Machine learning methods in predicting the risk of malignant transformation of oral potentially malignant disorders: A systematic review

BACKGROUND

Oral Potentially Malignant Disorders (OPMDs) refer to a heterogenous group of clinical presentations with heightened rate of malignant transformation. Identification of risk levels in OPMDs is crucial to determine the need for active intervention in high-risk patients and routine follow-up in low-risk ones. Machine learning models has shown tremendous potential in several areas of dentistry that strongly suggest its application to estimate rate of malignant transformation of precancerous lesions.

METHODS

A comprehensive literature search was performed on Pubmed/MEDLINE, Web of Science, Scopus, Embase, Cochrane Library database to identify articles including machine learning models and algorithms to predict malignant transformation in OPMDs. Relevant bibliographic data, study characteristics, and outcomes were extracted for eligible studies. Quality of the included studies was assessed through the IJMEDI checklist.

RESULTS

Fifteen articles were found suitable for the review as per the PECOS criteria. Amongst all studies, highest sensitivity (100%) was recorded for U-net architecture, Peaks Random forest model, and Partial least squares discriminant analysis (PLSDA). Highest specificity (100%) was noted for PLSDA. Range of overall accuracy in risk prediction was between 95.4% and 74%.

CONCLUSION

Machine learning proved to be a viable tool in risk prediction, demonstrating heightened sensitivity, automation, and improved accuracy for predicting transformation of OPMDs. It presents an effective approach for incorporating multiple variables to monitor the progression of OPMDs and predict their malignant potential. However, its sensitivity to dataset characteristics necessitates the optimization of input parameters to maximize the efficiency of the classifiers.

Predicting the Need for Blood Transfusions in Cardiac Surgery: A Comparison between Machine Learning Algorithms and Established Risk Scores in the Brazilian Population

INTRODUCTION

Blood transfusion is a common practice in cardiac surgery, despite its well-known negative effects. To mitigate blood transfusion-associated risks, identifying patients who are at higher risk of needing this procedure is crucial. Widely used risk scores to predict the need for blood transfusions have yielded unsatisfactory results when validated for the Brazilian population.

METHODS

In this retrospective study, machine learning (ML) algorithms were compared to predict the need for blood transfusions in a cohort of 495 cardiac surgery patients treated at a Brazilian reference service between 2019 and 2021. The performance of the models was evaluated using various metrics, including the area under the curve (AUC), and compared to the commonly used Transfusion Risk and Clinical Knowledge (TRACK) and Transfusion Risk Understanding Scoring Tool (TRUST) scoring systems.

RESULTS

The study found that the model had the highest performance, achieving an AUC of 0.7350 (confidence interval [CI]: 0.7203 to 0.7497). Importantly, all ML algorithms performed significantly better than the commonly used TRACK and TRUST scoring systems. TRACK had an AUC of 0.6757 (CI: 0.6609 to 0.6906), while TRUST had an AUC of 0.6622 (CI: 0.6473 to 0.6906).

CONCLUSION

The findings of this study suggest that ML algorithms may offer a more accurate prediction of the need for blood transfusions than the traditional scoring systems and could enhance the accuracy of predicting blood transfusion requirements in cardiac surgery patients. Further research could focus on optimizing and refining ML algorithms to improve their accuracy and make them more suitable for clinical use.

Incorporating longitudinal history of risk factors into atherosclerotic cardiovascular disease risk prediction using deep learning

Background

It is increasingly clear that longitudinal risk factor levels and trajectories are related to risk for atherosclerotic cardiovascular disease (ASCVD) above and beyond single measures. Currently used in clinical care, the Pooled Cohort Equations (PCE) are based on regression methods that predict ASCVD risk based on cross-sectional risk factor levels. Deep learning (DL) models have been developed to incorporate longitudinal data for risk prediction but its benefit for ASCVD risk prediction relative to the traditional Pooled Cohort Equations (PCE) remain unknown.

Objective

To develop a ASCVD risk prediction model that incorporates longitudinal risk factors using deep learning.

Methods

Our study included 15,565 participants from four cardiovascular disease cohorts free of baseline ASCVD who were followed for adjudicated ASCVD. Ten-year ASCVD risk was calculated in the training set using our benchmark, the PCE, and a longitudinal DL model, Dynamic-DeepHit. Predictors included those incorporated in the PCE: sex, race, age, total cholesterol, high density lipid cholesterol, systolic and diastolic blood pressure, diabetes, hypertension treatment and smoking. The discrimination and calibration performance of the two models were evaluated in an overall hold-out testing dataset.

Results

Of the 15,565 participants in our dataset, 2,170 (13.9%) developed ASCVD. The performance of the longitudinal DL model that incorporated 8 years of longitudinal risk factor data improved upon that of the PCE [AUROC: 0.815 (CI: 0.782-0.844) vs 0.792 (CI: 0.760-0.825)] and the net reclassification index was 0.385. The brier score for the DL model was 0.0514 compared with 0.0542 in the PCE.

Conclusion

Incorporating longitudinal risk factors in ASCVD risk prediction using DL can improve model discrimination and calibration.

