Survival prognostic factors in patients with acute myeloid leukemia using machine learning techniques

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

BACKGROUND

Machine Learning (ML) has demonstrated potential in enhancing care in adult oncology. However, its application in paediatric haematological malignancies is still emerging, necessitating a comprehensive review of its capabilities and limitations in this area.

METHODS

A literature search was conducted through Ovid. Studies included focused on ML models in paediatric patients with haematological malignancies. Studies were categorised into thematic groups for analysis.

RESULTS

Twenty studies, primarily on leukaemia, were included in this review. Studies were organised into thematic categories such as prognoses, treatment responses and toxicity predictions. Prognostic studies showed AUC scores between 0.685 and 0.929, indicating moderate-high predictive accuracy. Treatment response studies demonstrated AUC scores between 0.840 and 0.875, reflecting moderate accuracy. Toxicity prediction studies reported high accuracy with AUC scores from 0.870 to 0.927. Only five studies (25%) performed external validation. Significant heterogeneity was noted in ML tasks, reporting formats, and effect measures across studies, highlighting a lack of standardised reporting and challenges in data comparability.

CONCLUSION

The clinical applicability of these ML models remains limited by the lack of external validation and methodological heterogeneity. Addressing these challenges through standardised reporting and rigorous external validation is needed to translate ML from a promising research tool into a reliable clinical practice component.

IMPACT

Key message: Machine Learning (ML) significantly enhances predictive models in paediatric haematological cancers, offering new avenues for personalised treatment strategies. Future research should focus on developing ML models that can integrate with real-time clinical workflows. Addition to literature: Provides a comprehensive overview of current ML applications and trends. It identifies limitations to its applicability, including the limited diversity in datasets, which may affect the generalisability of ML models across different populations.

IMPACT

Encourages standardisation and external validation in ML studies, aiming to improve patient outcomes through precision medicine in paediatric haematological oncology.

Predictive Models for Long Term Survival of AML Patients Treated with Venetoclax and Azacitidine or 7+3 Based on Post Treatment Events and Responses: Retrospective Cohort Study

BACKGROUND

The treatment of acute myeloid leukemia (AML) in older or unfit patients typically involves a regimen of venetoclax plus azacitidine (ven/aza). Toxicity and treatment responses are highly variable following treatment initiation and clinical decision-making continually evolves in response to these as treatment progresses. To improve clinical decision support (CDS) following treatment initiation, predictive models based on evolving and dynamic toxicities, disease responses, and other features should be developed.

OBJECTIVE

This study aims to generate machine learning (ML)-based predictive models that incorporate individual predictors of overall survival (OS) for patients with AML, based on clinical events occurring after the initiation of ven/aza or 7+3 regimen.

METHODS

Data from 221 patients with AML, who received either the ven/aza (n=101 patients) or 7+3 regimen (n=120 patients) as their initial induction therapy, were retrospectively analyzed. We performed stratified univariate and multivariate analyses to quantify the association between toxicities, hospital events, and short-term disease responses and OS for the 7+3 and ven/aza subgroups separately. We compared the estimates of confounders to assess potential effect modifications by treatment. 17 ML-based predictive models were developed. The optimal predictive models were selected based on their predictability and discriminability using cross-validation. Uncertainty in the estimation was assessed through bootstrapping.

