Automatic Annotation of PubMed Articles with MeSH Qualifiers.

The Medical Subject Headings (MeSH) is a comprehensive indexing vocabulary used to label millions of books and articles on PubMed. The MeSH annotation of a document consists of one or more descriptors, the main headings, and of qualifiers, subheadings specific to a descriptor. Currently, there are more than 34 million documents on PubMed, which are manually tagged with MeSH terms. In this paper, we describe a machine-learning procedure that, given a document and its MeSH descriptors, predicts the respective qualifiers. In our experiment, we restricted the dataset to documents with the Heart Transplantation descriptor and we only used the PubMed abstracts. We trained binary classifiers to predict qualifiers of this descriptor using logistic regression with a tfidf vectorizer and a fine-tuned DistilBERT model. We carried out a small-scale evaluation of our models with the Mortality qualifier on a test set consisting of 30 articles (15 positives and 15 negatives). This test set was then manually re-annotated by a cardiac surgeon, expert in thoracic transplantation. On this re-annotated test set, we obtained macroaveraged F1 scores of 0.81 for the logistic regression model and of 0.85 for the DistilBERT model. Both scores are higher than the macroaveraged F1 score of 0.76 from the initial PubMed manual annotation. Our procedure would be easily extensible to all the MeSH descriptors with sufficient training data and, we believe, would enable human annotators to complete the indexing work more easily.Clinical Relevance-Selecting relevant articles is important for clinicians and researchers, but also often a challenge, especially in complex subspecialties such as heart transplantation. In this study, a machine-learning model outperformed PubMed's manual annotation, which is promising for improved quality in information retrieval.

Enhanced survival prediction using explainable artificial intelligence in heart transplantation.

The most limiting factor in heart transplantation is the lack of donor organs. With enhanced prediction of outcome, it may be possible to increase the life-years from the organs that become available. Applications of machine learning to tabular data, typical of clinical decision support, pose the practical question of interpretation, which has technical and potential ethical implications. In particular, there is an issue of principle about the predictability of complex data and whether this is inherent in the data or strongly dependent on the choice of machine learning model, leading to the so-called accuracy-interpretability trade-off. We model 1-year mortality in heart transplantation data with a self-explaining neural network, which is benchmarked against a deep learning model on the same development data, in an external validation study with two data sets: (1) UNOS transplants in 2017-2018 (n = 4750) for which the self-explaining and deep learning models are comparable in their AUROC 0.628 [0.602,0.654] cf. 0.635 [0.609,0.662] and (2) Scandinavian transplants during 1997-2018 (n = 2293), showing good calibration with AUROCs of 0.626 [0.588,0.665] and 0.634 [0.570, 0.698], respectively, with and without missing data (n = 982). This shows that for tabular data, predictive models can be transparent and capture important nonlinearities, retaining full predictive performance.

Using Operative Reports to Predict Heart Transplantation Survival.

Heart transplantation is a difficult procedure compared with other surgical operations, with a greater outcome uncertainty such as late rejection and death. We can model the success of heart transplants from predicting factors such as the age, sex, diagnosis, etc., of the donor and recipient. Although predictions can mitigate the uncertainty on the transplantation outcome, their accuracy is far from perfect. In this paper, we describe a new method to predict the outcome of a transplantation from textual operative reports instead of traditional tabular data. We carried out an experiment on 300 surgical reports to determine the survival rates at one year and five years. Using a truncated TF-IDF vectorization of the texts and logistic regression, we could reach a macro Fl of 59.1 %, respectively, 54.9% with a five-fold cross validation. While the size of the corpus is relatively small, our experiments show that the operative textual sources can discriminate the transplantation outcomes and could be a valuable additional input to existing prediction systems. Clinical Relevance- Heart transplantation involves a significant number of written reports including in the preoperative examinations and operative documentation. In this paper, we show that these written reports can predict the outcome of the transplantation at one and five years with macro 1s of 59.1 % and 54.9 %, respectively and complement existing prediction methods.

