Sakyajit Bhattacharya

Predicting Complications in Critical Care Using Heterogeneous Clinical Data

Patients in hospitals, particularly in critical care, are susceptible to many complications affecting morbidity and mortality. Digitized clinical data in electronic medical records can be effectively used to develop machine learning models to identify patients at risk of complications early and provide prioritized care to prevent complications. However, clinical data from heterogeneous sources within hospitals pose significant modeling challenges. In particular, unstructured clinical notes are a valuable source of information containing regular assessments of the patients condition but contain inconsistent abbreviations and lack the structure of formal documents. Our contributions in this paper are twofold. First, we present a new preprocessing technique for extracting features from informal clinical notes that can be used in a classification model to identify patients at risk of developing complications. Second, we explore the use of collective matrix factorization, a multi-view learning technique, to model heterogeneous clinical data-text-based features in combination with other measurements, such as clinical investigations, comorbidites, and demographic data. We present a detailed case study on postoperative respiratory failure using more than 700 patient records from the MIMIC II database. Our experiments demonstrate the efficacy of our preprocessing technique in extracting discriminatory features from clinical notes as well as the benefits of multi-view learning to combine clinical measurements with text data for predicting complications.

Classification with imbalance: A similarity-based method for predicting respiratory failure

Binary classification based methods are commonly used for designing predictive models in healthcare. A common problem in many healthcare datasets is that of imbalance, where there are far more observations in one class than the other during training. In such conditions, most classifiers do not have good predictive accuracy with respect to the under-represented class. We design a new similarity-based classifier to learn from imbalanced datasets, wherein input features are transformed using similarity with respect to a chosen subset of training points. We empirically demonstrate the superiority of our algorithm over state-of-the-art methods for imbalanced data classification in real and synthetic datasets. We also illustrate the application of our classifier in predicting Acute Respiratory Failure (ARF), a critical complication in Intensive Care Units (ICU), using semi-structured text contained in nursing notes recorded during a patients ICU stay. Our experiments, on more than 800 patient records show that using our new classifier to learn from text- based features can effectively be used to predict ARF and, potentially, other complications in ICUs.

Predicting Postoperative Acute Respiratory Failure in critical care using nursing notes and physiological signals

Postoperative Acute Respiratory Failure (ARF) is a serious complication in critical care affecting patient morbidity and mortality. In this paper we investigate a novel approach to predicting ARF in critically ill patients. We study the use of two disparate sources of information - semi-structured text contained in nursing notes and investigative reports that are regularly recorded and the respiration rate, a physiological signal that is continuously monitored during a patients ICU stay. Unlike previous works that retrospectively analyze complications, we exclude discharge summaries from our analysis envisaging a real time system that predicts ARF during the ICU stay. Our experiments, on more than 800 patient records from the MIMIC II database, demonstrate that text sources within the ICU contain strong signals for distinguishing between patients who are at risk for ARF from those who are not at risk. These results suggest that large scale systems using both structured and unstructured data recorded in critical care can be effectively used to predict complications, which in turn can lead to preemptive care with potentially improved outcomes, mortality rates and decreased length of stay and cost.

A statistical model for stroke outcome prediction and treatment planning

Stroke is a major cause of mortality and long-term disability in the world. Predictive outcome models in stroke are valuable for personalized treatment, rehabilitation planning and in controlled clinical trials. We design a new multi-class classification model to predict outcome in the short-term, the putative therapeutic window for several treatments. Our model addresses the challenges of class imbalance, where the training data is dominated by samples of a single class, and highly correlated predictor and outcome variables, which makes learning the effects of treatments on the outcome difficult. Empirically our model outperforms the best-known previous predictive models and can infer the most effective treatments in improving outcome that have been independently validated in clinical studies.

A dual boundary classifier for predicting acute hypotensive episodes in critical care

An Acute Hypotensive Episode (AHE) is the sudden onset of a sustained period of low blood pressure and is one among the most critical conditions in Intensive Care Units (ICU). Without timely medical care, it can lead to an irreversible organ damage and death. By identifying patients at risk for AHE early, adequate medical intervention can save lives and improve patient outcomes. In this paper, we design a novel dual–boundary classification based approach for identifying patients at risk for AHE. Our algorithm uses only simple summary statistics of past Blood Pressure measurements and can be used in an online environment facilitating real–time updates and prediction. We perform extensive experiments with more than 4,500 patient records and demonstrate that our method outperforms the previous best approaches of AHE prediction. Our method can identify AHE patients two hours in advance of the onset, giving sufficient time for appropriate clinical intervention with nearly 80% sensitivity and at 95% specificity, thus having very few false positives.

Clinical Decision Support for Stroke Using Multi---view Learning Based Models for NIHSS Scores

Cerebral stroke is a leading cause of physical disability and death in the world. The severity of a stroke is assessed by a neurological examination using a scale known as the NIH stroke scale NIHSS. As a measure of stroke severity, the NIHSS score is widely adopted and has been found to also be useful in outcome prediction, rehabilitation planning and treatment planning. In many applications, such as in patient triage in under---resourced primary health care centres and in automated clinical decision support tools, it would be valuable to obtain the severity of stroke with minimal human intervention using simple parameters like age, past conditions and blood investigations. In this paper we propose a new model for predicting NIHSS scores which, to our knowledge, is the first statistical model for stroke severity. Our multi---view learning approach can handle data from heterogeneous sources with mixed data distributions binary, categorical and numerical and is robust against missing values --- strengths that many other modeling techniques lack. In our experiments we achieve better predictive accuracy than other commonly used methods.

CloudRank: A statistical modelling framework for characterizing user behaviour towards targeted cloud management

A rank clustering system, CloudRank, is proposed that takes into account cloud user preference data to characterize cloud user behaviour and also identify (an initially unknown set of) groups of users with similar behaviour in an unsupervised manner. The user groups are determined based on fitting mixture models on the cloud user preference observations. A preference can be anything that a system designer would like to include to characterize high-level user requirements such as demands on performance, cost, security, availability, etc. CloudRank can be useful for: (i) cloud providers to target their service offerings according to the user groups (i.e. customer segments) through appropriate customization of services pertaining to the user groups typical requirements; (ii) recommendation systems or a marketplace (that enables inter-operability among different providers) to determine which offerings best suit certain user groups; and (iii) prediction of any new users behaviour based on their preference information. Results on realistic feedbacks from internal cloud service providers show an average of 80% accuracy of the proposed unsupervised technique. When compared with a supervised technique, i.e. when the number of user groups are known beforehand, the error is within 15%, thus making it a promising technique for realistic deployments, particularly when there is no prior knowledge regarding the clusters.