Source

The pooled database is hosted by EORTC. The funding sources had no role in the design of this research project; collection, analysis, or interpretation of the data; or writing of the article. S.L., G.I., and J.B. had full access to the raw data. S.L. had the final responsibility for the decision to submit for publication.

RESULTS

Description of the Database

For the majority of patients, the primary tumor was either lung (36%), colon (28%), or breast (11%; Table 1). A subset of 15,620 patients (67%) received treatment with TCAs either as a single agent (n = 5,776 [37%]), in combination with other TCAs (n = 1,139 [7%]), or with chemotherapy (n = 8,705 [56%]). A summary of available TCAs, classified according to their mechanism of action, is available in the Appendix Table A2.

TABLE 1.

Patients by Disease Category and Treatment Class

We identified 20,643 patients with at least one target lesion at baseline for additional analysis (Fig 1). Of the 2,616 patients who were not considered, 1,344 were excluded because no baseline assessment was performed (response assessments were collected only in that trial for a subset of patients; Appendix Table A1), 2,367 patients had no follow-up data available after the baseline assessment, 177 patients had their last complete tumor assessment (CR, PR, or SD) less than 4 weeks from baseline; and 36 patients had PD reported within 3 weeks from baseline. Data on 194 patients treated with immunotherapy (interleukin-21 or interferon alfa) were excluded from this analysis because immunotherapy is part of the separate ongoing initiative of iRECIST.³ Finally, data on 647 patients treated with a placebo or best supportive care were not included, which resulted in a primary analysis data set that contained information on 17,222 patients (Fig 1). A detailed description of this data set is available in the Appendix Tables A4 to A9. For two studies with targeted agents, no survival information was available (Appendix Table A9); therefore, 17,049 patients contributed to analyses related to overall survival (Fig 1). Table 2 lists the available information for specific subgroups of interest (selected for sufficient patient information from more than one trial arm), which were considered in the analyses presented here and in the Appendix.

FIG 1.

Summary of the number of patients available for analysis. (*) No valid outcome using all target lesions: no follow-up data postbaseline, n = 2,367; complete response, partial response, or stable disease less than 28 days (4 weeks) from baseline (mostly because at least one target lesion was not measured after baseline), n = 177; progressive disease within 21 days from baseline, n = 36. (†) Treatments not selected for analysis: placebo/best supportive care (Pl/BSC), n = 647; immunotherapy arms (interleukin-21 and interferon alfa), n = 194. ChT, chemotherapy; GIST, GI stromal tumor; TCA, targeted cancer agent.

TABLE 2.

Patients in Subgroups of Interest Considered for the Analyses

Variability of Within-Patient Lesions: Impact of Number of Target Lesions

For all possible combinations of target lesions in a patient, RECIST 1.1 outcome was assessed. By number of selected target lesions, we then determined how many different response categories could be assigned to a patient. For example, for a patient with five lesions (two with individual outcome PR, two SD, and one PD), the number of response categories for one selected lesion is three. Figure 2 shows, by treatment class and increasing number of selected target lesions, that the number of different response categories in which a patient could be classified decreases as the number of target lesions to be selected increases. There does not seem to be much difference between the graphs focused on patients treated with TCAs and those treated with chemotherapy or a combination of TCAs and chemotherapy. Regardless of class of treatment, as of five selected target lesions, no patients were categorized in more than two response categories. In addition, for those patients with at least five target lesions, more than 80% had a stable response assessment regardless of the selected lesions or treatment. This observation was confirmed by more detailed analyses that looked into within-patient variability (Appendix Tables A10 to A28). Therefore, there does not seem to be more within-patient variability when treated with TCAs compared with chemotherapy, at least not in the clinical trials considered in this database. The variability reduces as the number of lesions used for the response assessment increases, and stabilizes at five or more target lesions, but it does so similarly across treatment categories. The currently used rule for a maximum of five target lesions specified by RECIST 1.1 continues to cover most of the observed variability whether patients are treated with TCAs, chemotherapy, or a combination of both (Appendix).

FIG 2.

