Efficacy of Reduced-Intensity Chemotherapy With Oxaliplatin and Capecitabine on Quality of Life and Cancer Control Among Older and Frail Patients With Advanced Gastroesophageal Cancer

Key Points Question Do older and/or frail patients with advanced gastroesophageal cancer benefit from less intensive palliative chemotherapy, and can a formal geriatric assessment assist treatment decision-making? Findings This phase 3 randomized clinical trial including 559 patients with advanced gastroesophageal cancer found that reducing the intensity of chemotherapy provided an improved patient experience with no significant detriment in cancer control. Baseline frailty, quality of life, and neutrophil/lymphocyte ratio (an inflammation marker) were predictive of outcome and may contribute to treatment decisions. Meaning Decision-making for older and/or frail patients with advanced cancer can be enhanced using geriatric assessment; such patients generally benefit from reducing the intensity of chemotherapy.

C ancer is most common in older people. In North America and Europe, gastroesophageal cancer is the third most common cause of cancer death, with more than half these deaths in people over 75 years, 1 many of whom are frail and with comorbidities. But evidence guiding treatment for vulnerable older patients is poor: standard chemotherapy regimens were developed in trials involving predominantly nonfrail, noncomorbid patients of median age less than 65, [2][3][4][5] and although selected older people participated they cannot be assumed to fully represent the older population.
In 2011, MRC FOCUS2, 6 a national randomized trial designed for frail and older patients with colorectal cancer, was reported. It used reduced doses of chemotherapy and introduced a novel composite end point, Overall Treatment Utility (OTU), combining clinical efficacy, tolerability, and the patient's own assessment of treatment value and acceptability. In the same year we surveyed 50 gastrointestinal oncologists in the UK 7 : 49 reported routinely treating older patients with gastroesophageal cancer with standard schedules at reduced doses, or omitting agents. There was wide variation in practice and no use of objective geriatric assessment (GA) to guide decisions. This led to 321GO, a randomized feasibility trial in which older and/or frail patients with gastroesophageal cancer received 80% doses of the standard 3-drug schedule epirubicin/oxaliplatin/capecitabine (EOCap), 2 the same treatment omitting epirubicin (OCap), or the same treatment omitting both epirubicin and oxaliplatin (Cap). The best balance of benefits and tolerability was achieved with the OCap doublet. 7 This trial allowed further development of the OTU end point. 7,8 The GO2 randomized clinical trial takes the OCap schedule from 321GO 7 and compares 3 dose levels, seeking the best balance of efficacy and patient experience. For patients with uncertainty regarding whether to use chemotherapy at all, an alternative randomization compares the lowest dose level vs supportive care alone. We ask whether a baseline GA may aid personalized dose selection and perhaps identify patients unlikely to benefit from chemotherapy. More broadly, GO2 aims to stimulate researchers across all cancer types to evaluate patient-centered assessment, dosing, and outcome measurement for vulnerable patients.

Study Design and Participants
The GO2 trial (ISRCTN44687907) was an academic, multicenter, open-label randomized trial, approved by the UK National Research Ethics Service, overseen by independent Trial Steering and Data Monitoring & Ethics Committees. All participants gave fully informed written consent. The study is closed and follow-up is complete. The authors assume responsibility for accuracy, completeness, and fidelity to the trial protocol (Supplement 1) and statistical analysis plan (Supplement 2).
Eligibility criteria are detailed in the Trial Protocol. P at i e nt s h a d l o c a l l y a d v a n c e d a n d /o r m e t a s t at i c gastroesophageal cancer that was not pretreated. In the absence of established objective frailty thresholds, and given the complex interrelations of frailty and advanced age, we used oncologists' clinical judgment in selecting patients. The key eligibility criterion was that the oncologist considered full-dose standard combination chemotherapy (at that time epirubicin/oxaliplatin/capecitabine 2 or cisplatin/fluorouracil/trastuzumab 9 ) unsuitable because of the patient's advanced age and/or frailty. It was made clear to patients and clinicians that GO2 was a trial for older patients, but there were no chronological age limits. Furthermore, since scoring of performance status (PS) by oncologists in older patients is inconsistent, 10 a fixed PS threshold was not specified, but patients had to be considered fit for any of the treatments in their selected randomization. Patients with moderate renal/hepatic dysfunction could be entered with dose adjustment to compensate for reduced clearance. Medical comorbidity was allowed provided the oncologist did not consider this to preclude chemotherapy. Response Evaluation Criteria in Solid Tumorsassessable disease was not mandatory. 11