PREDICTORS OF AMBULATORY STATUS AT ONE YEAR FOLLOWING MAJOR LOWER EXTREMITY AMPUTATION

Nearly 60,000 major lower extremity amputations (AKA/BKA) are performed annually in the United States. We created a simple risk score for predicting ambulation at one-year following AKA/BKA. We queried the Vascular Quality Initiative amputation database for patients who underwent above-knee (AKA) or below-knee (BKA) amputation (2013-2018). The primary endpoint was ambulation at one year either independently or with assistance. The cohort was divided into 80% derivation and 20% validation. Using the derivation set, a multivariable model identified preoperatively available independent predictors of 1-year ambulation and an integer-based risk-score was created. Scores were calculated to assign patients to risk groups - low, medium, or high chance of being ambulatory at one year. Internal validation was performed by applying the risk score to the validation set. Of 8,725 AKA/BKA, 2055 met inclusion criteria - excluded: 2644 nonambulatory prior to amputation, 3753 missing 1-year follow-up ambulatory status. The majority - n=1366, 66% - were BKAs. The indications were CLTI; 47%, ischemic tissue loss; 9%, ischemic rest pain; 35%, infection/neuropathic; 9%, acute limb ischemia. Ambulation at one year was higher for BKA than AKA: 67%, versus 50%, p<0.0001. In the final prediction model, contralateral BKA/AKA was the strongest predictor of NON-ambulation. The score provided reasonable discrimination (c-statistic=0.65) and was well calibrated (Hosmer-Lemeshow p=0.24). 62% of patients who were ambulatory preoperatively remained ambulatory at one year. An integer-based risk score can stratify patients according to chance of ambulation at one year after major amputation and may be useful for preoperative patient counseling and selection.

Multimodal Risk Prediction with Physiological Signals, Medical Images and Clinical Notes

The broad adoption of electronic health records (EHRs) provides great opportunities to conduct healthcare research and solve various clinical problems in medicine. With recent advances and success, methods based on machine learning and deep learning have become increasingly popular in medical informatics. Combining data from multiple modalities may help in predictive tasks. To assess the expectations of multimodal data, we introduce a comprehensive fusion framework designed to integrate temporal variables, medical images, and clinical notes in Electronic Health Record (EHR) for enhanced performance in downstream predictive tasks. Early, joint, and late fusion strategies were employed to effectively combine data from various modalities. Model performance and contribution scores show that multimodal models outperform uni-modal models in various tasks. Additionally, temporal signs contain more information than CXR images and clinical notes in three explored predictive tasks. Therefore, models integrating different data modalities can work better in predictive tasks.

Prediction of acute organophosphate poisoning severity using machine learning techniques

Poisoning with organophosphate compounds is a significant public health risk, especially in developing countries. Considering the importance of early and accurate prediction of organophosphate poisoning prognosis, the aim of this study was to develop a machine learning-based prediction model to predict the severity of organophosphate poisoning. The data of patients with organophosphate poisoning were retrospectively extracted and split into training and test sets in a ratio of 70:30. The feature selection was done by least absolute shrinkage and selection operator method. Selected features were fed into five machine learning techniques, including Histogram Boosting Gradient, eXtreme Gradient Boosting, K-Nearest Neighborhood, Support Vector Machine (SVM) (kernel = linear), and Random Forest. The Scikit-learn library in Python programming language was used to implement the models. Finally, the performance of developed models was measured using ten-fold cross-validation methods and some evaluation criteria with 95 % confidence intervals. A total of 1237 patients were used to train and test the machine learning models. According to the criteria determining severe organophosphate poisoning, 732 patients were assigned to group 1 (patients with mild to moderate poisoning) and 505 patients were assigned to group 2 (patients with severe poisoning). With an AUC value of 0.907 (95 % CI 0.89-0.92), the model developed using XGBoost outperformed other models. Feature importance evaluation found that venous blood gas-pH, white blood cells, and plasma cholinesterase activity were the top three variables that contribute the most to the prediction performance of the prognosis in patients with organophosphate poisoning. XGBoost model yield an accuracy of 90.1 % (95 % CI 0.891-0.918), specificity of 91.4 % (95 % CI 0.90-0.92), a sensitivity of 89.5 % (95 % CI 0.87-0.91), F-measure of 91.2 % (95 % CI 0.90-0.921), and Kappa statistic of 91.2 % (95 % CI 0.90-0.92). The machine learning-based prediction models can accurately predict the severity of organophosphate poisoning. Based on feature selection techniques, the most important predictors of organophosphate poisoning were VBG-pH, white blood cell count, plasma cholinesterase activity, VBG-BE, and age. The best algorithm with the highest predictive performance was the XGBoost classifier.

The Natural History of Asymptomatic Gallstones: A Longitudinal Study and Prediction Model

BACKGROUND & AIMS

Despite the high prevalence of asymptomatic gallstones (AGs), there are limited data on their natural history. We aimed to determine the rate of symptom development in a contemporary population, determine factors associated with progression to symptomatic gallstones (SGs), and develop a clinical prediction model.

METHODS

We used a retrospective cohort design. The time to first SG was shown using Kaplan-Meier curves. Multivariable competing risk (death) regression analysis was used to identify variables associated with SGs. A prediction model for the development of SGs after 10 years was generated and calibration curves were plotted. Participants were patients with AGs based on ultrasound or computed tomography from the general medical population.

RESULTS

From 1996 to 2016, 22,257 patients (51% female) with AGs were identified; 14.5% developed SG with a median follow-up period of 4.6 years. The cumulative incidence was 10.1% (±0.22%) at 5 years, 21.5% (±0.39%) at 10 years, and 32.6% (±0.83%) at 15 years. In a multivariable model, the strongest predictors of developing SGs were female gender (hazard ratio [HR], 1.50; 95% CI, 1.39-1.61), younger age (HR per 5 years, 1.15; 95% CI, 1.14-1.16), multiple stones (HR, 2.42; 95% CI, 2.25-2.61), gallbladder polyps (HR, 2.55; 95% CI, 2.14-3.05), large stones (HR, 2.03; 95% CI, 1.80-2.29), and chronic hemolytic anemia (HR, 1.90; 95% CI, 1.33-2.72). The model showed good discrimination (C-statistic, 0.70) and calibration.

