Original Study Health-related Quality of Life in the Phase III LUME-Colon 1 Study: Comparison and Interpretation of Results From EORTC QLQ-C30 Analyses Heinz-Josef Lenz,1 Guillem Argiles,2 Takayuki Yoshino,3 Sara Lonardi,4 Alfredo Falcone,5 María Luisa Limón,6 Alberto Sobrero,7 Claudia Hastedt,8 Barbara Peil,9 Florian Voss,9 Ingolf Griebsch,10 Eric Van Cutsem11 Abstract Based on European Organization for Research and Treatment of Cancer Quality of Life Questionnaire (EORTC QLQ-C30) outputs from the LUME-Colon 1 study, we compared and discussed different statistical methods for evaluating health-related quality of life data in oncology clinical trials. The different analyses consistently showed that patients’ overall global health status/quality of life status was not impaired by active treatment with nintedanib versus placebo, and that patients perceived some benefits with nintedanib compared with placebo. Introduction: We used European Organization for Research and Treatment of Cancer Quality of Life Questionnaire (EORTC QLQ-C30) data from the LUME-Colon 1 study to illustrate different methods of statistical analysis for health- related quality of life (HRQoL), and compared the results. Patients and Methods: Patients were randomized 1:1 to receive nintedanib 200 mg twice daily plus best supportive care (n ¼ 386) or matched placebo plus best supportive care (n ¼ 382). Five methods (mean treatment difference averaged over time, using a mixed-effects growth curve model; mixed-effects models for repeated measurements (MMRM); time-to-deterioration (TTD); status change; and responder analysis) were used to analyze EORTC QLQ-C30 global health status (GHS)/QoL and scores from functional scales. Results: Overall, GHS/QoL and physical functioning deteriorated over time. Mean treatment difference slightly favored nintedanib over placebo for physical functioning (adjusted mean, 2.66; 95% confidence interval [CI], 0.97-4.34) and social functioning (adjusted mean, 2.62; 95% CI, 0.66-4.47). GHS/QoL was numerically better with nintedanib versus placebo (adjusted mean, 1.61; 95% CI, 0.004 to 3.27). MMRM analysis had similar results, with better physical functioning in the nintedanib group at all timepoints. There was no significant delay in GHS/QoL deterioration (10%) and physical functioning (16%) with nintedanib versus placebo (TTD analysis). Status change analysis showed a higher proportion of patients with markedly improved GHS/QoL and physical functioning in the nintedanib versus placebo groups. Responder analysis showed a similar, less pronounced pattern. Conclusion: Analyses of EORTC QLQ-C30 data showed that HRQoL was not impaired by treatment with nintedanib versus placebo. Analysis and interpretation of HRQoL endpoints should consider symptom type and severity and course of disease. Clinical Colorectal Cancer, Vol. 18, No. 4, 269-79 ª 2019 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Keywords: HRQoL, MMRM, Nintedanib, QoL, Time to deterioration 1 8 Division of Medical Oncology, USC Norris Comprehensive Cancer Center, Los TA CNS Retinopathies Emerging Areas, Boehringer Ingelheim International GmbH, Angeles, CA Ingelheim, Germany 2 9 Medical Oncology Department, Vall d’Hebron University Hospital and Institute of Biostatistics and Data Sciences, Boehringer Ingelheim Pharma GmbH & Co. KG, Oncology (VHIO), Universitat Autònoma de Barcelona, Barcelona, Spain Ingelheim, Germany 3 10 Department of Gastroenterology and Gastrointestinal Oncology, National Cancer TA Oncology, Boehringer Ingelheim International GmbH, Ingelheim, Germany 11 Center Hospital East, Chiba, Japan Digestive Oncology, University Hospitals Leuven and KULeuven, Leuven, Belgium 4 Phase 1 Trial Unit and Medical Oncology Unit 1, Istituto Oncologico Veneto IRCCS, Padova, Italy Submitted: Nov 26, 2018; Revised: Aug 1, 2019; Accepted: Aug 27, 2019; Epub: Sep 5 Department of Oncology, University of Pisa, Pisa, Italy 4, 2019 6 Medical Oncology Department, Hospital Universitario Virgen del Rocio, Sevilla, Spain Address for correspondence: Heinz-Josef Lenz, MD, FACP, USC Norris Comprehensive 7 Cancer Center, 1441 Eastlake Ave, NOR 3456, Los Angeles, CA 90089-9173 Medical Oncology 1, Ospedale Policlinico San Martino IRCCS, Genova, Italy E-mail contact: lenz@usc.edu - 269 1533-0028/ª 2019 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). https://doi.org/10.1016/j.clcc.2019.08.005 Clinical Colorectal Cancer December 2019 LUME-Colon 1 QoL Introduction alanine aminotransferase (8%) and aspartate aminotransferase (8%) Patient-reported outcomes (PROs) are an important consider- levels and fatigue. The most common AEs leading to treatment ation in the delivery of clinical care, clinical trial design, and discontinuation in the nintedanib group were fatigue (n ¼ 8), comparative effectiveness research in oncology. Regulatory bodies followed by asthenia, decreased appetite, and malignant neoplasm such as the United States Food and Drug Administration and the progression (all n ¼ 6).22 The LUME-Colon 1 study incorporated European Medicines Agency encourage the inclusion of patient- assessments of PROs, including HRQoL, as a further endpoint. focused clinical outcome assessments in clinical trials to support The objectives of this evaluation were: (1) to perform different drug approvals and labeling claims.1 PROs are also integrated into exploratory analyses of EORTC QLQ-C30 data from the LUME- health technology assessments to reflect clinical benefit from the Colon 1 study and (2) to discuss possible interpretations of these patient’s perspective.2 different evaluations and compare and contrast the various methods In trials that include patients with advanced cancer, including used. those with metastatic colorectal cancer (CRC), the assessment and analysis of PROs present several challenges. Generally, there is an Patients and Methods overall deterioration in patient health and quality of life (QoL) Study Design rather than an improvement, owing to the clinical course of the The study design of LUME-Colon 1 has been reported previ- disease.3 In addition, because of disease progression, there is a high ously.22 LUME-Colon 1 enrolled