Adversarial reinforcement learning for dynamic treatment regimes.

Treatment recommendation, as a critical task of delivering effective interventions according to patient state and expected outcome, plays a vital role in precision medicine and healthcare management. As a well-suited tactic to learn optimal policies of recommender systems, reinforcement learning is promising to address the challenge of treatment recommendation. However, existing solutions mostly require frequent interactions between treatment recommender systems and clinical environment, which are expensive, time-consuming, and even infeasible in clinical practice. In this study, we present a novel model-based offline reinforcement learning approach to optimize a treatment policy by utilizing patient treatment trajectories in Electronic Health Records (EHRs). Specifically, a patient treatment trajectory simulator is firstly constructed based on the ground-truth trajectories in EHRs. Thereafter, the constructed simulator is utilized to model the online interactions between the treatment recommender system and clinical environment. In this way, the counterfactual trajectories can be generated. To alleviate the bias deriving from the ground-truth and the counterfactual trajectories, an adversarial network is incorporated into the proposed model, such that a large space of treatment actions can be explored with the scaled rewards. The proposed model is evaluated on a simulated dataset and a real-world dataset. The experimental results demonstrate that the proposed model is superior to other methods, giving rise to a new solution for dynamic treatment regimes and beyond.

Knowledge-aware patient representation learning for multiple disease subtypes.

Learning latent representations of patients with a target disease is a core problem in a broad range of downstream applications, such as clinical endpoint prediction. The suffering of patients may have multiple subtypes with certain similarities and differences, which need to be addressed for learning effective patient representation to facilitate the downstream tasks. However, existing studies either ignore the distinction of disease subtypes to learn disease-level representations, or neglect the correlations between subtypes and only learn disease subtype-level representations, which affects the performance of patient representation learning. To alleviate this problem, we studied how to effectively integrate data from all disease subtypes to improve the representation of each subtype. Specifically, we proposed a knowledge-aware shared-private neural network model to explicitly use disease-oriented knowledge and learn shared and specific representations from the disease and its subtype perspectives. To evaluate the feasibility of the proposed model, we conducted a particular downstream task, i.e., clinical endpoint prediction, on the basis of the learned patient presentations. The results on the real-world clinical datasets demonstrated that our model could yield a significant improvement over state-of-the-art models.

On learning disentangled representations for individual treatment effect estimation.

OBJECTIVE: Estimating the individualized treatment effect (ITE) from observational data is a challenging task due to selection bias, which results from the distributional discrepancy between different treatment groups caused by the dependence between features and assigned treatments. This dependence is induced by the factors related to the treatment assignment. We hypothesize that features consist of three types of latent factors: outcome-specific factors, treatment-specific factors and confounders. Then, we aim to reduce the influence of treatment-related factors, i.e., treatment-specific factors and confounders, on outcome prediction to mitigate the effects of selection bias. METHOD: We present a novel representation learning model in which both the main task of outcome prediction and the auxiliary task of classifying the treatment assignment are used to learn the outcome-oriented and treatment-oriented latent representations, respectively. However, since the confounders are related to both treatment assignment and outcome, it is still contained in the representations. To further reduce influence of the confounders contained in both representations, individualized orthogonal regularization is incorporated into the proposed model. The orthogonal regularization forces the outcome-oriented and treatment-oriented latent representations of an individual to be vertical in the inner product space, meaning they are orthogonal with each other, and the common information of confounder is reduced. Such that the ITE can be estimated more precisely without the effects of selection bias. RESULT: We evaluate our proposed model on a semi-simulated dataset and a real-world dataset. The experimental results demonstrate that the proposed model achieves competitive or better performance compared with the performances of the state-of-the-art models. CONCLUSION: The proposed method is well performed on ITE estimation with the ability to reduce selection bias thoroughly by incorporating an auxiliary task and adopting orthogonal regularization to disentangle the latent factors. SIGNIFICANCE: This paper offers a novel method of reducing selection bias in estimating the ITE from observational data by disentangled representation learning.

Knowledge-aware multi-center clinical dataset adaptation: Problem, method, and application.