Response categories (at 12 weeks) by number of selected target lesions: number of different response categories a patient could be classified in on the basis of all possible selections of target lesions by number of selected lesions. The number of response categories for all possible combinations of target lesions in a patient per RECIST 1.1 outcome was assessed. By number of selected target lesions, we then determined how many different response categories could be assigned to a patient. For example, for a patient with five lesions (two with individual outcome of a partial response, two with stable disease, and one with progressive disease), the number of response categories for one selected lesion is three. TCA, targeted cancer agent.

Association With Survival

The results of the landmark analysis, adjusted for baseline tumor load, occurrence of new lesions, or unequivocal progression of nontarget disease (at 12 weeks), are summarized by treatment class in Figure 3, which shows a gradual relationship between percentage change observed in the sum of target lesions measured from baseline to 12 weeks and overall survival, with a larger percentage improvement associated with a better outcome. The category of no change from baseline (0%) is a difficult category that most likely occurs because of a tendency to record the same measure when lesions have not changed much. Furthermore, the estimated hazard ratios (HRs) and their corresponding 95% CIs show a distinction between the impact of tumor shrinkage and PD according to RECIST 1.1 on overall survival. Similar results were observed when looking at the different subgroups (Appendix).

FIG 3.

Forest plots by treatment class of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. (A) Targeted cancer agents (TCAs; n = 4,097), (B) TCAs and chemotherapy (n = 6,398), and (C) chemotherapy (n = 4,587). Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

A Time-Dependent Analysis Approach

The HRs estimated from the time-dependent Cox proportional hazards regression model are listed in Table 3 for the model that contained tumor growth rate in millimeters per week (other results are reported in Appendix Tables A29 to A40). The results of the chemotherapy subgroup confirm the results seen from the previous analysis of the RECIST 1.1 database⁶ (Appendix). Despite being highly significant as a result of the mere size of the subgroups, the modeling of target lesion tumor growth rate did not show a marked improvement in survival prediction compared with the other components, whereby again, strong effects of the occurrence of new lesions and progression of nontarget disease are seen.

TABLE 3.

Time-Dependent Multivariable Cox Proportional Hazards Regression Analysis of Overall Survival Using Components of Response and Progression According to Response Evaluation Criteria in Solid Tumors (RECIST) 1.1

Nevertheless, although tumor growth shows an important impact on survival of 2 mm/wk for chemotherapy, it does so even more on the survival of patients treated with TCAs. This result seems to be driven mainly by patients treated with single-agent signal transduction inhibitors, more specifically, in a study that contributed information of 838 patients with GI stromal tumors treated with imatinib.⁷ This is most likely related to this group of patients having a longer follow-up in terms of measurements. In the other subgroups, the predictive effect of the tumor growth rate was found to be more comparable with that observed in the chemotherapy subgroup. Finally, as in previous analyses, a gradual improvement of survival is seen as the best percentage change from baseline increases, regardless of treatment class.

DISCUSSION

There has been some concern that RECIST 1.1 may not be applicable to TCAs because the mechanism of action of these types of agents may result in different response patterns. Therefore, it has been suggested that TCAs require different criteria for response assessment. The database considered in this article is a unique source of data to study this. To our knowledge, it is the largest individual patient database to date with detailed tumor measurements from case report forms of patients treated with TCAs, which focuses mainly on signal transduction inhibitors and angiogenesis inhibitors either as single agents or in combination with chemotherapy or other TCAs. Extensive analyses of this database contained in the Appendix do not seem to support the assumption that more mixed responses are seen in patients treated with TCAs (either pooled or in subclasses). Therefore, RECIST 1.1 is good for evaluating tumor response to TCAs independent of subclass of TCA or tumor type. In all subgroups, progression per RECIST over the first 12 evaluation weeks as well as on study was a strong prognostic factor of worse survival. The analyses do not suggest any refinement for the definition of progression for TCAs specifically or for RECIST 1.1 in general.

