Personalized Risk Prediction Research Articles

Personalized Risk Prediction for 30‐Day Readmissions With Venous Thromboembolism Using Machine Learning

The aim of the study was to develop and validate machine learning models to predict the personalized risk for 30-day readmission with venous thromboembolism (VTE). This study was a retrospective, observational study. We extracted and preprocessed the structured electronic health records (EHRs) from a single academic hospital. Then we developed and evaluated three prediction models using logistic regression, the balanced random forest model, and the multilayer perceptron. The study sample included 158,804 total admissions; VTE-positive cases accounted for 2,080 admissions from among 1,695 patients (1.31%). Based on the evaluation results, the balanced random forest model outperformed the other two risk prediction models. This study delivered a high-performing, validated risk prediction tool using machine learning and EHRs to identify patients at high risk for VTE after discharge. The risk prediction model developed in this study can potentially guide treatment decisions for discharged patients for better patient outcomes.

A neural system dynamics modeling platform and its applications in randomized controlled trial data analysis

Conventional statistical methods used in clinical trials lack the ability to predict patient specific risk and very often do not consider the effects of time varying interventions. The aim of this study is to test a novel artificial neural network based model for clinical trial analysis to represent the continuous time evolution of risk. The novel methodology tested utilizes system dynamics and artificial neural networks. This methodology was applied to analyze data from 2,221 patients with acute myocardial infarction enrolled in a well characterized randomized study, the HORIZONS-AMI trial. Outcomes analyzed included: (1) target lesion revascularization (TLR) and (2) stent thrombosis. The proposed neural system dynamics (NSD) model was compared against traditional Cox Proportional Hazards (Cox PH) and Cox neural network (NN) models. Model performance was evaluated using C-statistic at 1, 2 -and 3- years follow-up and model-based simulation studies were performed to examine the effect of different variables on the predicted risk of TLR and stent thrombosis. The NSD model achieved comparable performance to Cox models for TLR. For stent thrombosis, the NSD model outperformed Cox models (1-year C-statistic: Stent thrombosis – Cox PH: 0.60, Cox NN: 0.66, neural SD model: 0.69). The neural SD model identified clinically relevant variables such as stent count, stent type and multiple lesions treated as significant predictors for TLR; and stent count and peak platelet count for stent thrombosis. Simulations illustrated change in predicted TLR risk maximal for the neural SD model, compared to other models. For stent thrombosis, simulation scenario illustrated predicted event rate doubling when clopidogrel is discontinued at 6 months compared to extended use. We have demonstrated an alternative analytical methodology that combines system dynamics and artificial neural networks to analyze results from a randomized trial. This novel approach can incorporate patient-specific longitudinal data and provide personalized risk prediction. • Incorporation of non-linear interactions and time-varying data improves personalized risk prediction. • Neural System Dynamics (NSD) model combines neural networks with system dynamics. • NSD model represents continuous time evolution of risk. • NSD model captures clinically relevant risk predictors of Target Lesion Revascularization and Stent Thrombosis. • NSD model based simulations provide more personalized risk trajectories with actionable insights using follow-up intervention information.

A Multi-Agent Approach for Personalized Hypertension Risk Prediction

Hypertension is a global health problem and a leading factor in severe and life-threatening cardiovascular diseases (CVD) and stroke. The onset is dependent on individual lifestyle choices, and no single root cause of the condition exists. Various machine learning solutions are proposed for the early diagnosis of hypertension and its prediction, but they are based on standard guidelines and do not provide personalized solutions. Current models mainly rely on batch learning methods and do not readily learn the new incoming data. There is also a lack of an intelligent technique for handling anomalies in data, which leads to unreliable prediction results. In this paper, an integrated multi-agent-based hypertension risk prediction system is proposed that detects and computes missing values in the time series and provides personalized hypertension risk predictions. The proposed solution incorporates Gaussian mixture models for enhancing the input data, and an Online Infinite Echo State Gaussian Process (OIESGP) is used to obtain real-time prediction distribution of blood pressure. The prediction system readily absorbs new incoming data, and the model is updated to learn any new patterns in the data. The hypertension risk score is estimated using the Framingham hypertension risk estimator, and a 4-year hypertension risk is computed. The prediction performance of the proposed model is evaluated on blood pressure data gathered from the Malaysian population using mean absolute error, mean square error, and root-mean-square errors. The experimental results indicate that the proposed prediction model exhibits greater prediction accuracy than existing state-of-the-art online prediction methods.

