The use of machine learning for the identification of peripheral artery disease and future mortality risk

Prediction of Distal Aortic Enlargement after Proximal Repair of Aortic Dissection Using Machine Learning

OBJECTIVES

This study aimed to construct a risk prediction model for distal aortic enlargement in patients with type B aortic dissection (TBAD) treated with proximal thoracic endovascular aortic repair (TEVAR).

METHODS

From June 2010 to June 2016, patients with TBAD who underwent proximal TEVAR were retrospectively analyzed. A total of 38 clinical and imaging variables were collected. Univariable logistic regression was conducted to explore potential risk factors associated with distal aortic enlargement. Elastic net regression was employed to select significantly influential variables. Then, machine learning algorithms (logistic regression (LR), artificial neutral network (ANN), random forest and support vector machine) were applied to build risk prediction models. The area under the receiver operating characteristic curve (AUC), sensitivity and specificity were used to evaluate the performance of these models.

RESULTS

A total of 503 patients were enrolled in this study. During the follow-up, 105 (20.9%) patients were identified as having distal aortic enlargement, and 69 (13.7%) patients were found to have distal aortic aneurysm formation. Five patients were identified with aortic rupture. True lumen collapse and multi-false lumens were two potential risk factors for distal aortic enlargement after proximal repair of TBAD. The LR model performed the best in predicting distal aortic enlargement, with the highest sensitivity (96.7%) and an AUC of 0.773. The best model for predicting distal aneurysm formation was the ANN model, which yielded the highest AUC (0.876) and a specificity of 79.1%.

CONCLUSIONS

Machine learning approaches can produce accurate predictions of distal aortic enlargement after proximal repair of TBAD, which potentially benefits subsequent management.

The Role of Circulating Biomarkers in Peripheral Arterial Disease

Peripheral arterial disease (PAD) of the lower extremities is a chronic illness predominantly of atherosclerotic aetiology, associated to traditional cardiovascular (CV) risk factors. It is one of the most prevalent CV conditions worldwide in subjects >65 years, estimated to increase greatly with the aging of the population, becoming a severe socioeconomic problem in the future. The narrowing and thrombotic occlusion of the lower limb arteries impairs the walking function as the disease progresses, increasing the risk of CV events (myocardial infarction and stroke), amputation and death. Despite its poor prognosis, PAD patients are scarcely identified until the disease is advanced, highlighting the need for reliable biomarkers for PAD patient stratification, that might also contribute to define more personalized medical treatments. In this review, we will discuss the usefulness of inflammatory molecules, matrix metalloproteinases (MMPs), and cardiac damage markers, as well as novel components of the liquid biopsy, extracellular vesicles (EVs), and non-coding RNAs for lower limb PAD identification, stratification, and outcome assessment. We will also explore the potential of machine learning methods to build prediction models to refine PAD assessment. In this line, the usefulness of multimarker approaches to evaluate this complex multifactorial disease will be also discussed.

Do clinical and paraclinical findings have the power to predict critical conditions of injured patients after traumatic injury resuscitation? Using data mining artificial intelligence

PURPOSE

The triage and initial care of injured patients and a subsequent right level of care is paramount for an overall outcome after traumatic injury. Early recognition of patients is an important case of such decision-making with risk of worse prognosis. This article is to answer if clinical and paraclinical signs can predict the critical conditions of injured patients after traumatic injury resuscitation.

METHODS

The study included 1107 trauma patients, 16 years and older. The patients were trauma victims of Levels I and II triage and admitted to the Rajaee (Emtiaz) Trauma Hospital, Shiraz, in 2014-2015. The cross-industry process for data mining methodology and modeling was used for assessing the best early clinical and paraclinical variables to predict the patients' prognosis. Five modeling methods including the support vector machine, K-nearest neighbor algorithms, Bagging and Adaboost, and the neural network were compared by some evaluation criteria.

RESULTS

Learning algorithms can predict the deterioration of injured patients by monitoring the Bagging and SVM models with 99% accuracy. The most-fitted variables were Glasgow Coma Scale score, base deficit, and diastolic blood pressure especially after initial resuscitation in the algorithms for overall outcome predictions.

