Data mining for censored time-to-event data: a Bayesian network model for predicting cardiovascular risk from electronic health record data

Predicting lung cancer survival prognosis based on the conditional survival bayesian network

Lung cancer is a leading cause of cancer deaths and imposes an enormous economic burden on patients. It is important to develop an accurate risk assessment model to determine the appropriate treatment for patients after an initial lung cancer diagnosis. The Cox proportional hazards model is mainly employed in survival analysis. However, real-world medical data are usually incomplete, posing a great challenge to the application of this model. Commonly used imputation methods cannot achieve sufficient accuracy when data are missing, so we investigated novel methods for the development of clinical prediction models. In this article, we present a novel model for survival prediction in missing scenarios. We collected data from 5,240 patients diagnosed with lung cancer at the Weihai Municipal Hospital, China. Then, we applied a joint model that combined a BN and a Cox model to predict mortality risk in individual patients with lung cancer. The established prognostic model achieved good predictive performance in discrimination and calibration. We showed that combining the BN with the Cox proportional hazards model is highly beneficial and provides a more efficient tool for risk prediction.

Predicting Groundwater PFOA Exposure Risks with Bayesian Networks: Empirical Impact of Data Preprocessing on Model Performance

The plethora of data on PFASs in environmental samples collected in response to growing concern about these chemicals could enable the training of machine-learning models for predicting exposure risks. However, differences in sampling and analysis methods across data sets must be reconciled through data preprocessing, and little information is available about how such manipulations affect the resulting models. This study evaluates how data preprocessing influences machine-learned Bayesian network models of PFOA in groundwater. We link 19 years of PFOA measurements from Minnesota, USA, to publicly available information about potential PFOA sources and factors that may influence their environmental fate. Nine different preprocessing methods were tested, and the resulting data sets were used to train models to predict the probability of PFOA ≥ 35 ppt, the 2017 Minnesota health advisory level. Different preprocessing approaches produced varying model structures with significantly different accuracies. Nonetheless, models showed similar relationships between predictor variables and PFOA exposure risks, and all models were relatively accurate, distinguishing wells at high risk from those at low risk for 82.0% to 89.0% of test data samples. There was a trade-off between data quality and model performance since a stricter data screening strategy decreased the sample size for model training.

Prognostic models for breast cancer: based on logistics regression and Hybrid Bayesian Network

BACKGROUND

To construct two prognostic models to predict survival in breast cancer patients; to compare the efficacy of the two models in the whole group and the advanced human epidermal growth factor receptor-2-positive (HER2+) subgroup of patients; to conclude whether the Hybrid Bayesian Network (HBN) model outperformed the logistics regression (LR) model.

METHODS

In this paper, breast cancer patient data were collected from the SEER database. Data processing and analysis were performed using Rstudio 4.2.0, including data preprocessing, model construction and validation. The L_DVBN algorithm in Julia0.4.7 and bnlearn package in R was used to build and evaluate the HBN model. Data with a diagnosis time of 2018(n = 23,384) were distributed randomly as training and testing sets in the ratio of 7:3 using the leave-out method for model construction and internal validation. External validation of the model was done using the dataset of 2019(n = 8128). Finally, the late HER2 + patients(n = 395) was selected for subgroup analysis. Accuracy, calibration and net benefit of clinical decision making were evaluated for both models.

RESULTS

The HBN model showed that seventeen variables were associated with survival outcome, including age, tumor size, site, histologic type, radiotherapy, surgery, chemotherapy, distant metastasis, subtype, clinical stage, ER receptor, PR receptor, clinical grade, race, marital status, tumor laterality, and lymph node. The AUCs for the internal validation of the LR and HBN models were 0.831 and 0.900; The AUCs for the external validation of the LR and HBN models on the whole population were 0.786 and 0.871; the AUCs for the external validation of the two models on the subgroup population were 0.601 and 0.813.

CONCLUSION

The accuracy, net clinical benefit, and calibration of the HBN model were better than LR model. The predictive efficacy of both models decreased and the difference was greater in advanced HER2 + patients, which means the HBN model had higher robustness and more stable predictive performance in the subgroup.