Adaptable utilization of clinical data collected from multiple centers, prompted by the need to overcome the shifts between the dataset distributions, and exploit these different datasets for potential clinical applications, has received significant attention in recent years. In this study, we propose a novel approach to this task by infusing an external knowledge graph (KG) into multi-center clinical data mining. Specifically, we propose an adversarial learning model to capture shared patient feature representations from multi-center heterogeneous clinical datasets, and employ an external KG to enrich the semantics of the patient sample by providing both clinical center-specific and center-general knowledge features, which are trained with a graph convolutional autoencoder. We evaluate the proposed model on a real clinical dataset extracted from the general cardiology wards of a Chinese hospital and a well-known public clinical dataset (MIMIC III, pertaining to ICU clinical settings) for the task of predicting acute kidney injury in patients with heart failure. The achieved experimental results demonstrate the efficacy of our proposed model.

Endpoint prediction of heart failure using electronic health records.

BACKGROUND: Heart failure (HF) is a serious condition associated with high morbidity and mortality rates. Effective endpoint prediction in patient treatment trajectories provides preventative information about HF prognosis, guides decision making about the type and intensity of care, and enables better understanding of provider performance. OBJECTIVE: We explored the potential of a large volume of electronic health records (EHRs) for endpoint prediction of HF. Specifically, a suite of patient features observed at the prediction time point were utilized as the auxiliary information during the training of the prediction model. MATERIAL AND METHOD: We extract the latent representation of patient treatment trajectory by equipping a recurrent neural network (RNN) with two learning strategies, namely adversarial learning and multi-task learning. As for the adversarial learning strategy, an adversarial learning scheme is used to differentiate the generated feature vector from the real one, while in the multi-task learning strategy, we consider the prediction of patient feature vector as an auxiliary task other than endpoint prediction. With such learning strategies, the extracted representation of patient treatment trajectory is particularly optimized for predicting HF endpoint, including HF-readmission, all-cause mortality and their combination (i.e., composite endpoint). RESULTS AND DISCUSSION: We evaluate the proposed approach on a real clinical dataset collected from a Chinese hospital. The experimental dataset contains 2102 HF patient treatment trajectories with 13,545 visits on the hospital. The area under the ROC curve (AUC) achieved by our best model in predicting composite endpoint is 0.744, which is better than that of state-of-the-art models, including the standard Long Short Term Memory (0.727), Gated Recurrent Unit (0.732), RETAIN(0.730). With respect to the prediction of HF-readmission and all-cause mortality, our method also shows better performance than benchmark models. CONCLUSION: The experimental results show that the proposed model can achieve competitive performance over state-of-the-art models in terms of endpoint prediction for HF, and reveal some suggestive hypotheses that could be validated by further investigations in the medical domain.

Interpretable clinical prediction via attention-based neural network.

BACKGROUND: The interpretability of results predicted by the machine learning models is vital, especially in the critical fields like healthcare. With the increasingly adoption of electronic healthcare records (EHR) by the medical organizations in the last decade, which accumulated abundant electronic patient data, neural networks or deep learning techniques are gradually being applied to clinical tasks by utilizing the huge potential of EHR data. However, typical deep learning models are black-boxes, which are not transparent and the prediction outcomes of which are difficult to interpret. METHODS: To remedy this limitation, we propose an attention neural network model for interpretable clinical prediction. In detail, the proposed model employs an attention mechanism to capture critical/essential features with their attention signals on the prediction results, such that the predictions generated by the neural network model can be interpretable. RESULTS: We evaluate our proposed model on a real-world clinical dataset consisting of 736 samples to predict readmissions for heart failure patients. The performance of the proposed model achieved 66.7 and 69.1% in terms of accuracy and AUC, respectively, and outperformed the baseline models. Besides, we displayed patient-specific attention weights, which can not only help clinicians understand the prediction outcomes, but also assist them to select individualized treatment strategies or intervention plans. CONCLUSIONS: The experimental results demonstrate that the proposed model can improve both the prediction performance and interpretability by equipping the model with an attention mechanism.

Treatment effect prediction with adversarial deep learning using electronic health records.