CONCLUSION

Data mining could help in triage, initial treatment, and further decision-making for outcome measures in trauma patients. Clinical and paraclinical variables after resuscitation could predict short-term outcomes much better than variables on arrival. With artificial intelligence modeling system, diastolic blood pressure after resuscitation has a greater association with predicting early mortality rather than systolic blood pressure after resuscitation. Artificial intelligence monitoring may have a role in trauma care and should be further investigated.

Machine learning-assisted decision-support models to better predict patients with calculous pyonephrosis

Background

To develop a machine learning (ML)-assisted model capable of accurately identifying patients with calculous pyonephrosis before making treatment decisions by integrating multiple clinical characteristics.

Methods

We retrospectively collected data from patients with obstructed hydronephrosis who underwent retrograde ureteral stent insertion, percutaneous nephrostomy (PCN), or percutaneous nephrolithotomy (PCNL). The study cohort was divided into training and testing datasets in a 70:30 ratio for further analysis. We developed 5 ML-assisted models from 22 clinical features using logistic regression (LR), LR optimized by least absolute shrinkage and selection operator (Lasso) regularization (Lasso-LR), support vector machine (SVM), extreme gradient boosting (XGBoost), and random forest (RF). The area under the curve (AUC) was applied to determine the model with the highest discrimination. Decision curve analysis (DCA) was used to investigate the clinical net benefit associated with using the predictive models.

Results

A total of 322 patients were included, with 225 patients in the training dataset, and 97 patients in the testing dataset. The XGBoost model showed good discrimination with the AUC, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of 0.981, 0.991, 0.962, 1.000, 1.000, and 0.989, respectively, followed by SVM [AUC =0.985, 95% confidence interval (CI): 0.970–1.000], Lasso-LR (AUC =0.977, 95% CI: 0.958–0.996), LR (AUC =0.936, 95% CI: 0.905–0.968), and RF (AUC =0.920, 95% CI: 0.870–0.970). Validation of the model showed that SVM yielded the highest AUC (0.977, 95% CI: 0.952–1.000), followed by Lasso-LR (AUC =0.959, 95% CI: 0.921–0.997), XGBoost (AUC =0.958, 95% CI: 0.902–1.000), LR (AUC =0.932, 95% CI: 0.878–0.987), and RF (AUC =0.868, 95% CI: 0.779–0.958) in the testing dataset.

Conclusions

Our ML-based models had good discrimination in predicting patients with obstructed hydronephrosis at high risk of harboring pyonephrosis, and the use of these models may be greatly beneficial to urologists in treatment planning, patient selection, and decision-making.

Identification of potentially undiagnosed patients with nontuberculous mycobacterial lung disease using machine learning applied to primary care data in the UK

Nontuberculous mycobacterial lung disease (NTMLD) is a rare lung disease often missed due to a low index of suspicion and unspecific clinical presentation. This retrospective study was designed to characterise the prediagnosis features of NTMLD patients in primary care and to assess the feasibility of using machine learning to identify undiagnosed NTMLD patients.IQVIA Medical Research Data (incorporating THIN, a Cegedim Database), a UK electronic medical records primary care database was used. NTMLD patients were identified between 2003 and 2017 by diagnosis in primary or secondary care or record of NTMLD treatment regimen. Risk factors and treatments were extracted in the prediagnosis period, guided by literature and expert clinical opinion. The control population was enriched to have at least one of these features.741 NTMLD and 112 784 control patients were selected. Annual prevalence rates of NTMLD from 2006 to 2016 increased from 2.7 to 5.1 per 100 000. The most common pre-existing diagnoses and treatments for NTMLD patients were COPD and asthma and penicillin, macrolides and inhaled corticosteroids. Compared to random testing, machine learning improved detection of patients with NTMLD by almost a thousand-fold with AUC of 0.94. The total prevalence of diagnosed and undiagnosed cases of NTMLD in 2016 was estimated to range between 9 and 16 per 100 000.This study supports the feasibility of machine learning applied to primary care data to screen for undiagnosed NTMLD patients, with results indicating that there may be a substantial number of undiagnosed cases of NTMLD in the UK.

