Gastric Cancer (2021) 24:445–456 https://doi.org/10.1007/s10120-020-01116-x ORIGINAL ARTICLE External validity of docetaxel triplet trials in advanced gastric cancer: are there patients who still benefit? Paula Jimenez‑Fonseca1 · Alberto Carmona‑Bayonas2,3 · Eva Martínez de Castro4 · Ana Custodio5 · Carles Pericay Pijaume6 · Raquel Hernandez7 · Gema Aguado8 · Natalia Castro Unanua9 · Juana María Cano10 · Flora López11 · Marcelo Garrido12 · Ana Fernández Montes13 · Laura Visa14 · Manuel Sánchez Cánovas15 · María Luisa Limón16 · Nieves Martínez Lago17 · Paola Pimentel18 · Alicia Hurtado19 · Aitor Azkárate20 · Federico Longo21 · Marc Diez22 · Aranzazu Arias‑Martinez23 · Tamara Sauri24 · Alfonso Martín Carnicero25 · Monserrat Mangas26 · Marta Martín Richard27 · Mónica Granja28 · Avinash Ramchandani29 · Carolina Hernández Pérez30 · Paula Cerdá31 · Aitziber Gil‑Negrete32 · Mariona Calvo33 · Rosario Vidal Tocino34 · Javier Gallego35 Received: 6 June 2020 / Accepted: 23 August 2020 / Published online: 24 September 2020 © The Author(s) 2020 Abstract Background The purpose of our study was to develop an online calculator to estimate the effect of docetaxel triplets (DPF) in first line of advanced gastric cancer (AGC), and to assess the external validity of docetaxel trials in individual patients. Methods The study includes patients with HER2(-) AGC treated with platin and fluoropyrimidine (PF) or with DPF in first line. Treatment effect and interactions were assessed using Bayesian accelerated failure time models. Result The series comprises 1376 patients; 238 treated with DPF and 1138 with PF between 2008 and 2019. DPF was associ- ated with increased progression-free survival (PFS) and overall survival (OS) with time ratio (TR) 1.27 (95% credible interval [CrI], 1.15–1.40), and TR 1.19 (95% CrI, 1.09–1.27), respectively. Serious adverse events were more common with DPF, particularly hematological effects (32% vs 22%). Younger participants received greater DPF dose density without achieving greater disease control, while severe toxicity was likewise higher. DPF yielded superior OS in Lauren intestinal (TR 1.27, 95% CrI, 1.08–1.11) vs diffuse subtype (TR 1.17, 95% CrI, 1.09–1.24) and the probability of increasing OS > 15% was 90% vs 67% in each subtype, respectively. The effect dwindles over time, which can be attributed to pathological changes and clinical practice changes. Conclusion Our study confirms the effect of DPF is highly dependent on several clinical–pathological variables, with discreet and gradually declining benefit over platinum doublets in later years, at the expense of increased toxicity. These results may help to underpin the idea that external validity of AGC trials should be revised regularly. Keywords Bayesian model · Chemotherapy · Gastric cancer · Docetaxel · Survival Introduction It is a complex concept whose determinants are based more on clinical rather than statistical expertise [1] and on factors External validity or the inferential generalizability of rand- such as selection criteria, pathological traits, or access to omized clinical trials (RCTs) refers to the appropriateness of new therapies [2]. Therefore, in light of the ever-changing extrapolating trial outcomes from one population to another. condition of clinical practice and target populations, this inferential generalizability of past RCTs should be revis- Electronic supplementary material The online version of this ited every so often [3–5]. The apparent dwindling effect of article (https​://doi.org/10.1007/s1012​0-020-01116​-x) contains docetaxel-containing triple-agent regimens (DPF) in the supplementary material, which is available to authorized users. first-line treatment of advanced gastric cancer (AGC) may be illustrative of this situation. * Paula Jimenez‑Fonseca palucaji@hotmail.com AGC continues to be a common neoplasm with high mor- tality, for which palliative chemotherapy is recommended [6]. Extended author information available on the last page of the article 13 Vol.:(0123456789) 446 P. Jimenez‑Fonseca et al. In 2006, the international TAX325 phase III RCT demon- that recruit consecutive cases of locally advanced, unresect- strated the benefit of docetaxel combined with platin-fluoro- able or metastatic adenocarcinoma of the stomach, gastroe- pyrimidine (DPF) in first line [7]. Nonetheless, the effect on sophageal junction, or distal esophagus [12, 16–24]. overall survival (OS) of DPF over platin-fluoropyrimidine was Eligibility cr iter ia for this analysis include modest (median 9.2 vs 8.6 months, hazard ratio (HR) 0.77, being > 18 years of age, cancer that does not overexpress P value = 0.02) and was attained at the cost of substantially HER2, and first-line treatment with at least one cycle of increased severe toxicity (69% vs 59%), respectively. This platin and 5FU (PF) with or without docetaxel [14]. Patients discreet benefit profile has diminished in successive RCTs [8, who had completed a prior neoadjuvant or adjuvant treat- 9]. In this regard, the 2017 update of Wagner’s meta-analysis, ment in the previous 6 months and those who had received with eight comparative studies, lowered the prospect of benefit taxanes as part of perioperative schedules were excluded. (HR 0.86, 95% confidence interval (CI) 0.78–0.95) [9]. More The data are managed through a website (http://www. recently, adding docetaxel to cisplatin and S-1, appraised in agame​nonst​udy.com/) that consists of filters and a system of the Japanese JCOG1013 phase III RCT, did not enhance OS queries to guarantee data reliability and control for missing (HR 0.99, 95% CI, 0.85–1.16), although it did increase adverse and inconsistent data, with telephone and online monitor- events (e.g., grade 3–4 neutropenia, 59% vs 32%) [8]. ing (PJF). This series of results revealing progressively smaller mag- The study was approved by a multicenter Research