BACKGROUND: Treatment effect prediction (TEP) plays an important role in disease management by ensuring that the expected clinical outcomes are obtained after performing specialized and sophisticated treatments on patients given their personalized clinical status. In recent years, the wide adoption of electronic health records (EHRs) has provided a comprehensive data source for intelligent clinical applications including the TEP investigated in this study. METHOD: We examined the problem of using a large volume of heterogeneous EHR data to predict treatment effects and developed an adversarial deep treatment effect prediction model to address the problem. Our model employed two auto-encoders for learning the representative and discriminative features of both patient characteristics and treatments from EHR data. The discriminative power of the learned features was further enhanced by decoding the correlational information between the patient characteristics and subsequent treatments by means of a generated adversarial learning strategy. Thereafter, a logistic regression layer was appended on the top of the resulting feature representation layer for TEP. RESULT: The proposed model was evaluated on two real clinical datasets collected from the cardiology department of a Chinese hospital. In particular, on acute coronary syndrome (ACS) dataset, the proposed adversarial deep treatment effect prediction (ADTEP) (0.662) exhibited 1.4, 2.2, and 6.3% performance gains in terms of the area under the ROC curve (AUC) over deep treatment effect prediction (DTEP) (0.653), logistic regression (LR) (0.648), and support vector machine (SVM) (0.621), respectively. As for heart failure (HF) case study, the proposed ADTEP also outperformed all benchmarks. The experimental results demonstrated that our proposed model achieved competitive performance compared to state-of-the-art models in tackling the TEP problem. CONCLUSION: In this work, we propose a novel model to address the TEP problem by utilizing a large volume of observational data from EHR. With adversarial learning strategy, our proposed model can further explore the correlational information between patient statuses and treatments to extract more robust and discriminative representation of patient samples from their EHR data. Such representation finally benefits the model on TEP. The experimental results of two case studies demonstrate the superiority of our proposed method compared to state-of-the-art methods.

Deep representation learning for individualized treatment effect estimation using electronic health records.

Utilizing clinical observational data to estimate individualized treatment effects (ITE) is a challenging task, as confounding inevitably exists in clinical data. Most of the existing models for ITE estimation tackle this problem by creating unbiased estimators of the treatment effects. Although valuable, learning a balanced representation is sometimes directly opposed to the objective of learning an effective and discriminative model for ITE estimation. We propose a novel hybrid model bridging multi-task deep learning and K-nearest neighbors (KNN) for ITE estimation. In detail, the proposed model firstly adopts multi-task deep learning to extract both outcome-predictive and treatment-specific latent representations from Electronic Health Records (EHR), by jointly performing the outcome prediction and treatment category classification. Thereafter, we estimate counterfactual outcomes by KNN based on the learned hidden representations. We validate the proposed model on a widely used semi-simulated dataset, i.e. IHDP, and a real-world clinical dataset consisting of 736 heart failure (HF) patients. The performance of our model remains robust and reaches 1.7 and 0.23 in terms of Precision in the estimation of heterogeneous effect (PEHE) and average treatment effect (ATE), respectively, on IHDP dataset, and 0.703 and 0.796 in terms of accuracy and F1 score respectively, on HF dataset. The results demonstrate that the proposed model achieves competitive performance over state-of-the-art models. In addition, the results reveal several findings which are consistent with existing medical domain knowledge, and discover certain suggestive hypotheses that could be validated through further investigations in the clinical domain.

Utilizing dynamic treatment information for MACE prediction of acute coronary syndrome.