Machine learning and atherosclerotic cardiovascular disease risk prediction in a multi-ethnic population

The pooled cohort equations (PCE) predict atherosclerotic cardiovascular disease (ASCVD) risk in patients with characteristics within prespecified ranges and has uncertain performance among Asians or Hispanics. It is unknown if machine learning (ML) models can improve ASCVD risk prediction across broader diverse, real-world populations. We developed ML models for ASCVD risk prediction for multi-ethnic patients using an electronic health record (EHR) database from Northern California. Our cohort included patients aged 18 years or older with no prior CVD and not on statins at baseline (n = 262,923), stratified by PCE-eligible (n = 131,721) or PCE-ineligible patients based on missing or out-of-range variables. We trained ML models [logistic regression with L2 penalty and L1 lasso penalty, random forest, gradient boosting machine (GBM), extreme gradient boosting] and determined 5-year ASCVD risk prediction, including with and without incorporation of additional EHR variables, and in Asian and Hispanic subgroups. A total of 4309 patients had ASCVD events, with 2077 in PCE-ineligible patients. GBM performance in the full cohort, including PCE-ineligible patients (area under receiver-operating characteristic curve (AUC) 0.835, 95% confidence interval (CI): 0.825-0.846), was significantly better than that of the PCE in the PCE-eligible cohort (AUC 0.775, 95% CI: 0.755-0.794). Among patients aged 40-79, GBM performed similarly before (AUC 0.784, 95% CI: 0.759-0.808) and after (AUC 0.790, 95% CI: 0.765-0.814) incorporating additional EHR data. Overall, ML models achieved comparable or improved performance compared to the PCE while allowing risk discrimination in a larger group of patients including PCE-ineligible patients. EHR-trained ML models may help bridge important gaps in ASCVD risk prediction.

Detection and Severity Assessment of Peripheral Occlusive Artery Disease via Deep Learning Analysis of Arterial Pulse Waveforms: Proof-of-Concept and Potential Challenges

Toward the ultimate goal of affordable and non-invasive screening of peripheral occlusive artery disease (PAD), the objective of this work is to investigate the potential of deep learning-based arterial pulse waveform analysis in detecting and assessing the severity of PAD. Using an established transmission line model of arterial hemodynamics, a large number of virtual patients associated with PAD of a wide range of severity and the corresponding arterial pulse waveform data were created. A deep convolutional neural network capable of detecting and assessing the severity of PAD based on the analysis of brachial and ankle arterial pulse waveforms was constructed, evaluated for efficacy, and compared with the state-of-the-art ankle-brachial index (ABI) using the virtual patients. The results suggested that deep learning may diagnose PAD more accurately and robustly than ABI. In sum, this work demonstrates the initial proof-of-concept of deep learning-based arterial pulse waveform analysis for affordable and convenient PAD screening as well as presents challenges that must be addressed for real-world clinical applications.

Resolving Clinical Phenotypes into Endotypes in Allergy: Molecular and Omics Approaches

Allergic diseases are highly complex with respect to pathogenesis, inflammation, and response to treatment. Current efforts for allergic disease diagnosis have focused on clinical evidence as a binary outcome. Although outcome status based on clinical phenotypes (observable characteristics) is convenient and inexpensive to measure in large studies, it does not adequately provide insight into the complex molecular determinants of allergic disease. Individuals with similar clinical diagnoses do not necessarily have similar disease etiologies, natural histories, or responses to treatment. This heterogeneity contributes to the ineffective response to treatment leading to an annual estimated cost of $350 billion in the USA alone. There has been a recent focus to deconvolute the clinical heterogeneity of allergic diseases into specific endotypes using molecular and omics approaches. Endotypes are a means to classify patients based on the underlying pathophysiological mechanisms involving distinct functions or treatment response. The advent of high-throughput molecular omics, immunophenotyping, and bioinformatics methods including machine learning algorithms is facilitating the development of endotype-based diagnosis. As we move to the next decade, we should truly start treating clinical endotypes not clinical phenotype. This review highlights current efforts taking place to improve allergic disease endotyping via molecular omics profiling, immunophenotyping, and machine learning approaches in the context of precision diagnostics in allergic diseases. Graphical Abstract.