CONCLUSIONS

In general medical patients with AGs, symptoms developed at approximately 2% per year. A predictive model with good calibration could be used to inform patients of their risk of SGs.

The Measurement and Interpretation of Fibroblast Growth Factor 23 (FGF23) Concentrations

Two decades after the discovery of the hormone FGF23, we know more about phosphate homeostasis as it turned out that FGF23 is the central hormone that regulates this. Hereditary hypophosphatemic rickets and tumor-induced osteomalacia could by then be explained, by autonomous FGF23 production, and the nephrology field was excited by this new marker as it turned out to be independently associated with mortality in people treated by hemodialysis. This led to the development of several immunoassays to be able to measure FGF23 in blood. In the past years we learned that FGF23 is a rather stable peptide, the precision of the assays is acceptable but assays are not standardized and therefore not comparable. This means that reference values and cutoff values need to be assay specific. For several assays reference values have been established and gender and age did not seem of high importance. The phosphate content of the diet, which can be culturally dependent, however, should be taken into account when interpreting results, but to what extent is not totally clear. Currently, clinical application of the immunoassays is established in the diagnosis of hereditary hypophosphatemic rickets and diagnosis and follow-up of tumor-induced osteomalacia. Definite conclusions on the usefulness of the FGF23 measurement in people with CKD either as a marker for risk prediction or a as target for treatment remains to be determined. The latter applications would require dedicated prospective clinical trials, which may take years, before providing answers. To improve the standardization of the FGF23 assays and to shed light on the biological functions that fragments might have we might aim for an LC-MS/MS-based method to quantify both intact and fragmented FGF23. In this literature review we will summarize the current knowledge on the physiological role of FGF23, its quantification, and the clinical usefulness of its determination.

Exploring the incremental utility of circulating biomarkers for robust risk prediction of incident atrial fibrillation in European cohorts using regressions and modern machine learning methods

AIMS

To identify robust circulating predictors for incident atrial fibrillation (AF) using classical regressions and machine learning (ML) techniques within a broad spectrum of candidate variables.

METHODS AND RESULTS

In pooled European community cohorts (n = 42 280 individuals), 14 routinely available biomarkers mirroring distinct pathophysiological pathways including lipids, inflammation, renal, and myocardium-specific markers (N-terminal pro B-type natriuretic peptide [NT-proBNP], high-sensitivity troponin I [hsTnI]) were examined in relation to incident AF using Cox regressions and distinct ML methods. Of 42 280 individuals (21 843 women [51.7%]; median [interquartile range, IQR] age, 52.2 [42.7, 62.0] years), 1496 (3.5%) developed AF during a median follow-up time of 5.7 years. In multivariable-adjusted Cox-regression analysis, NT-proBNP was the strongest circulating predictor of incident AF [hazard ratio (HR) per standard deviation (SD), 1.93 (95% CI, 1.82-2.04); P < 0.001]. Further, hsTnI [HR per SD, 1.18 (95% CI, 1.13-1.22); P < 0.001], cystatin C [HR per SD, 1.16 (95% CI, 1.10-1.23); P < 0.001], and C-reactive protein [HR per SD, 1.08 (95% CI, 1.02-1.14); P = 0.012] correlated positively with incident AF. Applying various ML techniques, a high inter-method consistency of selected candidate variables was observed. NT-proBNP was identified as the blood-based marker with the highest predictive value for incident AF. Relevant clinical predictors were age, the use of antihypertensive medication, and body mass index.

CONCLUSION

Using different variable selection procedures including ML methods, NT-proBNP consistently remained the strongest blood-based predictor of incident AF and ranked before classical cardiovascular risk factors. The clinical benefit of these findings for identifying at-risk individuals for targeted AF screening needs to be elucidated and tested prospectively.

Performance of multilabel machine learning models and risk stratification schemas for predicting stroke and bleeding risk in patients with non-valvular atrial fibrillation

BACKGROUND

Appropriate anticoagulant therapy for patients with atrial fibrillation (AF) requires assessment of stroke and bleeding risks. However, risk stratification schemas such as CHA2DS2-VASc and HAS-BLED have modest predictive capacity for patients with AF. Multilabel machine learning (ML) techniques may improve predictive performance and support decision-making for anticoagulant therapy. We compared the performance of multilabel ML models with the currently used risk scores for predicting outcomes in AF patients.

METHODS

This was a retrospective cohort study of 9670 patients, mean age 76.9 years, 46% women, who were hospitalized with non-valvular AF, and had 1-year follow-up. The outcomes were ischemic stroke (167), major bleeding (430) admissions, all-cause death (1912) and event-free survival (7387). Discrimination and calibration of ML models were compared with clinical risk scores by area under the curve (AUC). Risk stratification was assessed using net reclassification index (NRI).

RESULTS

Multilabel gradient boosting classifier chain provided the best AUCs for stroke (0.685 95% CI 0.676, 0.694), major bleeding (0.709 95% CI 0.703, 0.716) and death (0.765 95% CI 0.763, 0.768) compared to multi-layer neural networks and classifier chain using support vector machine. It provided modest performance improvement for stroke compared to AUC of CHA2DS2-VASc (0.652, NRI = 3.2%, p-value = 0.1), but significantly improved major bleeding prediction compared to AUC of HAS-BLED (0.522, NRI = 22.8%, p-value < 0.05). It also achieved greater discriminant power for death compared with AUC of CHA2DS2-VASc (0.606, p-value < 0.05). ML models identified additional risk features such as hemoglobin level, renal function.