BACKGROUND: Main adverse cardiac events (MACE) are essentially composite endpoints for assessing safety and efficacy of treatment processes of acute coronary syndrome (ACS) patients. Timely prediction of MACE is highly valuable for improving the effects of ACS treatments. Most existing tools are specific to predict MACE by mainly using static patient features and neglecting dynamic treatment information during learning. METHODS: We address this challenge by developing a deep learning-based approach to utilize a large volume of heterogeneous electronic health record (EHR) for predicting MACE after ACS. Specifically, we obtain the deep representation of dynamic treatment features from EHR data, using the bidirectional recurrent neural network. And then, the extracted latent representation of treatment features can be utilized to predict whether a patient occurs MACE in his or her hospitalization. RESULTS: We validate the effectiveness of our approach on a clinical dataset containing 2930 ACS patient samples with 232 static feature types and 2194 dynamic feature types. The performance of our best model for predicting MACE after ACS remains robust and reaches 0.713 and 0.764 in terms of AUC and Accuracy, respectively, and has over 11.9% (1.2%) and 1.9% (7.5%) performance gain of AUC (Accuracy) in comparison with both logistic regression and a boosted resampling model presented in our previous work, respectively. The results are statistically significant. CONCLUSIONS: We hypothesize that our proposed model adapted to leverage dynamic treatment information in EHR data appears to boost the performance of MACE prediction for ACS, and can readily meet the demand clinical prediction of other diseases, from a large volume of EHR in an open-ended fashion.

Evidential MACE prediction of acute coronary syndrome using electronic health records.

BACKGROUND: Major adverse cardiac event (MACE) prediction plays a key role in providing efficient and effective treatment strategies for patients with acute coronary syndrome (ACS) during their hospitalizations. Existing prediction models have limitations to cope with imprecise and ambiguous clinical information such that clinicians cannot reach to reliable MACE prediction results for individuals. METHODS: To remedy it, this study proposes a hybrid method using Rough Set Theory (RST) and Dempster-Shafer Theory (DST) of evidence. In details, four state-of-the-art models, including one traditional ACS risk scoring model, i.e., GRACE, and three machine learning based models, i.e., Support Vector Machine, L1-Logistic Regression, and Classification and Regression Tree, are employed to generate initial MACE prediction results, and then RST is applied to determine the weights of the four single models. After that, the acquired prediction results are assumed as basic beliefs for the problem propositions and in this way, an evidential prediction result is generated based on DST in an integrative manner. RESULTS: Having applied the proposed method on a clinical dataset consisting of 2930 ACS patient samples, our model achieves 0.715 AUC value with competitive standard deviation, which is the best prediction results comparing with the four single base models and two baseline ensemble models. CONCLUSIONS: Facing with the limitations in traditional ACS risk scoring models, machine learning models and the uncertainties of EHR data, we present an ensemble approach via RST and DST to alleviate this problem. The experimental results reveal that our proposed method achieves better performance for the problem of MACE prediction when compared with the single models.

Using neural attention networks to detect adverse medical events from electronic health records.

The detection of Adverse Medical Events (AMEs) plays an important role in disease management in ensuring efficient treatment delivery and quality improvement of health services. Recently, with the rapid development of hospital information systems, a large volume of Electronic Health Records (EHRs) have been produced, in which AMEs are regularly documented in a free-text manner. In this study, we are concerned with the problem of AME detection by utilizing a large volume of unstructured EHR data. To address this challenge, we propose a neural attention network-based model to incorporate the contextual information of words into AME detection. Specifically, we develop a context-aware attention mechanism to locate salient words with respect to the target AMEs in patient medical records. And then we combine the proposed context attention mechanism with the deep learning tactic to boost the performance of AME detection. We validate our proposed model on a real clinical dataset that consists of 8845 medical records of patients with cardiovascular diseases. The experimental results show that our proposed model advances state-of-the-art models and achieves competitive performance in terms of AME detection.

Probabilistic modeling personalized treatment pathways using electronic health records.