Eth- nitude poses the clinician with a twofold question. The first is ics Committee of all the Autonomous Communities and whether the effect of docetaxel varies on the basis of time fac- participating hospitals and was classified as a prospective, tors, geographical or epidemiological variables. In the interim postmarketing surveillance study by the Spanish Agency of between these studies, changes in clinical practice have been Medicines and Medical Devices (AEMPS), and was not oth- reported that may have modified the effect. Chief among these erwise involved in it. All participants still alive at the time of changes is the introduction of trastuzumab in tumors that data collection provided written, signed, informed consent. amplify or overexpress human epidermal growth factor recep- tor-2 (HER2) in 2010 [10]. Given that these neoplasms are no longer treated with docetaxel-based triplet, candidates for Variables DPF include ever since an abundance of diffuse histological subtype—precisely the ones that are most chemo-refractory The main endpoint, OS, was defined as the time between [10–12]. In addition to this, in 2014, the RAINBOW RCT treatment initiation and demise from any cause, censoring confirmed the benefit of paclitaxel and ramucirumab in second patients lost to follow-up. Progression-free survival (PFS) line, which could dilute the advantage of the docetaxel-based was defined as the interval between start of treatment and triplet in first line [13]. Thus, clinical guidelines now tend to progression, as per the RECIST 1.1 criteria, death, or last recommend the dual-agent platin and fluoropyrimidine regi- follow-up. Relative dose intensity (RDI) was expressed in men for HER2-negative AGC, with DPF as an option in fit percentages and defined as the dose intensity (the amount of individuals who require tumor response [14, 15]. drug per unit of time, expressed as mg/m2 weekly) adminis- The second question is whether there are still any patients tered with respect to the planned amount for each schedule. who could benefit from the DPF strategy in first line. This To limit confounding bias, nine clinical, pathological, and is germane, given the paucity of data regarding effect-mod- laboratory confounding factors were selected, on the basis ifying factors based on key variables such as histopatho- of theoretical criteria and the group’s experience from pre- logic subtype or age [6]. Our study seeks to shed light on vious studies [25]. These covariates were performance sta- both questions. We have, therefore, attempted to reproduce tus (ECOG-PS), Lauren histological classification subtype, the trends observed in RCTs in a national registry of AGC. histological grade, ascites, stage (III unresectable vs IV), Moreover, we have constructed an online calculator that liver tumor burden, type of platin (cisplatin vs oxaliplatin), depicts how individual characteristics or clinical–pathologi- neutrophil/lymphocyte ratio, age, and year of treatment [26]. cal variables have modified the effect that the addition of To limit the confounding bias, two measures of tumor docetaxel has had on survival endpoints. load have been contemplated: the number of organs involved and liver tumor burden. Patients’ baseline computerized tomographies were re-evaluated by the clinicians and liver Method tumor burden was categorized as: 0 metastasis, < 25%, 25–50%, 51–75%, and > 75% of liver volume affected by Patients and study design tumor tissue. The number of organs involved was defined as the number of organs (e.g., liver, lung, skeleton, lymph The participants are from the AGAMENON registry in nodes, unresected gastric primary, etc.) affected by the dis- which 34 Spanish and one Chilean center participate and ease, regardless of the number of metastases contained in 13 External validity of docetaxel triplet trials in advanced gastric cancer: are there patients… 447 each organ. Lymph metastases in distant basins were con- percentage of diffuse tumors, signet ring cells, ascites, and sidered different organs. greater tumor burden). At the time of analysis, with a median follow-up of Statistics 35.3 months, 1208 progression events and 1128 deaths had been reported. Median PFS and OS for the entire sample Therapeutic effect was evaluated via a Bayesian parametric were 5.8 (95% CI, 5.5–6.1) and 10.2 (95% CI, 9.6–10.8) accelerated failure time (AFT) model with lognormal distri- months, respectively. bution; this model assumes that the effect of the covariates is to accelerate or decelerate the course of illness, making Evaluation of the main effect, dose intensity, them suitable when the assumption of proportional haz- and safety profile ards is not met [23]. Its coefficients have an intuitive, direct interpretation in its exponentiated form, as time ratios (TR). Median PFS was 6.5 (95% CI, 5.6–7.3) vs 5.7 months Thus, a time ratio equal to 2 for a binary predictor means (5.4–6.12) and median OS was 11 (95% CI, 9.8–10.4) vs that the median of time-to-event is doubled in the presence 10.1 months (9.4–10.7) for DPF vs PF, respectively. In the of this variable. With this model, the interaction between multivariable model, DPF was associated with increased DPF, age, histopathologic subtype, and year of treatment PFS with TR 1.27 (95% credible interval [CrI], 1.15–1.40) was examined. This approach also enables historical exter- and also improved OS with TR 1.19 (95% CrI, 1.09–1.27). nal data to be incorporated as priors [27–30]. The prior for In a sensitivity analysis, the results of the Bayesian model the therapeutic effect was based on Wagner’s meta-anal- are robust, regardless of geographical location. ysis, for OS ~ N(0.15, 0.045) and for PFS ~ N(0.27, 0.07) The analysis of administered doses indicates that incor- [9]. The interactions were appraised skeptically, e.g., ~ N(0, porating docetaxel was at the expense of a discreet