RESULTS

The cumulative incidence of posttreatment toxicities varies between the ven/aza and 7+3 regimen. A variety of laboratory features and clinical events during the first 30 days were differentially associated with OS for the two treatments. An initial transfer to intensive care unit (ICU) worsened OS for 7+3 patients (aHR 1.18, 95% CI 1.10-1.28), while ICU readmission adversely affected OS for those on ven/aza (aHR 1.24, 95% CI 1.12-1.37). At the initial follow-up, achieving a morphologic leukemia free state (MLFS) did not affect OS for ven/aza (aHR 0.99, 95% CI 0.94-1.05), but worsened OS following 7+3 (aHR 1.16, 95% CI 1.01-1.31) compared to that of complete remission (CR). Having blasts over 5% at the initial follow-up negatively impacted OS for both 7+3 (P<.001) and ven/aza (P<.001) treated patients. A best response of CR and CR with incomplete recovery (CRi) was superior to MLFS and refractory disease after ven/aza (P<.001), whereas for 7+3, CR was superior to CRi, MLFS, and refractory disease (P<.001), indicating unequal outcomes. Treatment-specific predictive models, trained on 120 7+3 and 101 ven/aza patients using over 114 features, achieved survival AUCs over 0.70.

CONCLUSIONS

Our findings indicate that toxicities, clinical events, and responses evolve differently in patients receiving ven/aza compared with that of 7+3 regimen. ML-based predictive models were shown to be a feasible strategy for CDS in both forms of AML treatment. If validated with larger and more diverse data sets, these findings could offer valuable insights for developing AML-CDS tools that leverage posttreatment clinical data.

Machine learning and integrative multi-omics network analysis for survival prediction in acute myeloid leukemia

BACKGROUND

Acute myeloid leukemia (AML) is the most common malignant myeloid disorder in adults and the fifth most common malignancy in children, necessitating advanced technologies for outcome prediction.

METHOD

This study aims to enhance prognostic capabilities in AML by integrating multi-omics data, especially gene expression and methylation, through network-based feature selection methodologies. By employing artificial intelligence and network analysis, we are exploring different methods to build a machine learning model for predicting AML patient survival. We evaluate the effectiveness of combining omics data, identify the most informative method for network integration and compare the performance with standard feature selection methods.

RESULTS

Our findings demonstrate that integrating gene expression and methylation data significantly improves prediction accuracy compared to single omics data. Among network integration methods, our study identifies the best approach that improves informative feature selection for predicting patient outcomes in AML. Comparative analyses demonstrate the superior performance of the proposed network-based methods over standard techniques.

CONCLUSIONS

This research presents an innovative and robust methodology for building a survival prediction model tailored to AML patients. By leveraging multilayer network analysis for feature selection, our approach contributes to improving the understanding and prognostic capabilities in AML and laying the foundation for more effective personalized therapeutic interventions in the future.

Immunophenotypic clustering in paediatric acute myeloid leukaemia

Acute myeloid leukaemia (AML) is a highly heterogeneous disease, exhibiting diverse subtypes according to the characteristics of tumour cells. The immunophenotype is one of the aspects acquired routinely through flow cytometry in the diagnosis of AML. Here, we characterized the antigen expression in paediatric AML cases across both morphological and molecular genetic subgroups. We discovered a subgroup of patients with unfavourable prognosis that can be immunologically characterized, irrespective of morphological FAB results or genetic aberrations. Cox regression analysis unveiled key antigens influencing the prognosis of AML patients. In terms of underlying genotypes, we observed that the antigenic profiles and outcomes of one specific group, primarily composed of CBFA2T3::GLIS2 and FUS::ERG, were analogous to the reported RAM phenotype. Overall, our data highlight the significance of immunophenotype to tailor treatment for paediatric AML.

Identification of hub genes and potential molecular mechanisms related to drug sensitivity in acute myeloid leukemia based on machine learning