adults with histologically or dropout rate, which results in incomplete data sets.3 The dropout cytologically confirmed metastatic or locally advanced CRC that was may often differ between treatment arms, which complicates the not amenable to curative surgery and/or radiotherapy. Patients were issue of missing data.4 Finally, health-related quality of life required to have progressed on approved standard therapies or to (HRQoL) scales are often multidimensional and include several have experienced unacceptable toxicity.22 Study participants were subscales, which can lead to complex result patterns.5 randomized 1:1 to receive treatment with either nintedanib 200 mg The European Organization for Research and Treatment of Cancer bid plus BSC (n ¼ 386) or placebo bid plus BSC (n ¼ 382) in 21- Quality of Life Questionnaire (EORTC QLQ-C30) is often used to day cycles.22 In this study, BSC was defined as the best palliative collect data on HRQoL and symptom burden in patients with cancer.6 care, excluding specific anticancer treatments, according to the in- However, these data can be analyzed using a variety of methods, as vestigator’s decision. exemplified in different clinical studies of CRC (see Supplemental Table 1 in the online version).7-18 These statistical methods can be PRO Measures broadly allocated to 2 types: those that assess group means and those PROs, including HRQoL, were evaluated using EORTC QLQ- that assess clinically relevant changes at the level of the patient. The C30 and the EuroQoL 5-dimensions questionnaire (EQ-5D). The first type of method ‒ group mean methods ‒ treat the data as EQ-5D was included for health economic analyses, and so the continuous endpoints, and include longitudinal models such as mixed- analyses in this report will only focus on the EORTC QLQ-C30 effects growth curve models and mixed-effects models for repeated component. The EORTC QLQ-C30 consists of multi-item scales measurements (MMRM). The second type of method requires defi- and single-item measures as follows: 5 functional scales (physical, nition of a minimal clinically important difference (MCID) at the level role, emotional, cognitive, and social functioning)25; 3 symptom of the individual patient, including time-to-deterioration (TTD), scales (fatigue, pain, and nausea/vomiting)25; 6 single questions distribution of patients who improved, patients who were stable or (assessing dyspnea, appetite loss, sleep disturbance, constipation, worsened from baseline, and responder analysis.19 diarrhea, and the perceived financial impact of disease and treat- Nintedanib is an oral, twice-daily (bid), triple angiokinase in- ment)25; and global health status (GHS)/quality of life (QoL) hibitor that targets vascular endothelial growth factor receptors 1 to scale.25 3, platelet-derived growth factor receptors a/b, and fibroblast We report the GHS/QoL scale (consisting of 2 items) and the 5 growth factor receptors 1 to 3, as well as RET and Flt3 functional scales (see Supplemental Table 2 in the online version).25 receptors.20,21 Questionnaires were completed at screening, on day 1 of each 21- The randomized, double-blind, placebo-controlled, phase III day cycle, at the end of treatment visit, 28 days after the last LUME-Colon 1 study (ClinicalTrials.gov identifier treatment, and at follow-up for disease progression. Scoring fol- NCT02149108) investigated the efficacy and safety of nintedanib lowed the EORTC QLQ-C30 scoring algorithm.25 A linear trans- plus best supportive care (BSC) versus placebo plus BSC for the formation was used to standardize raw scores from 0 to 100.25 A treatment of patients with metastatic CRC refractory to standard higher score on the GHS/QoL scale and the functioning scales therapies.22 Co-primary endpoints were progression-free survival indicated a higher (‘better’) level of functioning and HRQoL.25 (PFS) and overall survival (OS).22 PFS by independent central re- view was significantly longer with nintedanib than with placebo Statistical Analyses (hazard ratio [HR], 0.58; 95% confidence interval [CI], 0.49-0.69; All analyses were descriptive and exploratory. P values were P < .0001),22 whereas OS was not (HR, 1.01; 95% CI, 0.86-1.19; nominal and not adjusted for multiple testing. In line with the P ¼ .8659).22 The safety profile of nintedanib was consistent with primary PFS and OS analysis of the trial, all models were adjusted safety data previously reported in early-phase CRC studies.23,24 In for the following stratification factors: regorafenib pretreatment; LUME-Colon 1, the most frequent grade  3 adverse events (AEs) time from onset of metastatic disease; and region. Five different in the nintedanib group were liver-related AEs, mainly increased methods were used to analyze the EORTC QLQ-C30 data. 270 - Clinical Colorectal Cancer December 2019 Heinz-Josef Lenz et al Methods Using Group Means allowed the model to incorporate time profiles without assuming a These methods analyze the scales as continuous endpoints (lon- specific shape over time, in contrast to the piecewise linear shape in gitudinal analyses), compute the mean of each treatment arm, and the mixed-effects growth curve model. The analysis was restricted to compare these means (Table 1). We used 2 of these methods to weeks 3, 6, 9, and 12. An unstructured covariance matrix was used analyze data from LUME-Colon 1: the mixed-effects growth curve to model the within-patient measurements. model and the MMRM. Methods Assessing Clinically Relevant Changes at the Mean Treatment Difference for GHS/QoL and Functional Scale Patient Level Scores Based on Mixed-effects Growth Curve Models. This method These methods show differences between treatment arms based assessed changes in scores over time, with the average profile over on the proportion of patients with an on-treatment improvement or time for each endpoint described by a piecewise linear regression a delay in worsening of the disease (Table 1). model. For this particular study, weeks 3, 6, and 9 were selected to allow the slope to change; many patients were anticipated to drop Deteriorations Analyzed as TTD. In this evaluation, TTD was out by week 12 owing