Estimating the health-related quality of life of kidney stone patients: initial results from the Wisconsin Stone Quality of Life Machine-Learning Algorithm (WISQOL-MLA).

OBJECTIVE: To build the Wisconsin Stone Quality of Life Machine-Learning Algorithm (WISQOL-MLA) to predict urolithiasis patients' health-related quality of life (HRQoL) based on demographic, symptomatic and clinical data collected for the validation of the Wisconsin Stone Quality-of-Life (WISQOL) questionnaire, an HRQoL measurement tool designed specifically for patients with kidney stones. MATERIAL AND METHODS: We used data from 3206 stone patients from 16 centres. We used gradient-boosting and deep-learning models to predict HRQoL scores. We also stratified HRQoL scores by quintile. The dataset was split using a standard 70%/10%/20% training/validation/testing ratio. Regression performance was evaluated using Pearson's correlation. Classification was evaluated with an area under the receiver-operating characteristic curve (AUROC). RESULTS: Gradient boosting obtained a test correlation of 0.62. Deep learning obtained a correlation of 0.59. Multivariate regression achieved a correlation of 0.44. Quintile stratification of all patients in the WISQOL dataset obtained an average test AUROC of 0.70 for the five classes. The model performed best in identifying the lowest (0.79) and highest quintiles (0.83) of HRQoL. Feature importance analysis showed that the model weighs in clinically relevant factors to estimate HRQoL, such as symptomatic status, body mass index and age. CONCLUSIONS: Harnessing the power of the WISQOL questionnaire, our initial results indicate that the WISQOL-MLA can adequately predict a stone patient's HRQoL from readily available clinical information. The algorithm adequately relies on relevant clinical factors to make its HRQoL predictions. Future improvements to the model are needed for direct clinical applications.

Clinical Applications of Machine Learning for Urolithiasis and Benign Prostatic Hyperplasia: A Systematic Review.

Introduction: Previous systematic reviews related to machine learning (ML) in urology often overlooked the literature related to endourology. Therefore, we aim to conduct a more focused systematic review examining the use of ML algorithms for the management of benign prostatic hyperplasia (BPH) or urolithiasis. In addition, we are the first group to evaluate these articles using the Standardized Reporting of Machine Learning Applications in Urology (STREAM-URO) framework. Methods: Searches of MEDLINE, Embase, and the Cochrane CENTRAL databases were conducted from inception through July 12, 2021. Keywords included those related to ML, endourology, urolithiasis, and BPH. Two reviewers screened the citations that were eligible for title, abstract, and full-text screening, with conflicts resolved by a third reviewer. Two reviewers extracted information from the studies, with discrepancies resolved by a third reviewer. The data collected were then qualitatively synthesized by consensus. Two reviewers evaluated each article according to the STREAM-URO checklist with discrepancies resolved by a third reviewer. Results: After identifying 459 unique citations, 63 articles were retained for data extraction. Most articles consisted of tabular (n = 32) and computer vision (n = 23) tasks. The two most common problem types were classification (n = 40) and regression (n = 12). In general, most studies utilized neural networks as their ML algorithm (n = 36). Among the 63 studies retrieved, 58 were related to urolithiasis and 5 focused on BPH. The urolithiasis studies were designed for outcome prediction (n = 20), stone classification (n = 18), diagnostics (n = 17), and therapeutics (n = 3). The BPH studies were designed for outcome prediction (n = 2), diagnostics (n = 2), and therapeutics (n = 1). On average, the urolithiasis and BPH articles met 13.8 (standard deviation 2.6), and 13.4 (4.1) of the 26 STREAM-URO framework criteria, respectively. Conclusions: The majority of the retrieved studies effectively helped with outcome prediction, diagnostics, and therapeutics for both urolithiasis and BPH. While ML shows great promise in improving patient care, it is important to adhere to the recently developed STREAM-URO framework to ensure the development of high-quality ML studies.

Recurrent disease progression networks for modelling risk trajectory of heart failure.