Background: Acute myeloid leukemia (AML) is the most common form of leukemia among adults and is characterized by uncontrolled proliferation and clonal expansion of hematopoietic cells. There has been a significant improvement in the treatment of younger patients, however, prognosis in the elderly AML patients remains poor. Methods: We used computational methods and machine learning (ML) techniques to identify and explore the differential high-risk genes (DHRGs) in AML. The DHRGs were explored through multiple in silico approaches including genomic and functional analysis, survival analysis, immune infiltration, miRNA co-expression and stemness features analyses to reveal their prognostic importance in AML. Furthermore, using different ML algorithms, prognostic models were constructed and validated using the DHRGs. At the end molecular docking studies were performed to identify potential drug candidates targeting the selected DHRGs. Results: We identified a total of 80 DHRGs by comparing the differentially expressed genes derived between AML patients and normal controls and high-risk AML genes identified by Cox regression. Genetic and epigenetic alteration analyses of the DHRGs revealed a significant association of their copy number variations and methylation status with overall survival (OS) of AML patients. Out of the 137 models constructed using different ML algorithms, the combination of Ridge and plsRcox maintained the highest mean C-index and was used to build the final model. When AML patients were classified into low- and high-risk groups based on DHRGs, the low-risk group had significantly longer OS in the AML training and validation cohorts. Furthermore, immune infiltration, miRNA coexpression, stemness feature and hallmark pathway analyses revealed significant differences in the prognosis of the low- and high-risk AML groups. Drug sensitivity and molecular docking studies revealed top 5 drugs, including carboplatin and austocystin-D that may significantly affect the DHRGs in AML. Conclusion: The findings from the current study identified a set of high-risk genes that may be used as prognostic and therapeutic markers for AML patients. In addition, significant use of the ML algorithms in constructing and validating the prognostic models in AML was demonstrated. Although our study used extensive bioinformatics and machine learning methods to identify the hub genes in AML, their experimental validations using knock-out/-in methods would strengthen our findings.

A review of the utility of prognostic tools in predicting 6-month mortality in cancer patients, conducted in the context of voluntary assisted dying

BACKGROUND

Eligibility to access the Victorian voluntary assisted dying (VAD) legislation requires that people have a prognosis of 6 months or less (or 12 months or less in the setting of a neurodegenerative diagnosis). Yet prognostic determination is frequently inaccurate and prompts clinician discomfort. Based on functional capacity and clinical and biochemical markers, prognostic tools have been developed to increase the accuracy of life expectancy predictions.

AIMS

This review of prognostic tools explores their accuracy to determine 6-month mortality in adults when treated under palliative care with a primary diagnosis of cancer (the diagnosis of a large proportion of people who are requesting VAD).

METHODS

A systematic search of the literature was performed on electronic databases Medline, Embase and Cinahl.

RESULTS

Limitations of prognostication identified include the following: (i) prognostic tools still provide uncertain prognoses; (ii) prognostic tools have greater accuracy predicting shorter prognoses, such as weeks to months, rather than 6 months; and (iii) functionality was often weighted significantly when calculating prognoses. Challenges of prognostication identified include the following: (i) the area under the curve (a value that represents how well a model can distinguish between two outcomes) cannot be directly interpreted clinically and (ii) difficulties exist related to determining appropriate thresholds of accuracy in this context.

CONCLUSIONS

Prognostication is a significant aspect of VAD, and the utility of the currently available prognostic tools appears limited but may prompt discussions about prognosis and alternative means (other than prognostic estimates) to identify those eligible for VAD.

Prediction of in-hospital hypokalemia using machine learning and first hospitalization day records in patients with traumatic brain injury

AIMS

Hypokalemia is a common complication following traumatic brain injury, which may complicate treatment and lead to unfavorable outcomes. Identifying patients at risk of hypokalemia on the first day of admission helps to implement prophylactic treatment, reduce complications, and improve prognosis.

METHODS

This multicenter retrospective study was performed between January 2017 and December 2020 using the electronic medical records of patients admitted due to traumatic brain injury. A propensity score matching approach was adopted with a ratio of 1:1 to overcome overfitting and data imbalance during subgroup analyses. Five machine learning algorithms were applied to generate a best-performed prediction model for in-hospital hypokalemia. The internal fivefold cross-validation and external validation were performed to demonstrate the interpretability and generalizability.