to disease progression. All available data were considered to be the time elapsed until a minimal clinically used to fit the model. important deterioration (ie, of 10 points from baseline) occurred.26 Each model included 2 random effects: intercept and slope. To fit Cox regression, log-rank tests, and Kaplan-Meier curves were used the model, the actual timepoints of the assessments were used. The as standard analysis methods. A stratified log-rank test, using the mean score for each HRQoL endpoint was calculated from the area same stratification as for PFS, was used to investigate whether a under the estimated growth curve up to the median follow-up time, delay in progression translated into a delayed worsening in HRQoL. and was interpreted as the average HRQoL score until the median Patients who died before their condition deteriorated were assumed follow-up time per group. to have deteriorated at the time of death. If a patient dropped out without experiencing deterioration or death, they were censored in Mean Treatment Difference for GHS/QoL and Functional Scale this analysis. This analysis assumed that censoring was independent Scores Based on MMRM. This longitudinal model also assessed of deterioration in HRQoL. changes in scores over time. It shows a pairwise comparison of mean scores between treatment arms for each timepoint/visit. The model Improvements Analyzed as Status Change (Improved, Stable, or included fixed categorical effects of treatment, stratification factors, Worsened). Status change assessed the proportion of patients who visit, treatment-by-visit interaction, baseline, and baseline-by-visit improved (even if this was temporary), stabilized, or worsened interaction. The effects of visit and treatment-by-visit interaction without any improvement for each EORTC QLQ-C30 scale. Table 1 Summary of Methods That Can Be Used to Assess EORTC QLQ-C30 MCID One Event vs. Methods Description Required Over Time Methods using group means Mixed-effects growth curve models  Mean score over time, typically assessed at selected timepoints No Over time MMRM  Estimates fixed effects: a visit effect, an arm effect, and a treatment-by- No Over time visit interaction. Random effects (individual trends) are also introduced28  This model accounts for the association of repeated measures made on the same patient28 Methods assessing clinically relevant changes at the patient level TTD  Survival-based analyses19 Yes One event  Describes time from baseline until first instance of patient deterioration of at least one MCID unit3  Patients with no deterioration before they are lost to follow-up are censored at the time of the last follow-up or the last HRQoL assessment3 TUDD  Survival-based analysis19 Yes One event  Describes time from baseline until definitive deterioration (ie, after which point there is no improvement of > 1 MCID unit compared with baseline value or if the patient dropped out after deterioration, resulting in missing data)3 Status change  Analysis of proportion of patients improving, worsening, or being stable, defined Yes One event by change  MCID  Comparison of percentage of patients being stable and/or improved vs. worsened Responder analysis  Comparison of proportion of responders, defined as patients with Yes Over time mean improvement of  1 MCID during follow-up time of the study30 Time to improvement  Time until first improvement  MCID Yes One event Abbreviations: EORTC QLQ-C30 ¼ European Organization for the Research and Treatment of Cancer Quality of Life Questionnaire; HRQoL ¼ health-related quality of life; MCID ¼ minimal clinically important difference; MMRM ¼ mixed-effects model for repeated measurements; TTD ¼ time to deterioration; TUDD ¼ time until definitive deterioration. Clinical Colorectal Cancer December 2019 - 271 LUME-Colon 1 QoL Patients were categorized as having perceived improvement at least Mean Treatment Difference for GHS/QoL and Functional once (improvement of  10 units from baseline at any time during Scale Scores the study), regardless of potential worsening at other timepoints; Baseline mean scores for GHS/QoL and physical functioning worsened (decrease of  10 units and no improvement at any time were comparable between treatment arms. Baseline mean scores for during the study); or remained stable (neither showing improve- GHS/QoL were 64.9  19.6 versus 65.3  20.5 for the nintedanib ment nor deterioration of  10 units at any time during the study). and placebo arms, respectively. Baseline mean scores for physical This endpoint was analyzed as a binary endpoint, comparing the functioning were 80.1  18.4 in the nintedanib arm and proportion of patients with improvements between treatment 79.8  17.9 in the placebo arm. groups using logistic regression. Mean treatment difference (ie, the difference in the area under the curve in the mixed-effects growth curve model, adjusted for Responder Analysis. In contrast to the status change analysis, the stratification factors) favored nintedanib plus BSC over placebo plus responder analysis was based on the average change from baseline BSC for physical functioning and social functioning. Numerical over the entire observation period. A responder was defined as a improvements were reported for GHS/QoL, role, and emotional patient who achieved an average  10-point increase in score from and cognitive functioning (Figure 1). baseline over the duration of follow-up. The responder analysis compared the proportions of patients in each treatment arm, as was MMRM Analysis done for status change analysis. Numerical improvements in the GHS/QoL scores and all func- tioning scores were reported in the nintedanib group versus the Results placebo group (Figures 2 and 3). Only physical functioning showed Patient Population and Baseline Characteristics a significant between-group difference at all timepoints (Figure 3). Patient demographics and baseline clinical characteristics have Overall, a worsening in GHS/QoL and physical functioning was been reported previously and were mostly balanced between treat- observed from weeks 3 to 12; this illustrates the expected, natural ment groups.22 course of the disease (Figure 