decrease 0.05) for the histological subtype, to discourage subgroup in RDI for platins and fluoropyrimidines, although the mean effects deemed extreme [7]. An online calculator was built dose intensity remained above 80% in most cases (Annex to obtain the model’s predictions. The trace and density plots Table 2). This occurred more intensely in the elderly. Thus, for Markov chain Monte Carlo (MCMC) samples denoted in DCX/DCF (docetaxel, cisplatin, capecitabine/5FU) adequate convergence (Gelman–Rubin Rhat measure < 1.1 schemes, cisplatin RDI was 81% vs 72%, and docetaxel RDI for all the parameters). A frequentist AFT model for OS is was 81% vs 72% in individuals <75 vs> 75 years, respec- also shown in Annex Figure 1 as comparison. tively. The use of cisplatin-based regimens is less common Continuous variables were assessed by restricted cubic with age; the turning point occurs around the age of 65 splines. Covariates with > 25% missing data were discarded (Annex Figure 2). and multiple imputation was applied (fully conditional Insofar as safety is concerned, DPF increased serious specification, on 20 imputed datasets) in the rest. The prob- adverse events compared to PF, particularly hematologi- ability of dichotomous outcomes was appraised by logistic cal events (32% vs 22%), although it was also associated regression. All analyses were performed with the R v3.1.6 with more diarrhea, stomatitis, alopecia, or asthenia (Figs. 2 software package, with the mice, splines, and brms libraries & 3). The attempt to intensify therapy in younger subjects [31–34]. yielded the highest rates of severe toxicity with DPF across the board. Accordingly, the probability of grade 3/4 toxicity for DPF was 40%, 37%, and 34%, for people aged 40, 60, Results and 80 years. By contrast, for PF, it was 31%, 31%, and 35%, respectively. There were fewer PF dose reductions, with Patients 83% vs 79% oxaliplatin RDI in participants < vs ≥ 75 years, respectively. The series comprises 1376 patients; 238 treated with DPF and 1138 with PF between 2008 and 2019. Annex Table 1 Conditional effects based on individual traits details the chemotherapy schedules used. Baseline character- istics are displayed in Table 1. The most salient differences We then fitted an AFT model to estimate the probability of include the fact that patients treated with PF (vs DPF) tend the effect of DPF depending on individual traits. An online to be older (median 66 vs 58 years), with worse functional calculator has been designed to obtain these estimations status (ECOG-PS ≥ 2, 17% vs 7%), predominance of intes- gradually (see the online calculator: https​://www.progn​ tinal subtype (34% vs 26%), and less of a propensity toward ostic​tools​.es/Agame​nonTr​iplet​/inici​o.aspx). Annex Figure 3 peritoneal disease (45% vs 53%). Figure 1 illustrates how shows three examples of the use of this calculator under younger individuals are more likely to receive DPF and Wagner’s meta-analysis prior. have tumors with distinct clinical–pathological traits (higher 13 448 P. Jimenez‑Fonseca et al. Table 1  Baseline characteristics Baseline characteristics Total, n = 1376 PF, n = 1138 DPF, n = 238 of patients Age, median (range) 65 (20–89) 66 (20–89) 58 (22–88) Sex, female 450 (33%) 367 (32%) 83 (35%) ECOG-PS 0 282 (20%) 217 (19%) 65 (27%) 1 889 (65%) 732 (64%) 157 (66%) ≥2 205 (15%) 189 (17%) 16 (7%) Primary tumor site Esophagus 114 (9%) 102 (9%) 12 (5%) GEJ 158 (11%) 132 (12%) 26 (11%) Stomach 1104 (80%) 904 (79%) 200 (84%) Histological grade 1 120 (9%) 110 (10%) 10 (4%) 2 328 (25%) 337 (29%) 44 (18%) 3 563 (42%) 450 (40%) 113 (48%) Not available 312 (24%) 241 (21%) 71 (30%) Lauren classification Diffuse 641 (47%) 526 (46%) 115 (49%) Intestinal 444 (32%) 383 (34%) 61 (26%) Not available 291 (21%) 229 (20%) 62 (25%) Signet ring cells 433 (31%) 362 (32%) 71 (30%) Tumor stage at diagnosis, locally 59 (4%) 41 (4%) 18 (7%) advanced unresectable Metastases sites Ascites 344 (25%) 288 (25%) 56 (24%) Peritoneal 634 (46%) 508 (45%) 126 (53%) Bone 137 (10%) 112 (10%) 25 (11%) Lung 167 (12%) 158 (14%) 9 (4%) Liver 486 (35%) 430 (38%) 56 (24%) Burden of liver disease > 50% 238 (17%) 207 (18%) 31 (13%) Number of metastatic sites > 2 349 (25%) 291 (26%) 58 (24%) Platin Oxaliplatin 894 (65%) 818 (72%) 76 (32%) Cisplatin 482 (35%) 320 (28%) 162 (68%) Primary tumor resection 441 (32%) 360 (32%) 81 (34%) CEA, ng/ml <5 653 (47%) 521 (46%) 132 (55%) 5–10 131 (10%) 111 (10%) 20 (8%) 10–30 121 (9%) 102 (9%) 29 (12%) > 30 232 (17%) 197 (17%) 35 (16%) No available 239 (17%) 207 (18%) 22 (9%) Albumin, g/dl > 35 g/dl 921 (67%) 757 (67%) 164 (69%) 30–35 218 (16%) 187 (16%) 31 (13%) < 30 114 (8%) 99 (9%) 15 (6%) No available 123 (9%) 95 (8%) 28 (12%) DPF, docetaxel, platinum, fluoropyrimidine; ECOG-PS, Eastern Cooperative Group Performance Status; GEJ, gastroesophageal junction; PF, platinum, fluoropyrimidine The model suggests a subgroup effect based on histo- benefit from DPF in any of the subtypes cannot be ruled out pathological subtype. Thus, the posterior probability of ben- under this supposition (Fig. 4 and Table 2). efit differs depending on histology. Nevertheless, a discreet 13 External validity of docetaxel triplet trials in advanced gastric cancer: are there patients… 449 Fig. 1  Probability of clinical–pathological variables and treatment The estimations are derived from logistic regressions that evaluate pattern based on age. DPF, docetaxel, platinum, fluoropyrimidine; the non-linear effect of age (restricted cubic splines with 3 knots) on ECOG-PS, Eastern Cooperative Group Performance Status. NOTE: binary variables Insofar as age is concerned, middle-aged individu- the treatment of intestinal subtype tumors (Annex Figure 4). als (range 50–70 years) were those who benefitted from Since 2014, this registry is gradually collecting more treat- DPF (Fig. 5). Regardless of treatment, the