RESULTS

A total of 4445 TBI patients were recruited for analysis and model generation. Hypokalemia occurred in 46.55% of recruited patients and the incidences of mild, moderate, and severe hypokalemia were 32.06%, 12.69%, and 1.80%, respectively. Hypokalemia was associated with increased mortality, while severe hypokalemia cast greater impacts. The logistic regression algorithm had the best performance in predicting decreased serum potassium and moderate-to-severe hypokalemia, with an AUC of 0.73 ± 0.011 and 0.74 ± 0.019, respectively. The prediction model was further verified using two external datasets, including our previous published data and the open-assessed Medical Information Mart for Intensive Care database. Linearized calibration curves showed no statistical difference (p > 0.05) with perfect predictions.

CONCLUSIONS

The occurrence of hypokalemia following traumatic brain injury can be predicted by first hospitalization day records and machine learning algorithms. The logistic regression algorithm showed an optimal predicting performance verified by both internal and external validation.

Survival Prediction of Children Undergoing Hematopoietic Stem Cell Transplantation Using Different Machine Learning Classifiers by Performing Chi-Square Test and Hyperparameter Optimization: A Retrospective Analysis

Bone marrow transplant (BMT) is an effective surgical treatment for bone marrow-related disorders. However, several associated risk factors can impair long-term survival after BMT. Machine learning (ML) technologies have been proven useful in survival prediction of BMT receivers along with the influences that limit their resilience. In this study, an efficient classification model predicting the survival of children undergoing BMT is presented using a public dataset. Several supervised ML methods were investigated in this regard with an 80-20 train-test split ratio. To ensure prediction with minimal time and resources, only the top 11 out of the 59 dataset features were considered using Chi-square feature selection method. Furthermore, hyperparameter optimization (HPO) using the grid search cross-validation (GSCV) technique was adopted to increase the accuracy of prediction. Four experiments were conducted utilizing a combination of default and optimized hyperparameters on the original and reduced datasets. Our investigation revealed that the top 11 features of HPO had the same prediction accuracy (94.73%) as the entire dataset with default parameters, however, requiring minimal time and resources. Hence, the proposed approach may aid in the development of a computer-aided diagnostic system with satisfactory accuracy and minimal computation time by utilizing medical data records.

Comparing machine learning algorithms to predict 5-year survival in patients with chronic myeloid leukemia

INTRODUCTION

Chronic myeloid leukemia (CML) is a myeloproliferative disorder resulting from the translocation of chromosomes 19 and 22. CML includes 15-20% of all cases of leukemia. Although bone marrow transplant and, more recently, tyrosine kinase inhibitors (TKIs) as a first-line treatment have significantly prolonged survival in CML patients, accurate prediction using available patient-level factors can be challenging. We intended to predict 5-year survival among CML patients via eight machine learning (ML) algorithms and compare their performance.

METHODS

The data of 837 CML patients were retrospectively extracted and randomly split into training and test segments (70:30 ratio). The outcome variable was 5-year survival with potential values of alive or deceased. The dataset for the full features and important features selected by minimal redundancy maximal relevance (mRMR) feature selection were fed into eight ML techniques, including eXtreme gradient boosting (XGBoost), multilayer perceptron (MLP), pattern recognition network, k-nearest neighborhood (KNN), probabilistic neural network, support vector machine (SVM) (kernel = linear), SVM (kernel = RBF), and J-48. The scikit-learn library in Python was used to implement the models. Finally, the performance of the developed models was measured using some evaluation criteria with 95% confidence intervals (CI).

RESULTS

Spleen palpable, age, and unexplained hemorrhage were identified as the top three effective features affecting CML 5-year survival. The performance of ML models using the selected-features was superior to that of the full-features dataset. Among the eight ML algorithms, SVM (kernel = RBF) had the best performance in tenfold cross-validation with an accuracy of 85.7%, specificity of 85%, sensitivity of 86%, F-measure of 87%, kappa statistic of 86.1%, and area under the curve (AUC) of 85% for the selected-features. Using the full-features dataset yielded an accuracy of 69.7%, specificity of 69.1%, sensitivity of 71.3%, F-measure of 72%, kappa statistic of 75.2%, and AUC of 70.1%.