BACKGROUND: Modeling personalized treatment pathways plays an important role in understanding essential/critical treatment behaviors performed on patients during their hospitalizations and thus provides the opportunity for the improvement of better health service delivery in treatment pathways. OBJECTIVE: Unlike traditional business process mining, modeling personalized treatment pathways is more challenging because they are typically case-specific. Although several studies have been devoted to modeling patient treatment pathways, limited efforts have been made on the extraction of latent semantics and their transitions behind patient treatment pathways, which are often ambiguous and poorly understood. METHODS: In this article, we propose an extension of the Hidden Markov Model to mine and model personalized treatment pathways by extracting latent treatment topics and identifying their sequential dependencies in pathways, in the form of probabilistic distributions and transitions of patients' raw Electronic Health Record (EHR) data. RESULTS: We evaluated the proposed model on 48,024 patients with cardiovascular diseases. A total of 15 treatment topics and their typical transition routes were discovered from EHR data that contained 1,391,251 treatment events with 2786 types of interventions and that were evaluated by ten clinicians manually. The obtained p-values are 0.000146 and 0.009106 in comparison with both Latent Dirichlet Allocation and Sequent Naïve Bayes models, respectively; this outcome indicate that our approach achieves a better understanding of human evaluators on modeling personalized treatment pathway than that of benchmark models. CONCLUSION: The experimental results on a real-world data set clearly suggest that the proposed model has efficiency in mining and modeling personalized treatment pathways. We argue that the discovered treatment topics and their transition routes, as actionable knowledge that represents the practice of treating individual patients in their clinical pathways, can be further exploited to help physicians better understand their specialty and learn from previous experiences for treatment analysis and improvement.

MACE prediction of acute coronary syndrome via boosted resampling classification using electronic medical records.

OBJECTIVES: Major adverse cardiac events (MACE) of acute coronary syndrome (ACS) often occur suddenly resulting in high mortality and morbidity. Recently, the rapid development of electronic medical records (EMR) provides the opportunity to utilize the potential of EMR to improve the performance of MACE prediction. In this study, we present a novel data-mining based approach specialized for MACE prediction from a large volume of EMR data. METHODS: The proposed approach presents a new classification algorithm by applying both over-sampling and under-sampling on minority-class and majority-class samples, respectively, and integrating the resampling strategy into a boosting framework so that it can effectively handle imbalance of MACE of ACS patients analogous to domain practice. The method learns a new and stronger MACE prediction model each iteration from a more difficult subset of EMR data with wrongly predicted MACEs of ACS patients by a previous weak model. RESULTS: We verify the effectiveness of the proposed approach on a clinical dataset containing 2930 ACS patient samples with 268 feature types. While the imbalanced ratio does not seem extreme (25.7%), MACE prediction targets pose great challenge to traditional methods. As these methods degenerate dramatically with increasing imbalanced ratios, the performance of our approach for predicting MACE remains robust and reaches 0.672 in terms of AUC. On average, the proposed approach improves the performance of MACE prediction by 4.8%, 4.5%, 8.6% and 4.8% over the standard SVM, Adaboost, SMOTE, and the conventional GRACE risk scoring system for MACE prediction, respectively. CONCLUSIONS: We consider that the proposed iterative boosting approach has demonstrated great potential to meet the challenge of MACE prediction for ACS patients using a large volume of EMR.

On mining latent topics from healthcare chat logs.

BACKGROUND: Public and internet-based social media such as online healthcare-oriented chat groups provide a convenient channel for patients and people concerned about health to communicate and share information with each other. The chat logs of an online healthcare-oriented chat group can potentially be used to extract latent topics, to encourage participation, and to recommend relevant healthcare information to users. OBJECTIVE: This paper addresses the use of online healthcare chat logs to automatically discover both underlying topics and user interests. METHOD: We present a new probabilistic model that exploits healthcare chat logs to find hidden topics and changes in these topics over time. The proposed model uses separate but associated hidden variables to explore both topics and individual interests such that it can provide useful insights to the participants of online healthcare chat groups about their interests in terms of weighted topics or vice versa. RESULTS: We evaluate the proposed model on a real-world chat log by comparing its performance to benchmark topic models, i.e., latent Dirichlet allocation (LDA) and Author Topic Model (ATM), on the topic extraction task. The chat log is obtained from an online chat group of pregnant women, which consists of 233,452 chat word tokens contributed by 118 users. Both detected individual interests and underlying topics with their progressive information over time are demonstrated. The results show that the performance of the proposed model exceeds that of the benchmark models. CONCLUSION: The experimental results illustrate that the proposed model is a promising method for extracting healthcare knowledge from social media data.

Incorporating comorbidities into latent treatment pattern mining for clinical pathways.