Randomization
If the clinician and/or patient considered chemotherapy definitely indic ated, patients entered the CHEMO-INTENSITY randomization and were allocated (1:1:1) to OCap dose Level A, B, or C. Level A treatment was oxaliplatin 130 mg/m 2 on day 1 and capecitabine 625 mg/m 2 twice daily on days 1-21, on a 21-day cycle; Level B treatment was 80% of Level A doses; and Level C treatment was 60% of Level A doses. If the patient and clinician agreed that best supportive care alone (BSC) would be an acceptable alternative, patients could instead enter the CHEMO-BSC randomization, with allocation (1:1) to OCap Level C or BSC. Randomization used an automated telephone/web system and validated minimization algorithm, with age, PS, metastases, histology, renal function, planned trastuzumab use, and center as stratification factors. The treatment allocation was not masked from study investigators or patients.

Key Points
Question Do older and/or frail patients with advanced gastroesophageal cancer benefit from less intensive palliative chemotherapy, and can a formal geriatric assessment assist treatment decision-making? Findings This phase 3 randomized clinical trial including 559 patients with advanced gastroesophageal cancer found that reducing the intensity of chemotherapy provided an improved patient experience with no significant detriment in cancer control. Baseline frailty, quality of life, and neutrophil/lymphocyte ratio (an inflammation marker) were predictive of outcome and may contribute to treatment decisions.
Meaning Decision-making for older and/or frail patients with advanced cancer can be enhanced using geriatric assessment; such patients generally benefit from reducing the intensity of chemotherapy.

Procedures
The GA, aligned with the European Organization for Research and Treatment of Cancer (EORTC) Elderly Minimum Data set, 12 was administered after consent but before randomization; results were not communicated to the clinician. It included G8, 13 Instrumental Activities of Daily Living (IADL), 14 Timed Up and Go test, 15 EORTC QLQ-C30/OG25, 16,17 and EQ-5D and visual analogue scale (EQ-VAS). 18 Frailty was scored by assessing impairment (yes/no) in nine domains (weight loss, mobility, falls, neuropsychiatric, physical functioning, social functioning, mood, fatigue, and polypharmacy) and participants were categorized as not frail (0-1/9 domain impaired), mildly frail (2/9 domains), or severely frail (≥3/9 domains). 19 Patients with estimated glomerular filtration rate (eGFR) of 30 to 50 mL/min or bilirubin 1.5 to 2 times the upper limit of normal received 75% of their allocated dose of capecitabine. Patients with ERBB2 (formerly HER2)-positive cancers could additionally receive trastuzumab. Imaging was repeated every 9 weeks, and chemotherapy stopped in the event of radiological or clinical progression, unacceptable toxic effects, or patient choice. Patients allocated to BSC had access to specialist palliative care, pain and psychosocial services, blood transfusions, nutritional support, radiotherapy, stenting, or surgical procedures as indicated; chemotherapy, although not planned, was allowable if it later became indicated.
Overall Treatment Utility was scored once, 9 weeks after starting chemotherapy. It comprised computed tomography (CT) and clinical assessment of cancer progression status; toxic effects (Common Terminology Criteria for Adverse Events, CT-CAE) and serious adverse events (SAEs); quality of life (QL, as scored with QLQ-C30 Global Health Status subscale), and patient value/acceptability, scored from 2 questions posed in a questionnaire before patients received their scan results: "Since you started chemotherapy, how worthwhile do you think your treatment has been?" and "How much has your chemotherapy interfered with your normal daily activities?" both scored "not at all/a little/quite a bit/very much." Overall Treatment Utility was not measurable in patients allocated BSC.
Treatment beyond 9 weeks continued until CT progression or clinical/patient decision. Longitudinal QL comprised weekly EQ-VAS and every 3 weeks EQ-5D and QLQ-C30 Fatigue Subscale during chemotherapy, then then once every 9 weeks until a year from randomization.