CONCLUSIONS

Accurate prediction of the survival likelihood of CML patients can inform caregivers to promote patient prognostication and choose the best possible treatment path. While external validation is required, our developed models will offer customized treatment and may guide the prescription of personalized medicine for CML patients.

Machine Learning–Based Exploratory Clinical Decision Support for Newly Diagnosed Patients With Acute Myeloid Leukemia Treated With 7 + 3 Type Chemotherapy or Venetoclax/Azacitidine

PURPOSE

There are currently limited objective criteria to help assist physicians in determining whether an individual patient with acute myeloid leukemia (AML) is likely to do better with induction with either standard 7 + 3 chemotherapy or targeted therapy with venetoclax plus azacitidine. The study goal was to address this need by developing exploratory clinical decision support methods.

PATIENTS AND METHODS

Univariable and multivariable analysis as well as comparison of a range of machine learning (ML) predictors were performed using cohorts of 120 newly diagnosed 7 + 3-treated AML patients compared with 101 venetoclax plus azacitidine–treated patients.

RESULTS

A variety of features in the two patient cohorts were identified that may potentially correlate with short- and long-term outcomes, toxicities, and other considerations. A subset of these diagnostic features was then used to develop ML-based predictors with relatively high areas under the curve of short- and long-term outcomes, hospital stays, transfusion requirements, and toxicities for individual patients treated with either venetoclax/azacitidine or 7 + 3.

CONCLUSION

Potential ML-based approaches to clinical decision support to help guide individual patients with newly diagnosed AML to either 7 + 3 or venetoclax plus azacitidine induction therapy were identified. Larger cohorts with separate test and validation studies are necessary to confirm these initial findings.

Evaluation of the Stellae-123 prognostic gene expression signature in acute myeloid leukemia

Risk stratification in acute myeloid leukemia (AML) has been extensively improved thanks to the incorporation of recurrent cytogenomic alterations into risk stratification guidelines. However, mortality rates among fit patients assigned to low or intermediate risk groups are still high. Therefore, significant room exists for the improvement of AML prognostication. In a previous work, we presented the Stellae-123 gene expression signature, which achieved a high accuracy in the prognostication of adult patients with AML. Stellae-123 was particularly accurate to restratify patients bearing high-risk mutations, such as ASXL1, RUNX1 and TP53. The intention of the present work was to evaluate the prognostic performance of Stellae-123 in external cohorts using RNAseq technology. For this, we evaluated the signature in 3 different AML cohorts (2 adult and 1 pediatric). Our results indicate that the prognostic performance of the Stellae-123 signature is reproducible in the 3 cohorts of patients. Additionally, we evidenced that the signature was superior to the European LeukemiaNet 2017 and the pediatric clinical risk scores in the prediction of survival at most of the evaluated time points. Furthermore, integration with age substantially enhanced the accuracy of the model. In conclusion, Stellae-123 is a reproducible machine learning algorithm based on a gene expression signature with promising utility in the field of AML.

Acute Myeloid Leukemia: New Multiomics Molecular Signatures and Implications for Systems Medicine Diagnostics and Therapeutics Innovation