MOTIVATION: Recurrent neural networks (RNN) are powerful frameworks to model medical time series records. Recent studies showed improved accuracy of predicting future medical events (e.g., readmission, mortality) by leveraging large amount of high-dimensional data. However, very few studies have explored the ability of RNN in predicting long-term trajectories of recurrent events, which is more informative than predicting one single event in directing medical intervention. METHODS: In this study, we focus on heart failure (HF) which is the leading cause of death among cardiovascular diseases. We present a novel RNN framework named Deep Heart-failure Trajectory Model (DHTM) for modelling the long-term trajectories of recurrent HF. DHTM auto-regressively predicts the future HF onsets of each patient and uses the predicted HF as input to predict the HF event at the next time point. Furthermore, we propose an augmented DHTM named DHTM+C (where "C" stands for co-morbidities), which jointly predicts both the HF and a set of acute co-morbidities diagnoses. To efficiently train the DHTM+C model, we devised a novel RNN architecture to model disease progression implicated in the co-morbidities. RESULTS: Our deep learning models confers higher prediction accuracy for both the next-step HF prediction and the HF trajectory prediction compared to the baseline non-neural network models and the baseline RNN model. Compared to DHTM, DHTM+C is able to output higher probability of HF for high-risk patients, even in cases where it is only given less than 2 years of data to predict over 5 years of trajectory. We illustrated multiple non-trivial real patient examples of complex HF trajectories, indicating a promising path for creating highly accurate and scalable longitudinal deep learning models for modeling the chronic disease.

Mining heterogeneous clinical notes by multi-modal latent topic model.

Latent knowledge can be extracted from the electronic notes that are recorded during patient encounters with the health system. Using these clinical notes to decipher a patient's underlying comorbidites, symptom burdens, and treatment courses is an ongoing challenge. Latent topic model as an efficient Bayesian method can be used to model each patient's clinical notes as "documents" and the words in the notes as "tokens". However, standard latent topic models assume that all of the notes follow the same topic distribution, regardless of the type of note or the domain expertise of the author (such as doctors or nurses). We propose a novel application of latent topic modeling, using multi-note topic model (MNTM) to jointly infer distinct topic distributions of notes of different types. We applied our model to clinical notes from the MIMIC-III dataset to infer distinct topic distributions over the physician and nursing note types. Based on manual assessments made by clinicians, we observed a significant improvement in topic interpretability using MNTM modeling over the baseline single-note topic models that ignore the note types. Moreover, our MNTM model led to a significantly higher prediction accuracy for prolonged mechanical ventilation and mortality using only the first 48 hours of patient data. By correlating the patients' topic mixture with hospital mortality and prolonged mechanical ventilation, we identified several diagnostic topics that are associated with poor outcomes. Because of its elegant and intuitive formation, we envision a broad application of our approach in mining multi-modality text-based healthcare information that goes beyond clinical notes. Code available at https://github.com/li-lab-mcgill/heterogeneous_ehr.

Inferring multimodal latent topics from electronic health records.

Electronic health records (EHR) are rich heterogeneous collections of patient health information, whose broad adoption provides clinicians and researchers unprecedented opportunities for health informatics, disease-risk prediction, actionable clinical recommendations, and precision medicine. However, EHRs present several modeling challenges, including highly sparse data matrices, noisy irregular clinical notes, arbitrary biases in billing code assignment, diagnosis-driven lab tests, and heterogeneous data types. To address these challenges, we present MixEHR, a multi-view Bayesian topic model. We demonstrate MixEHR on MIMIC-III, Mayo Clinic Bipolar Disorder, and Quebec Congenital Heart Disease EHR datasets. Qualitatively, MixEHR disease topics reveal meaningful combinations of clinical features across heterogeneous data types. Quantitatively, we observe superior prediction accuracy of diagnostic codes and lab test imputations compared to the state-of-art methods. We leverage the inferred patient topic mixtures to classify target diseases and predict mortality of patients in critical conditions. In all comparison, MixEHR confers competitive performance and reveals meaningful disease-related topics.

Application of Machine Learning and Grocery Transaction Data to Forecast Effectiveness of Beverage Taxation.

Sugar Sweetened Beverages (SSB) are the primary source of artificially added sugar and have a casual association with chronic diseases. Taxation of SSB has been proposed, but limited evidence exists to guide this public health policy. Grocery transaction data, with price, discounting and other information for beverage products, present an opportunity to evaluate the likely effects of taxation policy. Sales are often non-linearly associated with price and are affected by the prices of multiple competing brands. We evaluated the predictive performance of Boosted Decision Tree Regression (B-DTR) and Deep Neural Networks (DNN) that account for the non-linearity and competition across brands, and compared their performance to a benchmark regression, the Least Absolute Shrinkage and Selection Operator (LASSO). B-DTR and DNN showed a lower Mean Squared Error (MSE) of prediction in the sales of most major SSB brands in comparison to LASSO, indicating a superior accuracy in predicting the effectiveness of SSB taxation. We demonstrated the application of machine learning methods and large transactional data from grocery stores to forecast the effectiveness food taxation.