2). There were significant between-group differences in physical Completion of EORTC QLQ-C30 Questionnaires functioning in favor of the nintedanib group at weeks 3, 6, 9, and During the first 12 cycles of treatment, EORTC QLQ-C30 12 (Figure 3B).Treatment differences were similar between weeks 3 questionnaires were completed by  87.5% of patients. In the to 6 and 9 to 12, respectively, with larger differences in magnitude nintedanib plus BSC arm and the placebo plus BSC arm, 82.7% reported later in therapy (ie, at weeks 9 and 12) (Figure 3B). and 80.4% of patients, respectively, completed questionnaires at the Between-treatment group differences in emotional functioning end of treatment. scores followed a similar trend to those observed for physical Figure 1 Longitudinal Analysis: Mean Treatment Difference for GHS/QoL and Functional Scale Scores Abbreviations: CI ¼ confidence interval; GHS ¼ global health status; Q ¼ question; QoL ¼ quality of life. 272 - Clinical Colorectal Cancer December 2019 Heinz-Josef Lenz et al Figure 2 MMRM Analysis: Adjusted Mean Score of GHS/QoL and Physical Functioning. Figure Without Break in Axis Is Shown Inset. Baseline Values Are From Longitudinal Analysis and Are Shown for Reference. Error Bars Are ± SE Abbreviations: GHS ¼ global health status; MMRM ¼ mixed-effects model for repeated measurements; QoL ¼ quality of life. functioning, although between-group differences were only signifi- other functional scales, with cognitive and social functioning also cant at week 9 (see Supplemental Table 3 in the online version) showing a significant difference between treatment groups (Figure 3D). Between-group differences in cognitive functioning (Figure 6). and social functioning are shown in Figures 3E and F, respectively. In contrast to other endpoints, there were no significant between- Responder Analysis treatment group differences in role functioning at any of the On average over the treatment period, only a small proportion of timepoints (see Supplemental Table 3 in the online version) patients perceived an improvement in HRQoL and functioning (Figure 3C). scores (5.3%-13.8% in the placebo arm and 6.9%-17.2% in the nintedanib arm) (see Supplemental Table 4 in the online version). TTD Nevertheless, the nintedanib group had a somewhat higher response There was a general trend in favor of the nintedanib group rate in GHS/QoL and physical functioning scores than the placebo compared with the placebo group for all functional scales apart from group, with an odds ratio equating to an improvement of 46% and role functioning; there was also a trend in favor of the nintedanib 32%, respectively (see Supplemental Table 4 in the online version). group for GHS/QoL (Figures 4 and 5). Cognitive functioning There were higher responder rates with nintedanib versus placebo in showed a significant difference between treatment arms (Figure 5). the cognitive and social functioning scales than in other functional scales, equating to a relative improvement of 54% and 29%, Status Change (Proportion of Patients Improved vs. Not respectively (see Supplemental Table 4 in the online version). Improved [Stable or Worsened]) However, the confidence intervals included 1 for all scores. The proportion of patients with improved GHS/QoL and physical functioning was significantly higher in the nintedanib Discussion group than in the placebo group (Figure 6), with an odds ratio The LUME-Colon 1 study failed to meet both co-primary equating to an improvement of 56% and 54%, respectively. There endpoints: there was no survival benefit and a significant but was a general trend in favor of nintedanib versus placebo for all modest increase in PFS with nintedanib versus placebo. PRO data Clinical Colorectal Cancer December 2019 - 273 LUME-Colon 1 QoL Figure 3 MMRM Analysis: Mean Treatment Difference. Plots Shown Are Placebo-corrected Means With Upper and Lower Confidence Intervals Abbreviations: GHS ¼ global health status; MMRM ¼ mixed-effects model for repeated measurements; QoL ¼ quality of life. from the study, reported here, indicate that patients’ overall Overall, results from the different methods of analysis showed an GHS/QoL status and physical functioning were not impaired, and apparent benefit of treatment with nintedanib versus placebo in were improved in some respects, by treatment with nintedanib various PROs, in particular in physical functioning. The longitu- compared with placebo. dinal analysis using the mixed-effects growth curve model included 274 - Clinical Colorectal Cancer December 2019 Heinz-Josef Lenz et al Figure 4 Kaplan-Meier Curves of Time to Deterioration of GHS/QoL Score (A) and Physical Functioning Score (B) Abbreviations: CI ¼ confidence interval; GHS ¼ global health status; HR ¼ hazard ratio; QoL ¼ quality of life. the first 9 weeks of treatment. Results showed that average physical items of the physical functioning assessment indicate that patients functioning was better with nintedanib than with placebo during experienced delayed deterioration in their ability to undertake this time. MMRM confirmed this pattern through weeks 3, 6, 9, strenuous activity (eg, heavy lifting) and take a walk outside, and in and 12, with larger improvements at weeks 9 and 12. The ‘status their need for help when eating and dressing (see Supplemental change’ analysis, which focused on temporary but clinically relevant Figure 3 in the online version). Looking at individual items in changes in individual patients at specific timepoints during therapy, this way can contribute to our understanding of patients’ experience supported the outputs of the longitudinal analysis. It showed that while on treatment. there were statistically significant and clinically meaningful im- The variety of statistical methods used to analyze data obtained provements in physical functioning, GHS/QoL, and cognitive and using EORTC QLQ-C30 can make it difficult to interpret the likely social functioning with nintedanib versus placebo. impact of those data in the clinic.27 This lack of standards for In contrast to the status change analysis, the responder analysis analyzing PRO and HRQoL data has been