Value of a Machine Learning Approach for Predicting Clinical Outcomes in Young Patients With Hypertension

Risk stratification of young patients with hypertension remains challenging. Generally, machine learning (ML) is considered a promising alternative to traditional methods for clinical predictions because it is capable of processing large amounts of complex data. We, therefore, explored the feasibility of an ML approach for predicting outcomes in young patients with hypertension and compared its performance with that of approaches now commonly used in clinical practice. Baseline clinical data and a composite end point—comprising all-cause death, acute myocardial infarction, coronary artery revascularization, new-onset heart failure, new-onset atrial fibrillation/atrial flutter, sustained ventricular tachycardia/ventricular fibrillation, peripheral artery revascularization, new-onset stroke, end-stage renal disease—were evaluated in 508 young patients with hypertension (30.83±6.17 years) who had been treated at a tertiary hospital. Construction of the ML model, which consisted of recursive feature elimination, extreme gradient boosting, and 10-fold cross-validation, was performed at the 33-month follow-up evaluation, and the model’s performance was compared with that of the Cox regression and recalibrated Framingham Risk Score models. An 11-variable combination was considered most valuable for predicting outcomes using the ML approach. The C statistic for identifying patients with composite end points was 0.757 (95% CI, 0.660–0.854) for the ML model, whereas for Cox regression model and the recalibrated Framingham Risk Score model it was 0.723 (95% CI, 0.636–0.810) and 0.529 (95% CI, 0.403–0.655). The ML approach was comparable with Cox regression for determining the clinical prognosis of young patients with hypertension and was better than that of the recalibrated Framingham Risk Score model.

[Artificial intelligence in vascular surgery and vascular medicine]

New digital technologies will also gain in importance in vascular surgery. There is a wide field of potential applications. Simulation-based training of endovascular procedures can lead to improvement in procedure-specific parameters and reduce fluoroscopy and procedural times. The use of intraoperative image-guided navigation and robotics also enables a reduction of the radiation dose. Artificial intelligence can be used for risk stratification and individualization of treatment approaches. Health apps can be used to improve the follow-up care of patients.

Nationwide prediction of type 2 diabetes comorbidities

Identification of individuals at risk of developing disease comorbidities represents an important task in tackling the growing personal and societal burdens associated with chronic diseases. We employed machine learning techniques to investigate to what extent data from longitudinal, nationwide Danish health registers can be used to predict individuals at high risk of developing type 2 diabetes (T2D) comorbidities. Leveraging logistic regression-, random forest- and gradient boosting models and register data spanning hospitalizations, drug prescriptions and contacts with primary care contractors from >200,000 individuals newly diagnosed with T2D, we predicted five-year risk of heart failure (HF), myocardial infarction (MI), stroke (ST), cardiovascular disease (CVD) and chronic kidney disease (CKD). For HF, MI, CVD, and CKD, register-based models outperformed a reference model leveraging canonical individual characteristics by achieving area under the receiver operating characteristic curve improvements of 0.06, 0.03, 0.04, and 0.07, respectively. The top 1,000 patients predicted to be at highest risk exhibited observed incidence ratios exceeding 4.99, 3.52, 1.97 and 4.71 respectively. In summary, prediction of T2D comorbidities utilizing Danish registers led to consistent albeit modest performance improvements over reference models, suggesting that register data could be leveraged to systematically identify individuals at risk of developing disease comorbidities.

Comparing Logistic Regression Models with Alternative Machine Learning Methods to Predict the Risk of Drug Intoxication Mortality

(1) Medical research has shown an increasing interest in machine learning, permitting massive multivariate data analysis. Thus, we developed drug intoxication mortality prediction models, and compared machine learning models and traditional logistic regression. (2) Categorized as drug intoxication, 8,937 samples were extracted from the Korea Centers for Disease Control and Prevention (2008-2017). We trained, validated, and tested each model through data and compared their performance using three measures: Brier score, calibration slope, and calibration-in-the-large. (3) A chi-square test demonstrated that mortality risk statistically significantly differed according to severity, intent, toxic substance, age, and sex. The multilayer perceptron model (MLP) had the highest area under the curve (AUC), and lowest Brier score in training and validation phases, while the logistic regression model (LR) showed the highest AUC (0.827) and lowest Brier score (0.0307) in the testing phase. MLP also had the second-highest AUC (0.816) and second-lowest Brier score (0.003258) in the testing phase, demonstrating better performance than the decision-making tree model. (4) Given the complexity of choosing tuning parameters, LR proved competitive when using medical datasets, which require strict accuracy.