Tumor shrinkage during the first 12 evaluation weeks as well as on study was found to be a strong prognostic factor for better outcome, regardless of treatment class or tumor type. A gradual pattern of improvement of HRs was seen with increasing percentage change from baseline for patients treated with chemotherapy with (Fig 3B) or without (Fig 3C) TCAs, but the pattern was less clear for patients treated with TCAs only (Fig 3A). Depth of response, therefore, does not necessarily lead to better overall survival in this patient population, in contrast to previous reports in non–small-cell lung cancer and metastatic colorectal cancer.^8,9

Although the database is unique in size, it is also heterogeneous (different lines of treatment, different tumor types, and different phases of clinical research) and limited to information that pertains directly to the assessment of each patient’s tumor load. In this setting, subsequent therapy could be an important confounder of the relationship between response to treatment and long-term outcomes, such as overall survival. However, in the absence of such information, it is difficult to account for its effect.

To study the association with long-term outcomes, such as overall survival, we have used both landmark analyses and Cox models with time-varying covariables. The first has the advantage of accounting for lead time bias but results in a loss of information because only patients who survive beyond the landmark can be taken into account. The latter can capture the effect of covariables over time, but it is difficult to visualize and interpret the resulting effect estimates because these are more complex than those from a simple Cox model. Nevertheless, both analyses support the general message of this report. In conclusion, on the basis of these analyses of a large data warehouse, the RECIST Working Group recommends that RECIST 1.1 can also be used for tumor response measurements during treatment with targeted cancer drugs.

Appendix

In 2011, the RECIST Working Group launched the first calls for a data warehouse, and the final database was successfully compiled at the European Organisation for Research and Treatment of Cancer Headquarters on the basis of 50 phase II and phase III trials with clinical data from patients treated with targeted cancer agents (TCAs) alone or in combination with chemotherapy compounds (Table A1).

General Considerations About the Data Coding

Table A2 lists the TCAs according to their mechanism of action (as per National Cancer Institute guidelines: http://www.cancer.gov/about-cancer/treatment/types/targeted-therapies/targeted-therapies-fact-sheet). Table A3 lists the categories which were considered for the classification of the site of tumour lesions.

Detailed Description of Analysis Data Set

Tables A4 to A9 and Figure A1 list detailed descriptions of the analysis data set.

Variability Assessments

Measurements of the potential target lesions are the focus this section. Nontarget disease or new lesions are not taken into account. We investigated the impact of the number of target lesions to be selected on the variability in response assessment from baseline up to 12 weeks. For 633 of 17,222 patients, the first assessment was beyond the time window. These patients are not included in the analyses.

The percentage change from baseline is a continuous variable calculated as: {[(sum of largest diameters at baseline) − (sum of largest diameters at week 12)] / (sum of largest diameters at baseline)} × 100. A percentage change greater than 0 indicates a decrease of the size of the lesion (≥ 30% would correspond to a partial response). A percentage change less than 0 indicates an increase of the size of the lesion (≤ −20% would correspond to progressive disease). For simplicity of reporting, the treatment categories single TCA and two TCAs were grouped into one category, TCAs. This analysis did not take into account that some lesions belong to the same patient.

By treatment category.

There is some more variability in percentage change determined in lesions treated with TCAs, which are listed in Table A11 and Figure A2.

Variance of the percentage change.

A summary of the analysis of the variance per patient of percentage change from baseline in single lesions by treatment category is listed in Table A12. Patients with only one potential target lesion did not contribute to this analysis because it was not possible to calculate variance. Furthermore, we determined the average variability of the percentage change per patient over all possible combinations of target lesions (Figure A3). The average response variance plot can be used to identify the number of target lesions for which the variability is minimal and starts to stabilize. The standardized average response variance plot presents the variability standardized relative to the baseline (in this case based on 1 target lesion), so that each line represents for each increase in number of target lesions how much variability relative to the “baseline” we are able to reduce, i.e. standardized mean of variance = standardized mean of variance = mean of variance (selected lesions from two to nine) / mean variance (one selected lesion). This could be very informative if the variability differs a lot between the different treatment categories at the start. Table A13 lists the number of patients who contributed to this analysis.

Until four selected lesions, the average response variance seems to be higher in the TCA category than in the other. From five selected lesions, the difference between those treatment categories decreases.

Number of response categories by treatment according to the number of selected lesions.