elderly had ments with paclitaxel, with or without ramucirumab, as greater probabilities of tumor control with any treatment second-line therapy (23% vs 8% after and before 2014, at 3 months, but benefited less from DPF in terms of PFS. respectively, Annex Figure 5). Nor is PFS increased in the younger range of ages with doc- etaxel-containing triplets, given that they have more chemo- refractory tumors (Fig. 1). Discussion Frequentist and Bayesian models based on weakly informative priors are shown in Annex Figure 1 and Annex We have updated in a national AGC registry the external Table 3. validity of the international bibliography regarding first- line therapies containing docetaxel [7–9]. Overall, we have Analysis of timing on therapeutic effect observed that the use of triplets with docetaxel increased PFS by 27% and OS by 19% in patients with AGC. This The magnitude of effect of DPF on OS has fallen over improvement in survival endpoints is consistent with those the years. In 2013, TR was 1.21 (95% CrI, 1.09–1.33) as of the data from the TAX325 trial and Wagner’s meta-anal- opposed to 1.13 (95% CrI, 0.97–1.32) in 2019 (neutral ysis [7, 9], although it has come about at the expense of prior). PFS analyses revealed similar trends. Over time, increased severe toxicity. DPF is a less popular option in this series, particularly for 13 450 P. Jimenez‑Fonseca et al. Fig. 2  Most Frequent On − Therapy Adverse Events Sorted by Relative Risk. AE, adverse event; AST, Aspartate Aminotransferase; CI, confi- dence interval; DPF, docetaxel, platinum, fluoropyrimidine; n, number; PF, platinum, fluoropyrimidine The AGAMENON registry reproduces the temporal trend docetaxel-containing triple-agent regimens depending on seen in successive RCTs with DPF in first line [8, 9], mak- time variation and clinical–pathological factors. Due to the ing it a worthwhile instrument to explore the reasons for existence of heterogeneity of effects, the calculator aims this weakening of effect and ascertain those patient groups to estimate the most likely effect of DPF as a function of for whom DPF combinations are still effective. Our results covariates. The model we have adjusted here points toward appear to confirm a reduction of therapeutic effect associated the clinical profile with greater possibilities of gaining from with DPF from 2015 onward, and contributes explanatory DPF being fit subjects, in particular, aged 50–70, with intes- information regarding the fluctuation of those effects over tinal tumors, and intermediate tumor burden, as the online time [7, 8]. While our data indicate that the most plausible calculator illustrates. However, the data from our registry explanation is the introduction of taxanes in second line, show that younger individuals failed to attain better out- other aspects must be born in mind, such as the increased comes with DPF, as can be seen in the calculator, but did popularity of the oxaliplatin-containing doublet, which suffer a higher rate of serious adverse events. The reason is observed in the literature [35, 36]. Moreover, in recent can be found in the higher prevalence of aggressive, chemo- years, DPF is administered more in diffuse, treatment- refractory tumor phenotypes in this age range (e.g., diffuse resistant tumors, which may be due in part to the use of tumors, signet ring cells, peritoneal carcinomatosis, etc.). In PF and trastuzumab in HER2 + cancers (most of which are these cases, DPF did not seemingly help to prevent progres- intestinal subtype). In fact, the Japanese JCOG1013 RCT sion and the result in terms of survival endpoints was more recruited patients with HER2-negative AGC, which possibly modest than in subjects aged 50–70 years who benefitted increased the rate of diffuse tumors, whereas theTAX325 the most with this regimen [26]. Indeed, the various histo- RCT did not factor in HER2 status [7, 8]. pathological subtypes of gastric cancer respond differently We have elaborated an online calculator to offer an intui- to chemotherapy [11, 12]. Thus, the intestinal subtype, more tive depiction of the baseline prognosis and the effect of sensitive to DPF, is more frequent in elderly individuals, 13 External validity of docetaxel triplet trials in advanced gastric cancer: are there patients… 451 Fig. 3  Most Frequent On − Therapy grade 3/4 Adverse Events Sorted by Relative Risk. AE, adverse event; AST, Aspartate Aminotransferase; CI, confidence interval; DPF, docetaxel, platinum, fluoropyrimidine; n, number; PF, platinum, fluoropyrimidine paradoxically the group that tolerates this therapy worse have sought to modify the triplet to enhance its tolerability when administered at standard recommended doses. Con- [38–41]. A single-phase II randomized trial (NCT00737373) versely, diffuse tumors are typical of young patients [11, has compared a docetaxel-containing triplet (FLOT) with 26]. Interestingly, most AGC RCTs have not stratified for 5FU and oxaliplatin (FLO) in an elderly population. It histology, nor have they appraised interactions between detected significantly increased toxicity and worse quality of therapeutic effect and different clinical–pathological vari- life, without evidence of a gain in OS in this context [42]. On ables [7]. For this reason, the online calculator is useful for the whole, these data led to the assumption that the reduc- the clinician, and adds evidence to what we already know tion of benefit in older ages was related to tolerance [14, 43]. about these combinations. In our study, modifications in elderly patients were associ- In light of this information, the AGAMENON registry ated with diminished therapeutic effect, similar to the AIO points to the value of regularly re-examining and updating group’s trial, casting doubt on the usefulness of this strategy. the external validity of RCTs, as well as to conduct geo- Our study has certain