Trans-ancestry genome-wide association meta-analysis of prostate cancer identifies new susceptibility loci and informs genetic risk prediction.

Prostate cancer is a highly heritable disease with large disparities in incidence rates across ancestry populations. We conducted a multiancestry meta-analysis of prostate cancer genome-wide association studies (107,247 cases and 127,006 controls) and identified 86 new genetic risk variants independently associated with prostate cancer risk, bringing the total to 269 known risk variants. The top genetic risk score (GRS) decile was associated with odds ratios that ranged from 5.06 [95% confidence interval (CI) 4.84–5.29] for men of European ancestry to 3.74 [95% CI 3.36–4.17] for men of African ancestry. Men of African ancestry were estimated to have a mean GRS that was 2.18-times higher [95% CI 2.14–2.22], and men of East Asian ancestry 0.73-times lower [95% CI 0.71–0.76], than men of European ancestry. These findings support the role of germline variation contributing to population differences in prostate cancer risk, with the GRS offering an approach for personalized risk prediction.

The Liability Threshold Model for Predicting the Risk of Cardiovascular Disease in Patients with Type 2 Diabetes: A Multi-Cohort Study of Korean Adults.

Personalized risk prediction for diabetic cardiovascular disease (DCVD) is at the core of precision medicine in type 2 diabetes (T2D). We first identified three marker sets consisting of 15, 47, and 231 tagging single nucleotide polymorphisms (tSNPs) associated with DCVD using a linear mixed model in 2378 T2D patients obtained from four population-based Korean cohorts. Using the genetic variants with even modest effects on phenotypic variance, we observed improved risk stratification accuracy beyond traditional risk factors (AUC, 0.63 to 0.97). With a cutoff point of 0.21, the discrete genetic liability threshold model consisting of 231 SNPs (GLT231) correctly classified 87.7% of 2378 T2D patients as high or low risk of DCVD. For the same set of SNP markers, the GLT and polygenic risk score (PRS) models showed similar predictive performance, and we observed consistency between the GLT and PRS models in that the model based on a larger number of SNP markers showed much-improved predictability. In silico gene expression analysis, additional information was provided on the functional role of the genes identified in this study. In particular, HDAC4, CDKN2B, CELSR2, and MRAS appear to be major hubs in the functional gene network for DCVD. The proposed risk prediction approach based on the liability threshold model may help identify T2D patients at high CVD risk in East Asian populations with further external validations.

The Irish dual-energy X-ray absorptiometry (DXA) Health Informatics Prediction (HIP) for Osteoporosis Project

PurposeThe purpose of the Irish dual-energy X-ray absorptiometry (DXA) Health Informatics Prediction (HIP) for Osteoporosis Project is to create a large retrospective cohort of adults in Ireland to examine the...

Abstract PO-146: Multiethnic GWAS meta-analysis identifies novel variants and informs genetic risk prediction for prostate cancer across populations