CONCLUSIONS

Multilabel ML models can outperform clinical risk stratification scores for predicting the risk of major bleeding and death in non-valvular AF patients.

Efficient Evaluation of Prediction Rules in Semi-Supervised Settings under Stratified Sampling

In many contemporary applications, large amounts of unlabeled data are readily available while labeled examples are limited. There has been substantial interest in semi-supervised learning (SSL) which aims to leverage unlabeled data to improve estimation or prediction. However, current SSL literature focuses primarily on settings where labeled data is selected uniformly at random from the population of interest. Stratified sampling, while posing additional analytical challenges, is highly applicable to many real world problems. Moreover, no SSL methods currently exist for estimating the prediction performance of a fitted model when the labeled data is not selected uniformly at random. In this paper, we propose a two-step SSL procedure for evaluating a prediction rule derived from a working binary regression model based on the Brier score and overall misclassification rate under stratified sampling. In step I, we impute the missing labels via weighted regression with nonlinear basis functions to account for stratified sampling and to improve efficiency. In step II, we augment the initial imputations to ensure the consistency of the resulting estimators regardless of the specification of the prediction model or the imputation model. The final estimator is then obtained with the augmented imputations. We provide asymptotic theory and numerical studies illustrating that our proposals outperform their supervised counterparts in terms of efficiency gain. Our methods are motivated by electronic health record (EHR) research and validated with a real data analysis of an EHR-based study of diabetic neuropathy.

Characteristics Associated with 5-year Fracture Risk vs. 5-year Mortality Risk Among Late-life Men

BACKGROUND

Identifying late-life men who might benefit from treatment to prevent fracture is challenging given high mortality. Our objective was to evaluate risks of clinical fracture, hip fracture, and mortality prior to fracture among men ≥80 years.

METHODS

Study participants included 3,145 community-dwelling men (mean [SD] age 83 [2.8] years) from the Osteoporotic Fractures in Men (MrOS) Study. We used separate multivariable Fine-Gray competing risk models with pre-specified risk factors [age, hip bone mineral density (BMD), recent fracture (<5 years), fall history (previous year), and multimorbidity (# conditions)] to estimate sub-distribution hazard ratios and absolute 5-year risks of any clinical fracture and mortality prior to clinical fracture. Secondary analysis considered hip fracture.

RESULTS

There were 414 incident clinical fractures and 595 deaths without prior fracture within 5 years. BMD, fall history, and recent fracture were strong predictors of clinical fracture. Age and multimorbidity were strong predictors of mortality before fracture. After accounting for competing risks, age, BMD, and fall history were each associated with both risk of hip fracture and mortality before hip fracture. Model discrimination varied from 0.65 (mortality before fracture) to 0.79 (hip fracture). Estimated mortality differed substantially among men with similar clinical fracture risk due to modest correlation between fracture risk and competing mortality risk=0.37.

CONCLUSIONS

In late-life men, strong risk factors for clinical fracture and hip fracture include fall history, BMD, and recent fracture. Osteoporosis drug treatment decisions may be further enhanced by consideration of fracture risk versus overall life expectancy.

Risk Prediction of Pancreatic Cancer in Patients With Abnormal Morphologic Findings Related to Chronic Pancreatitis: A Machine Learning Approach

BACKGROUND AND AIMS

A significant factor contributing to poor survival in pancreatic cancer is the often late stage at diagnosis. We sought to develop and validate a risk prediction model to facilitate the distinction between chronic pancreatitis-related vs potential early pancreatic ductal adenocarcinoma (PDAC)-associated changes on pancreatic imaging.

METHODS

In this retrospective cohort study, patients aged 18-84 years whose abdominal computed tomography/magnetic resonance imaging reports indicated duct dilatation, atrophy, calcification, cyst, or pseudocyst between January 2008 and November 2019 were identified. The outcome of interest is PDAC in 3 years. More than 100 potential predictors were extracted. Random survival forests approach was used to develop and validate risk models. Multivariable Cox proportional hazard model was applied to estimate the effect of the covariates on the risk of PDAC.

RESULTS

The cohort consisted of 46,041 (mean age 66.4 years). The 3-year incidence rate was 4.0 (95% confidence interval CI 3.6-4.4)/1000 person-years of follow-up. The final models containing age, weight change, duct dilatation, and either alkaline phosphatase or total bilirubin had good discrimination and calibration (c-indices 0.81). Patients with pancreas duct dilatation and at least another morphological feature in the absence of calcification had the highest risk (adjusted hazard ratio [aHR] = 14.15, 95% CI 8.7-22.6), followed by patients with calcification and duct dilatation (aHR = 7.28, 95% CI 4.09-12.96), and patients with duct dilation only (aHR = 6.22, 95% CI 3.86-10.03), compared with patients with calcifications alone as the reference group.

CONCLUSION

The study characterized the risk of pancreatic cancer among patients with 5 abnormal morphologic findings based on radiology reports and demonstrated the ability of prediction algorithms to provide improved risk stratification of pancreatic cancer in these patients.

Developing a Siamese Network for UTIs Risk Prediction in Immobile Patients Undergoing Stroke

Stroke patients tend to suffer from immobility, which increases the possibility of post-stroke complications. Urinary tract infections (UTIs) are one of the complications as an independent predictor of poor prognosis of stroke patients. However, the incidence of new UTIs onsets during hospitalization was rare in most datasets with a prevalence of 4%. This imbalanced data distribution sets obstacles to establishing an accurate prediction model. Our study aimed to develop an effective prediction model to identify UTIs risk in immobile stroke patients, and (2) to compare its prediction performance with traditional machine learning models. We tackled this problem by building a Siamese Network leveraging commonly used clinical features to identifying patients with UTIs risk. Model derivation and validation were based on a nationwide dataset including 3982 Chinese patients. Results showed that the Siamese Network performed better than traditional machine learning models in imbalanced datasets (Sensitivity: 0.810; AUC: 0.828).