Acute myeloid leukemia (AML) is a common, complex, and multifactorial malignancy of the hematopoietic system. AML diagnosis and treatment outcomes display marked heterogeneity and patient-to-patient variations. To date, AML-related biomarker discovery research has employed single omics inquiries. Multiomics analyses that reconcile and integrate the data streams from multiple levels of the cellular hierarchy, from genes to proteins to metabolites, offer much promise for innovation in AML diagnostics and therapeutics. We report, in this study, a systems medicine and multiomics approach to integrate the AML transcriptome data and reporter biomolecules at the RNA, protein, and metabolite levels using genome-scale biological networks. We utilized two independent transcriptome datasets (GSE5122, GSE8970) in the Gene Expression Omnibus database. We identified new multiomics molecular signatures of relevance to AML: miRNAs (e.g., mir-484 and miR-519d-3p), receptors (ACVR1 and PTPRG), transcription factors (PRDM14 and GATA3), and metabolites (in particular, amino acid derivatives). The differential expression profiles of all reporter biomolecules were crossvalidated in independent RNA-Seq and miRNA-Seq datasets. Notably, we found that PTPRG holds important prognostication potential as evaluated by Kaplan-Meier survival analyses. The multiomics relationships unraveled in this analysis point toward the genomic pathogenesis of AML. These multiomics molecular leads warrant further research and development as potential diagnostic and therapeutic targets.

Evaluating the frequency, prognosis and survival of RUNX1 and ASXL1 mutations in patients with acute myeloid leukaemia in northeastern Iran

To evaluate the frequency and prognosis of runt‐related transcription factor 1 (RUNX1) and additional sex combs like‐1 (ASXL1) mutations in acute myeloid leukaemia (AML) patients in northeastern Iran. This cross‐sectional study was performed on 40 patients with AML (including 35 patients with denovo AML and five patients with secondary AML) from February 2018 to February 2021. All patients were followed up for 36 months. We evaluated the frequency and survival rate of RUNX1 and ASXL1 mutations in AML patients. To detect mutations, peripheral blood samples and bone marrow aspiration were taken from all participants. One male patient (2.5%) had RUNX1 mutations and four cases (10%; 3 females vs. 1 male) had ASXL1 mutations. The survival rates of AML patients after 1, 3, 6, 9, 12, 24 and 36 months were 98%, 90%, 77%, 62%, 52%, 27% and 20%, respectively. There was a significant relationship between the occurrence of ASXL1 mutations and the survival of patients with AML (p = 0.027). Also, there was a significant relationship between the incidence of death and haemoglobin levels in patients with AML (p = 0.045). Thus, with an increase of one unit in patients' haemoglobin levels, the risk of death is reduced by 16.6%. Patients with AML had a high mortality rate, poor therapy outcome and low survival rate. ASXL1 and RUNX1 mutations are associated with a worse prognosis in patients with newly diagnosed AML. Also, we witnessed that the prevalence of ASXL1 to RUNX1 mutations was higher in northeastern Iran compared with other regions.

Survival prediction in acute myeloid leukemia using gene expression profiling

BACKGROUND

Acute myeloid leukemia (AML) is a genetically heterogeneous blood disorder. AML patients are associated with a relatively poor overall survival. The objective of this study was to establish a machine learning model to accurately perform the prognosis prediction in AML patients.

METHODS

We first screened for prognosis-related genes using Kaplan-Meier survival analysis in The Cancer Genome Atlas dataset and validated the results in the Oregon Health & Science University dataset. With a random forest model, we built a prognostic risk score using patient's age, TP53 mutation, ELN classification and normalized 197 gene expression as predictor variable. Gene set enrichment analysis was implemented to determine the dysregulated gene sets between the high-risk and low-risk groups. Similarity Network Fusion (SNF)-based integrative clustering was performed to identify subgroups of AML patients with different clinical features.

RESULTS

The random forest model was deemed the best model (area under curve value, 0.75). The random forest-derived risk score exhibited significant association with shorter overall survival in AML patients. The gene sets of pantothenate and coa biosynthesis, glycerolipid metabolism, biosynthesis of unsaturated fatty acids were significantly enriched in phenotype high risk score. SNF-based integrative clustering indicated three distinct subsets of AML patients in the TCGA cohort. The cluster3 AML patients were characterized by older age, higher risk score, more frequent TP53 mutations, higher cytogenetics risk, shorter overall survival.

CONCLUSIONS

The random forest-based risk score offers an effective method to perform prognosis prediction for AML patients.