Development and External Validation of a Deep Learning Algorithm for Prognostication of Cardiovascular Outcomes

Background and Objectives

We aim to explore the additional discriminative accuracy of a deep learning (DL) algorithm using repeated-measures data for identifying people at high risk for cardiovascular disease (CVD), compared to Cox hazard regression.

Methods

Two CVD prediction models were developed from National Health Insurance Service-Health Screening Cohort (NHIS-HEALS): a Cox regression model and a DL model. Performance of each model was assessed in the internal and 2 external validation cohorts in Koreans (National Health Insurance Service-National Sample Cohort; NHIS-NSC) and in Europeans (Rotterdam Study). A total of 412,030 adults in the NHIS-HEALS; 178,875 adults in the NHIS-NSC; and the 4,296 adults in Rotterdam Study were included.

Results

Mean ages was 52 years (46% women) and there were 25,777 events (6.3%) in NHIS-HEALS during the follow-up. In internal validation, the DL approach demonstrated a C-statistic of 0.896 (95% confidence interval, 0.886–0.907) in men and 0.921 (0.908–0.934) in women and improved reclassification compared with Cox regression (net reclassification index [NRI], 24.8% in men, 29.0% in women). In external validation with NHIS-NSC, DL demonstrated a C-statistic of 0.868 (0.860–0.876) in men and 0.889 (0.876–0.898) in women, and improved reclassification compared with Cox regression (NRI, 24.9% in men, 26.2% in women). In external validation applied to the Rotterdam Study, DL demonstrated a C-statistic of 0.860 (0.824–0.897) in men and 0.867 (0.830–0.903) in women, and improved reclassification compared with Cox regression (NRI, 36.9% in men, 31.8% in women).

Conclusions

A DL algorithm exhibited greater discriminative accuracy than Cox model approaches.

Trial Registration

ClinicalTrials.gov Identifier: NCT02931500

A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models

OBJECTIVES

The objective of this study was to compare performance of logistic regression (LR) with machine learning (ML) for clinical prediction modeling in the literature.

STUDY DESIGN AND SETTING

We conducted a Medline literature search (1/2016 to 8/2017) and extracted comparisons between LR and ML models for binary outcomes.

RESULTS

We included 71 of 927 studies. The median sample size was 1,250 (range 72-3,994,872), with 19 predictors considered (range 5-563) and eight events per predictor (range 0.3-6,697). The most common ML methods were classification trees, random forests, artificial neural networks, and support vector machines. In 48 (68%) studies, we observed potential bias in the validation procedures. Sixty-four (90%) studies used the area under the receiver operating characteristic curve (AUC) to assess discrimination. Calibration was not addressed in 56 (79%) studies. We identified 282 comparisons between an LR and ML model (AUC range, 0.52-0.99). For 145 comparisons at low risk of bias, the difference in logit(AUC) between LR and ML was 0.00 (95% confidence interval, -0.18 to 0.18). For 137 comparisons at high risk of bias, logit(AUC) was 0.34 (0.20-0.47) higher for ML.

CONCLUSION

We found no evidence of superior performance of ML over LR. Improvements in methodology and reporting are needed for studies that compare modeling algorithms.

Finding missed cases of familial hypercholesterolemia in health systems using machine learning

Familial hypercholesterolemia (FH) is an underdiagnosed dominant genetic condition affecting approximately 0.4% of the population and has up to a 20-fold increased risk of coronary artery disease if untreated. Simple screening strategies have false positive rates greater than 95%. As part of the FH Foundation's FIND FH initiative, we developed a classifier to identify potential FH patients using electronic health record (EHR) data at Stanford Health Care. We trained a random forest classifier using data from known patients (n = 197) and matched non-cases (n = 6590). Our classifier obtained a positive predictive value (PPV) of 0.88 and sensitivity of 0.75 on a held-out test-set. We evaluated the accuracy of the classifier's predictions by chart review of 100 patients at risk of FH not included in the original dataset. The classifier correctly flagged 84% of patients at the highest probability threshold, with decreasing performance as the threshold lowers. In external validation on 466 FH patients (236 with genetically proven FH) and 5000 matched non-cases from the Geisinger Healthcare System our FH classifier achieved a PPV of 0.85. Our EHR-derived FH classifier is effective in finding candidate patients for further FH screening. Such machine learning guided strategies can lead to effective identification of the highest risk patients for enhanced management strategies.

