Instructions
The model was developed from a registry of 1376 adult patients with locally advanced, unresectable or metastatic distal esophageal, GEJ, or gastric cancer treated between 2008 and 2019 at 34 Spanish teaching hospitals.
The registry’s basic eligibility criteria included patients who were fit enough to receive conventional first-line polychemotherapy validated in phase III studies, regardless of management taking place within the context of normal healthcare or in a clinical trial.
A Bayesian approach was used that yields gradual responses and estimates the continuous probability of effect size. This enables historical external data to be incorporated as priors. In Bayesian statistics, priors summarize the mathematical properties of our previous beliefs about the variables of interest, making it useful to conduct analyses under several perspectives that foster pragmatic, nuanced conclusions. Thus, the model use a specific prior for the therapeutic effect based on Wagner’s meta-analysis, for OS ~ N(0.15, 0.045) and for PFS ~ N(0.27, 0.07). Under this premise, the interactions were appraised skeptically (e.g., ~ N(0, 0.05) for the Lauren subtype; ~ N(0, 0.5) for age, ~ N(0, 0.2) for year) to discourage subgroup effects deemed extreme. The corroboration using this method improved accuracy and limited bias, by introducing a range of plausible perspectives about the previous knowledge. This perspective was used to build this online calculator.
Therapeutic effect was evaluated via a Bayesian parametric accelerated failure time (AFT) model with lognormal distribution; this model assumes that the effect of the covariates is to accelerate or decelerate the course of illness, making them suitable when the assumption of proportional hazards is not respected. This model was used, since its coefficients have an intuitive, direct interpretation in its exponentiated form, such as time ratios (TR). Thus, a coefficient of 0.5 for a binary predictor means that the median time-to-event is halved in the presence of this variable. With this model, the interaction between treatment, age, histopathologic subtype, and year of treatment was examined.
To assess convergence, we used trace plots for Markov chain Monte Carlo (MCMC) samples and the Gelman-Rubin (Rhat) measure for all variables. Continuous variables were assessed by restricted cubic splines. Covariates with >20% missing data were discarded and multiple imputation was applied (fully conditional specification, on 20 imputed datasets) in the rest. All analyses were performed with the R v3.1.6 software package, with the mice, splines, and brms libraries.
The scope of the model’s applicability must always be confirmed and attention must be paid to unusual risk factors that may be important in a minority of patients, as well as those that might eventually develop during the course of the disease, since only baseline variables have been contemplated.
Predictions are merely approximate, so decisions should be taken at the discretion of the attending physician.
References
- Jimenez-Fonseca P, Alberto Carmona-Bayonas A, Martínez de Castro E, …, Gallego J. Confirmation of the external validity of clinical trials in ever-changing target populations and time conditions: the example of docetaxel in advanced gastric cancer. April 2020. Epud ahead of print.
- Wagner AD, Syn NLX, Moehler M, Grothe W, Yong WP, Tai B, et al. Chemotherapy for advanced gastric cancer. Cochrane Database Syst Rev. 2017;8(8):CD004064.
- Van Cutsem E, Moiseyenko VM, Tjulandin S, Majlis A, Constenla M, Boni C, et al. Phase III study of docetaxel and cisplatin plus fluorouracil compared with cisplatin and fluorouracil as first-line therapy for advanced gastric cancer: a report of the V325 Study Group. J Clin Oncol. 2006;24(31):4991–7.
Disclaimer: This tool is intended for use by healthcare professionals only. Patients with advanced gastric cancer should seek medical care. Physicians and other healthcare professionals who use the Agamenon Triplet calculator should exercise their own clinical judgment. This app does not provide professional advice. We have exercised great caution in creating this app, based on real-world data. However, medical standards and practice may vary as new information becomes available and professionals should consult a variety of medical sources. While we estimate survival probability for advanced gastric cancer patients undergoing first-line chemotherapy, we in no way endorse any particular management strategy. Your reliance upon predictions obtained through this app is solely at your own risk. We assume no liability or responsibility whatsoever for damage or injury (including death) to you or anyone else from the use of this tool.