Simulating the Outcome of Heart Allocation Policies Using Deep Neural Networks.

We created a system to simulate the heart allocation process in a transplant queue, using a discrete event model and a neural network algorithm, which we named the Lund Deep Learning Transplant Algorithm (LuDeLTA). LuDeLTA is utilized to predict the survival of the patients both in the queue and after transplant. We tried four different allocation policies: wait time, clinical rules and allocating the patients using either LuDeLTA or The International Heart Transplant Survival Algorithm (IHTSA) model. Both IHTSA and LuDeLTA were used to evaluate the results. The predicted mean survival for allocating according to wait time was about 4,300 days, clinical rules 4,300 days and using neural networks 4,700 days.

Improving prediction of heart transplantation outcome using deep learning techniques.

The primary objective of this study is to compare the accuracy of two risk models, International Heart Transplantation Survival Algorithm (IHTSA), developed using deep learning technique, and Index for Mortality Prediction After Cardiac Transplantation (IMPACT), to predict survival after heart transplantation. Data from adult heart transplanted patients between January 1997 to December 2011 were collected from the UNOS registry. The study included 27,860 heart transplantations, corresponding to 27,705 patients. The study cohorts were divided into patients transplanted before 2009 (derivation cohort) and from 2009 (test cohort). The receiver operating characteristic (ROC) values, for the validation cohort, computed for one-year mortality, were 0.654 (95% CI: 0.629-0.679) for IHTSA and 0.608 (0.583-0.634) for the IMPACT model. The discrimination reached a C-index for long-term survival of 0.627 (0.608-0.646) for IHTSA, compared with 0.584 (0.564-0.605) for the IMPACT model. These figures correspond to an error reduction of 12% for ROC and 10% for C-index by using deep learning technique. The predicted one-year mortality rates for were 12% and 22% for IHTSA and IMPACT, respectively, versus an actual mortality rate of 10%. The IHTSA model showed superior discriminatory power to predict one-year mortality and survival over time after heart transplantation compared to the IMPACT model.

Predicting the outcome for patients in a heart transplantation queue using deep learning.

Heart transplantations have made it possible to extend the median survival time to 12 years for patients with end-stage heart diseases. This operation is unfortunately limited by the availability of donor organs and patients have to wait on average about 200 days in a waiting list before being operated. This waiting time varies considerably across the patients. In this paper, we studied the outcome for patients entering a transplantation waiting list using deep learning techniques. We implemented a model in the form of two-layer neural networks and we predicted the outcome as still waiting, transplanted or dead in the waiting list, at three different time points: 180 days, 365 days, and 730 days. As data source, we used the United Network for Organ Sharing (UNOS) registry, where we extracted adult patients (>17 years) from January 2000 to December 2011. We trained our model using the Keras framework, and we report F1 macro scores of respectively 0.674, 0.680, and 0.680 compared to a baseline of 0.271. We also applied a backward elimination procedure, using our neural network, to extract the 10 most significant parameters predicting the patient status for the three different time points.

Selection of an optimal feature set to predict heart transplantation outcomes.

Heart transplantation (HT) is a life saving procedure, but a limited donor supply forces the surgeons to prioritize the recipients. The understanding of factors that predict mortality could help the doctors with this task. The objective of this study is to find locally optimal feature sets to predict survival of HT patients for different time periods. To this end, we applied logistic regression together with a greedy forward and backward search. As data source, we used the United Network for Organ Sharing (UNOS) registry, where we extracted adult patients (>17 years) from January 1997 to December 2008. As methods to predict survival, we used the Index for Mortality Prediction After Cardiac Transplantation (IMPACT) and the International Heart Transplant Survival Algorithm (IHTSA). We used the LIBLINEAR library together with the Apache Spark cluster computing framework to carry out the computation and we found feature sets for 1, 5, and 10 year survival for which we obtained area under the ROC curves (AUROC) of 68%, 68%, and 76%, respectively.