Predicting Future Cardiovascular Events in Patients With Peripheral Artery Disease Using Electronic Health Record Data

BACKGROUND

Patients with peripheral artery disease (PAD) are at risk of major adverse cardiac and cerebrovascular events. There are no readily available risk scores that can accurately identify which patients are most likely to sustain an event, making it difficult to identify those who might benefit from more aggressive intervention. Thus, we aimed to develop a novel predictive model-using machine learning methods on electronic health record data-to identify which PAD patients are most likely to develop major adverse cardiac and cerebrovascular events.

METHODS AND RESULTS

Data were derived from patients diagnosed with PAD at 2 tertiary care institutions. Predictive models were built using a common data model that allowed for utilization of both structured (coded) and unstructured (text) data. Only data from time of entry into the health system up to PAD diagnosis were used for modeling. Models were developed and tested using nested cross-validation. A total of 7686 patients were included in learning our predictive models. Utilizing almost 1000 variables, our best predictive model accurately determined which PAD patients would go on to develop major adverse cardiac and cerebrovascular events with an area under the curve of 0.81 (95% CI, 0.80-0.83).

CONCLUSIONS

Machine learning algorithms applied to data in the electronic health record can learn models that accurately identify PAD patients at risk of future major adverse cardiac and cerebrovascular events, highlighting the great potential of electronic health records to provide automated risk stratification for cardiovascular diseases. Common data models that can enable cross-institution research and technology development could potentially be an important aspect of widespread adoption of newer risk-stratification models.

Leveraging the Electronic Health Record to Create an Automated Real-Time Prognostic Tool for Peripheral Arterial Disease

Background Automated individualized risk prediction tools linked to electronic health records ( EHR s) are not available for management of patients with peripheral arterial disease. The goal of this study was to create a prognostic tool for patients with peripheral arterial disease using data elements automatically extracted from an EHR to enable real-time and individualized risk prediction at the point of care. Methods and Results A previously validated phenotyping algorithm was deployed to an EHR linked to the Rochester Epidemiology Project to identify peripheral arterial disease cases from Olmsted County, MN, for the years 1998 to 2011. The study cohort was composed of 1676 patients: 593 patients died over 5-year follow-up. The c-statistic for survival in the overall data set was 0.76 (95% confidence interval [CI], 0.74-0.78), and the c-statistic across 10 cross-validation data sets was 0.75 (95% CI, 0.73-0.77). Stratification of cases demonstrated increasing mortality risk by subgroup (low: hazard ratio, 0.35 [95% CI, 0.21-0.58]; intermediate-high: hazard ratio, 2.98 [95% CI, 2.37-3.74]; high: hazard ratio, 8.44 [95% CI, 6.66-10.70], all P<0.0001 versus the reference subgroup). An equation for risk calculation was derived from Cox model parameters and β estimates. Big data infrastructure enabled deployment of the real-time risk calculator to the point of care via the EHR . Conclusions This study demonstrates that electronic tools can be deployed to EHR s to create automated real-time risk calculators to predict survival of patients with peripheral arterial disease. Moreover, the prognostic model developed may be translated to patient care as an automated and individualized real-time risk calculator deployed at the point of care.

A comparison of logistic regression models with alternative machine learning methods to predict the risk of in-hospital mortality in emergency medical admissions via external validation

We compare the performance of logistic regression with several alternative machine learning methods to estimate the risk of death for patients following an emergency admission to hospital based on the patients' first blood test results and physiological measurements using an external validation approach. We trained and tested each model using data from one hospital (n = 24,696) and compared the performance of these models in data from another hospital (n = 13,477). We used two performance measures - the calibration slope and area under the receiver operating characteristic curve. The logistic model performed reasonably well - calibration slope: 0.90, area under the receiver operating characteristic curve: 0.847 compared to the other machine learning methods. Given the complexity of choosing tuning parameters of these methods, the performance of logistic regression with transformations for in-hospital mortality prediction was competitive with the best performing alternative machine learning methods with no evidence of overfitting.

Predicting hospital and emergency department utilization among community-dwelling older adults: Statistical and machine learning approaches

OBJECTIVE

The objective of this study was to compare the performance of several commonly used machine learning methods to traditional statistical methods for predicting emergency department and hospital utilization among patients receiving publicly-funded home care services.