In healthcare organizational settings, the design of a clinical pathway (CP) is challenging since patients following a particular pathway may have not only one single first-diagnosis but also several typical comorbidities, and thus it requires different disciplines involved to put together their partial knowledge about the overall pathway. Although many data mining techniques have been proposed to discover latent treatment information for CP analysis and reconstruction from a large volume of clinical data, they are specific to extract nontrivial information about the therapy and treatment of the first-diagnosis. The influence of comorbidities on adopting essential treatments is crucial for a pathway but has seldom been explored. This study proposes to extract latent treatment patterns that characterize essential treatments for both first-diagnosis and typical comorbidities from the execution data of a pathway. In particular, we propose a generative statistical model to extract underlying treatment patterns, unveil the latent associations between diagnosis labels (including both first-diagnosis and comorbidities) and treatments, and compute the contribution of comorbidities in these patterns. The proposed model extends latent Dirichlet allocation with an additional layer for diagnosis modeling. It first generates a set of latent treatment patterns from diagnosis labels, followed by sampling treatments from each pattern. We verify the effectiveness of the proposed model on a real clinical dataset containing 12,120 patient traces, which pertain to the unstable angina CP. Three treatment patterns are discovered from data, indicating latent correlations between comorbidities and treatments in the pathway. In addition, a possible medical application in terms of treatment recommendation is provided to illustrate the potential of the proposed model. Experimental results indicate that our approach can discover not only meaningful latent treatment patterns exhibiting comorbidity focus, but also implicit changes of treatments of first-diagnosis due to the incorporation of typical comorbidities potentially.

Outcome Prediction in Clinical Treatment Processes.

Clinical outcome prediction, as strong implications for health service delivery of clinical treatment processes (CTPs), is important for both patients and healthcare providers. Prior studies typically use a priori knowledge, such as demographics or patient physical factors, to estimate clinical outcomes at early stages of CTPs (e.g., admission). They lack the ability to deal with temporal evolution of CTPs. In addition, most of the existing studies employ data mining or machine learning methods to generate a prediction model for a specific type of clinical outcome, however, a mathematical model that predicts multiple clinical outcomes simultaneously, has not yet been established. In this study, a hybrid approach is proposed to provide a continuous predictive monitoring service on multiple clinical outcomes. More specifically, a probabilistic topic model is applied to discover underlying treatment patterns of CTPs from electronic medical records. Then, the learned treatment patterns, as low-dimensional features of CTPs, are exploited for clinical outcome prediction across various stages of CTPs based on multi-label classification. The proposal is evaluated to predict three typical classes of clinical outcomes, i.e., length of stay, readmission time, and the type of discharge, using 3492 pieces of patients' medical records of the unstable angina CTP, extracted from a Chinese hospital. The stable model was characterized by 84.9% accuracy and 6.4% hamming-loss with 3 latent treatment patterns discovered from data, which outperforms the benchmark multi-label classification algorithms for clinical outcome prediction. Our study indicates the proposed approach can potentially improve the quality of clinical outcome prediction, and assist physicians to understand the patient conditions, treatment inventions, and clinical outcomes in an integrated view.

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.

Discovery of clinical pathway patterns from event logs using probabilistic topic models.

Discovery of clinical pathway (CP) patterns has experienced increased attention over the years due to its importance for revealing the structure, semantics and dynamics of CPs, and to its usefulness for providing clinicians with explicit knowledge which can be directly used to guide treatment activities of individual patients. Generally, discovery of CP patterns is a challenging task as treatment behaviors in CPs often have a large variability depending on factors such as time, location and patient individual. Based on the assumption that CP patterns can be derived from clinical event logs which usually record various treatment activities in CP executions, this study proposes a novel approach to CP pattern discovery by modeling CPs using mixtures of an extension to the Latent Dirichlet Allocation family that jointly models various treatment activities and their occurring time stamps in CPs. Clinical case studies are performed to evaluate the proposed approach via real-world data sets recording typical treatment behaviors in patient careflow. The obtained results demonstrate the suitability of the proposed approach for CP pattern discovery, and indicate the promise in research efforts related to CP analysis and optimization.