What contribution can genetics make to predict the risk of Alzheimer's disease?

Alzheimer's disease (AD) is the most common neurodegenerative disorder. Although its etiology remains incompletely understood, genetic variants are important contributors. The prediction of AD risk through individual genetic variants is an important topic of research that may have individual and societal consequences when preventive treatments will become available. However, the genetic substratum of AD is heterogeneous. In addition to the extremely rare and fully penetrant pathogenic variants of the PSEN1, PSEN2 or APP genes causing autosomal dominant AD, a large spectrum of risk factors have been identified in complex forms, including the common risk factor APOEɛ4, which is associated with a moderate-to-high risk, common polymorphisms associated with a modest individual risk, and a plethora of rare variants in genes like SORL1, TREM2 or ABCA7 with moderate to high-magnitude effect. Understanding how these genetic factors contribute to AD risk in a given individual, in additional to non-genetic factors, remains a challenge. Over the last 10 years, age-related penetrance curves have progressively incorporated advances in the knowledge of AD genetics, from APOE to common polygenic components and, currently, SORL1 rare variants, which represents an important step towards precision medicine in AD. In this review, we present the complex genetic architecture of AD and we expose the prediction of AD risk according to its underlying genetic component.

Risk prediction of atrial fibrillation in the community combining biomarkers and genetics

AIMS

Classical cardiovascular risk factors (CVRFs), biomarkers, and common genetic variation have been suggested for risk assessment of atrial fibrillation (AF). To evaluate their clinical potential, we analysed their individual and combined ability of AF prediction.

METHODS AND RESULTS

In N = 6945 individuals of the FINRISK 1997 cohort, we assessed the predictive value of CVRF, N-terminal pro B-type natriuretic peptide (NT-proBNP), and 145 recently identified single-nucleotide polymorphisms (SNPs) combined in a developed polygenic risk score (PRS) for incident AF. Over a median follow-up of 17.8 years, n = 551 participants (7.9%) developed AF. In multivariable-adjusted Cox proportional hazard models, NT-proBNP [hazard ratio (HR) of log transformed values 4.77; 95% confidence interval (CI) 3.66-6.22; P < 0.001] and the PRS (HR 2.18; 95% CI 1.88-2.53; P < 0.001) were significantly related to incident AF. The discriminatory ability improved asymptotically with increasing numbers of SNPs. Compared with a clinical model, AF risk prediction was significantly improved by addition of NT-proBNP and the PRS. The C-statistic for the combination of CVRF, NT-proBNP, and the PRS reached 0.83 compared with 0.79 for CVRF only (P < 0.001). A replication in the Dutch Prevention of REnal and Vascular ENd-stage Disease (PREVEND) cohort revealed similar results. Comparing the highest vs. lowest quartile, NT-proBNP and the PRS both showed a more than three-fold increased AF risk. Age remained the strongest risk factor with a 16.7-fold increased risk of AF in the highest quartile.

CONCLUSION

The PRS and the established biomarker NT-proBNP showed comparable predictive ability. Both provided incremental predictive value over standard clinical variables. Further improvements for the PRS are likely with the discovery of additional SNPs.

Breast Arterial Calcification: A Potential Biomarker for Atherosclerotic Cardiovascular Disease Risk?

PURPOSE OF REVIEW

We aimed to summarize the current evidence regarding the association between breast arterial calcification (BAC) and atherosclerotic cardiovascular disease (ASCVD) in women and discuss the potential role of BAC in the risk stratification and preventive approaches for ASCVD.

RECENT FINDINGS

BAC has emerged as a potential women-specific risk marker for ASCVD. Although BAC presents as a medial calcification of the arteries, notably different from the intimal atherosclerotic process, current evidence supports a correlation between BAC and ASCVD risk factors or subclinical and clinical ASCVD, such as coronary artery disease or stroke. As millions of women undergo mammograms each year, the potential clinical application of BAC in enhanced ASCVD risk estimation, with no additional cost or radiation, has tremendous appeal. Although further research regarding optimal risk assessment and management in women with BAC is required, the presence of BAC should prompt healthy cardiovascular lifestyle modifications.

Development of a Comprehensive Antenatal Risk Assessment Tool to Predict Adverse Maternal and Perinatal Outcomes in Rural Areas: An Exploratory Study

Objective: To develop a comprehensive antenatal risk assessment tool to predict adverse maternal and early perinatal outcomes in a rural setting.

Materials and methods: Cross-sectional study among women admitted for delivery in a rural maternity hospital, south India. Risk factors from Rotterdam Reproductive Risk Reduction (R4U) scorecard and social factors relevant to Indian rural context were included in questionnaire. Maternal and perinatal outcomes were obtained from in-patient records. Logistic regression of risk factors associated with adverse outcomes and weighted scores assigned using beta-coefficients. Cut-off score to predict adverse outcome was derived using Receiver Operator Characteristic Curve (ROC Curve) and Likelihood ratios.

Results: Adjusted odds for adverse outcome highest for small for gestational age by ultrasound scan [OR=7.4 (1.4-36.5)], tobacco chewing [OR=5.6 (1.8–28.5)] and hypertensive disorders of pregnancy [OR=3.5 (1.9-9.6)]. After assigning weighted scores, the 74-item antenatal risk assessment tool had a maximum possible score of 86. Risk score was calculated for all subjects. Cut-off score to predict adverse outcome was 4, using ROC curve, with a sensitivity of 98%, a specificity of 21% and positive likelihood ratio of 1.23 (1.10-1.37).