recognized, and a con- had a stronger criterion of a mean improvement of  10 points sortium initiative (the Setting International Standards in Analysing averaged over all timepoints in the study. This analysis did not show Patient-Reported Outcomes and Quality of Life Endpoints Data) has a statistically significant difference between treatment groups, been established. This consortium, directed by the EORTC, aims to perhaps reflecting the short time to disease progression in this study establish recommendations on how to standardize the analysis of population. TTD in physical functioning was numerically longer HRQoL and other PRO data in randomized cancer drug trials.27 with nintedanib than with placebo, although there was no signifi- The outputs of the different statistical methods used to analyze cant difference between treatment arms. Data from the individual EORTC QLQ-C30 in this study demonstrate that there are Clinical Colorectal Cancer December 2019 - 275 LUME-Colon 1 QoL Figure 5 TTD: GHS/QoL and Functioning Scales. Results Presented on a Logarithmic Scale to Base 4 Abbreviations: CI ¼ confidence interval; GHS ¼ global health status; HR ¼ hazard ratio; Q ¼ question; QoL ¼ quality of life; TTD ¼ time to deterioration. different ways to evaluate what patients experience in everyday life relevant criteria for comparing group means are needed, which are while on treatment. Interpretation of these different analyses often more difficult to substantiate than a MCID on the individual requires context: in this study, it is important to note the rapid patient level. Another disadvantage is that during long-term follow- deterioration in QoL and physical functioning experienced by pa- up, the assignment of timepoints to visits might be challenging if tients in both treatment arms, and the short PFS. Clinical inter- clinic visits are delayed. pretation of the results of these methods requires us to bear in mind the strengths and limitations of each one, as follows. TTD Among the strengths of TTD is its use of MCID. For well- Mixed-effects Growth Curve and MMRM Models established questionnaires such as the EORTC QLQ-C30, widely These analyses offer an overall comparison of the treatment arms, accepted MCIDs are available. This makes TTD a clinically relevant taking the longitudinal course of the disease and treatment into measure that is easy to interpret and apply. TTD is a particularly account. Both models assume that data are missing at random. This meaningful analysis if a deterioration in HRQoL is expected long assumption is violated if patients who drop out have worse HRQoL term and the therapy tested is expected to delay progression and scores than those who continue in the study, which is a common deterioration of HRQoL. Analysis of TTD mirrors the analysis of scenario. However, a key advantage of these methods is that group PFS, complementing the interpretation of this common endpoint in means can be analyzed without a defined MCID. Furthermore, all oncology trials. Finally, TTD analyses can be adjusted using the information from the questionnaire can be analyzed directly, censoring to take into account missing data, patient deaths, and without the need to classify the collected data as an improvement or a changes in patients’ individual perception of their QoL over time.28 deterioration. In addition, mixed-effects growth curve models are able One of the disadvantages of TTD analyses is the need to clearly to incorporate the actual timepoint of a measurement, even if there define what constitutes deterioration.28 Another point is that the are many assessments during long-term follow-up, whereas the analysis may be less robust if patients drop out because of pro- MMRM uses the planned timepoints of the assessments. MMRM is gressive disease but do not experience deterioration according to the regarded as a well-established and robust approach, particularly if MCID; this would mean that censoring could introduce bias into there are missing data (as long as those data are missing at random).28 the analysis. Furthermore, when interpreting the results of a TTD One disadvantage of mixed-effects growth curve and MMRM analysis, it is necessary to keep in mind that only the time to first models is that the outputs can be difficult to interpret: clinically deterioration is taken into account ‒ information on long-term 276 - Clinical Colorectal Cancer December 2019 Heinz-Josef Lenz et al Figure 6 Status Change: Proportion of Patients With Improvement in Functional Scales and GHS/QoL Scores. 1Odds Ratio >1 Favors Nintedanib; 2P Value Corresponds to Patients Who Improved vs. Patients Who Did Not Improve (ie, Stable Plus Worsened Categories Combined) Abbreviations: GHS ¼ global health status; QoL ¼ quality of life. treatment effects after the first deterioration is not considered. Disadvantages of responder analyses include that the definition of Hence, TTD analysis might be most appropriate if the first dete- response needs to be based on a validated MCID.29 In addition, this rioration is of interest, or if the natural course of disease suggests a type of analysis has reduced power compared with an analysis based continuous deterioration. Finally, TTD analyses assume that on continuous data; also, it is more challenging to achieve a censoring is non-informative; if data are not missing at random response with longer follow-up if there is a fast deterioration in before a deterioration occurs, this assumption is violated. HRQoL owing to the underlying disease. Status Change General Points on the Interpretation of Different Analyses The main advantage of the status change method is its simplicity: In some cases, the interpretation of methods that are based on it looks at clinically meaningful changes in HRQoL at the level of clinically relevant changes can be misleading. For example, patients each individual patient. Depending on how status change is defined, who experience a different QoL profile over time could have the same even a single measured change from baseline can be sufficient to TTD if they reach the MCID at the same timepoint (see Supplemental classify a patient as ‘improved.’ In severe, incurable diseases, the Figure 1 in the online version). Alternatively, a patient whose disease overall course of the disease is one of deterioration, and so a status status deteriorates rapidly but then improves could have a shorter TTD change analysis can reveal at least temporary improvements that may than a patient whose QoL remains stable for some time but then be experienced by some patients. deteriorates (see Supplemental Figure 2 in the online version). Simi- A potential disadvantage of this method is that it is susceptible to larly, a patient with a short improvement in QoL who drops out bias in favor of the treatment arm that had a longer follow-up. because of disease progression and then has a rapid deterioration in Status change analyses are therefore most appropriate in studies HRQoL after the end of follow-up could be categorized as a that have similar expected observation periods in all treatment ‘responder.’ The potential for misinterpretation highlights the groups. importance of analyzing HRQoL data using different, complementary methods to obtain an accurate picture of patients’ experiences. Responder Analysis Analysis of QoL endpoints needs to be meaningful in context, Responder analyses require a mean improvement over all time- which includes consideration of the type and severity of symptoms points of at least the MCID. It is more stringent than the status and the course of treatment and disease. In this study, the patient change analysis, and does not just take into account temporary population had late-stage CRC, with a high symptom burden, fast improvements. It is appropriate if a mean improvement can be deterioration of QoL, and short survival. Delay of progression and expected over the entire course of study treatment. deterioration while maintaining HRQoL and functioning is the Clinical Colorectal Cancer December 2019 - 277 LUME-Colon 1 QoL primary goal of therapy. Given the considerations listed in this Servier; has had an advisory role for Amgen, Bayer, Bristol-Myers Discussion, analysis of TTD, complemented by a longitudinal Squibb, Lilly, Merck, Roche, Sanofi, and Servier; and received analysis and an analysis of status change, may be most useful for honoraria from Celgene, Lilly, Merck, Roche, and Servier. A.S. has characterizing patients’ experience on study treatment. had an advisory role for Amgen, Bayer, Celgene, Roche, Merck Serono, Sanofi, and Servier, and has attended a speakers’ bureau for Conclusions Amgen, AstraZeneca, Bayer, Bristol-Myers Squibb, Celgene, Lilly, PRO data from LUME-Colon 1 showed that the overall GHS/ Merck Serono, Roche, Sanofi, and Takeda. E.V.C.’s institution has QoL status was not impaired by active treatment with nintedanib, received research funds from Amgen, Bayer, Boehringer Ingelheim, and that patients perceived some HRQoL benefits with nintedanib Celgene, Ipsen, Lilly, Merck, Merck KGaA, Novartis, Roche, compared with placebo. This is in line with the modest PFS benefit Sanofi, and Servier and has had an advisory role for Bayer, Lilly, observed in the study. Careful consideration and comparison of Roche, and Servier. C.H., B.P., F.V., and I.G. are employees of different methods of analysis of EORTC QLQ C-30 are required in Boehringer Ingelheim. M.L.L. has nothing to declare. settings such as metastatic refractory CRC in order to support clinical interpretation and application of findings. Supplemental Data Supplemental tables and figures accompanying this article can be Clinical Practice Points found in the online version at https://doi.org/10.1016/j.clcc.2019.  PROs are an essential part of clinical trials, and in particular play 08.005. a key role in interpreting the benefit of a novel treatment from the patient’s perspective. References  We used EORTC QLQ-C30 data from the LUME-Colon 1 1. Committee for Medicinal Products for Human Use, European Medicines Agency. study to compare and contrast the different statistical methods Reflection paper on the use of patient reported outcome (PRO) measures in oncology studies. Draft. June 17, 2014, Available at: www.ema.europa.eu/docs/ that can be used to evaluate these types of data. The different en_GB/document_library/Scientific_guideline/2014/06/WC500168852.pdf. analyses complement each other. Accessed: August 31, 2017. 2. Brettschneider C, Luhmann D, Raspe H. Informative value of patient reported  These analyses show that, in LUME-Colon 1, overall GHS/QoL outcomes (PRO) in health technology assessment (HTA). GMS Health Technol status was not impaired by active treatment with nintedanib Assess 2011; 7:Doc01. 3. Anota A, Hamidou Z, Paget-Bailly S, et al. Time to health-related quality of life compared with placebo. In fact, patients even perceived some score deterioration as a modality of longitudinal analysis for health-related quality benefits with nintedanib compared with placebo, in particular of life studies in oncology: do we need RECIST for quality of life to achieve standardization? Qual Life Res 2015; 24:5-18. with regard to physical functioning. 4. Bell ML, Kenward MG, Fairclough DL, Horton NJ. Differential dropout and bias  However, there was no survival benefit and only a modest in- in randomised controlled trials: when it matters and when it may not. BMJ 2013; 346:e8668. crease in PFS with nintedanib versus placebo in this study. 5. Pe M, Dorme L, Coens C, et al, Setting International Standards in Analyzing  Careful consideration and comparison of different methods of Patient-Reported Outcomes and Quality of Life Endpoints Data Consortium (SISAQOL). Statistical analysis of patient-reported outcome data in randomised analysis of EORTC QLQ C-30 are required in settings such as controlled trials of locally advanced and metastatic breast cancer: a systematic re- metastatic refractory CRC in order to support clinical interpre- view. Lancet Oncol 2018; 19:e459-69. 6. Giesinger JM, Kieffer JM, Fayers PM, et al, EORTC Quality of Life Group. tation and application of findings. Replication and validation of higher order models demonstrated that a summary score for the EORTC QLQ-C30 is robust. J Clin Epidemiol 2016; 69:79-88. 