Survival Modelling For Data From Combined Cohorts: Opening the Door to Meta Survival Analyses and Survival Analysis using Electronic Health Records

Non-parametric estimation of the survival function using observed failure time data depends on the underlying data generating mechanism, including the ways in which the data may be censored and/or truncated. For data arising from a single source or collected from a single cohort, a wide range of estimators have been proposed and compared in the literature. Often, however, it may be possible, and indeed advantageous, to combine and then analyze survival data that have been collected under different study designs. We review non-parametric survival analysis for data obtained by combining the most common types of cohort. We have two main goals: (i) To clarify the differences in the model assumptions, and (ii) to provide a single lens through which some of the proposed estimators may be viewed. Our discussion is relevant to the meta analysis of survival data obtained from different types of study, and to the modern era of electronic health records.

Forecasting COVID19 parameters using time-series: KSA, USA, Spain, and Brazil comparative case study

Many countries are suffering from the COVID19 pandemic. The number of confirmed cases, recovered, and deaths are of concern to the countries having a high number of infected patients. Forecasting these parameters is a crucial way to control the spread of the disease and struggle with the pandemic. This study aimed at forecasting the number of cases and deaths in KSA using time-series and well-known statistical forecasting techniques including Exponential Smoothing and Linear Regression. The study is extended to forecast the number of cases in the main countries such that the US, Spain, and Brazil (having a large number of contamination) to validate the proposed models (Drift, SES, Holt, and ETS). The forecast results were validated using four evaluation measures. The results showed that the proposed ETS (resp. Drift) model is efficient to forecast the number of cases (resp. deaths). The comparison study, using the number of cases in KSA, showed that ETS (with RMSE reaching 18.44) outperforms the state-of-the art studies (with RMSE equal to 107.54). The proposed forecasting model can be used as a benchmark to tackle this pandemic in any country.

Influencing factors of Barthel index scores among the community-dwelling elderly in Hong Kong: a random intercept model

Background

Barthel Index (BI) is one of the most widely utilized tools for assessing functional independence in activities of daily living. Most existing BI studies used populations with specific diseases (e.g., Alzheimer’s and stroke) to test prognostic factors of BI scores; however, the generalization of these findings was limited when the target populations varied.

Objectives

The aim of the present study was to utilize electronic health records (EHRs) and data mining techniques to develop a generic procedure for identifying prognostic factors that influence BI score changes among community-dwelling elderly.

Methods

Longitudinal data were collected from 113 older adults (81 females; mean age = 84 years, SD = 6.9 years) in Hong Kong elderly care centers. Visualization technologies were used to align annual BI scores with individual EHRs chronologically. Linear mixed-effects (LME) regression was conducted to model longitudinal BI scores based on socio-demographics, disease conditions, and features extracted from EHRs.

Results

The visualization presented a decline in BI scores changed by time and health history events. The LME model yielded a conditional R² of 84%, a marginal R² of 75%, and a Cohen’s f² of 0.68 in the design of random intercepts for individual heterogeneity. Changes in BI scores were significantly influenced by a set of socio-demographics (i.e., sex, education, living arrangement, and hobbies), disease conditions (i.e., dementia and diabetes mellitus), and EHRs features (i.e., event counts in allergies, diagnoses, accidents, wounds, hospital admissions, injections, etc.).

Conclusions

The proposed visualization approach and the LME model estimation can help to trace older adults’ BI score changes and identify the influencing factors. The constructed long-term surveillance system provides reference data in clinical practice and help healthcare providers manage the time, cost, data and human resources in community-dwelling settings.

A comprehensive scoping review of Bayesian networks in healthcare: Past, present and future

No comprehensive review of Bayesian networks (BNs) in healthcare has been published in the past, making it difficult to organize the research contributions in the present and identify challenges and neglected areas that need to be addressed in the future. This unique and novel scoping review of BNs in healthcare provides an analytical framework for comprehensively characterizing the domain and its current state. A literature search of health and health informatics literature databases using relevant keywords found 3810 articles that were reduced to 123. This was after screening out those presenting Bayesian statistics, meta-analysis or neural networks, as opposed to BNs and those describing the predictive performance of multiple machine learning algorithms, of which BNs were simply one type. Using the novel analytical framework, we show that: (1) BNs in healthcare are not used to their full potential; (2) a generic BN development process is lacking; (3) limitations exist in the way BNs in healthcare are presented in the literature, which impacts understanding, consensus towards systematic methodologies, practice and adoption; and (4) a gap exists between having an accurate BN and a useful BN that impacts clinical practice. This review highlights several neglected issues, such as restricted aims of BNs, ad hoc BN development methods, and the lack of BN adoption in practice and reveals to researchers and clinicians the need to address these problems. To map the way forward, the paper proposes future research directions and makes recommendations regarding BN development methods and adoption in practice.