limitations inherent in a registry graphical validations of the outcomes, particularly when the study, including missing values for histological subtype in target population changes over time [10–12] or efficacious 21% of the cases. The procedures of multiple imputation off-trial therapies entail rethinking treatment strategy [13, decrease bias, but in this case, they are associated with 37]. more conservative estimations. Second, one of the criti- Among elderly subjects in the AGAMENON registry, cisms of Bayesian models is the subjectivism in selecting DPF was seen to be administered with dose reductions and priors. However, the use of a perspective based on objec- modifications. This mitigated the risk of toxicity in the tive external data lessens this problem, and allows gradual, elderly and revealed that most of the adaptations took place pragmatic answers to be attained. In this regard, Bayes- at the beginning, in line with the published clinical trials that ian analyses have added advantages over the classical 13 452 P. Jimenez‑Fonseca et al. Fig. 4  Posterior probabilities for the therapeutic effect of DPF vs models with therapeutic effect by covariate interactions. Color coding PF on overall survival depending on clinical–pathological vari- on the panels highlights the subgroups with greater effect (blue). A ables. CrI, credible interval. We applied specific priors for the main time ratio of more than 1 for the covariate implies that this slows or effect ~ N(0.15, 0.045), and skeptical priors for interactions (see meth- prolongs the time to event, whereas a time ratio of less than 1 indi- ods). The density plots must be interpreted as posterior probability of cates that an event is more likely to occur earlier. Thus, a time ratio the effect under each prior. Thus, the more to the right the area, the equal to 2 would mean that the median of time to event is doubled in greater the benefit and the narrower it becomes, the greater the accu- patients treated with DPF racy of the estimate. The results are derived from multivariable AFT frequentist models [28], including greater accuracy of disease point to FLOT being less active in tumors with estimations and less bias. Based on the Wagner’s previ- diffuse histology, which is consistent with our results [11]. ous results [9], our analysis is capable of capturing a dis- In short, our data confirm the need to update the appli- creet effect in the diffuse subtype, as well as a result for cability of RCTs, such as TAX325 [7], from time to time, the intestinal subtype in line with the conclusions of the or others that will fall behind as clinical practice changes TAX325 RCT [7]. However, the hypothesis about the spe- [44], especially in view of the biological diversity of this cific effect of DPF on diffuse tumors would have to be elu- disease. Ultimately, our study confirms the benefit of the cidated in further RCTs, which we believe to be unlikely docetaxel triple-drug regimen as first-line treatment in the at this time. With respect to the validity of our results, the real-world setting, attesting to the greater applicability in reader must be aware that the registry still contains a lim- middle-aged individuals with non-diffuse tumors. However, ited number of subjects treated with FLOT, which is also in the best-case scenario, the benefit is modest and comes at a specific limitation in Wagner’s meta-analysis, used as a the expense of increased toxicity. Current clinical practice prior in our Bayesian model [9]. Despite the fact that the guidelines endorse the use of dual-agent schedules, whereas impact of FLOT in seniors is as yet uncertain, the Bayes- triplets with docetaxel are deemed a useful alternative for fit ian model adjusted here contemplates the type of platin, patients in the event that an urgent tumor response is needed age, and year of treatment, among other factors. While or in locally advanced, unresectable tumors [14, 45, 46]. more experience with this scheme in advanced tumors is The need to design new RCTs in AGC separating the evalu- still needed, the data currently available in perioperative ation of effects based on clinical–pathological variables is 13 External validity of docetaxel triplet trials in advanced gastric cancer: are there patients… 453 Table 2  Evaluation of potentially age-dependent effect-modifying factors for OS Variables Specific informative prior Time ratio (95%, CrI) Posterior probability of effect Posterior probability size > 15% (TR > 1.15) of effect size > 30% (TR > 1.30) ECOG-PS ≥ 2 1.16 (0.99–1.36) 54% 1% ECOG-PS 0-1 1.18 (1.10–1.25) 75% 13% Grade 3 1.19 (1.06–1.36) 72% 14% Grade 2 1.20 (1.05–1.40) 73% 22% Grade 1 1.17 (1.09–1.25) 69% 1% Intestinal subtype 1.27 (1.08–1.11) 90% 40% Diffuse subtype 1.17 (1.09–1.24) 67% 0 Organs involved, 4 1.22 (1.04–1.43) 73% 26% Organs involved, 3 1.18 (1.02–1.39) 65% 17% Organs involved, 2 1.27 (1.10–1.44) 88% 37% Organs involved, 1 1.17 (1.09–1.24) 67% 0 Burden of liver disease > 75% 1.16 (0.98–1.37) 53% 13% Burden of liver disease 51–75% 1.19 (1.02–1.40) 66% 20% Burden of liver disease 25–50% 1.16 (0.99–1.36) 54% 12% Burden of liver disease < 25% 1.18 (1.10–1.25) 74% 1% CrI, Bayesian credible interval; ECOG-PS, Eastern Cooperative Group performance status; ROPE, region of practical equivalence (effect = 0 ± 10%); TR, time ratio; OS = overall survival. The time ratios were derived from bayesian AFT lognormal models with treatment-by- covariate interactions; these models are multivariable (14 confounding factors, see Methods). The specific informative prior encompasses using the adapted evidence from Wagner’s 2017 meta-analysis (normal prior with mean = 0.15, standard deviation = 0.045 for therapeutic effect) with moderately skeptical priors for interactions (see “Methods”) A time ratio of more than 1 for the covariate implies that this slows or prolongs the time to event, whereas a time ratio of less than 1 indicates that an event is more likely to occur earlier. Thus, a time ratio equal to 2 would mean that the median of time to event is doubled in patients treated with DPF The posterior probability of effect sizes > 15% or 30% (TR > 1.15 or 1.30) denotes the actual probability of achieving a benefit of that magnitude (15 or 30%) or greater Fig. 5  Therapeutic effect on the basis of age. CrI, credible inter- val. Note: Results are derived from a multivariable AFT Bayesian model, with effect by age interaction 13 454 P. Jimenez‑Fonseca et al. crucial given these data. This information should be taken ation​-and-updat​ing-of-a-progn​ostic​-1-Royst​on-Parma​r/5ba4a​ into account when choosing treatment strategy in the context 569ff​a416f​740a1​06e49​15608​37074​ce44b​. 