Outcomes
End points, conforming to the joint EORTC/Alliance/ International Society of Geriatric Oncology (SIOG) Statement, 20 are defined in the trial protocol (Supplement 1) and statistical analysis plan (Supplement 2). In the CHEMO-INTENSITY randomization the primary end point was investigatordetermined progression-free survival (PFS). 21 The key secondary end point was OTU. A score of "Good OTU" requires no radiological or clinical evidence of cancer progression, no major toxic effects (a serious adverse reaction [SAR], or any grade ≥3 non-hematological toxicity), no significant deterioration in QL (≥16 percentage-points drop in EORTC Global QL subscale 22 ) and no adverse responses to patient value/acceptability ques-tions ("not at all" worthwhile or "very much" interference). Poor OTU denotes evidence of cancer progression and at least 1 other negative factor (toxic effects, SAE, QL deterioration, or poor value/acceptability), or the patient has died. Intermediate OTU means either cancer progression without any other negative factor or negative factors without cancer progression. Other secondary end points were toxic effects; symptoms (QLQ-C30/OG250); QL; RECIST response 11 ; overall survival (OS); and quality-adjusted survival. Fatigue was scored using QLQ-C30 with time-to-deterioration from randomization to a deterioration of 16 percentage points or more. In the CHEMO-BSC randomization, the primary end point was OS; secondary end points were patient-reported fatigue and QL.

Statistical Analysis
In the CHEMO-INTENSITY randomization, reducing the dose of chemotherapy was hypothesized to provide a better patient experience without major detriment in PFS. The trial therefore used a PFS noninferiority design but with a relatively nonstringent boundary, set following careful discussion at a forum of patients and clinicians, where acceptable absolute PFS/OS losses were considered as a trade-off against toxicity. Patients were prepared to sacrifice 6 weeks or more of PFS in return for reduced treatment toxic effects, but clinicians were more conservative and the trial was eventually powered to exclude 34 days or greater reduction in median PFS from a predicted 134 days, equivalent to hazard ratio (HR) at or over 1.34. With 1-sided 5% significance and 80% power, this required 284 events or more per pairwise comparison, requiring recruitment of 501 patients or more. In the CHEMO-BSC randomization, chemotherapy was hypothesized to improve OS; however, given that the uptake of this randomization was not predicable at the time of designing the trial, the sample size could not be predetermined and only exploratory analysis was planned.
Efficacy analyses were by intention to treat (ITT); safety and toxic effect analyses in patients who received 1 or more dose of protocol therapy. Kaplan-Meier methods were used to estimate survivor functions for time-to-event end points. 23 Cox proportional hazards regression adjusted for minimization factors were used to estimate HRs and CIs, using 1-sided 95% CIs for the noninferiority comparison. 24 Proportional hazards were assessed using a permutation test of martingale residuals. No violations were observed. Overall Treatment Utility comparisons used ordinal logistic regression adjusting for minimization factors to estimate odds ratios (ORs) and 95% CIs. 25 For QL subscales, we found no evidence against the missing-at-random assumption using descriptive and logistic regression analysis, so we applied multiple imputation by chained equations (MICE). 26 We compared allocated groups for QL and symptom subscales using multiple linear regression adjusted for the baseline subscale and minimization factors (excluding center). For fatigue we compared groups using multilevel repeated mixed-model analyses allowing for time effects, treatment-time interactions, baseline fatigue (fixed effects), and patient and patient-time interaction (random effects). These models were also used to estimate treatment effects and 95% CI. We performed sensitivity analyses for complete cases. Subgroup analysis used the same model methods as primary and secondary end point comparisons, prespecified for potentially prognostic variables: patient characteristics/ minimization factors, frailty, QL/symptoms, and laboratory tests. Tests for heterogeneity correspond to 1 degree of freedom for 2-category subgroups (or continuous scales), 2 degrees of freedom for 3-category subgroups, and so on. Following univariate analyses with ordinal logistic regression, we performed multivariable analysis using backward elimination including all variables, irrespective of univariate result. Finally, we constructed a nomogram using a transformation of the linear predictor to a scale estimating the probability of each OTU outcome.
P values for superiority comparisons are 2-sided and considered significant at an overall significance level of 5%. All other analyses are described in the statistical analysis plan (Supplement 2). For analysis, SAS version 9.4 (SAS Institute) and R version 3.2.3 (R Project) were used.