Abstract Prostate cancer (PCa) is a highly heritable disease with large disparities in incidence rates across racial and ethnic populations. The inadequate representation of diverse populations in current genome-wide association studies (GWAS) limits the translational potential of findings to the world’s populations and could result in biased risk prediction, further exacerbating health disparities. To improve our understanding of genetic risk of PCa, we conducted a multiethnic meta-analysis of PCa GWAS using 107,247 cases and 127,006 controls from the Prostate Cancer Association Group to Investigate Cancer-Associated Alterations in the Genome (PRACTICAL) consortium, including 85,554 cases and 91,972 controls of European ancestry, 10,368 cases and 10,986 controls of African ancestry, 8,611 cases and 18,809 controls of Asian ancestry, and 2,714 cases and 5,239 controls of Hispanic ancestry. We identified 269 genetic risk variants independently associated with PCa risk, 86 of which were novel. To understand the aggregate effect of the 269 variants, we constructed a genetic risk score (GRS) using the multiethnic weights of the risk variants. Compared to men in the average 40-60% GRS category, the estimated OR for men in the top GRS decile (90-100%) was 5.06 [95% CI 4.84-5.29] for men of European ancestry, 3.74 [95% CI 3.36-4.17] for men of African ancestry, 4.47 [95% CI 3.52-5.68] for men of Asian ancestry, and 4.15 [95% CI 3.33-5.17] for men of Hispanic ancestry. Men of African ancestry were estimated to have a 2.18-times higher mean GRS [95% CI 2.14-2.22], and men of Asian ancestry 0.73-times lower [95% CI 0.71- 0.76], than men of European ancestry. Age significantly modified the GRS association with PCa risk, such that in men of European ancestry, the top decile GRS category was associated with an OR of 6.71 [95 % CI 5.99-7.52] for men ages 55 years or younger and 4.39 [95% CI 4.19-4.60] for men older than 55 years. Similarly, in men of African ancestry, the top decile GRS category was associated with an OR of 4.70 [95% CI 3.65-6.07] for men ages 55 years or younger and 3.37 [95% CI 2.99-3.80] for men older than 55 years. We found that 51% of aggressive cases of European ancestry and 45% of aggressive cases of African ancestry were within the top 20% of the GRS. The lifetime absolute risk of PCa for men in the top decile of the GRS reached 38% for both African Americans [95% CI 36-41%] and Whites [95% CI 37-39%], 31% [95% CI 27-36%] for Hispanics and 26% [95% CI 22-30%] for Asians. For comparison, we constructed a genome-wide GRS, including the 269 variants and variants associated with PCa with P<1.0 × 10-5. ORs calculated with genome-wide GRS were similar and had nearly identical discriminative ability as the 269 GRS in independent samples of 6,852 cases and 193,117 controls of European ancestry and 1,586 cases and 1,047 controls of African ancestry. These findings support the role of germline variation contributing to racial and ethnic disparities in PCa risk, with the GRS offering an approach for personalized risk prediction across populations. Citation Format: David V. Conti, Burcu F. Darst, Lilit Moss, Edward J. Saunders, Xin Sheng, Alisha Chou, Tokhir Dadaev, Sonja I. Berndt, Stephen K. Van Den Eeden, Stephen J. Chanock, Michael B. Cook, Hidewaki Nakagawa, John S. Witte, Rosalind A. Eeles, Zsofia Kote-Jarai, Christopher A. Haiman. Multiethnic GWAS meta-analysis identifies novel variants and informs genetic risk prediction for prostate cancer across populations [abstract]. In: Proceedings of the AACR Virtual Conference: Thirteenth AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2020 Oct 2-4. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2020;29(12 Suppl):Abstract nr PO-146.

Abstract 17348: Cardiovascular Risk Prediction in Diabetes From Machine Learning: The ACCORD Study

Background: Atherosclerotic cardiovascular disease (ASCVD) risk prediction in persons with type 2 diabetes (T2DM) using existing calculators is imprecise. We aimed to develop a machine-learning (ML) model for prediction of ASCVD events in adults with T2DM. Methods: We utilized subjects with T2DM from the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial without known CVD and calculated their 10-year ASCVD risk using the ACC/AHA pooled cohort risk calculator (PCRC) predicting the composite outcome of myocardial infarction, non-fatal stroke and cardiovascular death using age, gender, race, systolic blood pressure (SBP), antihypertensive medication use, total cholesterol, high-density lipoprotein cholesterol, current smoking status and diabetes mellitus status. We developed an ASCVD risk calculator based on Random Forest (RF) ML algorithms using follow-up data from ACCORD with the same 9 predictors. 5-fold stratified random split was applied as cross-validation strategies. Results: A total of 6581 T2DM participants without baseline ASCVD were included in our final sample with a median follow up of 9.1 years. The performance of PCRC was modest with an AUC=0.604. In contrast, the ML model had much better performance with a RF AUC=0.866. The figure shows the rank of feature importance (%) from random forest modeling (from high to low): age, systolic blood pressure, total cholesterol, HDL-C, female gender, White ethnicity, current smoker, hypertension treatment. Conclusion: The ML ASCVD Risk Calculator outperforms the AHA/ACC PCRC in predicting ASCVD outcomes among those with T2DM from the ACCORD trial. Age, SBP, total cholesterol and HDL-C were the most important features in ASCVD prediction among those with T2DM. Future studies need to validate these and other ML algorithms and to explore their applicability in guidelines.

SEAIR Framework Accounting for a Personalized Risk Prediction Score: Application to the Covid-19 Epidemic

The aim of the present work is to provide an SEAIR framework which takes a personalized risk prediction score as an additional input. Each individual is categorized depending on his actual status with respect to the disease - moderate or severe symptoms -, and the level of risk predicted - low or high. This idea leads to a 4-fold extension of the ODE model in classical SEAIR. This model offers the possibility for policy-makers to explore differentiated containment strategies, by varying sizes for the low risk segment and varying dates for 'progressive release' of the population, while exploring the discriminative capacity of the risk score, for instance through its AUC. Differential contact rates for low-risk/high-risk compartments are also included in the model. The demo allows to select contact rates and time-depending exit strategies. The hard-coded parameters correspond to the data for the Covid-19 epidemic in France, and the risk refers to the probability of being admitted in ICU upon infection. Some examples of simulations are provided.