5. Dekkers OM, von Elm E, Algra A, Romijn JA, Vandenbroucke JP. of the growing recommendation of dual-agent schedules in How to assess the external validity of therapeutic trials: a concep- first line and the continuum of care of AGC. tual approach. Int J Epidemiol. 2010;39:89–94. 6. Rawla P, Barsouk A. Epidemiology of gastric cancer: global Acknowledgments Priscilla Chase Duran for editing and translating trends, risk factors and prevention. Prz Gastroenterol. 2019;14:26. the manuscript. Miguel Vaquero, Natalia Cateriano and the IRICOM 7. Van Cutsem E, Moiseyenko VM, Tjulandin S, Majlis A, Constenla S.L. team for the support of the website registry. M, Boni C, et al. Phase III study of docetaxel and cisplatin plus fluorouracil compared with cisplatin and fluorouracil as first-line Authors’ contributions PJF, ACB and JG developed the project, ana- therapy for advanced gastric cancer: a report of the V325 Study lyzed the data and drafted the manuscript. The other authors recruited Group. J Clin Oncol. 2006;24:4991–7. patients and provided clinical information, comments, and improve- 8. Yamada Y, Boku N, Mizusawa J, Iwasa S, Kadowaki S, Nakayama ments to the manuscript. All authors participated in the interpretation N, et al. Docetaxel plus cisplatin and S-1 versus cisplatin and S-1 and discussion of data, and the critical review of the manuscript. in patients with advanced gastric cancer (JCOG1013): an open- label, phase 3, randomised controlled trial. Lancet Gastroenterol Hepatol. 2019;4:501–10. Funding None to declare; this is an academic study. The study was 9. Wagner AD, Syn NLX, Moehler M, Grothe W, Yong WP, Tai supported by the authors themselves. B, et al. Chemotherapy for advanced gastric cancer. Cochrane Database Syst Rev. 2017;8:CD004064. Availability of data and materials All the data generated or analyzed 10. Bang YJ, Van Cutsem E, Feyereislova A, Chung HC, Shen L, in this study are included in the manuscript or the supplementary Sawaki A, et al. Trastuzumab in combination with chemother- information. apy versus chemotherapy alone for treatment of HER2-positive advanced gastric or gastro-oesophageal junction cancer (ToGA): Compliance with ethical standards a phase 3, open-label, randomised controlled trial. Lancet. 2010;376:687–97. 11. Al-Batran S-E, Hofheinz RD, Pauligk C, Kopp H-G, Haag GM, Conflict of interest The authors state that they have no conflict of inter- Luley KB, et al. Histopathological regression after neoadjuvant est related to this study. docetaxel, oxaliplatin, fluorouracil, and leucovorin versus epiru- bicin, cisplatin, and fluorouracil or capecitabine in patients with Ethic approval and consent to participate All procedures followed resectable gastric or gastro-oesophageal junction adenocarcinoma were in accordance with the ethical standards of the committee respon- (FLOT4-AIO). Lancet Oncol. 2016;17:1697–708. sible for human experimentation (institutional and national) and with 12. Jiménez Fonseca P, Carmona-Bayonas A, Hernández R, Custo- the Helsinki Declaration of 1964 and later versions. Informed consent dio A, Cano JM, Lacalle A, et al. Lauren subtypes of advanced was obtained from all patients before being included in the study. This gastric cancer influence survival and response to chemotherapy: work is original and has not been previously presented elsewhere. real-world data from the AGAMENON National Cancer Registry. Br J Cancer. 2017;117:775–82. Open Access This article is licensed under a Creative Commons Attri- 13. Wilke H, Muro K, Van Cutsem E, Oh S-C, Bodoky G, Shimada Y, bution 4.0 International License, which permits use, sharing, adapta- et al. Ramucirumab plus paclitaxel versus placebo plus paclitaxel tion, distribution and reproduction in any medium or format, as long in patients with previously treated advanced gastric or gastro- as you give appropriate credit to the original author(s) and the source, oesophageal junction adenocarcinoma (RAINBOW): a double- provide a link to the Creative Commons licence, and indicate if changes blind, randomised phase 3 trial. Lancet Oncol. 2014;15:1224–35. were made. The images or other third party material in this article are 14. National Comprehensive Cancer Network. NCCN clinical prac- included in the article’s Creative Commons licence, unless indicated tice guidelines for gastric cancer. NCCN Guidel. (cited 2019 Mar otherwise in a credit line to the material. If material is not included in 11). p. Varsion 2.2020. https​://www.nccn.org/profe​ssion​als/physi​ the article’s Creative Commons licence and your intended use is not cian_gls/pdf/gastr​ic.pdf. permitted by statutory regulation or exceeds the permitted use, you will 15. Muro K, Van Cutsem E, Narita Y, Pentheroudakis G, Baba E, Li need to obtain permission directly from the copyright holder. To view a J, et al. Pan-Asian adapted ESMO Clinical Practice Guidelines copy of this licence, visit http://creat​iveco​mmons​.org/licen​ses/by/4.0/. for the management of patients with metastatic gastric cancer: a JSMO–ESMO initiative endorsed by CSCO, KSMO, MOS, SSO and TOS. Ann Oncol. 