Results
A total of 514 patients entered the CHEMO-INTENSITY and 45 entered the CHEMO-BSC randomization, between Janu-ary 2014 and November 2017, at 61 UK medical centers ( Figure 1; eTable 1 in Supplement 3); cutoff date was February 2019. Slow recruitment in some centers was attributed to patients opting for the lower dose levels off-trial. Populations were well-balanced within each randomization but differed between the two, with CHEMO-BSC patients having higher rates of poor PS and severe frailty, driven especially by impaired ADL, low mood and social care requirements ( Table 1; eTable 2 in Supplement 3). Of the total 559 patients, 44 (8%) were neither frail nor aged over 75 years, reflecting the flexibility of the selection criteria (eTable 3 in Supplement 3).

CHEMO-INTENSITY Randomization
In the ITT analysis, both lower doses satisfied the prespecified trial definition of noninferior PFS compared with Level A; neither CI crosses the HR boundary of 1.34 ( Figure 2).  . Level C produced more good (43%) and fewer poor (29%) OTU outcomes than A or B, but these differences were not statistically significant (eFigure 1 in Supplement 3). Other patient experience end points also trended toward better outcomes w ith lower doses (eTables 4-9 in Supplement 3). Time-to-deterioration of fatigue favored Level C (C vs A HR = 0.88 [95% CI, 0.65-1.19]). Global QL (EORTC Core Quality of Life questionnaire, QLQ-C30, and EuroQoL-5 Dimension, EQ-5D) improved between baseline and 9 weeks with Levels B and C, but not A (eFigures 4-7 in Supplement 3). Longitudinal fatigue and QL showed no major differences. Cancer symptoms improved between baseline and 9 weeks in all arms similarly (eFigures 8-12 in Supplement 3).
The rate of toxic effects fell markedly with reducing dose levels ( Table 2). Consequently, treatment delivery was more reliable: during the first three 3-week cycles, comparing Levels A, B, and C, a dose reduction was required in 63 (39%), 39 (24%), and 21 (13%) patients, respectively; 53

CHEMO-BSC Randomization
A total of 45 patients entered the CHEMO-BSC randomization. In those allocated chemotherapy, toxicity was higher than in patients allocated the same dose level in the CHEMO-
In the CHEMO-INTENSITY randomization (n = 514), interaction was seen between the multivariable predictive score and dose level (P = .01) with greater incremental benefit of lower-dose treatment in patients with better baseline scores: thus a patient with score = 1 allocated to Level C has 68% prob-ability of good, 20% intermediate, and 12% poor OTU, but if the same patient is allocated Level A these probabilities are 41%, 30%, and 29%. No baseline score was identified as predicting better OTU with higher-dose treatment (eTables 12-14 in Supplement 3).