7. Yamaguchi K, Ando M, Ooki A, et al. Quality of life analysis in patients with RAS wild-type metastatic colorectal cancer treated with first-line cetuximab plus Acknowledgments chemotherapy. Clin Colorectal Cancer 2017; 16:e29-37. 8. Pinto C, Di Fabio F, Rosati G, et al. Observational study on quality of life, safety, and This work was supported by Boehringer Ingelheim. Medical effectiveness of first-line cetuximab plus chemotherapy in KRAS wild-type metastatic writing assistance, supported financially by Boehringer Ingelheim, colorectal cancer patients: the ObservEr Study. Cancer Med 2016; 5:3272-81. 9. Quidde J, Hegewisch-Becker S, Graeven U, et al. Quality of life assessment in was provided by Syneos Health Communications during the patients with metastatic colorectal cancer receiving maintenance therapy after first- preparation of this manuscript. line induction treatment: a preplanned analysis of the phase III AIO KRK 0207 trial. Ann Oncol 2016; 27:2203-10. 10. Hamidou Z, Chibaudel B, Hebbar M, et al. Time to definitive health-related Disclosure quality of life score deterioration in patients with resectable metastatic colorectal cancer treated with FOLFOX4 versus sequential dose-dense FOLFOX7 followed H.-J.L. has had an advisory role for Bayer, Merck Serono, Pfizer, by FOLFIRI: the MIROX randomized phase III trial. PLoS One 2016; 11: and Roche, and has received honoraria from Bayer, Boehringer e0157067. 11. Yoshino T, Komatsu Y, Yamada Y, et al. Randomized phase III trial of regorafenib Ingelheim, Merck Serono, and Roche. G.A. has received honoraria in metastatic colorectal cancer: analysis of the CORRECT Japanese and non- for advisory roles, travel grants and research grants (past five years) Japanese subpopulations. Invest New Drugs 2015; 33:740-50. 12. Qin S, Kim TW, Yau TC, et al. Effects of regorafenib therapy on health-related from Hoffman La-Roche, Bristol Myers Squibb, Bayer, Servier, quality of life (HRQoL) in patients with metastatic colorectal cancer (mCRC) in Amgen, Merck Serono and Menarini; and his institution has the phase III CONCUR trial. J Clin Oncol 2015; 33(3 Suppl):697. 13. Thaler J, Karthaus M, Mineur L, et al. Skin toxicity and quality of life in patients received honoraria due to his investigator contribution in clinical with metastatic colorectal cancer during first-line panitumumab plus FOLFIRI trials from Bayer, Servier, Novartis, Boehringer Ingelheim, Boston treatment in a single-arm phase II study. BMC Cancer 2012; 12:438. 14. Au HJ, Karapetis CS, O’Callaghan CJ, et al. Health-related quality of life in pa- Pharmaceuticals, Hoffman La-Roche and Genentech. T.Y.’s insti- tients with advanced colorectal cancer treated with cetuximab: overall and KRAS- tution has received research funds from Boehringer Ingelheim and specific results of the NCIC CTG and AGITG CO.17 Trial. J Clin Oncol 2009; 27:1822-8. GlaxoSmithKline. S.L. has received research funds from Sanofi; had 15. Tol J, Koopman M, Cats A, et al. Chemotherapy, bevacizumab, and cetuximab in an advisory role with Bayer and Amgen; and attended a speakers’ metastatic colorectal cancer. N Engl J Med 2009; 360:563-72. 16. Sobrero AF, Maurel J, Fehrenbacher L, et al. EPIC: phase III trial of cetuximab bureau for Roche and Lilly. A.F. has received research funds paid to plus irinotecan after fluoropyrimidine and oxaliplatin failure in patients with his institute from Amgen, Bayer, Merck, Roche, Sanofi, MSD, and metastatic colorectal cancer. J Clin Oncol 2008; 26:2311-9. 278 - Clinical Colorectal Cancer December 2019 Heinz-Josef Lenz et al 17. Rosati G, Cordio S, Tucci A, et al. Phase II trial of oxaliplatin and tegafur/uracil 25. Fayers P, Aaronson N, Bjordal K, et al. The EORTC QLQ-C30 Scoring Manual. and oral folinic acid for advanced or metastatic colorectal cancer in elderly patients. 3rd ed. Brussels: European Organisation for Research and Treatment of Cancer; Oncology 2005; 69:122-9. 2001. 18. Grothey A, Van Cutsem E, Sobrero A, et al, CORRECT Study Group. Regor- 26. Osoba D, Bezjak A, Brundage M, Pater J, National Cancer Institute of Canada afenib monotherapy for previously treated metastatic colorectal cancer (COR- Clinical Trials Group. Evaluating health-related quality of life in cancer clinical RECT): an international, multicentre, randomised, placebo-controlled, phase 3 trials: the National Cancer Institute of Canada Clinical Trials Group experience. trial. Lancet 2013; 381:303-12. Value Health 2007; 10(Suppl 2):S138-45. 19. Williams P, Fofana F, de la Loge C, et al. Methods for Analyzing Patient-reported 27. Bottomley A, Pe M, Sloan J, et al, Setting International Standards in Analyzing Outcomes in Oncology Trials. Poster presented at ISPOR 2016, Vienna, Austria. Patient-Reported Outcomes and Quality of Life Endpoints Data (SISAQOL) Poster PRM233. consortium. Analysing data from patient-reported outcome and quality of life 20. Hilberg F, Roth GJ, Krssak M, et al. BIBF 1120: triple angiokinase inhibitor with endpoints for cancer clinical trials: a start in setting international standards. Lancet sustained receptor blockade and good antitumor efficacy. Cancer Res 2008; 68:4774-82. Oncol 2016; 17:e510-4. 21. Boehringer Ingelheim. Data on File 2018. 28. Bonnetain F, Fiteni F, Efficace F, Anota A. Statistical challenges in the analysis of 22. Van Cutsem E, Yoshino T, Lenz HJ, et al. Nintedanib for the treatment of patients health-related quality of life in cancer clinical trials. J Clin Oncol 2016; 34:1953-6. with refractory metastatic colorectal cancer (LUME-Colon 1): a phase III, interna- 29. Snapinn SM, Jiang Q. Responder analyses and the assessment of a clinically tional, randomized, placebo-controlled study. Ann Oncol 2018; 29:1955-63. relevant treatment effect. Trials 2007; 8:31. 23. Mross K, Buchert M, Frost A, et al. Vascular effects, efficacy and safety of nin- 30. Lin Y. Robust inference for responder analysis: innovative clinical trial design using tedanib in patients with advanced, refractory colorectal cancer: a prospective phase a minimum p-value approach. Contemp Clin Trials Commun 2016; 3:65-9. I subanalysis. BMC Cancer 2014; 14:510. 