2018;30:19–33. 16. Carmona-Bayonas A, Jiménez-Fonseca P, Custodio A, Sánchez Cánovas M, Hernández R, Pericay C, et al. Anthracycline-based References triplets do not improve the efficacy of platinum-fluoropyrimidine doublets in first-line treatment of advanced gastric cancer: real- 1. Rothwell PM. Commentary: external validity of results of rand- world data from the AGAMENON National Cancer Registry. omized trials: disentangling a complex concept. Int J Epidemiol. Gastric Cancer. 2017;14:1379–88. 2010;39:94–6. 17. Carmona-Bayonas A, Jiménez-Fonseca P, Lorenzo MLS, Ram- 2. Rothwell PM. Factors that can affect the external validity of ran- chandani A, Martínez EA, Custodio A, et al. On the effect of domised controlled trials. PLOS Clin Trial. 2006;1(1):e9. triplet or doublet chemotherapy in advanced gastric cancer: results 3. Rothwell PM. External validity of randomised controlled trials:“to from a National Cancer Registry. J Natl Compr Cancer Netw. whom do the results of this trial apply?”. Lancet. 2005;365:82–93. 2016;14:1379–88. 4. Royston P, Parmar MKB, Altman DG. External validation and 18. Jiménez-Fonseca P, Carmona-Bayonas A, Lorenzo MLS, Plazas updating of a prognostic survival model. Hub Trials Methodol. JG, Custodio A, Hernández R, et al. Prognostic significance of Res. 2010. https​://www.seman​ticsc​holar​.org/paper​/Exter​nal-valid​ performing universal HER2 testing in cases of advanced gastric cancer. Gastric Cancer. 2016;20:465–74. 13 External validity of docetaxel triplet trials in advanced gastric cancer: are there patients… 455 19. Custodio A, Carmona-Bayonas A, Fonseca PJ, Sánchez ML, 36. Al-Batran S-E, Hartmann JT, Probst S, Schmalenberg H, Hol- Antonio Viudez R, Hernández JM, et al. Nomogram-based pre- lerbach S, Hofheinz R, et al. Phase III trial in metastatic gas- diction of survival in patients with advanced oesophagogastric troesophageal adenocarcinoma with fluorouracil, leucovorin plus adenocarcinoma receiving first-line chemotherapy: a multi- either oxaliplatin or cisplatin: a study of the Arbeitsgemeinschaft center prospective study in the era of trastuzumab. Br J Cancer. Internistische Onkologie. J Clin Oncol. 2008;26:1435–42. 2017;116:1526–35. 37. Moons KGM, Kengne AP, Grobbee DE, Royston P, Vergouwe Y, 20. Visa L, Jiménez-Fonseca P, Martínez EA, Hernández R, Cus- Altman DG, et al. Risk prediction models: II. External validation, todio A, Garrido M, et al. Efficacy and safety of chemotherapy model updating, and impact assessment. Heart. 2012;98:691–8. in older versus non-older patients with advanced gastric can- 38. Shah MA, Janjigian YY, Stoller R, Shibata S, Kemeny M, cer: a real-world data, non-inferiority analysis. J Geriatr Oncol. Krishnamurthi S, et al. Randomized multicenter phase II study 2017;9:254–64. of modified docetaxel, cisplatin, and fluorouracil (DCF) versus 21. Carmona-Bayonas A, Jiménez-Fonseca P, Echavarria I, Cánovas DCF plus growth factor support in patients with metastatic gastric MS, Aguado G, Gallego J, et al. Surgery for metastases for esoph- adenocarcinoma: a study of the US Gastric Cancer Consortium. J ageal-gastric cancer in the real world: data from the AGAMENON Clin Oncol. 2015;33:3874–9. national registry. Eur J Surg Oncol. 2018;44:1191–8. 39. Al-Batran S-E, Hartmann JT, Hofheinz R, Homann N, Rethwisch 22. Viúdez A, Carmona-Bayonas A, Gallego J, Lacalle A, Hernández V, Probst S, et al. Biweekly fluorouracil, leucovorin, oxaliplatin, R, Cano JM, et al. Optimal duration of first-line chemotherapy for and docetaxel (FLOT) for patients with metastatic adenocarci- advanced gastric cancer: data from the AGAMENON registry. noma of the stomach or esophagogastric junction: a phase II trial Clin Transl Oncol. 2020;22:734–50. of the Arbeitsgemeinschaft Internistische Onkologie. Ann Oncol. 23. Carmona-Bayonas A, Jiménez-Fonseca P, et al. Multistate 2008;19:1882–7. models: accurate and dynamic methods to improve predictions 40. Van Cutsem E, Boni C, Tabernero J, Massuti B, Middleton G, of thrombotic risk in patients with cancer. Thromb Haemost. Dane F, et al. Docetaxel plus oxaliplatin with or without fluoro- 2019;119:1849–59. uracil or capecitabine in metastatic or locally recurrent gastric 24. Cotes Sanchís A, Gallego J, Hernandez R, Arrazubi V, Custo- cancer: a randomized phase II study. Ann Oncol. 2014;26:149–56. dio A, Cano JM, et al. Second-line treatment in advanced gastric 41. Rosenberg AJ, Rademaker A, Hochster HS, Ryan T, Hensing cancer: data from the Spanish AGAMENON registry. PLoS One. T, Shankaran V, et al. Docetaxel, oxaliplatin, and 5-fluorouracil 2020;15:e0235848. https:​ //doi.org/10.1371/journ​al.pone.023584​ 8. (DOF) in metastatic and unresectable gastric/gastroesophageal 25. Harrell F. Regression modeling strategies: with applications to lin- junction adenocarcinoma: a phase ii study with long-term follow- ear models, logistic and ordinal regression, and survival analysis. up. Oncologist. 2019;24:1039–e642. 2nd ed. New York: Springer; 2015. 42. Al-Batran S-E, Pauligk C, Homann N, Hartmann JT, Moehler 26. Crew KD, Neugut AI. Epidemiology of gastric cancer. World J M, Probst S, et al. The feasibility of triple-drug chemotherapy Gastroenterol. 2006;12:354. combination in older adult patients with oesophagogastric can- 27. Wijeysundera DN, Austin PC, Hux JE, Beattie WS, Laupacis cer: a randomised trial of the Arbeitsgemeinschaft Internistische A. Bayesian statistical inference enhances the interpretation of Onkologie (FLOT65+). Eur J Cancer. 2013;49:835–42. contemporary randomized controlled trials. J Clin Epidemiol. 43. Lorenzen S, Stahl M, Hofheinz R-D, Al-Batran S-E, Lordick 2009;62:13–21. F. Influence of taxanes on treatment sequence in gastric cancer. 28. Van de Schoot R, Kaplan D, Denissen J, Asendorpf JB, Neyer FJ, Oncol Res Treat. 