Table 1 Characteristics of Included Studies Exploring Machine Learning in Paediatric Haematological Malignancies.

From: Machine learning in paediatric haematological malignancies: a systematic review of prognosis, toxicity and treatment response models

Study	Cancer Type	Task/ Problem	Input Variable	Output Variable	ML Method	No. of Patients	Cross validation	External Validation	Highest AUC	Other Comparative Statical Scores Used
Prognosis and Relapse/ Recurrence Studies
He et al. ¹²	AML	Prognostic prediction	Expression levels of pyroptosis-related genes	Risk score	LASSO	NA	Not specified	E-MTAB-1216 dataset	0.893	HR = 2.04
He et al. ¹³	ALL	Survival prediction	Clinical characteristics, immunophenotype, genetic data	EFS predictive model	LASSO	1693	10-fold CV	Not specified	0.822	C-index = 0.81
Cui et al. ¹⁴	Leukaemia (sub-type not specified)	Survival prediction	Clinical characteristics	Survival predictive model	Bayesian inference	17539	Not specified	Not specified		C-index = 0.93
Zheng et al. ¹⁵	AML	Prognostic prediction	Gene expression data of m6A-related lncRNAs	OS predictive model	LASSO	646	Not specified	TCGA database	0.685	C-index = 0.82
Bohannan et al. ¹⁶	ALL	Survival prediction	Genomic data	EFS predictive model	RF	156	Training (70%) and testing (30%) cohorts used	Not specified	0.929	HR = 5.41 C-index = 0.82
Gao et al. ¹⁷	B-cell ALL	Survival prediction	Clinical characteristics	OS predictive model	LASSO	1316	C-index	TARGET database	0.898	C-index = 0.87
Lin et al. ¹⁸	B-cell ALL	Relapse prediction	NAD+ metabolism-related genes	Relapse predictive model	RF	NA	Mentioned but not specified	Not specified	0.8031
Pan et al. ¹⁹	ALL	Relapse prediction	Sociodemographic, clinical, immunological, and cytogenetic data	Relapse predictive model	RF, SVM, LR & DT	570	10-fold CV	Independent test set of 84 patients	0.904	Accuracy = 82.9%
Treatment Response Studies
Gbadamosi et al. ²⁰	AML	GO-related response prediction	Genetic data	Treatment response outcome model	LASSO	301	1000-fold CV	Not specified		OS = 0.676 HR = 0.565
Pedreira et al. ²¹	ALL	Treatment intensity decision support model	Clinical data	Treatment decision model	NN	158	Leave-one-out	Not specified		RHR = 98% ROR = 21% RSR = 0%
Gal et al. ²²	AML	Complete remission prediction	Gene expression data	Complete remission predictive model	K-NN, SVM & RF	473	5-fold CV	Not specified	0.840
Kashef et al. ²³	ALL	Treatment prediction	Clinical data	Complete remission predictive model	GBM, RF, GLM	241	5-fold CV	Not specified	0.8725
Kashef et al. ²⁴	ALL	Treatment outcome classification	Clinical characteristics & treatment related toxicity	Classification of treatment outcomes model	DT, SVM, RF,LDA,MLR, GBM,	241	10-fold CV	Not specified	0.870	Accuracy = 94.9%
Treatment Toxicity Studies
Al-Fahad et al. ²⁵	ALL	Treatment toxicity prediction	MRI-derived information	Classification of cognitive abilities	LASSO	200	Training (80%) and testing (20%) cohorts used	Not specified	0.870
Ramalingam et al. ²⁶	ALL	Treatment toxicity prediction	Genotypes of SLC19A1, MTHFR, TYMS, and cytogenetic data	Methotrexate-related toxicities	MDR	115	10-fold CV	Not specified		OR = 5.71 -2 Log Likelihood of Reduced Model = 97.104
Zhan et al. ²⁷	B-cell ALL	Treatment toxicity prediction	SNPs in 16 genes, clinical characteristics and methotrexate delayed clearance	Predictive models for the risk of neutropenia and fever	RF with ADASYN SVM, DT	139	Training (70%) and testing (30%) cohorts used	Not specified	0.927
Tram et al. ²⁸	Lymphoma	Treatment toxicity prediction	CT image	Risk of treatment-related late effects	NN	100	5-fold CV	Against human raters		Dice value = 0.988 HR = 3.1
Theruvath et al. ²⁹	Lymphoma	Dosing prediction	PET/MRI scan data	Enhanced PET/MRI images	NN	20	Not specified	Against neural network called SubtlePET	1	K statistic = 1
Disease Susceptibility/ Diagnosis Studies
Mahmood et al. ³⁰	ALL	Classifying risk factors	Clinical, genomic and socio-environmental data	Risk score for ALL	CART, RF, GBM & DT	50	10-fold CV	Not specified		Accuracy = 99.83%
Kulis et al. ³¹	B-Cell Precursor ALL	Classifying risk factors	Antigens measured through flow cytometry	Identification of specific genetic aberrations	GBM	818	5-fold CV	Not specified		OR = 16.90

ADASYN Adaptive Synthetic, ALL Acute Lymphoblastic Leukaemia, AML Acute Myeloid Leukaemia, AUC Area Under the Curve, CART Classification and Regression Tree, CV Cross-validation, DT Decision Tree, GBM Gradient Boosting Model, GLM Generalised Linear Model, GO gemtuzumab ozogamicin, HR Hazard Ratio, LncRNA Long noncoding RNA, K-NN k-nearest neighbour, LASSO Least Absolute Shrinkage and Selection Operator, LDA Linear Discriminant Analysis, LR Linear Regression, MDR Multifactor Dimensionality Reduction, NN Neural Network, OR Odds Ratio, RF Random Forest, RHR Rate of High Risk, ROR Rate of Overestimated Risk, RSR Rate of Subestimation Risk, SNP Single Nucleotide Polymorphism, SVM Support Vector Machine

Back to article page

Table 1 Characteristics of Included Studies Exploring Machine Learning in Paediatric Haematological Malignancies.

Search

Quick links