31. Lang I, Kohne CH, Folprecht G, et al. Quality of life analysis in patients 24. Van Cutsem E, Prenen H, D’Haens G, et al. A phase I/II, open-label, randomised with KRAS wild-type metastatic colorectal cancer treated first-line with study of nintedanib plus mFOLFOX6 versus bevacizumab plus mFOLFOX6 in cetuximab plus irinotecan, fluorouracil and leucovorin. Eur J Cancer 2013; 49: first-line metastatic colorectal cancer patients. Ann Oncol 2015; 26:2085-91. 439-48. Clinical Colorectal Cancer December 2019 - 279 LUME-Colon 1 QoL Supplemental Data Supplemental Figure 1 Individual Patient Profiles Over Time With Same TTD Abbreviations: BL ¼ baseline; MID ¼ minimal important difference; TTD ¼ time to deterioration. 279.e1 - Clinical Colorectal Cancer December 2019 Heinz-Josef Lenz et al Supplemental Figure 2 Individual Patient Profiles With Short and Long TTD Abbreviations: BL ¼ baseline; MID ¼ minimal important difference; TTD ¼ time to deterioration. Supplemental Figure 3 TTD: Physical Functioning Scale e Individual Items Abbreviations: CI ¼ confidence interval; EORTC QLQ-C30 ¼ European Organization for Research and Treatment of Cancer Quality of Life Questionnaire; HR ¼ hazard ratio; Q ¼ question; TTD ¼ time to deterioration. Clinical Colorectal Cancer December 2019 - 279.e2 279.e3 LUME-Colon 1 QoL - Clinical Colorectal Cancer December 2019 Supplemental Table 1 Analysis of the EORTC QLQ-C30 in CRC Trials of Systemic Drug Therapy Difference Between Arms Change From Baseline Longitudinal Study at Specified Timepoints to Specified Timepoint TTD TUDD Analysis Responder Status Change 7 Yamaguchi et al, 2017 and Lang et al, 2013 31 3 3 3 3 Cetuximab plus chemotherapy: CRYSTAL study Pinto et al, 20168 3 3 Cetuximab plus chemotherapy: ObservEr Study Quidde et al, 20169 3 3 3 3 Fluoropyrimidine plus bevacizumab: AIO KRK 0207 trial Hamidou et al, 201610 FOLFOX4 versus sequential dose-dense 3 FOLFOX7 followed by FOLFIRI: MIROX trial Grothey et al, 201318 and Yoshino et al, 201511 3 3 Regorafenib: CORRECT trial Qin et al, 201512 3 3 Regorafenib: CONCUR trial Thaler et al, 201213 3 3 Panitumumab plus FOLFIRI Au et al, 200914 3 3 3 3 Cetuximab: NCIC CTG and AGITG CO.17 trial Tol et al, 200915 3 Cetuximab plus chemotherapy and bevacizumab Sobrero et al, 200816 3 3 Cetuximab plus irinotecan: EPIC trial Rosati et al, 200517 3 Oxaliplatin plus tegafur/uracil and oral folinic acid Abbreviations: CRC ¼ colorectal cancer; EORTC QLQ-C30 ¼ European Organization for Research and Treatment of Cancer Quality of Life Questionnaire; TTD ¼ time to deterioration; TUDD ¼ time until definitive deterioration. Heinz-Josef Lenz et al Supplemental Table 2 Predefined HRQoL Measures of Interest in LUME-Colon 1 Scale Item/Question GHS/QoLa 1 How would you rate your overall health during the past week? 2 How would you rate your overall QoL during the past week? Physical functioningb 1 Do you have any trouble doing strenuous activities, like carrying a heavy shopping bag or a suitcase? 2 Do you have any trouble taking a long walk? 3 Do you have any trouble taking a short walk outside of the house? 4 Do you need to stay in bed or on a chair during the day? 5 Do you need help with eating, dressing, washing yourself, or using the toilet? Abbreviations: GHS ¼ global health status; HRQoL ¼ health-related quality of life; QoL ¼ quality of life. a For both questions, the patient can circle a number between 1 (very poor) and 7 (excellent). b There are 4 response categories: 1 ¼ not at all, 2 ¼ a little, 3 ¼ quite a bit, and 4 ¼ very much. Questionnaires were completed at timepoints specified in the protocol: at screening, day 1 of each 21-day cycle, at the end of treatment visit, at the end of the residual period, and at follow-up for disease progression. Supplemental Table 3 MMRM Analysis Difference From Placebo Endpoint Adjusted Mean (SE) 95% CI GHS/QoL (Q29, Q30) Week 3 2.5 (1.3) 0.0 to 4.9 Week 6 2.5 (1.4) 0.3 to 5.3 Week 9 0.2 (1.9) 3.4 to 3.9 Week 12 0.3 (2.2) 4.0 to 4.6 Physical functioning (Q1-Q5) Week 3 2.6 (1.0) 0.7 to 4.5 Week 6 3.0 (1.4) 0.4 to 5.7 Week 9 5.9 (1.7) 2.6 to 9.3 Week 12 4.5 (2.2) 0.2 to 8.8 Role functioning (Q6-Q7) Week 3 1.0 (1.6) 2.1 to 4.2 Week 6 1.6 (1.9) 2.0 to 5.3 Week 9 1.3 (2.3) 3.2 to 5.8 Week 12 1.2 (2.9) 4.4 to 6.8 Emotional functioning (Q21-Q24) Week 3 0.5 (1.2) 1.8 to 2.8 Week 6 0.6 (1.4) 2.1 to 3.3 Week 9 4.2 (1.6) 1.1 to 7.4 Week 12 3.8 (2.0) 0.0 to 7.7 Cognitive functioning (Q20, Q25) Week 3 1.6 (1.1) 0.7 to 3.8 Week 6 1.0 (1.3) 1.6 to 3.6 Week 9 5.7 (1.6) 2.7 to 8.8 Week 12 2.1 (2.0) 1.9 to 6.1 Social functioning (Q26, Q27) Week 3 2.9 (1.5) 0.0 to 5.8 Week 6 3.8 (1.7) 0.5 to 7.2 Week 9 4.1 (2.2) 0.3 to 8.4 Week 12 1.3 (2.6) 3.8 to 6.4 Results are from an MMRM model. Fixed effects include treatment, stratification factors, visit, treatment-by-visit interaction, baseline, and baseline-by-visit interaction. Number of patients contributing to the MMRM model in each treatment group: placebo, n ¼ 356; nintedanib, n ¼ 360. Abbreviations: CI ¼ confidence interval; GHS ¼ global health status; MMRM ¼ mixed-effects model for repeated measurements; Q ¼ question; QoL ¼ quality of life; SE ¼ standard error. Clinical Colorectal Cancer December 2019 - 279.e4 LUME-Colon 1 QoL Supplemental Table 4 Responder Analysis Number of Responders (%) Nintedanib vs. Placebo Endpoint Placebo (N [ 356) a Nintedanib (Na [ 360) Odds Ratio (95% CI) GHS/QoL GHS/QoL (Q29, Q30) 35 (9.8) 50 (13.9) 1.46 (0.92-2.31) Functional scales Physical functioning (Q1-Q5) 19 (5.3) 25 (6.9) 1.32 (0.71-2.45) Role functioning (Q6, Q7) 39 (11.0) 40 (11.1) 1.01 (0.63-1.61) Emotional functioning (Q21-Q24) 45 (12.6) 46 (12.8) 1.00 (0.64-1.55) Cognitive functioning (Q20, Q25) 31 (8.7) 46 (12.8) 1.54 (0.95-2.49) Social functioning (Q26, Q27) 49 (13.8) 62 (17.2) 1.29 (0.86-1.94) Results are from fitting a generalized linear regression model adjusted for treatment and stratifications factors used at randomization and a logit link (odds ratio). An odds ratio > 1 indicates a benefit to nintedanib. Abbreviations: CI ¼ confidence interval; GHS ¼ global health status; Q ¼ question; QoL ¼ quality of life. a Number of patients with baseline and at least one post-baseline assessment. 279.e5 - Clinical Colorectal Cancer December 2019