2019;43(1–2):1–6. Van Aken MAG. A gentle introduction to Bayesian analysis: appli- 44. Tabernero J, Van Cutsem E, Bang Y-J, Fuchs CS, Wyrwicz L, Lee cations to developmental research. Child Dev. 2014;85:842–60. KW, et al. Pembrolizumab with or without chemotherapy versus 29. Bland JM, Altman DG. Bayesians and frequentists. Br Med J. chemotherapy for advanced gastric or gastroesophageal junction 1998;317:1151–60. (G/GEJ) adenocarcinoma: The phase III KEYNOTE-062 study. 30. Bendtsen M. A gentle introduction to the comparison between American Society of Clinical Oncology; 2019. null hypothesis testing and Bayesian analysis: reanalysis of two 45. Al-Batran S-E, Homann N, Pauligk C, Illerhaus G, Martens UM, randomized controlled trials. J Med Internet Res. 2018;20:e10873. Stoehlmacher J, et al. Effect of neoadjuvant chemotherapy fol- 31. Harrell Jr F, Frank E, Maintaner Frank E. Package ‘rms’. 2015 lowed by surgical resection on survival in patients with limited (cited 2020 Jan 1). p. 229. http://cran.r-proje​ct.org/web/packa​ges/ metastatic gastric or gastroesophageal junction cancer: the AIO- rms/index​.html. FLOT3 trial. JAMA Oncol. 2017;3:1237–44. 32. Bürkner P-C. Brms: An R package for Bayesian multilevel 46. Smyth EC, Verheij M, Allum W, Cunningham D, Cervantes A, models using Stan. J Stat Softw Foundation Open Access Stat. Arnold D. Gastric cancer: ESMO Clinical Practice Guidelines for 2017;80:1–28. diagnosis, treatment and follow-up. Ann Oncol. 2016;27:v38–49. 33. Stoyan G. A Practical Guide to Splines. Applied Mathematical Sciences. Springer‐Verlag, editor. Berlin‐Heidelberg‐New York; Publisher’s Note Springer Nature remains neutral with regard to 1980. jurisdictional claims in published maps and institutional affiliations. 34. Buuren S van, Groothuis-Oudshoorn K. Mice: multivariate impu- tation by chained equations in R. J Stat Softw. 2010;1–68 35. Cunningham D, Starling N, Rao S, Iveson T, Nicolson M, Coxon F, et al. Capecitabine and oxaliplatin for advanced esophagogas- tric cancer. N Engl J Med. 2008;358:36–46. 13 456 P. Jimenez‑Fonseca et al. Affiliations Paula Jimenez‑Fonseca1 · Alberto Carmona‑Bayonas2,3 · Eva Martínez de Castro4 · Ana Custodio5 · Carles Pericay Pijaume6 · Raquel Hernandez7 · Gema Aguado8 · Natalia Castro Unanua9 · Juana María Cano10 · Flora López11 · Marcelo Garrido12 · Ana Fernández Montes13 · Laura Visa14 · Manuel Sánchez Cánovas15 · María Luisa Limón16 · Nieves Martínez Lago17 · Paola Pimentel18 · Alicia Hurtado19 · Aitor Azkárate20 · Federico Longo21 · Marc Diez22 · Aranzazu Arias‑Martinez23 · Tamara Sauri24 · Alfonso Martín Carnicero25 · Monserrat Mangas26 · Marta Martín Richard27 · Mónica Granja28 · Avinash Ramchandani29 · Carolina Hernández Pérez30 · Paula Cerdá31 · Aitziber Gil‑Negrete32 · Mariona Calvo33 · Rosario Vidal Tocino34 · Javier Gallego35 1 19 Medical Oncology Department, Hospital Universitario Medical Oncology Department, Hospital Universitario Central de Asturias, ISPA, Avenida Roma s/n, Oviedo, Spain Fundación Alcorcón, Madrid, Spain 2 20 Hematology and Medical Oncology Department, Hospital Medical Oncology Department, Hospital Universitario Son Universitario Morales Meseguer, University of Murcia, Espases, Mallorca, Spain IMIB, CP13/00126, PI17/0050 (ISCIII& FEDER), Murcia, 21 Medical Oncology Department, Hospital Universitario Spain Ramón and Cajal, Madrid, Spain 3 Fundación Séneca (04515/GERM/06), Murcia, Spain 22 Medical Oncology Department, Hospital Universitario Vall 4 Medical Oncology Department, Hospital Universitario d’Hebron, Barcelona, Spain Marqués de Valdecilla, IDIVAL, Santander, Spain 23 Pharmacy Department, Hospital Universitario Central de 5 Medical Oncology Department, Hospital Universitario La Asturias, Oviedo, Spain Paz, CIBERONC CB16/12/00398, Madrid, Spain 24 Medical Oncology Department, Hospital Clinic, Barcelona, 6 Medical Oncology Department, Hospital Universitario Parc Spain Tauli, Sabadell, Spain 25 Medical Oncology Department, Hospital San Pedro, 7 Medical Oncology Department, Hospital Universitario de Logroño, Spain Canarias, Tenerife, Spain 26 Medical Oncology Department, Hospital Galdakao-Usansolo, 8 Medical Oncology Department, Hospital Universitario Galdakao‑Usansolo, Spain Gregorio Marañón, Madrid, Spain 27 Medical Oncology Department, Hospital Universitario Santa 9 Medical Oncology Department, Complejo Hospitalario de Creu i Sant Pau, Barcelona, Spain Navarra, Pamplona, Spain 28 Medical Oncology Department, Hospital Universitario 10 Medical Oncology Department, Hospital General Clínico San Carlos, Madrid, Spain Universitario de Ciudad Real, Ciudad Real, Spain 29 Medical Oncology Department, Hospital Universitario 11 Medical Oncology Department, Hospital Universitario Doce Insular de Gran Canaria, Las Palmas de Gran Canaria, Spain de Octubre, Madrid, Spain 30 Medical Oncology Department, Hospital Universitario 12 Medical Oncology Department, Pontificia Universidad Nuestra Señora de la Candelaria, Tenerife, Spain Católica de Chile, Santiago de Chile, Chile 31 Medical Oncology Department, Centro Médico Tecknon, 13 Medical Oncology Department, Complejo Hospitalario de Barcelona, Spain Orense, Orense, Spain 32 Medical Oncology Department, Hospital Universitario de 14 Medical Oncology Department, Hospital Universitario El Donostia, San Sebastián, Spain Mar, Barcelona, Spain 33 Medical Oncology Department, Catalan Institute 15 Hematology and Medical Oncology Department, Hospital of Oncology, L’Hospitalet, Spain Universitario Morales Meseguer, Murcia, Spain 34 Medical Oncology Department, Complejo Asistencial 16 Medical Oncology Department, Hospital Universitario Universitario de Salamanca, IBSAL, Salamanca, Spain Virgen del Rocío, Seville, Spain 35 Medical Oncology Department, Hospital General 17 Medical Oncology Department, Complejo Hospitalario Universitario de Elche, Elche, Spain Universitario de A Coruña, Coruña, Spain 18 Medical Oncology Department, Hospital Santa Lucía, Cartagena, Spain 13