A predictive Bayesian network that risk stratifies patients undergoing Barrett's surveillance for personalized risk of developing malignancy.

BackgroundBarrett’s esophagus is strongly associated with esophageal adenocarcinoma. Considering costs and risks associated with invasive surveillance endoscopies better methods of risk stratification are required to assist decision-making and move toward more personalised tailoring of Barrett’s surveillance.MethodsA Bayesian network was created by synthesizing data from published studies analysing risk factors for developing adenocarcinoma in Barrett’s oesophagus through a two-stage weighting process.ResultsData was synthesized from 114 studies (n = 394,827) to create the Bayesian network, which was validated against a prospectively maintained institutional database (n = 571). Version 1 contained 10 variables (dysplasia, gender, age, Barrett’s segment length, statin use, proton pump inhibitor use, BMI, smoking, aspirin and NSAID use) and achieved AUC of 0.61. Version 2 contained 4 variables with the strongest evidence of association with the development of adenocarcinoma in Barrett’s (dysplasia, gender, age, Barrett’s segment length) and achieved an AUC 0.90.ConclusionThis Bayesian network is unique in the way it utilizes published data to translate the existing empirical evidence surrounding the risk of developing adenocarcinoma in Barrett’s esophagus to make personalized risk predictions. Further work is required but this tool marks a vital step towards delivering a more personalized approach to Barrett’s surveillance.

Could personalised risk prediction for type 2 diabetes using polygenic risk scores direct prevention, enhance diagnostics, or improve treatment?

Over the past three decades, the number of people globally with diabetes mellitus has more than doubled. It is estimated that by 2030, 439 million people will be suffering from the disease, 90-95% of whom will have type 2 diabetes (T2D). In 2017, 5 million deaths globally were attributable to T2D, placing it in the top 10 global causes of death. Because T2D is a result of both genetic and environmental factors, identification of individuals with high genetic risk can help direct early interventions to prevent progression to more serious complications. Genome-wide association studies have identified ~400 variants associated with T2D that can be used to calculate polygenic risk scores (PRS). Although PRSs are not currently more accurate than clinical predictors and do not yet predict risk with equal accuracy across all ethnic populations, they have several potential clinical uses. Here, we discuss potential usages of PRS for predicting T2D and for informing and optimising interventions. We also touch on possible health inequality risks of PRS and the feasibility of large-scale implementation of PRS in clinical practice. Before PRSs can be used as a therapeutic tool, it is important that further polygenic risk models are derived using non-European genome-wide association studies to ensure that risk prediction is accurate for all ethnic groups. Furthermore, it is essential that the ethical, social and legal implications of PRS are considered before their implementation in any context.

Deep learning model for prediction of hepatocellular carcinoma in patients with HBV-related cirrhosis on antiviral therapy.

Background & AimsPersonalised risk prediction of the development of hepatocellular carcinoma (HCC) among patients with liver cirrhosis on potent antiviral therapy is important for targeted screening and individualised intervention. This study aimed to develop and validate a new model for risk prediction of HCC development based on deep learning, and to compare it with previously reported risk models.MethodsA novel deep-learning-based model was developed from a cohort of 424 patients with HBV-related cirrhosis on entecavir therapy with 2 residual blocks, including 7 layers of a neural network, and it was validated using an independent external cohort (n = 316). The deep-learning-based model was compared to 6 previously reported models (platelet, age, and gender-hepatitis B score [PAGE-B], Chinese University HCC score [CU-HCC], HCC-Risk Estimating Score in CHB patients Under Entecavir [HCC-RESCUE], age, diabetes, race, etiology of cirrhosis, sex, and severity HCC score [ADRESS-HCC], modified PAGE-B score [mPAGE], and Toronto HCC risk index [THRI]) using Harrell's concordance (c)-index.ResultsDuring a median 5.2 yr of follow-up (inter-quartile range 2.8–6.9 yr), 86 patients (20.3%) developed HCC. The deep-learning-based model had a Harrell's c-index of 0.719 in the derivation cohort and 0.782 in the validation cohort. Goodness of fit was confirmed by the Hosmer-Lemeshow test (p >0.05). Moreover, this model in the validation cohort had the highest c-index among the 6 previously reported models: PAGE-B (0.570), CU-HCC (0.548), HCC-RESCUE (0.577), ADRESS-HCC (0.551), mPAGE (0.598), and THRI (0.587) (all p <0.001). The misclassification rate of this model was 23.7% (model accuracy: 76.3%) in the validation group.ConclusionsThe deep-learning-based model had better performance than the previous models for predicting the HCC risk in patients with HBV-related cirrhosis on potent antivirals.Lay summaryFor early detection of hepatocellular carcinoma, it is important to maintain regular surveillance. However, there is currently no standard prediction model for risk stratification that can be used to establish a personalised surveillance strategy. We develop and validate a deep-learning-based model that showed better performance than previous models.

