Suicide is the second leading cause of death among 25-34 year olds and the third leading cause of death among 15-25 year olds in the United States. In the Emergency Department, where suicidal patients often present, estimating the risk of repeated attempts is generally left to clinical judgment. This paper presents our second attempt to determine the role of computational algorithms in u nderstanding a suicidal patient's thoughts, as represented by suicide notes. We focus on developing methods of natural language p rocessing that distinguish between genuine and elicited suicide notes. We hypothesize that machine learning algorithms can categorize suicide notes as well as mental health professionals and psychiatric physician trainees do. The data used are comprised of suicide notes from 33 suicide completers and matched to 33 elicited notes from healthy control group members. Eleven mental health profess ionals and 31 psychiatric trainees were asked to decide if a note was genuine or elicited. Their decisions were compared to nine different machine-learning algorithms. The results indicate that trainees accurately classified notes 49% of the time, mental health professionals accurately classified notes 63% of the time, and the best machine learning algorithm accurately classified the notes 78% of the time. This is an important step in developing an evidence-based predictor of repeated suicide attempts because it shows that natural language processing can aid in distinguishing between classes of suicidal notes.

This paper reports on a shared task involving the assignment of ICD-9-CM codes to radiology reports. Two features distinguished this task from previous shared tasks in the biomedical domain. One is that it resulted in the first freely distributable corpus of fully anonymized clinical text. This resource is permanently available and will (we hope) facilitate future research. The other key feature of the task is that it required categorization with respect to a large and commercially significant set of labels. The number of participants was larger than in any previous biomedical challenge task. We describe the data production process and the evaluation measures, and give a preliminary analysis of the results. Many systems performed at levels approaching the inter-coder agreement, suggesting that human-like performance on this task is within the reach of currently available technologies.

This paper reports on a shared task involving the assignment of emotions to suicide notes. Two features distinguished this task from previous shared tasks in the biomedical domain. One is that it resulted in the corpus of fully anonymized clinical text and annotated suicide notes. This resource is permanently available and will (we hope) facilitate future research. The other key feature of the task is that it required categorization with respect to a large set of labels. The number of participants was larger than in any previous biomedical challenge task. We describe the data production process and the evaluation measures, and give a preliminary analysis of the results. Many systems performed at levels approaching the inter-coder agreement, suggesting that human-like performance on this task is within the reach of currently available technologies.

What adolescents say when they think about or attempt suicide influences the medical care they receive. Mental health professionals use teenagers' words, actions, and gestures to gain insight into their emotional state and to prescribe what they believe to be optimal care. This prescription is often inconsistent among caregivers, however, and leads to varying outcomes. This variation could be reduced by applying machine learning as an aid in clinical decision support. We designed a prospective clinical trial to test the hypothesis that machine learning methods can discriminate between the conversation of suicidal and nonsuicidal individuals. Using semisupervised machine learning methods, the conversations of 30 suicidal adolescents and 30 matched controls were recorded and analyzed. The results show that the machines accurately distinguished between suicidal and nonsuicidal teenagers.

We hypothesize that machine-learning algorithms (MLA) can classify completer and simulated suicide notes as well as mental health professionals (MHP). Five MHPs classified 66 simulated or completer notes; MLAs were used for the same task. Results: MHPs were accurate 71% of the time; using the sequential minimization optimization algorithm (SMO) MLAs were accurate 78% of the time. There was no significant difference between the MLA and MPH classifiers. This is an important first step in developing an evidence based suicide predictor for emergency department use.

This paper reports on the results of an initiative to create and annotate a corpus of suicide notes that can be used for machine learning. Ultimately, the corpus included 1,278 notes that were written by someone who died by suicide. Each note was reviewed by at least three annotators who mapped words or sentences to a schema of emotions. This corpus has already been used for extensive scientific research.

ObjectiveThe constant progress in computational linguistic methods provides amazing opportunities for discovering information in clinical text and enables the clinical scientist to explore novel approaches to care. However, these new approaches need evaluation. We describe an automated system to compare descriptions of epilepsy patients at three different organizations: Cincinnati Children’s Hospital, the Children’s Hospital Colorado, and the Children’s Hospital of Philadelphia. To our knowledge, there have been no similar previous studies.Materials and methodsIn this work, a support vector machine (SVM)-based natural language processing (NLP) algorithm is trained to classify epilepsy progress notes as belonging to a patient with a specific type of epilepsy from a particular hospital. The same SVM is then used to classify notes from another hospital. Our null hypothesis is that an NLP algorithm cannot be trained using epilepsy-specific notes from one hospital and subsequently used to classify notes from another hospital better than a random baseline classifier. The hypothesis is tested using epilepsy progress notes from the three hospitals.ResultsWe are able to reject the null hypothesis at the 95% level. It is also found that classification was improved by including notes from a second hospital in the SVM training sample.Discussion and conclusionWith a reasonably uniform epilepsy vocabulary and an NLP-based algorithm able to use this uniformity to classify epilepsy progress notes across different hospitals, we can pursue automated comparisons of patient conditions, treatments, and diagnoses across different healthcare settings.

Understanding written or spoken language presumably involves spreading neural activation in the brain. This process may be approximated by spreading activation in semantic networks, providing enhanced representations that involve concepts not found directly in the text. The approximation of this process is of great practical and theoretical interest. Although activations of neural circuits involved in representation of words rapidly change in time snapshots of these activations spreading through associative networks may be captured in a vector model. Concepts of similar type activate larger clusters of neurons, priming areas in the left and right hemisphere. Analysis of recent brain imaging experiments shows the importance of the right hemisphere non-verbal clusterization. Medical ontologies enable development of a large-scale practical algorithm to re-create pathways of spreading neural activations. First concepts of specific semantic type are identified in the text, and then all related concepts of the same type are added to the text, providing expanded representations. To avoid rapid growth of the extended feature space after each step only the most useful features that increase document clusterization are retained. Short hospital discharge summaries are used to illustrate how this process works on a real, very noisy data. Expanded texts show significantly improved clustering and may be classified with much higher accuracy. Although better approximations to the spreading of neural activations may be devised a practical approach presented in this paper helps to discover pathways used by the brain to process specific concepts, and may be used in large-scale applications.