Comparison of machine learning techniques to handle imbalanced COVID-19 CBC datasets

The adverse inpatient medication event and frailty (AIME-frail) risk prediction model

BACKGROUND

Medication harm affects between 5 and 15% of hospitalised patients, with approximately half of the harm events considered preventable through timely intervention. The Adverse Inpatient Medication Event (AIME) risk prediction model was previously developed to guide a systematic approach to patient prioritisation for targeted clinician review, but frailty was not tested as a candidate predictor variable.

AIM

To evaluate the predictive performance of an updated AIME model, incorporating a measure of frailty, when applied to a new multisite cohort of hospitalised adult inpatients.

METHODS

A retrospective cohort study was conducted at two tertiary Australian hospitals on patients discharged between 1st January and April 31, 2020. Data were extracted from electronic medical records (EMRs) and clinical coding databases. Medication harm was identified using ICD-10 Y-codes and confirmed by senior pharmacist review of medical records. The Hospital Frailty Risk Score (HFRS) was calculated for each patient. Logistic regression analysis was used to construct a modified AIME model. Candidate variables of the original AIME model, together with new variables including HFRS were tested. Performance of the final model was reported using area under the curve (AUC) and decision curve analysis (DCA).

RESULTS

A total of 4089 patient admissions were included, with a mean age ± standard deviation (SD) of 64 years (±19 years), 2050 patients (50%) were males, and mean HFRS was 6.2 (±5.9). 184 patients (4.5%) experienced one or more medication harm events during hospitalisation. The new AIME-Frail risk model incorporated 5 of the original variables: length of stay (LOS), anti-psychotics, antiarrhythmics, immunosuppressants, and INR greater than 3, as well as 5 new variables: HFRS, anticoagulants, antibiotics, insulin, and opioid use. The AUC was 0.79 (95% CI: 0.76-0.83) which was superior to the original model (AUC = 0.70, 95% CI: 0.65-0.74) with a sensitivity of 69%, specificity of 81%, positive predictive value of 0.14 (95% CI: 0.10-0.17) and negative predictive value of 0.98 (95% CI: 0.97-0.99). The DCA identified the model as having potential clinical utility between the probability thresholds of 0.05-0.4.

CONCLUSION

The inclusion of a frailty measure improved the predictive performance of the AIME model. Screening inpatients using the AIME-Frail tool could identify more patients at high-risk of medication harm who warrant timely clinician review.

Comparing machine learning algorithms to predict COVID‑19 mortality using a dataset including chest computed tomography severity score data

Since the beginning of the COVID-19 pandemic, new and non-invasive digital technologies such as artificial intelligence (AI) had been introduced for mortality prediction of COVID-19 patients. The prognostic performances of the machine learning (ML)-based models for predicting clinical outcomes of COVID-19 patients had been mainly evaluated using demographics, risk factors, clinical manifestations, and laboratory results. There is a lack of information about the prognostic role of imaging manifestations in combination with demographics, clinical manifestations, and laboratory predictors. The purpose of the present study is to develop an efficient ML prognostic model based on a more comprehensive dataset including chest CT severity score (CT-SS). Fifty-five primary features in six main classes were retrospectively reviewed for 6854 suspected cases. The independence test of Chi-square was used to determine the most important features in the mortality prediction of COVID-19 patients. The most relevant predictors were used to train and test ML algorithms. The predictive models were developed using eight ML algorithms including the J48 decision tree (J48), support vector machine (SVM), multi-layer perceptron (MLP), k-nearest neighbourhood (k-NN), Naïve Bayes (NB), logistic regression (LR), random forest (RF), and eXtreme gradient boosting (XGBoost). The performances of the predictive models were evaluated using accuracy, precision, sensitivity, specificity, and area under the ROC curve (AUC) metrics. After applying the exclusion criteria, a total of 815 positive RT-PCR patients were the final sample size, where 54.85% of the patients were male and the mean age of the study population was 57.22 ± 16.76 years. The RF algorithm with an accuracy of 97.2%, the sensitivity of 100%, a precision of 94.8%, specificity of 94.5%, F1-score of 97.3%, and AUC of 99.9% had the best performance. Other ML algorithms with AUC ranging from 81.2 to 93.9% had also good prediction performances in predicting COVID-19 mortality. Results showed that timely and accurate risk stratification of COVID-19 patients could be performed using ML-based predictive models fed by routine data. The proposed algorithm with the more comprehensive dataset including CT-SS could efficiently predict the mortality of COVID-19 patients. This could lead to promptly targeting high-risk patients on admission, the optimal use of hospital resources, and an increased probability of survival of patients.