Bayesian networks in healthcare: What is preventing their adoption?

There has been much research effort expended toward the use of Bayesian networks (BNs) in medical decision-making. However, because of the gap between developing an accurate BN and demonstrating its clinical usefulness, this has not resulted in any widespread BN adoption in clinical practice. This paper investigates this problem with the aim of finding an explanation and ways to address the problem through a comprehensive literature review of articles describing BNs in healthcare. Based on the literature collection that has been systematically narrowed down from 3810 to 116 most relevant articles, this paper analyses the benefits, barriers and facilitating factors (BBF) for implementing BN-based systems in healthcare using the ITPOSMO-BBF framework. A key finding is that works in the literature rarely consider barriers and even when these were identified they were not connected to facilitating factors. The main finding is that the barriers can be grouped into: (1) data inadequacies; (2) clinicians' resistance to new technologies; (3) lack of clinical credibility; (4) failure to demonstrate clinical impact; (5) absence of an acceptable predictive performance; and (6) absence of evidence for model's generalisability. The facilitating factors can be grouped into: (1) data collection improvements; (2) software and technological improvements; (3) having interpretable and easy to use BN-based systems; (4) clinical involvement in the development or review of the model; (5) investigation of model's clinical impact; (6) internal validation of the model's performance; and (7) external validation of the model. These groupings form a strong basis for a generic framework that could be used for formulating strategies for ensuring BN-based clinical decision-support system adoption in frontline care settings. The output of this review is expected to enhance the dialogue among researchers by providing a deeper understanding for the neglected issue of BN adoption in practice and promoting efforts for implementing BN-based systems.

Bayesian Network as a Decision Tool for Predicting ALS Disease

Clinical diagnosis of amyotrophic lateral sclerosis (ALS) is difficult in the early period. But blood tests are less time consuming and low cost methods compared to other methods for the diagnosis. The ALS researchers have been used machine learning methods to predict the genetic architecture of disease. In this study we take advantages of Bayesian networks and machine learning methods to predict the ALS patients with blood plasma protein level and independent personal features. According to the comparison results, Bayesian Networks produced best results with accuracy (0.887), area under the curve (AUC) (0.970) and other comparison metrics. We confirmed that sex and age are effective variables on the ALS. In addition, we found that the probability of onset involvement in the ALS patients is very high. Also, a person’s other chronic or neurological diseases are associated with the ALS disease. Finally, we confirmed that the Parkin level may also have an effect on the ALS disease. While this protein is at very low levels in Parkinson’s patients, it is higher in the ALS patients than all control groups.

Accurate Prediction of Coronary Heart Disease for Patients With Hypertension From Electronic Health Records With Big Data and Machine-Learning Methods: Model Development and Performance Evaluation

BACKGROUND

Predictions of cardiovascular disease risks based on health records have long attracted broad research interests. Despite extensive efforts, the prediction accuracy has remained unsatisfactory. This raises the question as to whether the data insufficiency, statistical and machine-learning methods, or intrinsic noise have hindered the performance of previous approaches, and how these issues can be alleviated.

OBJECTIVE

Based on a large population of patients with hypertension in Shenzhen, China, we aimed to establish a high-precision coronary heart disease (CHD) prediction model through big data and machine-learning.

METHODS

Data from a large cohort of 42,676 patients with hypertension, including 20,156 patients with CHD onset, were investigated from electronic health records (EHRs) 1-3 years prior to CHD onset (for CHD-positive cases) or during a disease-free follow-up period of more than 3 years (for CHD-negative cases). The population was divided evenly into independent training and test datasets. Various machine-learning methods were adopted on the training set to achieve high-accuracy prediction models and the results were compared with traditional statistical methods and well-known risk scales. Comparison analyses were performed to investigate the effects of training sample size, factor sets, and modeling approaches on the prediction performance.

RESULTS

An ensemble method, XGBoost, achieved high accuracy in predicting 3-year CHD onset for the independent test dataset with an area under the receiver operating characteristic curve (AUC) value of 0.943. Comparison analysis showed that nonlinear models (K-nearest neighbor AUC 0.908, random forest AUC 0.938) outperform linear models (logistic regression AUC 0.865) on the same datasets, and machine-learning methods significantly surpassed traditional risk scales or fixed models (eg, Framingham cardiovascular disease risk models). Further analyses revealed that using time-dependent features obtained from multiple records, including both statistical variables and changing-trend variables, helped to improve the performance compared to using only static features. Subpopulation analysis showed that the impact of feature design had a more significant effect on model accuracy than the population size. Marginal effect analysis showed that both traditional and EHR factors exhibited highly nonlinear characteristics with respect to the risk scores.