Conclusion: This comprehensive antenatal risk assessment tool is easy to administer, specific to rural areas and can help community-level workers to screen, monitor, and refer high risk pregnancies for further management to prevent adverse maternal and perinatal outcomes. This may be considered a prototype towards developing more robust antenatal risk screening and outcome prediction in rural settings.

QTc interval prolongation and life-threatening arrhythmias during hospitalization in patients with COVID-19. Results from a multi-center prospective registry

Background

Prolonged QTc interval and life-threatening arrhythmias (LTA) are potential drug induced complications previously reported with antimalarial, antivirals and antibiotics.

Objectives

To evaluate prevalence and predictors of QTc interval prolongation and incidence of LTA during hospitalization for COVID-19 among patients with normal admission QTc.

Methods

110 consecutive patients were enrolled in a multicenter international registry. 12-lead ECG was performed at admission, after 7 and 14 days; QTc values were analyzed.

Results

Fifteen (14%) patients developed a prolonged-QTc (pQT) after 7 days (mean QTc increase 66±20msec, +16%, p<0.001); these patients were older, had higher basal heart rates, higher rates of paroxysmal atrial fibrillation, lower platelet count. QTc increase was inversely proportional to baseline QTc levels and leukocyte count and directly to basal heart rates(p<0.01).At multivariate stepwise analysis including age, male gender, paroxysmal atrial fibrillation, basal QTc values, basal heart rate and dual antiviral therapy, age(OR 1.06, 95% C.I. 1.00-1.13, p<0.05), basal heart rate(OR 1.07, 95% C.I. 1.02-1.13, p<0.01) and dual antiviral therapy(OR 12.46, 95% C.I. 2.09-74.20, p<0.1) were independent predictors of QT-prolongation.Incidence of LTA during hospitalization was 3.6%. One patient experienced cardiac arrest and three non-sustained ventricular tachycardia. LTAs were recorded after a median of 9 days from hospitalization and were associated with 50% of mortality rate.

Conclusions

After 7 days of hospitalization, 14% of patients with Covid-19 developed pQTc; age, basal heart rate and dual antiviral therapy were found as independent predictor of pQTc. Life threatening arrhythmias have an incidence of 3.6% and were associated with poor outcome.

Toward Genetic Prediction of Nonalcoholic Fatty Liver Disease Trajectories: PNPLA3 and Beyond

Nonalcoholic fatty liver disease (NAFLD) is on the verge of becoming the leading cause of liver disease. NAFLD develops at the interface between environmental factors and inherited predisposition. Genome-wide association studies, followed by exome-wide analyses, led to identification of genetic risk variants (eg, PNPLA3, TM6SF2, and SERPINA1) and key pathways involved in fatty liver disease pathobiology. Functional studies improved our understanding of these genetic factors and the molecular mechanisms underlying the trajectories from fat accumulation to fibrosis, cirrhosis, and cancer over time. Here, we summarize key NAFLD risk genes and illustrate their interactions in a 3-dimensional "risk space." Although NAFLD genomics sometimes appears to be "lost in translation," we envision clinical utility in trial design, outcome prediction, and NAFLD surveillance.

Pathways of Progression From Intraductal Papillary Mucinous Neoplasm to Pancreatic Ductal Adenocarcinoma Based on Molecular Features

BACKGROUND & AIMS

Intraductal papillary mucinous neoplasms (IPMNs) are regarded as precursors of pancreatic ductal adenocarcinomas (PDAs), but little is known about the mechanism of progression. This makes it challenging to assess cancer risk in patients with IPMNs. We investigated associations of IPMNs with concurrent PDAs by genetic and histologic analyses.

METHODS

We obtained 30 pancreatic tissues with concurrent PDAs and IPMNs, and 168 lesions, including incipient foci, were mapped, microdissected, and analyzed for mutations in 18 pancreatic cancer-associated genes and expression of tumor suppressors.

RESULTS

We determined the clonal relatedness of lesions, based on driver mutations shared by PDAs and concurrent IPMNs, and classified the lesions into 3 subtypes. Twelve PDAs contained driver mutations shared by all concurrent IPMNs, which we called the sequential subtype. This subset was characterized by less diversity in incipient foci with frequent GNAS mutations. Eleven PDAs contained some driver mutations that were shared with concurrent IPMNs, which we called the branch-off subtype. In this subtype, PDAs and IPMNs had identical KRAS mutations but different GNAS mutations, although the lesions were adjacent. Whole-exome sequencing and methylation analysis of these lesions indicated clonal origin with later divergence. Ten PDAs had driver mutations not found in concurrent IPMNs, called the de novo subtype. Expression profiles of TP53 and SMAD4 increased our ability to differentiate these subtypes compared with sequencing data alone. The branch-off and de novo subtypes had substantial heterogeneity among early clones, such as differences in KRAS mutations. Patients with PDAs of the branch-off subtype had a longer times of disease-free survival than patients with PDAs of the de novo or the sequential subtypes.

CONCLUSIONS

Detailed histologic and genetic analysis of PDAs and concurrent IPMNs identified 3 different pathways by which IPMNs progress to PDAs-we call these the sequential, branch-off, and de novo subtypes. Subtypes might be associated with clinical and pathologic features and be used to select surveillance programs for patients with IPMNs.