For each numerical combination of target lesions, we also determined how the selection of target lesions could affect the assessment of the response of a patient. For each numerical combination, we thus classified patients as either complete response (100%), partial response (100% to 30%), stable disease (30% to −20%), or progressive disease (≤ −20%), and we studied in how many different response categories a patient could be classified on the basis of the different selections of target lesions (Figure A4). For those patients with at least five target lesions, more than 80% had a stable response assessment, regardless of the selected lesions or the treatment.

By type of targeted cancer agent.

More variability in percentage change was observed in lesions treated with signal transduction inhibitors with or without chemotherapy. In addition, the median per-patient variance of percentage change is largest in signal transduction inhibitors combined with chemotherapy and lowest in angiogenesis inhibitors combined with chemotherapy. Nevertheless, for those patients with at least five target lesions, 80% or more had a stable response assessment regardless of the selected lesions or the treatment, except for single-agent angiogenesis inhibitors (Tables A14 to A16 and Figures A4 to A6).

Tumor types by treatment category.

For patients with lung cancer, there is more variability in percentage change determined in lesions from patients treated with TCAs than in those treated with single or combination chemotherapy (Tables A17 to A19). Although the mean per-patient variance of percentage change is largest for patients treated with TCAs, the median per-patient variance of percentage change is smallest, and less than 40% of patients showed mixed responses across the different lesions. In addition, for those patients with at least five target lesions, more than 80% of those with lung cancer treated with TCAs had a stable response assessment regardless of the selected lesions, the highest among the three treatment categories (Figures A7 to A10).

For patients with colorectal cancer, there is more variability in percentage change determined in lesions from patients treated with TCAs than in those treated with single or combination chemotherapy (Tables A20 to A22). The median and mean per-patient variance of percentage change is also largest in this subgroup. For patients with at least five target lesions, approximately 80% of those with colorectal cancer treated with TCAs had a stable response assessment regardless of the selected lesions, the lowest among the three treatment categories (Figures A11 to A14).

For patients with breast cancer, there is more variability in percentage change determined in lesions from patients treated with combinations of TCAs and chemotherapy than in patients treated with TCAs (although results from one study only) or chemotherapy alone in a small number of patients (Tables A23 to A25). Approximately 40% of patients treated with targeted therapies showed mixed responses across the different lesions versus slightly more than 50% of patients treated with combination therapies. The number of patients with at least five target lesions is small, so one has to be careful not to overinterpret the results (Figures A15 to A18).

Tables A26 to A28 list the summary statistics and patient contributions to the variability assessment for patients with GI stromal tumor cancer. Figures A19 to A22 summarize the additional variance assessments.

Association With Survival: Landmark Approach

The results of the landmark analyses, adjusted for baseline tumor load, occurrence of new lesions, or unequivocal progression of nontarget disease (at 12 weeks), are summarized in Figures A23-A39

Time-Dependent Survival Analysis

On the basis of the analysis by Litière et al,⁶ this section is dedicated to exploring whether the components of progression, which vary over time, can improve prediction of overall survival in our warehouse and whether this prediction differs by treatment category. For this analysis, target lesions were selected according to RECIST 1.1 (ie, per patient, a maximum of five lesions with a maximum of two per organ). Lymph nodes could only be selected as target lesions if the short axis was larger than 15 mm. They were considered to have returned to normal as soon as the short axis regressed to less than 10 mm. For target lesions located in the lymph nodes, a short axis less than 10 mm was considered normal. To assess this uniformly throughout the database, we opted for a pragmatic approach whereby all target lymph node short-axis measurements were subtracted by 10 mm (with 0 as lower bound). We determined at each measurement time the best target response as the best percentage improvement from baseline, tumor growth of target lesions as worst percentage change from nadir, tumor growth of target lesions as worst rate of increase from nadir (millimeters per week), presence of new lesions, and occurrence of nontarget progressive disease. Note that calculation of tumor growth as percentage change from nadir is not possible when the nadir is complete response (this results in a division by 0). Patients for whom this is the case are not taken into account in this analysis.