OR13-07 Crystal Bone: Personalized, Short-Term Fracture Risk Prediction with Natural Language Processing Methods

Fragility fractures due to osteoporosis are common and are associated with significant clinical, personal, and economic burden. Even after a fragility fracture, osteoporosis remains widely underdiagnosed and undertreated. Common fracture risk assessment tools, such as FRAX1 and Garvan,2 confer risk over the long term but do not provide short-term risk estimates necessary to identify very high-risk patients likely to fracture in the next 1–2 years. Furthermore, these tools utilize cross-sectional data representing a subset of all available clinical risk factors for risk prediction. Thus, these methods are generalized across patient populations and may not fully utilize patient histories commonly found in electronic health records (EHRs) that contain temporal information for thousands of unique features. The Optum® de-identified EHR dataset (2007–2018) provides an opportunity to use historical medical data to generate short-term, personalized fracture risk predictions for individual patients. We used the Optum® dataset to develop Crystal Bone, a method that applies machine learning techniques commonly used in natural language processing to the temporal nature of patient histories in order to predict fracture risk over a 1- to 2-year timeframe. Specifically, we repurposed deep-learning models typically applied to language-based prediction tasks in which the goal is to learn the meanings of words and sentences to classify them. Crystal Bone uses context-based embedding techniques to learn an equivalent “semantic” meaning of various medical events. Similar to how language models predict the next word in a given sentence or the topic of an overall document, Crystal Bone can predict that a patient’s future trajectory may contain a fracture or that the “signature” of the patient’s overall journey is similar to that of a typical fracture patient. We applied Crystal Bone to two datasets, one enriched for fracture patients and one representative of a typical hospital system. In both datasets, when predicting likelihood of fracture in the next 1–2 years, Crystal Bone had an area under the receiver operating characteristic (AUROC) score ranging from 72% to 83% on a test (hold-out) dataset. These results suggest performance similar to that of FRAX and Garvan, which have 10-year fracture risk prediction AUROC scores of 64.4% +/- 3.7%.3 Our results suggest that it is possible to use each patient’s unique medical history as it changes over time to predict patients at risk for fracture in 1–2 years. Furthermore, it is theoretically possible to integrate a model like Crystal Bone directly into an EHR system, enabling “hands-off” fracture risk prediction, which could lead to improved identification of patients at very high risk for fracture. 1Kanis JA, Osteoporos Int. 2012;23:2239–56. 2Rubin KH, J Bone Miner Res. 2013;28:1701–17. 3Leslie WD, Osteoporos Int. 2014;25:1–21.

Stabilization patterns and variability of hs-CRP, NT-proBNP and ST2 during 1 year after acute coronary syndrome admission: results of the BIOMArCS study.

Objectives Details of the biological variability of high-sensitivity C-reactive protein (hs-CRP), N-terminal prohormone of brain natriuretic peptide (NT-proBNP) and ST2 are currently lacking in patients with acute coronary syndrome (ACS) but are crucial knowledge when aiming to use these biomarkers for personalized risk prediction. In the current study, we report post-ACS kinetics and the variability of the hs-CRP, NT-proBNP and ST2. Methods BIOMArCS is a prospective, observational study with high frequency blood sampling during 1 year post-ACS. Using 1507 blood samples from 191 patients that remained free from adverse cardiac events, we investigated post-ACS kinetics of hs-CRP, NT-proBNP and ST2. Biological variability was studied using the samples collected between 6 and 12 months after the index ACS, when patients were considered to have stable coronary artery disease. Results On average, hs-CRP rose peaked at day 2 and rose well above the reference value. ST2 peaked immediately after the ACS but never rose above the reference value. NT-proBNP level rose on average during the first 2 days post-ACS and slowly declined afterwards. The within-subject variation and relative change value (RCV) of ST2 were relatively small (13.8%, RCV 39.7%), while hs-CRP (41.9%, lognormal RCV 206.1/-67.3%) and NT-proBNP (39.0%, lognormal RCV 185.2/-64.9%) showed a considerable variation. Conclusions Variability of hs-CRP and NT-proBNP within asymptomatic and clinically stable post-ACS patients is considerable. In contrast, within-patient variability of ST2 is low. Given the low within-subject variation, ST2 might be the most useful biomarker for personalizing risk prediction in stable post-ACS patients.