Risk Prediction Models for Acute Kidney Injury in Critically Ill Patients: Opus in Progressu

Acute kidney injury (AKI) is a complex systemic syndrome associated with high morbidity and mortality. Among critically ill patients admitted to intensive care units (ICUs), the incidence of AKI is as high as 50% and is associated with dismal outcomes. Thus, the development and validation of clinical risk prediction tools that accurately identify patients at high risk for AKI in the ICU is of paramount importance. We provide a comprehensive review of 3 clinical risk prediction tools that have been developed for incident AKI occurring in the first few hours or days following admission to the ICU. We found substantial heterogeneity among the clinical variables that were examined and included as significant predictors of AKI in the final models. The area under the receiver operating characteristic curves was ∼0.8 for all 3 models, indicating satisfactory model performance, though positive predictive values ranged from only 23 to 38%. Hence, further research is needed to develop more accurate and reproducible clinical risk prediction tools. Strategies for improved assessment of AKI susceptibility in the ICU include the incorporation of dynamic (time-varying) clinical parameters, as well as biomarker, functional, imaging, and genomic data.

Risk Prediction of Diabetic Nephropathy via Interpretable Feature Extraction from EHR Using Convolutional Autoencoder

This paper describes a technology for predicting the aggravation of diabetic nephropathy from electronic health record (EHR). For the prediction, we used features extracted from event sequence of lab tests in EHR with a stacked convolutional autoencoder which can extract both local and global temporal information. The extracted features can be interpreted as similarities to a small number of typical sequences of lab tests, that may help us to understand the disease courses and to provide detailed health guidance. In our experiments on real-world EHRs, we confirmed that our approach performed better than baseline methods and that the extracted features were promising for understanding the disease.

Developing and validating a risk prediction model for acute care based on frailty syndromes

OBJECTIVES

Population ageing may result in increased comorbidity, functional dependence and poor quality of life. Mechanisms and pathophysiology underlying frailty have not been fully elucidated, thus absolute consensus on an operational definition for frailty is lacking. Frailty scores in the acute medical care setting have poor predictive power for clinically relevant outcomes. We explore the utility of frailty syndromes (as recommended by national guidelines) as a risk prediction model for the elderly in the acute care setting.

SETTING

English Secondary Care emergency admissions to National Health Service (NHS) acute providers.

PARTICIPANTS

There were N=2,099,252 patients over 65 years with emergency admission to NHS acute providers from 01/01/2012 to 31/12/2012 included in the analysis.

PRIMARY AND SECONDARY OUTCOME MEASURES

Outcomes investigated include inpatient mortality, 30-day emergency readmission and institutionalisation. We used pseudorandom numbers to split patients into train (60%) and test (40%). Receiver operator characteristic (ROC) curves and ordering the patients by deciles of predicted risk was used to assess model performance. Using English Hospital Episode Statistics (HES) data, we built multivariable logistic regression models with independent variables based on frailty syndromes (10th revision International Statistical Classification of Diseases, Injuries and Causes of Death (ICD-10) coding), demographics and previous hospital utilisation. Patients included were those>65 years with emergency admission to acute provider in England (2012).

RESULTS

Frailty syndrome models exhibited ROC scores of 0.624-0.659 for inpatient mortality, 0.63-0.654 for institutionalisation and 0.57-0.63 for 30-day emergency readmission.

CONCLUSIONS

Frailty syndromes are a valid predictor of outcomes relevant to acute care. The models predictive power is in keeping with other scores in the literature, but is a simple, clinically relevant and potentially more acceptable measurement for use in the acute care setting. Predictive powers of the score are not sufficient for clinical use.

A multibiomarker risk score helps predict risk for Barrett's esophagus

BACKGROUND & AIMS

Risk prediction models for Barrett's esophagus (BE) have been developed using multiple demographic and clinical variables, but their predictive performance has been modest. Adding a multibiomarker risk score may improve discriminatory ability.

METHODS

We used data from 141 patients with definitive BE and 138 controls participating in a case-control study at the Michael E. DeBakey Veterans Affairs Medical Center (Houston, TX) (97% men, 65% of controls were white, and 89% of cases were white). We derived and compared 3 prediction models. Model 1 included only gastroesophageal reflux disease (GERD) frequency and duration; model 2 included GERD frequency and duration, age, sex, race, waist-to-hip ratio, and Helicobacter pylori status; and model 3 included the variables in model 2 as well as a multibiomarker risk score based on serum levels of interleukin (IL)12p70, IL6, IL8, IL10, and leptin. We assessed their predictive accuracy in terms of discrimination using the area under the receiver operating characteristic curve and calibration analyses.

RESULTS

The multibiomarker risk score was associated significantly with risk for BE. Compared with persons with a score of 0, persons with a score of 3 or higher had a greater than 10-fold increased risk for BE (biomarker risk score, ≥3; odds ratio, 11.9; 95% confidence interval, 4.06-34.9; P trend < .001). Risk prediction using the multibiomarker score in conjunction with demographic and clinical features improved discrimination compared with using only GERD frequency and duration (area under the receiver operating characteristic curve, 0.85 vs 0.74; P = .01).

CONCLUSIONS

Based on data from a case-control study of predominantly white male veterans, a risk prediction model including a multibiomarker score, derived from serum levels of cytokines and leptin, as well as GERD frequency and duration, age, sex, race, waist-to-hip ratio, and H pylori infection, can identify persons in this population with BE more accurately than previous methods.

Performance of the Cockcroft-Gault, MDRD and CKD-EPI Formulae in Non-Valvular Atrial Fibrillation: Which one Should be Used for Risk Stratification?