Discussion
The GO2 randomized clinical trial is the first large trial testing the relationship between treatment intensity and patientfocused outcomes in frail and/or older patients with cancer. Previous reports have studied older patients who were fit enough to enter all-comer trials, 27 or relied on traditional efficacy and safety end points. 28 The GO2 trial uses modern methods and studies patients-the older frail, older nonfrail, and younger frail-who rarely participate in trials. In response to calls to address the deficit in evidence guiding treatment of vulnerable cancer patients, 29 we offer GO2 as an exemplar of real-world, patient-centered research.
Lower-dose chemotherapy improved patients' experience without compromising anticancer control. This balance is captured in OTU, an objective measure of a virtual conversation between physician and patient and reflecting their joint assessment of treatment value: "With the benefit of hindsight, am I glad I recommended this treatment?" and "Am I glad that I accepted it?" The GO2 trial also demonstrates that a baseline GA can contribute to the doctor-patient decision by estimating an individual's probability of better or worse OTU.
In designing GO2, decisions were necessary for patient selection, treatment, and statistical design, all of which may be debated. For example, previous trials could not characterize patients who were not included, but it was precisely those patients who were to be selected for GO2. It was therefore nec- essary to use clinicians' experience, rather than an objective tool, to offer trial entry to patients they assessed as unsuited to full-dose combination chemotherapy but able to receive reduced-intensity treatment.
Even the highest dose in GO2, Level A, was less-thanstandard treatment, comprising just 2 drugs from the standard EOCap triplet. 2 Although this includes full-dose oxaliplatin, it is combined with low-dose continuous capecitabine rather than the intensive intermittent schedule typically used in doublet therapy. Level C therefore represents just 60% of 2 out of 3 drugs, around 40% of full standard dose intensity. It is also important to recognize that Level A, although a reference schedule for this trial, is not standard therapy; indeed, the stimulus for the trial was a survey showing that there is no standard for this population. 7 For this reason we did not apply the typical stringent noninferiority boundary demanded by regulators, but were instead able to work with patients and clinicians to carefully balance the competing needs for cancer control and good tolerability.
One mechanism for retaining cancer control despite lower doses is avoidance of toxicity-induced treatment reductions and stoppages. Toxic effects leading to treatment modification may be accepted by oncologists as part of standard oncology practice, but it represents a negative experience for patients and detracts from both quality of life and cancer control; and these impacts are particularly heightened in patients with poor baseline reserve. Only 32% patients starting Level A were able to receive 3 cycles without reduction or stoppage, compared with 58% with Level C.
The GO2 trial aimed to develop dose individualization guided by baseline geriatric assessment: we anticipated fitter patients would benefit from higher-dose treatment; however, we did not identify any group for whom the higher doses are preferable. Using the OTU outcome measure, reflecting the balance of benefits and harms, goes beyond conventional single-outcome models looking at survival or toxicity in isolation. 30 In so doing, GO2 challenges a pervading assumption of oncology: that within the bounds of tolerability more is better. We hope it will stimulate research exploring lower-dose chemotherapy, perhaps extending to younger and less frail patients. We hope also that those designing trials of novel agents, including registration studies, will consider the option of lower-dose chemotherapy as the reference or platform to which novel agents are added, to widen access to these trials.
The 3-month survival benefit seen in the CHEMO-BSC randomization, though nonsignificant in isolation as a consequence of small numbers and an imbalance in patient characteristics, concurs with previous data 3 1 and supports consideration of low-dose chemotherapy in vulnerable patients. This should, however, be interpreted alongside the baseline predictor, which helps identify patients at high risk of poor treatment utility, for whom BSC may be a preferable path.

Limitations
A limitation of GO2 is that our GA was purely observational. Implementation of these findings-and future researchshould embrace the newer concept of Comprehensive Geriatric Assessment (CGA): both identification of vulnerabilities and active remedial management to correct them. An outstanding research question is whether CGA-based prehabilitation will convert a patient from low to high probability of achieving good OTU.

Conclusion
The GO2 clinical trial shows that the goals of palliative chemotherapy in the older and/or frail population, including but not limited to cancer control, may be better achieved using treatment at doses well below those currently regarded as standard. Careful baseline geriatric health assessment in the oncology clinic can help predict the likelihood of achieving those goals, and so contribute to patients' and clinicians' treatment decisions. Assessing the outcome of cancer treatment should be multidimensional, including its value to patients and its adverse effects, and we recommend further development of OTU to capture this complexity. The GO2 trial offers a design paradigm for enhancing older patients' access to research and ensuring that our evidence base embraces the whole population that we serve.