Personalized risk prediction for breast cancer pre-screening using artificial intelligence and thermal radiomics.

Breast cancer is the leading cause of cancer deaths among women today. Survival rates in developing countries are around 50%-60% due to late detection. A personalized, accurate risk scoring method can help in targeting the right population for follow-up tests and enables early detection of breast abnormalities. Most of the available risk assessment tools use generic and weakly correlated features like age, weight, height etc. While a personalized risk scoring from screening modalities such as mammography and ultrasound could be helpful, these tests are limited to very few metropolitan hospitals in developing countries due to high capital cost, operational expenses and interpretation expertise needed for a large screening population. We propose and analyze a new personalized risk framework called Thermalytix Risk Score (TRS) to identify a high-risk target population for regular screening and enable early stage breast cancer detection at scale. This technique uses Artificial Intelligence (AI) over thermal images to automatically generate a breast health risk score. This risk score is mainly derived from two sub-scores namely, vascular score and hotspot score. A hotspot score signifies the abnormality seen from irregular asymmetric heat patterns seen on the skin surface, whereas vascular score predicts the presence of asymmetric vascular activity. These scores are generated using machine learning algorithms over medically interpretable parameters that describes the metabolic activity inside the breast tissue and indicate the presence of a possible malignancy even in asymptomatic women. The proposed personalized risk score was tested on 769 subjects in four breast cancer screening facilities. The subjects' age ranged from 18 to 82 years with a median of around 45 years. Out of the 769 subjects, 185 subjects were diagnosed with a breast malignancy by an expert radiologist after mammography, ultrasound and/or histopathology. Our personalized AI based risk score achieved an area under the receiver-operator curve (AUC) of 0.89 when compared to an age normalized risk score that showed an AUC of 0.68. We also found that if the computed risk score is used to place individuals into four risk groups, the likelihood of malignancy also increases monotonically with the risk grouping level. The proposed AI based personalized risk score uses breast thermal image patterns for risk computation and compares favorably to other generic risk estimation approaches. The proposed risk framework solution is automated, affordable, non-invasive, non-contact and radiation free and works for a wide age range of women from 18 to 82 years, including young women with dense breasts. The proposed score might be further used to assign subjects into one of the four risk groups and provide guidance on the periodicity of screening needed. In addition, the automatically annotated thermal images localizes the potential abnormal regions and might empower the physician to create a better personalized care.

Evaluating the Utility of Polygenic Risk Scores in Identifying High-Risk Individuals for Eight Common Cancers

BackgroundGenome-wide association studies have identified common genetic risk variants in many loci associated with multiple cancers. We sought to systematically evaluate the utility of these risk variants in identifying high-risk individuals for eight common cancers.MethodsWe constructed polygenic risk scores (PRS) using genome-wide association studies–identified risk variants for each cancer. Using data from 400 812 participants of European descent in a population-based cohort study, UK Biobank, we estimated hazard ratios associated with PRS using Cox proportional hazard models and evaluated the performance of the PRS in cancer risk prediction and their ability to identify individuals at more than a twofold elevated risk, a risk level comparable to a moderate-penetrance mutation in known cancer predisposition genes.ResultsDuring a median follow-up of 5.8 years, 14 584 incident case patients of cancers were identified (ranging from 358 epithelial ovarian cancer case patients to 4430 prostate cancer case patients). Compared with those at an average risk, individuals among the highest 5% of the PRS had a two- to threefold elevated risk for cancer of the prostate, breast, pancreas, colorectal, or ovary, and an approximately 1.5-fold elevated risk of cancer of the lung, bladder, or kidney. The areas under the curve ranged from 0.567 to 0.662. Using PRS, 40.4% of the study participants can be classified as having more than a twofold elevated risk for at least one site-specific cancer.ConclusionsA large proportion of the general population can be identified at an elevated cancer risk by PRS, supporting the potential clinical utility of PRS for personalized cancer risk prediction.