STUDY DESIGN AND SETTING

We conducted a population-based retrospective cohort study of publicly-funded home care recipients in the Hamilton-Niagara-Haldimand-Brant region of southern Ontario, Canada between 2014 and 2016. Gradient boosted trees, neural networks, and random forests were tested against two variations of logistic regression for predicting three outcomes related to emergency department and hospital utilization within six months of a comprehensive home care clinical assessment. Models were trained on data from years 2014 and 2015 and tested on data from 2016. Performance was compared using logarithmic score, Brier score, AUC, and diagnostic accuracy measures.

RESULTS

Gradient boosted trees achieved the best performance on all three outcomes. Gradient boosted trees provided small but statistically significant performance gains over both traditional methods on all three outcomes, while neural networks significantly outperformed logistic regression on two of three outcomes. However, sensitivity and specificity gains from using gradient boosted trees over logistic regression were only in the range of 1%-2% at several classification thresholds.

CONCLUSION

Gradient boosted trees and simple neural networks yielded small performance benefits over logistic regression for predicting emergency department and hospital utilization among patients receiving publicly-funded home care. However, the performance benefits were of negligible clinical importance.

Machine learning models in electronic health records can outperform conventional survival models for predicting patient mortality in coronary artery disease

Prognostic modelling is important in clinical practice and epidemiology for patient management and research. Electronic health records (EHR) provide large quantities of data for such models, but conventional epidemiological approaches require significant researcher time to implement. Expert selection of variables, fine-tuning of variable transformations and interactions, and imputing missing values are time-consuming and could bias subsequent analysis, particularly given that missingness in EHR is both high, and may carry meaning. Using a cohort of 80,000 patients from the CALIBER programme, we compared traditional modelling and machine-learning approaches in EHR. First, we used Cox models and random survival forests with and without imputation on 27 expert-selected, preprocessed variables to predict all-cause mortality. We then used Cox models, random forests and elastic net regression on an extended dataset with 586 variables to build prognostic models and identify novel prognostic factors without prior expert input. We observed that data-driven models used on an extended dataset can outperform conventional models for prognosis, without data preprocessing or imputing missing values. An elastic net Cox regression based with 586 unimputed variables with continuous values discretised achieved a C-index of 0.801 (bootstrapped 95% CI 0.799 to 0.802), compared to 0.793 (0.791 to 0.794) for a traditional Cox model comprising 27 expert-selected variables with imputation for missing values. We also found that data-driven models allow identification of novel prognostic variables; that the absence of values for particular variables carries meaning, and can have significant implications for prognosis; and that variables often have a nonlinear association with mortality, which discretised Cox models and random forests can elucidate. This demonstrates that machine-learning approaches applied to raw EHR data can be used to build models for use in research and clinical practice, and identify novel predictive variables and their effects to inform future research.

Use of machine learning to predict early biochemical recurrence after robot-assisted prostatectomy

OBJECTIVES

To train and compare machine-learning algorithms with traditional regression analysis for the prediction of early biochemical recurrence after robot-assisted prostatectomy.

PATIENTS AND METHODS

A prospectively collected dataset of 338 patients who underwent robot-assisted prostatectomy for localized prostate cancer was examined. We used three supervised machine-learning algorithms and 19 different training variables (demographic, clinical, imaging and operative data) in a hypothesis-free manner to build models that could predict patients with biochemical recurrence at 1 year. We also performed traditional Cox regression analysis for comparison.

RESULTS

K-nearest neighbour, logistic regression and random forest classifier were used as machine-learning models. Classic Cox regression analysis had an area under the curve (AUC) of 0.865 for the prediction of biochemical recurrence. All three of our machine-learning models (K-nearest neighbour (AUC 0.903), random forest tree (AUC 0.924) and logistic regression (AUC 0.940) outperformed the conventional statistical regression model. Accuracy prediction scores for K-nearest neighbour, random forest tree and logistic regression were 0.976, 0.953 and 0.976, respectively.

CONCLUSIONS

Machine-learning techniques can produce accurate disease predictability better that traditional statistical regression. These tools may prove clinically useful for the automated prediction of patients who develop early biochemical recurrence after robot-assisted prostatectomy. For these patients, appropriate individualized treatment options can improve outcomes and quality of life.