Overall survival was analyzed by treatment category using Cox proportional hazards regression modeling by adjusting for baseline sum and including these parameters as time-dependent covariables (Tables A29 to A40). The goodness of fit of these models was assessed by time-dependent versions of receiver operating characteristic curves and their areas under the curve (AUCs) with incident/dynamic definitions of sensitivity and specificity (Heagerty et al: Biometrics 61:92-105, 2005). The AUC provides a measure of the model’s discriminatory power whereby an AUC of 1 reflects a perfect test and an AUC of 0.5 reflects a predictive ability comparable to tossing a coin (Figures A40 to A43).

Time-Dependent Model With Tumor Growth as Percentage

Calculation of tumor growth as percentage change from nadir is not possible when the nadir is complete response (this results in a division by 0). Patients for whom this is the case are not taken into account in this analysis.

FIG A1.

Analysis dataset: extent of disease at baseline.

FIG A2.

Variability assessments: Percentage change from baseline to week 12 by treatment category.

FIG A3.

Variability assessments: Variance plots of Percentage change from baseline to week 12 by treatment category.

FIG A4.

Variability assessments: Percentage change from baseline to week 12 by type of TCA.

FIG A5.

Variability assessments: Variance plots of Percentage change from baseline to week 12 by type of TCA.

FIG A6.

Variability assessment: Response categories (at 12 weeks) by number of selected target lesions: number of different response categories a patient could be classified in on the basis of all possible selections of target lesions by number of selected lesions. The number of response categories for all possible combinations of target lesions in a patient per RECIST 1.1 outcome was assessed. By number of selected target lesions, we then determined how many different response categories could be assigned to a patient. For example, for a patient with five lesions (two with individual outcome of a partial response, two with stable disease, and one with progressive disease), the number of response categories for one selected lesion is three.

FIG A7.

Variability assessments: Lung Cancer - Percentage change from baseline to week 12 by treatment category.

FIG A8.

Variability assessments: Lung Cancer - Variance plots of Percentage change from baseline to week 12 by treatment category.

FIG A9.

Variability assessments: Lung Cancer - Variance plots of Percentage change from baseline to week 12 by treatment category.

FIG A10.

Variability assessment - Lung Cancer: Response categories (at 12 weeks) by number of selected target lesions: number of different response categories a patient could be classified in on the basis of all possible selections of target lesions by number of selected lesions. The number of response categories for all possible combinations of target lesions in a patient per RECIST 1.1 outcome was assessed. By number of selected target lesions, we then determined how many different response categories could be assigned to a patient. For example, for a patient with five lesions (two with individual outcome of a partial response, two with stable disease, and one with progressive disease), the number of response categories for one selected lesion is three.

FIG A11.

Variability assessments: Colon cancer - Percentage change from baseline to week 12 by treatment category.

FIG A12.

Variability assessments: Colon Cancer - Variance plots of Percentage change from baseline to week 12 by treatment category.

FIG A13.

Variability assessments: Colon Cancer - Variance plots of Percentage change from baseline to week 12 by treatment category.

FIG A14.

Variability assessment - Colon Cancer: Response categories (at 12 weeks) by number of selected target lesions: number of different response categories a patient could be classified in on the basis of all possible selections of target lesions by number of selected lesions. The number of response categories for all possible combinations of target lesions in a patient per RECIST 1.1 outcome was assessed. By number of selected target lesions, we then determined how many different response categories could be assigned to a patient. For example, for a patient with five lesions (two with individual outcome of a partial response, two with stable disease, and one with progressive disease), the number of response categories for one selected lesion is three.

FIG A15.

Variability assessments: Breast Cancer - Percentage change from baseline to week 12 by treatment category.

FIG A16.

Variability assessments: Breast Cancer - Variance plots of Percentage change from baseline to week 12 by treatment category.

FIG A17.

Variability assessments: Breast Cancer - Variance plots of Percentage change from baseline to week 12 by treatment category.

FIG A18.

Variability assessment - Breast Cancer: Response categories (at 12 weeks) by number of selected target lesions: number of different response categories a patient could be classified in on the basis of all possible selections of target lesions by number of selected lesions. The number of response categories for all possible combinations of target lesions in a patient per RECIST 1.1 outcome was assessed. By number of selected target lesions, we then determined how many different response categories could be assigned to a patient. For example, for a patient with five lesions (two with individual outcome of a partial response, two with stable disease, and one with progressive disease), the number of response categories for one selected lesion is three.