CONCLUSIONS

We demonstrated that accurate risk prediction of CHD from EHRs is possible given a sufficiently large population of training data. Sophisticated machine-learning methods played an important role in tackling the heterogeneity and nonlinear nature of disease prediction. Moreover, accumulated EHR data over multiple time points provided additional features that were valuable for risk prediction. Our study highlights the importance of accumulating big data from EHRs for accurate disease predictions.

A bibliometric analysis and visualization of medical data mining research

BACKGROUND

Data mining technology used in the field of medicine has been widely studied by scholars all over the world. But there is little research on medical data mining (MDM) from the perspectives of bibliometrics and visualization, and the research topics and development trends in this field are still unclear.

METHODS

This paper has applied bibliometric visualization software tools, VOSviewer 1.6.10 and CiteSpace V, to study the citation characteristics, international cooperation, author cooperation, and geographical distribution of the MDM.

RESULTS

A total of 1575 documents are obtained, and the most frequent document type is article (1376). SHAN NH is the most productive author, with the highest number of publications of 12, and the Gillies's article (750 times citation) is the most cited paper. The most productive country and institution in MDM is the USA (559) and US FDA (35), respectively. The Journal of Biomedical Informatics, Expert Systems with Applications and Journal of Medical Systems are the most productive journals, which reflected the nature of the research, and keywords "classification (790)" and "system (576)" have the strongest strength. The hot topics in MDM are drug discovery, medical imaging, vaccine safety, and so on. The 3 frontier topics are reporting system, precision medicine, and inflammation, and would be the foci of future research.

CONCLUSION

The present study provides a panoramic view of data mining methods applied in medicine by visualization and bibliometrics. Analysis of authors, journals, institutions, and countries could provide reference for researchers who are fresh to the field in different ways. Researchers may also consider the emerging trends when deciding the direction of their study.

Improved Interpretability of Machine Learning Model Using Unsupervised Clustering: Predicting Time to First Treatment in Chronic Lymphocytic Leukemia

PURPOSE

Time to event is an important aspect of clinical decision making. This is particularly true when diseases have highly heterogeneous presentations and prognoses, as in chronic lymphocytic lymphoma (CLL). Although machine learning methods can readily learn complex nonlinear relationships, many methods are criticized as inadequate because of limited interpretability. We propose using unsupervised clustering of the continuous output of machine learning models to provide discrete risk stratification for predicting time to first treatment in a cohort of patients with CLL.

PATIENTS AND METHODS

A total of 737 treatment-naïve patients with CLL diagnosed at Mayo Clinic were included in this study. We compared predictive abilities for two survival models (Cox proportional hazards and random survival forest) and four classification methods (logistic regression, support vector machines, random forest, and gradient boosting machine). Probability of treatment was then stratified.

RESULTS

Machine learning methods did not yield significantly more accurate predictions of time to first treatment. However, automated risk stratification provided by clustering was able to better differentiate patients who were at risk for treatment within 1 year than models developed using standard survival analysis techniques.

CONCLUSION

Clustering the posterior probabilities of machine learning models provides a way to better interpret machine learning models.

Using Data Analytics to Predict Hospital Mortality in Sepsis Patients

Predictive analytics can be used to anticipate the risks associated with some patients, and prediction models can be employed to alert physicians and allow timely proactive interventions. Recently, health care providers have been using different types of tools with prediction capabilities. Sepsis is one of the leading causes of in-hospital death in the United States and worldwide. In this study, the authors used a large medical dataset to develop and present a model that predicts in-hospital mortality among Sepsis patients. The predictive model was developed using a dataset of more than one million records of hospitalized patients. The independent predictors of in-hospital mortality were identified using the chi-square automatic interaction detector. The authors found that adding hospital attributes to the predictive model increased the accuracy from 82.08% to 85.3% and the area under the curve from 0.69 to 0.84, which is favorable compared to using only patients' attributes. The authors discuss the practical and research contributions of using a predictive model that incorporates both patient and hospital attributes in identifying high-risk patients.