Risk factors for progression of age‐related macular degeneration

PurposeAge‐related macular degeneration (AMD) is a degenerative disease of the macula, often leading to progressive vision loss. The rate of disease progression can vary among individuals and has been associated with multiple risk factors. In this review, we provide an overview of the current literature investigating phenotypic, demographic, environmental, genetic, and molecular risk factors, and propose the most consistently identified risk factors for disease progression in AMD based on these studies. Finally, we describe the potential use of these risk factors for personalised healthcare.Recent findingsWhile phenotypic risk factors such as drusen and pigment abnormalities become more important to predict disease progression during the course of the disease, demographic, environmental, genetic and molecular risk factors are more valuable at earlier disease stages. Demographic and environmental risk factors such as age and smoking are consistently reported to be related to disease progression, while other factors such as sex, body mass index (BMI) and education are less often associated. Of all known AMD variants, variants that are most consistently reported with disease progression are rs10922109 and rs570618 in CFH, rs116503776 in C2/CFB/SKIV2L, rs3750846 in ARMS2/HTRA1 and rs2230199 in C3. However, it seems likely that other AMD variants also contribute to disease progression but to a lesser extent. Rare variants have probably a large effect on disease progression in highly affected families. Furthermore, current prediction models do not include molecular risk factors, while these factors can be measured accurately in the blood. Possible promising molecular risk factors are High‐Density Lipoprotein Cholesterol (HDL‐C), Docosahexaenoic acid (DHA), eicosapentaenoic acid (EPA), zeaxanthin and lutein.SummaryPhenotypic, demographic, environmental, genetic and molecular risk factors can be combined in prediction models to predict disease progression, but the selection of the proper risk factors for personalised risk prediction will differ among individuals and is dependent on their current disease stage. Future prediction models should include a wider set of genetic variants to determine the genetic risk more accurately, and rare variants should be taken into account in highly affected families. In addition, adding molecular factors in prediction models may lead to preventive strategies and personalised advice.

Personalized risk prediction of symptomatic intracerebral hemorrhage after stroke thrombolysis using a machine-learning model.

Background:Personalized prediction of the risk of symptomatic intracerebral hemorrhage (sICH) after stroke thrombolysis is clinically useful. Machine-learning-based modeling may provide the personalized prediction of the risk of sICH after stroke thrombolysis.Methods:We identified 2578 thrombolysis-treated ischemic stroke patients between January 2013 and December 2016 from a multicenter database, where 70% were used to train models and the remaining 30% were used as the nominal test sets. Another 136 consecutive tissue plasminogen-activated-treated patients between January 2017 and December 2017 from our institute were enrolled as the independent test sets for clinical usability evaluation. Five machine-learning models were developed to predict the risk of sICH after stroke thrombolysis, and the receiving operating characteristic (ROC) was used to compare the prediction performance.Results:In total, 2237 cases were included in our study, of which 102 had sICH transformation (4.56%). Finally, the three-layer neuro network was selected with the best performance on nominal test sets (AUC = 0.82). The probability of the model score was further categorized into three risk ranks (18.97%, 5.63%, and 0.81%) according to the risk distribution. Implementing our system in clinical practice was associated with reduced computed tomography (CT)-to-treatment time (CTT; 41 min versus 52 min, p < 0.001). All sICH patients were correctly predicted to be within the high-sICH risk rank.Conclusions:The machine-learning-based modeling is feasible for providing personalized risk prediction of sICH after stroke thrombolysis, and is able to reduce the CTT. More data are needed to further optimize the model and improve the accuracy of prediction.

Progression to dementia in memory clinic patients with mild cognitive impairment and normal \u03b2-amyloid

BackgroundDetermination of β-amyloid (Aβ) positivity and likelihood of underlying Alzheimer’s disease (AD) relies on dichotomous biomarker cut-off values. Individuals with mild cognitive impairment (MCI) and Aβ within the normal range may still have a substantial risk of developing dementia, primarily of Alzheimer type. Their prognosis, as well as predictors of clinical progression, are not fully understood. The aim of this study was to explore the associations of cerebrospinal fluid (CSF) biomarkers (Aβ42, total tau, phosphorylated tau) and other characteristics, including modifiable vascular factors, with the risk of progression to dementia among patients with MCI and normal CSF Aβ42.MethodsThree hundred eighteen memory clinic patients with CSF and clinical data, and at least 1-year follow-up, were included. Patients had normal CSF Aβ42 levels based on clinical cut-offs. Cox proportional hazard models with age as time scale and adjusted for sex, education, and cognition (Mini-Mental State Examination) were used to investigate predictors of progression to dementia and Alzheimer-type dementia. Potential predictors included CSF biomarkers, cognitive performance (verbal learning and memory), apolipoprotein E (APOE) ε4 genotype, medial temporal lobe atrophy, family history of dementia, depressive symptoms, and vascular factors, including the Cardiovascular Risk Factors, Aging and Dementia (CAIDE) risk score. Predictive performance of patient characteristics was further explored with Harrell C statistic.ResultsLower normal Aβ42 and higher total tau and phosphorylated tau were associated with higher dementia risk, and the association was not driven by Aβ42 values close to cut-off. Additional predictors included poorer cognition, APOE ε4 genotype, higher systolic blood pressure, and lower body mass index, but not the CAIDE dementia risk score. Aβ42 individually and in combination with other CSF biomarkers improved the risk prediction compared to age and cognition alone. Medial temporal lobe atrophy or vascular factors did not increase the predictive performance.ConclusionsPossibility of underlying AD pathology and increased dementia risk should not be ruled out among MCI patients with CSF Aβ42 within the normal range. While cut-offs may be useful in clinical practice to identify high-risk individuals, personalized risk prediction tools incorporating continuous biomarkers may be preferable among individuals with intermediate risk. The role of modifiable vascular factors could be explored in this context.