FIG A19.

Variability assessments: GIST - Percentage change from baseline to week 12 by treatment category.

FIG A20.

Variability assessments: GIST - Variance plots of Percentage change from baseline to week 12 by treatment category.

FIG A21.

Variability assessments: GIST - Variance plots of Percentage change from baseline to week 12 by treatment category.

FIG A22.

Variability assessment - GIST: Response categories (at 12 weeks) by number of selected target lesions: number of different response categories a patient could be classified in on the basis of all possible selections of target lesions by number of selected lesions. The number of response categories for all possible combinations of target lesions in a patient per RECIST 1.1 outcome was assessed. By number of selected target lesions, we then determined how many different response categories could be assigned to a patient. For example, for a patient with five lesions (two with individual outcome of a partial response, two with stable disease, and one with progressive disease), the number of response categories for one selected lesion is three.

FIG A23.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A24.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A25.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A26.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A27.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A28.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A29.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A30.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A31.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A32.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A33.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A34.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A35.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A36.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A37.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A38.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A39.

Forest plots of a landmark analysis at 12 weeks of survival by percentage change from baseline of the sum of target lesions up to 12 weeks. Models stratified by study and adjusted for baseline sum of diameters, occurrence of new lesions, and progression of nontarget lesions.

FIG A40.

Time-dependent survival analysis: cumulative contributions to the area under the curve (AUC) of the different components of progression. (A) TCAs, (B) Chemotherapy, (C) TCA and chemotherapy. Abbreviation: TCA = targeted cancer agent.

FIG A41.

Time-dependent survival analysis: separate contributions to the area under the curve (AUC) of the different components of progression. (A) TCAs, (B) Chemotherapy, (C) TCA and chemotherapy. Abbreviation: TCA = targeted cancer agent.

FIG A42.

Time-dependent survival analysis: cumulative contributions to the area under the curve (AUC) of the different components of progression. (A) TCAs, (B) Chemotherapy, (C) TCA and chemotherapy. Abbreviation: TCA = targeted cancer agent”.

FIG A43.

Time-dependent survival analysis: separate contributions to the area under the curve (AUC) of the different components of progression. (A) TCAs, (B) Chemotherapy, (C) TCA and chemotherapy. Abbreviation: TCA = targeted cancer agent”.

TABLE A1.

Phase II and III Trials With Clinical Data From Patients Treated With TCAs or TCAs Plus Chemotherapy Compounds

TABLE A2.

Targeted Cancer Agents in the Warehouse

TABLE A3.

Lesion Site Classification

TABLE A4.

Analysis Dataset: Type of Treatment by Disease

TABLE A5.

Analysis Dataset: Description of TCAs

TABLE A6.

Analysis Dataset: Extent of Disease at Baseline

TABLE A7.

Analysis Dataset: Site of Target Lesions at Baseline

TABLE A8.

Analysis Dataset: Period Covered by the Tumor Measurement Assessments

TABLE A9.

Analysis Dataset: Survival by Tumor Type and Treatment Category

TABLE A10.

Variability Assessments: Patients in Subgroups of Interest Considered for the Analyses

TABLE A11.

Variability Assessments: Summary Statistics: Percentage Change From Baseline to Week 12 by Treatment Category

TABLE A12.

Variability Assessments: Distribution of the Per-Patient Variance of the Percentage Change from Baseline to Week 12, by Treatment Category

TABLE A13.

Variability Assessments: Patients Who Contributed to the Variability Assessment

TABLE A14.

Variability Assessments: Summary Statistics: Percentage Change From Baseline to 12 Weeks by Type of Targeted Cancer Agent

TABLE A15.

Variability Assessments: Distribution of the Per-Patient Variance of the Percentage Change from Baseline to Week 12, by Type of TCA

TABLE A16.

Variability Assessments: Patients Who Contributed to the Variability Assessment

TABLE A17.

Variability Assessments: Lung Cancer Summary Statistics: Percentage Change From Baseline to Week 12 by Treatment Category

TABLE A18.

Variability Assessments: Lung Cancer - Distribution of the Per-Patient Variance of the Percentage Change from Baseline to Week 12, by Treatment Category

TABLE A19.

