Development and validation of a predictive tool for oesophageal cancer patients: A Moroccan-based study

Despite treatments emergence concerning oesophageal cancer (EC), it remains the 8th deadliest cancers in Morocco among other cancers. For that we sought to investigate on Moroccan patients diagnosed with EC locally advanced and metastatic ones in order to establish a nomogram to predict the probability of death.

Data of demographic, clinical pathological, biologic features, and treatment records were collected for 264 EC patients newly diagnosed from 2012 to 2019 in the Medical Oncology Department at Mohammed VI University Hospital of Marrakech. the cohort was split to training and validation set with ratio 7:3. The independent factors related to death were determined using Logistic Regression Analysis. We then constructed and validated internally a nomogram to predict probability of death of EC patients. The model was subjected to 300 bootstrap to get the final model's bias-corrected predictive accuracy measurements. The accuracy of the model was measured using C-index. The calibration plots were plotted, and the discriminative ability of the model was measured using AUC.

Over 264 EC patients, 138 locally advanced and metastatic diagnosed with EC as primary malignancy tumor, with no more than 20 percent of missing value were interred into analysis. From 27 extracted EC patients characteristics, only 4 variables were the independent to death, including age at diagnosis, chemotherapy, radiotherapy, and lung metastasis. These 4 factors were therefore incorporated to build and validate internally the nomogram. the calibration curves of both training and validation cohorts, show optimal agreement between the actual and predictive probabilities. Moreover, the Area Under the Curve (AUC) of the Receiver Operating Curves Characteristics (ROC) obtained from both cohorts, show a good discriminative accuracy with an AUC of 0.89 and 0.91 for training and validation sets respectively.

Through this work, we tried to establish a new predictive tool that concern patient newly diagnosed with EC based on factors related to an African population.

The authors.

Has not received any funding.

All authors have declared no conflicts of interest.