A Machine Learning Algorithm to Improve Risk Assessment for Patients with Sickle Cell Disease

Sickle cell disease (SCD) is a clinical syndrome that encompasses several different genotypes, the 3 most common being homozygosity for the bS allele (HbSS), compound heterozygosity of HbS and HbC (HbSC), and compound heterozygosity of HbS and HbSb thalassemia (HbSb+ or HbSb0 thalassemia). Generally, patients with HbSS and HbSb0 thalassemia genotypes have the most severe clinical manifestations, while patients with HbSC and HbSβ+-thalassemia are thought to be less severe. Within each of these genotypic groups, however, there are also substantial phenotypic differences. This heterogeneity makes it difficult to quantify the severity of the disease process and to guide therapeutics. As more intensive, high risk and costly treatments such as hematopoietic stem cell transplant and gene therapy are developing, the ability to assess patients at highest risk of early mortality becomes increasingly important. Integrating varied clinical, laboratory, and imaging markers for personalized risk prediction has been difficult, however, newer machine learning methods for outcome prediction take a more agnostic approach than traditional statistical methods and can detect complex, non-linear relationships in the data. In this study, we sought to apply machine learning methods to a well-characterized cohort of SCD patients followed at the National Institutes of Health in order to identify clinically meaningful subgroups of patients at highest risk of mortality. Between 2006 and 2017, 601 patients (age 35±13 years, 51% female) underwent echocardiogram, standard laboratory markers and hemoglobin electrophoresis resulting in 61 candidate variables. Among these patients, 488 had HbSS, 12 HbSb0 thalassemia, 80 HbSC, 20 HbSb+ thalassemia. All-cause mortality was ascertained by proxy interview, through medical records, and through the CDC National Death Index. Average follow-up time was 5 years and 130 patients were deceased. A random survival forest (RSF) algorithm followed by nested model selection and AIC Cox regression analysis identified 13 predictors of mortality (estimated right ventricular systolic pressure, peak tricuspid regurgitant (TR) velocity, mitral E velocity, septal and posterior wall thickness, IVC diameter, right atrial area, BUN, alkaline phosphatase, N-terminal-pro brain natriuretic peptide (BNP), creatinine, potassium and bicarbonate). This model performed better than individual clinical and laboratory variables with a C-statistic of 0.822 (genotype 0.524, eGFR 0.624, NT-proBNP 0.686, TR velocity 0.703). K-means clustering grouped all patients into 3 main clusters with significant survival differences. Survival at 8 years for the entire group was 70%; for individual clusters, survival was 43% for cluster 1, 72% for cluster 2, and 88% for cluster 3 (Figure 1A). Since TR velocity is recognized as one of the most specific independent predictors of mortality, we compared our results with this parameter. There was a better stratification of mortality risk using the 7 strongest parameters from RSF compared with TR velocity alone (Figure 1B), particularly for longer term outcomes. In this cohort of 601 patients with SCD, machine learning methods were used to show the heterogeneity of this disorder and the ability to detect phenotypic clusters with different mortality profiles. Although there are many individual predictors of mortality, few methods other than assessment by an expert clinician can integrate all known variables in deeply phenotyped patients. RSF and cluster analysis was used in this cohort to analyze a large amount of data in order to identify seven variables that could stratify patients into groups with significantly different outcomes. The specificity of this approach was high (c-statistic 0.822) and better than that of individual markers of end-organ involvement. Disclosures No relevant conflicts of interest to declare.