Variability Assessments: Patients With Lung Cancer Who Contributed to the Variability Assessment

TABLE A20.

Variability Assessments: Colorectal Cancer Summary Statistics: Percentage Change From Baseline to Week 12 by Treatment Category

TABLE A21.

Variability Assessments: Colorectal Cancer - Distribution of the Per-Patient Variance of the Percentage Change from Baseline to Week 12, by Treatment Category

TABLE A22.

Variability Assessments: Patients With Colorectal Cancer Who Contributed to the Variability Assessment

TABLE A23.

Variability Assessments: Breast Cancer Summary Statistics: Percentage Change From Baseline to Week 12 by Treatment Category

TABLE A24.

Variability Assessments: Breast Cancer - Distribution of the Per-Patient Variance of the Percentage Change from Baseline to Week 12, by Treatment Category

TABLE A25.

Variability Assessments: Patients With Breast Cancer Who Contributed to the Variability Assessment

TABLE A26.

Variability Assessments: GIST Cancer Summary Statistics: Percentage Change From Baseline to 12 Weeks by Treatment Category

TABLE A27.

Variability Assessments: GIST - Distribution of the Per-Patient Variance of the Percentage Change from Baseline to Week 12, by Treatment Category

TABLE A28.

Variability Assessments: Patients With GIST Cancer Who Contributed to the Variability Assessment

TABLE A29.

Time-Dependent Model With Tumor Growth Rate by Treatment Category

TABLE A30.

Time-Dependent Model with Tumor Growth Rate by Type of TCA

TABLE A31.

Time-Dependent Model with Tumor Growth Rate by Tumor Type

TABLE A32.

Time-Dependent Model with Tumor Growth Rate for Lung Cancer Patients by Treatment Category

TABLE A33.

Time-Dependent Model with Tumor Growth Rate for Colorectal Cancer Patients by Treatment Category

TABLE A34.

Time-Dependent Model with Tumor Growth Rate for Breast Cancer Patients by Treatment Category

TABLE A35.

Time-Dependent Model with Percentage Growth Rate by Treatment Category

TABLE A36.

Time-Dependent Model with Percentage Growth Rate by Type of TCA

TABLE A37.

Time-Dependent Model with Percentage Growth Rate by Tumor Type

TABLE A38.

Time-Dependent Model with Percentage Growth Rate for Lung Cancer Patients by Treatment Category

TABLE A39.

Time-Dependent Model with Percentage Growth Rate for Colorectal Cancer Patients by Treatment Category

TABLE A40.

Time-Dependent Model with Percentage Growth Rate for Breast Cancer Patients by Treatment Category

AUTHOR CONTRIBUTIONS

Conception and design: Saskia Litière, Elisabeth G.E. De Vries, Jan Bogaerts, Janet Dancey, Nancy Lin, Yan Liu, Sumithra Mandrekar, Lawrence H. Schwartz, Lalitha Shankar, Lesley Seymour

Financial support: Lesley Seymour

Administrative support: Robert Ford, Lesley Seymour

Provision of study material or patients: Janet Dancey

Collection and assembly of data: Saskia Litière, Gaëlle Isaac, Elisabeth G.E. De Vries, Jan Bogaerts, Yan Liu, Lawrence H. Schwartz, Lesley Seymour

Data analysis and interpretation: Saskia Litière, Gaëlle Isaac, Elisabeth G.E. De Vries, Jan Bogaerts, Alice Chen, Janet Dancey, Robert Ford, Stephen Gwyther, Otto Hoekstra, Erich Huang, Nancy Lin, Sumithra Mandrekar, Lawrence H. Schwartz, Lalitha Shankar, Patrick Therasse, Lesley Seymour

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST AND DATA AVAILABILITY STATEMENT

RECIST 1.1 for Response Evaluation Apply Not Only to Chemotherapy-Treated Patients But Also to Targeted Cancer Agents: A Pooled Database Analysis

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc.