Research on the integrated system of case-based reasoning and Bayesian network

With the arrival of big data era, the integration of Case-Based Reasoning (CBR) and Bayesian Network (BN) has become increasingly a promising technology in implementing the intelligence of engineering application. To further improve the efficiency of the integrated system and make it adapt to the large number of parameter under big data, Within-Cross (WC) algorithm is proposed in this paper to assigned big data to each slave node of Hadoop platform for parallel data processing. The WC algorithm can greatly reduce the time costing of parallel data processing and thus improve the efficiency of the integrated system by fully using the computation resource of Hadoop platform. To further enhance the accuracy of the integrated system, this paper proposes a new method called Weighted Super Parameters of Dirichlet Distribution (WSPDD) algorithm to perform probability learning. The WSPDD algorithm gives a weight to each super parameter of Dirichlet Distribution of all problem parameters to adjust the result of probability statistics and then improve probability learning. Therefore, the accuracy of the integrated system is greatly improved. Finally, an application domain is taken as a case study to validate the proposed method.

A Systematic Review on Healthcare Analytics: Application and Theoretical Perspective of Data Mining

The growing healthcare industry is generating a large volume of useful data on patient demographics, treatment plans, payment, and insurance coverage—attracting the attention of clinicians and scientists alike. In recent years, a number of peer-reviewed articles have addressed different dimensions of data mining application in healthcare. However, the lack of a comprehensive and systematic narrative motivated us to construct a literature review on this topic. In this paper, we present a review of the literature on healthcare analytics using data mining and big data. Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, we conducted a database search between 2005 and 2016. Critical elements of the selected studies—healthcare sub-areas, data mining techniques, types of analytics, data, and data sources—were extracted to provide a systematic view of development in this field and possible future directions. We found that the existing literature mostly examines analytics in clinical and administrative decision-making. Use of human-generated data is predominant considering the wide adoption of Electronic Medical Record in clinical care. However, analytics based on website and social media data has been increasing in recent years. Lack of prescriptive analytics in practice and integration of domain expert knowledge in the decision-making process emphasizes the necessity of future research.

Systematic mapping study of data mining–based empirical studies in cardiology

Data mining provides the methodology and technology to transform huge amount of data into useful information for decision making. It is a powerful process to extract knowledge and discover new patterns embedded in large data sets. Data mining has been increasingly used in medicine, particularly in cardiology. In fact, data mining applications can greatly benefits all parts involved in cardiology such as patients, cardiologists and nurses. This article aims to perform a systematic mapping study so as to analyze and synthesize empirical studies on the application of data mining techniques in cardiology. A total of 142 articles published between 2000 and 2015 were therefore selected, studied and analyzed according to the four following criteria: year and channel of publication, research type, medical task and empirical type. The results of this mapping study are discussed and a list of recommendations for researchers and cardiologists is provided.

Utilizing Chinese Admission Records for MACE Prediction of Acute Coronary Syndrome

Background: Clinical major adverse cardiovascular event (MACE) prediction of acute coronary syndrome (ACS) is important for a number of applications including physician decision support, quality of care assessment, and efficient healthcare service delivery on ACS patients. Admission records, as typical media to contain clinical information of patients at the early stage of their hospitalizations, provide significant potential to be explored for MACE prediction in a proactive manner. Methods: We propose a hybrid approach for MACE prediction by utilizing a large volume of admission records. Firstly, both a rule-based medical language processing method and a machine learning method (i.e., Conditional Random Fields (CRFs)) are developed to extract essential patient features from unstructured admission records. After that, state-of-the-art supervised machine learning algorithms are applied to construct MACE prediction models from data. Results: We comparatively evaluate the performance of the proposed approach on a real clinical dataset consisting of 2930 ACS patient samples collected from a Chinese hospital. Our best model achieved 72% AUC in MACE prediction. In comparison of the performance between our models and two well-known ACS risk score tools, i.e., GRACE and TIMI, our learned models obtain better performances with a significant margin. Conclusions: Experimental results reveal that our approach can obtain competitive performance in MACE prediction. The comparison of classifiers indicates the proposed approach has a competitive generality with datasets extracted by different feature extraction methods. Furthermore, our MACE prediction model obtained a significant improvement by comparison with both GRACE and TIMI. It indicates that using admission records can effectively provide MACE prediction service for ACS patients at the early stage of their hospitalizations.