Novel Biomarker Prediction for Lung Cancer Using Random Forest Classifiers

Lung cancer is considered the most common and the deadliest cancer type. Lung cancer could be mainly of 2 types: small cell lung cancer and non-small cell lung cancer. Non-small cell lung cancer is affected by about 85% while small cell lung cancer is only about 14%. Over the last decade, functional genomics has arisen as a revolutionary tool for studying genetics and uncovering changes in gene expression. RNA-Seq has been applied to investigate the rare and novel transcripts that aid in discovering genetic changes that occur in tumours due to different lung cancers. Although RNA-Seq helps to understand and characterise the gene expression involved in lung cancer diagnostics, discovering the biomarkers remains a challenge. Usage of classification models helps uncover and classify the biomarkers based on gene expression levels over the different lung cancers. The current research concentrates on computing transcript statistics from gene transcript files with a normalised fold change of genes and identifying quantifiable differences in gene expression levels between the reference genome and lung cancer samples. The collected data is analysed, and machine learning models were developed to classify genes as causing NSCLC, causing SCLC, causing both or neither. An exploratory data analysis was performed to identify the probability distribution and principal features. Due to the limited number of features available, all of them were used in predicting the class. To address the imbalance in the dataset, an under-sampling algorithm Near Miss was carried out on the dataset. For classification, the research primarily focused on 4 supervised machine learning algorithms: Logistic Regression, KNN classifier, SVM classifier and Random Forest classifier and additionally, 2 ensemble algorithms were considered: XGboost and AdaBoost. Out of these, based on the weighted metrics considered, the Random Forest classifier showing 87% accuracy was considered to be the best performing algorithm and thus was used to predict the biomarkers causing NSCLC and SCLC. The imbalance and limited features in the dataset restrict any further improvement in the model's accuracy or precision. In our present study using the gene expression values (LogFC, P Value) as the feature sets in the Random Forest Classifier BRAF, KRAS, NRAS, EGFR is predicted to be the possible biomarkers causing NSCLC and ATF6, ATF3, PGDFA, PGDFD, PGDFC and PIP5K1C is predicted to be the possible biomarkers causing SCLC from the transcriptome analysis. It gave a precision of 91.3% and 91% recall after fine tuning. Some of the common biomarkers predicted for NSCLC and SCLC were CDK4, CDK6, BAK1, CDKN1A, DDB2.

Mortality Prediction of COVID-19 Patients Using Radiomic and Neural Network Features Extracted from a Wide Chest X-ray Sample Size: A Robust Approach for Different Medical Imbalanced Scenarios

Aim: The aim of this study was to develop robust prognostic models for mortality prediction of COVID-19 patients, applicable to different sets of real scenarios, using radiomic and neural network features extracted from chest X-rays (CXRs) with a certified and commercially available software. Methods: 1816 patients from 5 different hospitals in the Province of Reggio Emilia were included in the study. Overall, 201 radiomic features and 16 neural network features were extracted from each COVID-19 patient’s radiography. The initial dataset was balanced to train the classifiers with the same number of dead and survived patients, randomly selected. The pipeline had three main parts: balancing procedure; three-step feature selection; and mortality prediction with radiomic features through three machine learning (ML) classification models: AdaBoost (ADA), Quadratic Discriminant Analysis (QDA) and Random Forest (RF). Five evaluation metrics were computed on the test samples. The performance for death prediction was validated on both a balanced dataset (Case 1) and an imbalanced dataset (Case 2). Results: accuracy (ACC), area under the ROC-curve (AUC) and sensitivity (SENS) for the best classifier were, respectively, 0.72 ± 0.01, 0.82 ± 0.02 and 0.84 ± 0.04 for Case 1 and 0.70 ± 0.04, 0.79 ± 0.03 and 0.76 ± 0.06 for Case 2. These results show that the prediction of COVID-19 mortality is robust in a different set of scenarios. Conclusions: Our large and varied dataset made it possible to train ML algorithms to predict COVID-19 mortality using radiomic and neural network features of CXRs.