Background: Renal dysfunction is a strong predictor of adverse events in patients with atrial fibrillation (AF). The Cokcroft-Gault, Modification of Diet in Renal Disease (MDRD) and Chronic Kidney Disease-Epidemiology Collaboration (CKD-EPI) equations are available for estimating the glomerular filtration rate (GFR). No comparisons between these equations have yet been performed in patients with non-valvular AF concerning their mid-term prognostic performance. Methods: Cross-sectional study of 555 consecutive patients with non-valvular AF undergoing transesophageal echocardiogram. We tested the prognostic performance of the aforementioned GFR estimation formulae, namely their ability to predict all-cause mortality (primary endpoint) and major cardiac adverse or ischemic cerebrovascular events (secondary endpoints) during an average follow-up of 24 months. Results: Regarding the primary endpoint, Cockcroft-Gault (AUC=0.749±0.028) was superior to both MDRD (AUC=0.624±0.039) and CKD-EPI (AUC=0.641±0.034) [p<0.001 both comparisons] while CKD-EPI was superior to MDRD (p=0.011). Cockcroft-Gault was marginally superior to both MDRD (AUC=0.673±0.049 vs. AUC=0.586±0.054, p=0.041) and CKD-EPI (AUC=0.673±0.049 vs. AUC=0.604±0.054, p=0.063) in the prediction of ischemic cerebrovascular events, while no difference was found between CKD-EPI and MDRD. Concerning AUC for prediction of MACE, Cockcroft-Gault was superior to MDRD (p=0.009) and CKD-EPI (p=0.012), while CKD-EPI was similar to MDRD (p=0.215). Multivariate predictive models consistently included Cockcroft-Gault formula along with CHADS₂, excluding the other two equations. Measures of reclassification revealed a significant improvement in risk stratification for all studied endpoints with Cockcroft-Gault instead of CKD-EPI. Conclusions: In patients with non-valvular AF, the Cockcroft-Gault more appropriately classified individuals with respect to risk of all-cause mortality, ischaemic cerebrovascular event and major adverse cardiac event.

Serum Concentrations of Mannan-Binding Lectin (MBL) and MBL-Associated Serine Protease-2 and the Risk of Adverse Events in Pediatric Patients With Cancer and Fever in Neutropenia

BACKGROUND

It is unknown whether serum concentrations of mannan-binding lectin (MBL) and MBL-associated serine protease-2 (MASP-2) influence the risk of adverse events (AEs) in children with cancer presenting with fever in neutropenia (FN).

METHODS

Pediatric patients with cancer presenting with FN after non-myeloablative chemotherapy were observed in a prospective multicenter study. Mannan-binding lectin and MASP-2 were measured using commercially available enzyme-linked immunosorbent assay in serum taken at cancer diagnosis. Multiple FN episodes per patient were allowed. Associations of MBL and MASP-2 with AE in general, with bacteremia, and with serious medical complications (SMC) during FN were analyzed using mixed logistic regression.

RESULTS

Of 278 FN episodes, AE was reported in 84 (30%), bacteremia was reported in 42 (15%), and SMC was reported in 16 (5.8%). Median MBL was 2152 ng/mL (range, 7-10 060). It was very low (<100) in 11 (9%) patients, low (100-999) in 36 (29%) patients, and normal (≥1000) in 79 (63%) patients. Median MASP-2 was 410 ng/mL (range, 68-2771). It was low (<200) in 18 (14%) patients and normal in the remaining 108 (86%) patients. Mannan-binding lectin and MASP-2 were not significantly associated with AE or bacteremia. Normal versus low MBL was independently associated with a significantly higher risk of SMC (multivariate odds ratio, 12.8; 95% confidence interval, 1.01-163; P = .050).

CONCLUSIONS

Mannan-binding lectin and MASP-2 serum concentrations were not found to predict the risk to develop AEs or bacteremia during FN. Normal MBL was associated with an increased risk of SMC during FN. This finding, in line with earlier studies, does not support the concept of MBL supplementation in MBL-deficient children with cancer presenting with FN.

Customized Prediction of Short Length of Stay Following Elective Cardiac Surgery in Elderly Patients Using a Genetic Algorithm

Objective

To develop a customized short LOS (<6 days) prediction model for geriatric patients receiving cardiac surgery, using local data and a computational feature selection algorithm.

Design

Utilization of a machine learning algorithm in a prospectively collected STS database consisting of patients who received cardiac surgery between January 2002 and June 2011.

Setting

Urban tertiary-care center.

Participants

Geriatric patients aged 70 years or older at the time of cardiac surgery.

Interventions

None.

Measurements and Main Results

Predefined morbidity and mortality events were collected from the STS database. 23 clinically relevant predictors were investigated for short LOS prediction with a genetic algorithm (GenAlg) in 1426 patients. Due to the absence of an STS model for their particular surgery type, STS risk scores were unavailable for 771 patients. STS prediction achieved an AUC of 0.629 while the GenAlg achieved AUCs of 0.573 (in those with STS scores) and 0.691 (in those without STS scores). Among the patients with STS scores, the GenAlg features significantly associated with shorter LOS were absence of congestive heart failure (CHF) (OR = 0.59, p = 0.04), aortic valve procedure (OR = 1.54, p = 0.04), and shorter cross clamp time (OR = 0.99, p = 0.004). In those without STS prediction, short LOS was significantly correlated with younger age (OR = 0.93, p < 0.001), absence of CHF (OR = 0.53, p = 0.007), no preoperative use of beta blockers (OR = 0.66, p = 0.03), and shorter cross clamp time (OR = 0.99, p < 0.001).

Conclusion

While the GenAlg-based models did not outperform STS prediction for patients with STS risk scores, our local-data-driven approach reliably predicted short LOS for cardiac surgery types that do not allow STS risk calculation. We advocate that each institution with sufficient observational data should build their own cardiac surgery risk models.