A semi-mechanism approach based on MRI and proteomics for prediction of conversion from mild cognitive impairment to Alzheimer's disease

Mild cognitive impairment (MCI) is a precursor phase of Alzheimer’s disease (AD). As current treatments may be effective only at the early stages of AD, it is important to track MCI patients who will convert to AD. The aim of this study is to develop a high performance semi-mechanism based approach to predict the conversion from MCI to AD and improve our understanding of MCI-to-AD conversion mechanism. First, analysis of variance (ANOVA) test and lasso regression are employed to identify the markers related to the conversion. Then the Bayesian network based on selected markers is established to predict MCI-to-AD conversion. The structure of Bayesian network suggests that the conversion may start with fibrin clot formation, verbal memory impairment, eating pattern changing and hyperinsulinemia. The Bayesian network achieves a high 10-fold cross-validated prediction performance with 96% accuracy, 95% sensitivity, 65% specificity, area under the receiver operating characteristic curve of 0.82 on data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. The semi-mechanism based approach provides not only high prediction performance but also clues of mechanism for MCI-to-AD conversion.

Adapting machine learning techniques to censored time-to-event health record data: A general-purpose approach using inverse probability of censoring weighting

Models for predicting the probability of experiencing various health outcomes or adverse events over a certain time frame (e.g., having a heart attack in the next 5years) based on individual patient characteristics are important tools for managing patient care. Electronic health data (EHD) are appealing sources of training data because they provide access to large amounts of rich individual-level data from present-day patient populations. However, because EHD are derived by extracting information from administrative and clinical databases, some fraction of subjects will not be under observation for the entire time frame over which one wants to make predictions; this loss to follow-up is often due to disenrollment from the health system. For subjects without complete follow-up, whether or not they experienced the adverse event is unknown, and in statistical terms the event time is said to be right-censored. Most machine learning approaches to the problem have been relatively ad hoc; for example, common approaches for handling observations in which the event status is unknown include (1) discarding those observations, (2) treating them as non-events, (3) splitting those observations into two observations: one where the event occurs and one where the event does not. In this paper, we present a general-purpose approach to account for right-censored outcomes using inverse probability of censoring weighting (IPCW). We illustrate how IPCW can easily be incorporated into a number of existing machine learning algorithms used to mine big health care data including Bayesian networks, k-nearest neighbors, decision trees, and generalized additive models. We then show that our approach leads to better calibrated predictions than the three ad hoc approaches when applied to predicting the 5-year risk of experiencing a cardiovascular adverse event, using EHD from a large U.S. Midwestern healthcare system.

A probabilistic topic model for clinical risk stratification from electronic health records

BACKGROUND AND OBJECTIVE

Risk stratification aims to provide physicians with the accurate assessment of a patient's clinical risk such that an individualized prevention or management strategy can be developed and delivered. Existing risk stratification techniques mainly focus on predicting the overall risk of an individual patient in a supervised manner, and, at the cohort level, often offer little insight beyond a flat score-based segmentation from the labeled clinical dataset. To this end, in this paper, we propose a new approach for risk stratification by exploring a large volume of electronic health records (EHRs) in an unsupervised fashion.

METHODS

Along this line, this paper proposes a novel probabilistic topic modeling framework called probabilistic risk stratification model (PRSM) based on Latent Dirichlet Allocation (LDA). The proposed PRSM recognizes a patient clinical state as a probabilistic combination of latent sub-profiles, and generates sub-profile-specific risk tiers of patients from their EHRs in a fully unsupervised fashion. The achieved stratification results can be easily recognized as high-, medium- and low-risk, respectively. In addition, we present an extension of PRSM, called weakly supervised PRSM (WS-PRSM) by incorporating minimum prior information into the model, in order to improve the risk stratification accuracy, and to make our models highly portable to risk stratification tasks of various diseases.

RESULTS

We verify the effectiveness of the proposed approach on a clinical dataset containing 3463 coronary heart disease (CHD) patient instances. Both PRSM and WS-PRSM were compared with two established supervised risk stratification algorithms, i.e., logistic regression and support vector machine, and showed the effectiveness of our models in risk stratification of CHD in terms of the Area Under the receiver operating characteristic Curve (AUC) analysis. As well, in comparison with PRSM, WS-PRSM has over 2% performance gain, on the experimental dataset, demonstrating that incorporating risk scoring knowledge as prior information can improve the performance in risk stratification.

CONCLUSIONS

Experimental results reveal that our models achieve competitive performance in risk stratification in comparison with existing supervised approaches. In addition, the unsupervised nature of our models makes them highly portable to the risk stratification tasks of various diseases. Moreover, patient sub-profiles and sub-profile-specific risk tiers generated by our models are coherent and informative, and provide significant potential to be explored for the further tasks, such as patient cohort analysis. We hypothesize that the proposed framework can readily meet the demand for risk stratification from a large volume of EHRs in an open-ended fashion.