Machine learning to predict 5-year survival among pediatric Acute Myeloid Leukemia patients and development of OSPAM-C online survival prediction tool

Abstract

Background: Acute myeloid leukemia (AML) accounts for a fifth of childhood leukemia. Although survival rates for AML have greatly improved over the past few decades, they vary depending on demographic and AML type factors. Objectives: To predict the five-year survival among pediatric AML patients using machine learning algorithms and deploy the best performing algorithm as an online survival prediction tool. Materials and methods: Pediatric patients (0 to 14 years) with a microscopically confirmed AML were extracted from the Surveillance Epidemiology and End Results (SEER) database (2000-2016) and split into training and test datasets (80/20 ratio). Four machine learning algorithms (logistic regression, support vector machine, gradient boosting and K nearest neighbor) and a deep neural network algorithm were trained on features to predict five-year survival. Performances of the algorithms were compared and the best performing algorithm was deployed as an online prediction tool. Results: A total of 2,140 patients met our inclusion criteria. The gradient boosting algorithm was the best performer in terms of discrimination and predictive ability. It was deployed as the online survival prediction tool named OSPAM-C (https://ashis-das.shinyapps.io/ospam/). Conclusions: Our study provides a framework for the development and deployment of an online survival prediction tool for pediatric patients with AML. While external validation is needed, our survival prediction tool presents an opportunity to inform clinical decision-making for AML patients.

Semi-supervised learning in cancer diagnostics

“…The lack of reproducibility in research on artificial intelligence in general ( 44 ) is also likely to be a future issue in biomedical use-cases of SSL as unfortunately only a minority of studies provide publicly accessible code to support their results ( 11 , 19 , 27 , 28 , 30 , 38 , 40 ). As is evident from previous studies on SSL in oncology, use cases mainly include tumor entities with high prevalence such as breast ( 15 – 17 , 25 – 28 , 33 – 35 , 37 , 41 ), lung ( 18 , 22 , 23 , 33 , 34 , 38 , 41 ), and colorectal cancer ( 11 , 12 , 34 , 35 ) where single centers can amass sufficiently sized data sets to conduct SSL experiments. This is also reflected in the overwhelming absence of studies on SSL in hematology with only one single study ( 40 ) including any hematological neoplasm at all.…”

“…To grade breast cancer samples, Das et al. ( 17 ) employ a Generative Adversarial Network (GAN) where the discriminator uses an unsupervised model that is stacked over a supervised model with shared parameters to utilize both labeled and unlabeled samples. An Auxiliary Classifier GAN that divides lung cancer samples into malignant and benign patches which allows for subsequent pixel-based PD-L1 scoring is reported by Kapil et al.…”

“…Comparing SL and SSL, Al Azzam et al (16) report similar accuracies for SSL when using only half the number of labels needed for SL in breast cancer prediction from fine needle aspirates. To grade breast cancer samples, Das et al (17) employ a Generative Adversarial Network (GAN) where the discriminator uses an unsupervised model that is stacked over a supervised model with shared parameters to utilize both labeled and unlabeled samples. An Auxiliary Classifier GAN that divides lung cancer samples into malignant and benign patches which allows for subsequent pixel-based PD-L1 scoring is reported by Kapil et al (18) for non-small cell lung cancer tissue needle aspirates.…”

Design and Development of an Intelligent System for Predicting 5-Year Survival in Gastric Cancer

“…Presently, ML can predict breast cancer survivability in the primary stages. 34 - 36 Das et al 37 and Hauser et al 38 have compared selected ML methods to the survival prognosis of patients with leukemia. They have respectively found that the gradient boosting algorithms (BAs) such hist gradient boosting (HGB) with area under the curve (AUC) of 0.779 and XGBoost with AUC of 0.87 achieve the highest performance.…”