Enhanced neonatal surgical site infection prediction model utilizing statistically and clinically significant variables in combination with a machine learning algorithm

BACKGROUND

Machine-learning can elucidate complex relationships/provide insight to important variables for large datasets. This study aimed to develop an accurate model to predict neonatal surgical site infections (SSI) using different statistical methods.

METHODS

The 2012-2015 National Surgical Quality Improvement Program-Pediatric for neonates was utilized for development and validations models. The primary outcome was any SSI. Models included different algorithms: full multiple logistic regression (LR), a priori clinical LR, random forest classification (RFC), and a hybrid model (combination of clinical knowledge and significant variables from RF) to maximize predictive power.

RESULTS

16,842 patients (median age 18 days, IQR 3-58) were included. 542 SSIs (4%) were identified. Agreement was observed for multiple covariates among significant variables between models. Area under the curve for each model was similar (full model 0.65, clinical model 0.67, RF 0.68, hybrid LR 0.67); however, the hybrid model utilized the fewest variables (18).

CONCLUSIONS

The hybrid model had similar predictability as other models with fewer and more clinically relevant variables. Machine-learning algorithms can identify important novel characteristics, which enhance clinical prediction models.

Maternal exposure to ambient PM10 during pregnancy increases the risk of congenital heart defects: Evidence from machine learning models

Previous research suggested an association between maternal exposure to ambient air pollutants and risk of congenital heart defects (CHDs), though the effects of particulate matter ≤10μm in aerodynamic diameter (PM₁₀) on CHDs are inconsistent. We used two machine learning models (i.e., random forest (RF) and gradient boosting (GB)) to investigate the non-linear effects of PM₁₀ exposure during the critical time window, weeks 3-8 in pregnancy, on risk of CHDs. From 2009 through 2012, we carried out a population-based birth cohort study on 39,053 live-born infants in Beijing. RF and GB models were used to calculate odds ratios for CHDs associated with increase in PM₁₀ exposure, adjusting for maternal and perinatal characteristics. Maternal exposure to PM₁₀ was identified as the primary risk factor for CHDs in all machine learning models. We observed a clear non-linear effect of maternal exposure to PM₁₀ on CHDs risk. Compared to 40μgm^-3, the following odds ratios resulted: 1) 92μgm^-3 [RF: 1.16 (95% CI: 1.06, 1.28); GB: 1.26 (95% CI: 1.17, 1.35)]; 2) 111μgm^-3 [RF: 1.04 (95% CI: 0.96, 1.14); GB: 1.04 (95% CI: 0.99, 1.08)]; 3) 124μgm^-3 [RF: 1.01 (95% CI: 0.94, 1.10); GB: 0.98 (95% CI: 0.93, 1.02)]; 4) 190μgm^-3 [RF: 1.29 (95% CI: 1.14, 1.44); GB: 1.71 (95% CI: 1.04, 2.17)]. Overall, both machine models showed an association between maternal exposure to ambient PM₁₀ and CHDs in Beijing, highlighting the need for non-linear methods to investigate dose-response relationships.

Integrated Genomic Medicine: A Paradigm for Rare Diseases and Beyond

Individualized medicine, or the tailoring of therapeutic interventions to a patient's unique genetic, biochemical, physiological, exposure and behavioral profile, has been enhanced, if not enabled, by modern biomedical technologies such as high-throughput DNA sequencing platforms, induced pluripotent stem cell assays, biomarker discovery protocols, imaging modalities, and wireless monitoring devices. Despite successes in the isolated use of these technologies, however, it is arguable that their combined and integrated use in focused studies of individual patients is the best way to not only tailor interventions for those patients, but also shed light on treatment strategies for patients with similar conditions. This is particularly true for individuals with rare diseases since, by definition, they will require study without recourse to other individuals, or at least without recourse to many other individuals. Such integration and focus will require new biomedical scientific paradigms and infrastructure, including the creation of databases harboring study results, the formation of dedicated multidisciplinary research teams and new training programs. We consider the motivation and potential for such integration, point out areas in need of improvement, and argue for greater emphasis on improving patient health via technological innovations, not merely improving the technologies themselves. We also argue that the paradigm described can, in theory, be extended to the study of individuals with more common diseases.