Elisabeth G.E. De Vries

Consulting or Advisory Role: Pfizer (Inst), Sanofi (Inst), Daiichi Sankyo (Inst)

Research Funding: Amgen (Inst), Roche (Inst), Genentech (Inst), Chugai Pharma (Inst), Synthon (Inst), AstraZeneca (Inst), Radius Health (Inst), CytomX Therapeutics (Inst), Nordic Nanovector (Inst), Regeneron Pharmaceuticals (Inst), G1 Therapeutics (Inst)

Jan Bogaerts

Stock and Other Ownership Interests: Bristol-Myers Squibb

Travel, Accommodations, Expenses: AstraZeneca

Janet Dancey

Leadership: 3Ci

Honoraria: 3Ci

Consulting or Advisory Role: Roche Canada, Sanofi Canada

Research Funding: Pfizer (Inst), Merck (Inst), AstraZeneca (Inst), MedImmune (Inst), Novartis (Inst), Bristol-Myers Squibb (Inst), Roche (Inst)

Travel, Accommodations, Expenses: Sanofi Canada

Robert Ford

Consulting or Advisory Role: AbbVie, ACR Image Metrix, Acuta Capital Partners, Agensys, Alchemia, Amgen, Aptiv, Aragon Pharmaceuticals, ARIAD, Array BioPharma, Astellas Pharma, BeiGene, BioClinica, Biomedical Systems, Bristol-Myers Squibb, Celldex, Celgene, CELSION, Chiltern, Clovis Oncology, Covance, CTI BioPharma, CytRx, DNAtrix, Eisai, Eli Lilly, EMD Serono, Exelixis, GlaxoSmithKline, Genentech, Roche, Gilead Sciences, Ipsen, ICON, Ignyta, Immunocellular Therapeutics, Incyte, Janssen Pharmaceuticals, Kura Oncology, Kyowa Hakko Kirin, LifeBond, MacroGenics, MedImmune, Medivir, Merck, Merck KGaA, Merus, Mirati Therapeutics, Morphotek, Nektar, Novartis, Novocure, Odonate Therapeutics, OncoSec Medical, Oncothyreon, Ono Pharmaceutical, Optimer Biotechnology, OrbiMed Advisors, PAREXEL International, Pfizer, Puma Biotechnology, Quintiles, RedHill, Radiant Sage, Replimune, Sanofi, Samsung Bioepis, Savient Pharmaceuticals, Shanghai Junshi Biosciences, Spectrum Pharmaceuticals, Sun Pharmaceutical Industries, Syneos Health, Tesaro, TEVA Pharmaceuticals Industries, Tocagen, Tokai Pharmaceuticals, TRACON Pharmaceuticals, Vascular Biogenics, VBI Vaccines, Viralytics, Xcovery, Zeria Pharmaceutical, Loxo Oncology, Imaging Endpoints

Speakers’ Bureau: Medidata Solutions

Travel, Accommodations, Expenses: Medidata Solutions

Stephen Gwyther

Employment: Celgene (I)

Stock and Other Ownership Interests: Eli Lilly (I)

Travel, Accommodations, Expenses: Celgene (I)

Nancy Lin

Consulting or Advisory Role: Genentech, Roche, Seattle Genetics, Puma Biotechnology, Shionogi, Novartis

Research Funding: Genentech, Pfizer, Novartis, Array BioPharma

Patents, Royalties, Other Intellectual Property: Royalties for chapter in UptoDate regarding management of breast cancer brain metastases

Yan Liu

Employment: MEDIAN Technologies

Sumithra Mandrekar

Consulting or Advisory Role: Pfizer, Pique Therapeutics

Other Relationship: BeiGene

Lawrence H. Schwartz

Consulting or Advisory Role: Novartis, Roche

Research Funding: Eli Lilly (Inst), Merck Sharpe & Dohme (Inst), Millennium Pharamceuticals (Inst), Daiichi Sankyo (Inst)

Patents, Royalties, Other Intellectual Property: Varian Medical Systems

Patrick Therasse

Employment: SERVIER

Travel, Accommodations, Expenses: SERVIER

Lesley Seymour

Stock and Other Ownership Interests: AstraZeneca

Consulting or Advisory Role: Boehringer Ingelheim

Research Funding: AstraZeneca (Inst), Merck (Inst), Senhwa Biosciences (Inst)

No other potential conflicts of interest were reported.