Background: Dyspnoea is one of the emergency department’s (ED) most common and deadly chief complaints, but frequently misdiagnosed and mistreated. We aimed to design a diagnostic decision support which classifies dyspnoeic ED visits into acute heart failure (AHF), exacerbation of chronic obstructive pulmonary disease (eCOPD), pneumonia and “other diagnoses” by using deep learning and complete, unselected data from an entire regional health care system.

Methods: In this cross-sectional study, we included all dyspnoeic ED visits of patients ≥ 18 years of age at the two EDs in the region of Halland, Sweden, 07/01/2017–12/31/2019. Data from the complete regional health care system within five years prior to the ED visit were analysed. Gold standard diagnoses were defined as the subsequent in-hospital or ED discharge notes, and a subsample was manually reviewed by emergency medicine experts. A novel deep learning model, the clinical attention-based recurrent encoder network (CareNet), was developed. Cohort performance was compared to a simpler CatBoost model. A list of all variables and their importance for diagnosis was created. For each unique patient visit, the model selected the most important variables, analysed them and presented them to the clinician interpretably by taking event time and clinical context into account. AUROC, sensitivity and specificity were compared.

Findings: The most prevalent diagnoses among the 10,315 dyspnoeic ED visits were AHF (15.5%), eCOPD (14.0%) and pneumonia (13.3%). Median number of unique events, i.e., registered clinical data with time stamps, per ED visit was 1,095 (IQR 459–2,310). CareNet median AUROC was 87.0%, substantially higher than the CatBoost model´s (81.4%). CareNet median sensitivity for AHF, eCOPD, and pneumonia was 74.5%, 92.6%, and 54.1%, respectively, with a specificity set above 75.0, slightly inferior to that of the CatBoost baseline model. The model assembled a list of 1,596 variables by importance for diagnosis, on top were prior diagnoses of heart failure or COPD, daily smoking, atrial fibrillation/flutter, life management difficulties and maternity care. Each patient visit received their own unique attention plot, graphically displaying important clinical events for the diagnosis.

Interpretation: We designed a novel interpretable deep learning model for diagnosis in emergency department dyspnoea patients by analysing unselected data from a complete regional health care system. © 2024 Heyman et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Satellite imagery is a potent tool for estimating human wealth and poverty, especially in regions lacking reliable data. This study compares a range of poverty estimation approaches from satellite images, spanning from expert-based to fully machine learning-based methodologies. Human experts ranked clusters from the Tanzania DHS survey using high-resolution satellite images. Then expert-defined features were utilized in a machine learning algorithm to estimate poverty. An explainability method was applied to assess the importance and interaction of these features in poverty prediction. Additionally, a convolutional neural network (CNN) was employed to estimate poverty from medium-resolution satellite images of the same locations. Our analysis indicates that increased human involvement in poverty estimation diminishes accuracy compared to machine learning involvement, exemplified with the case of Tanzania. Expert defined features exhibited significant overlap and poor interaction when used together in a classifier. Conversely, the CNN-based approach outperformed human experts, demonstrating superior predictive capability with medium-resolution images. These findings highlight the importance of leveraging machine learning explainability methods to identify predictive elements that may be overlooked by human experts. This study advocates for the integration of emerging technologies with traditional methodologies to optimize data collection and analysis of poverty and welfare. © The Author(s) 2024.

In this work, we use an Evolutionary Algorithm (EA) to discover a novel Neural Network (NN) regression-based survival loss function with the aim of improving the C-index performance. Our contribution is threefold; firstly, we propose an evolutionary meta-learning algorithm SAGA$_{loss}$ for optimizing a neural-network regression-based loss function that maximizes the C-index; our algorithm consistently discovers specialized loss functions that outperform MSCE. Secondly, based on our analysis of the evolutionary search results, we highlight a non-intuitive insight that signifies the importance of the non-zero gradient for the censored cases part of the loss function, a property that is shown to be useful in improving concordance. Finally, based on this insight, we propose MSCE$_{Sp}$, a novel survival regression loss function that can be used off-the-shelf and generally performs better than the Mean Squared Error for censored cases. We performed extensive experiments on 19 benchmark datasets to validate our findings. © 2024 is held by the owner/author(s).

At the emergency department (ED), it is important to quickly and accurately determine which patients are likely to have a major adverse cardiac event (MACE). Machine learning (ML) models can be used to aid physicians in detecting MACE, and improving the performance of such models is an active area of research. In this study, we sought to determine if ML models can be improved by including a prior electrocardiogram (ECG) from each patient. To that end, we trained several models to predict MACE within 30 days, both with and without prior ECGs, using data collected from 19,499 consecutive patients with chest pain, from five EDs in southern Sweden, between the years 2017 and 2018. Our results indicate no improvement in AUC from prior ECGs. This was consistent across models, both with and without additional clinical input variables, for different patient subgroups, and for different subsets of the outcome. While contradicting current best practices for manual ECG analysis, the results are positive in the sense that ML models with fewer inputs are more easily and widely applicable in practice. © 2023 The Authors

Effectively identifying deviations in real-world medical time-series data is a critical endeavor, essential for early surveillance of disease outbreaks. This paper demonstrates the integration of time-series anomaly detection techniques to develop surveillance systems for disease outbreaks. Utilizing data from Sweden's telephone counseling service (1177), we first illustrate the trends in physical and mental symptoms recorded as contact reasons, offering valuable insights for outbreak detection. Subsequently, an advanced anomaly detection technique is applied incrementally to these time-series symptoms as univariate and multivariate approaches to assess the effectiveness of a machine learning-based method on early detection of the COVID-19 outbreak © 2024 The Authors.

The Concordance Index (C-index) is a commonly used metric in Survival Analysis for evaluating the performance of a prediction model. This paper proposes a decomposition of the C-index into a weighted harmonic mean of two quantities: one for ranking observed events versus other observed events, and the other for ranking observed events versus censored cases. This decomposition enables a more fine-grained analysis of the strengths and weaknesses of survival prediction methods. The usefulness of this decomposition is demonstrated through benchmark comparisons against state-of-the-art and classical models, together with a new variational generative neural-network-based method (SurVED), which is also proposed in this paper. Performance is assessed using four publicly available datasets with varying levels of censoring. The analysis using the C-index decomposition and synthetic censoring shows that deep learning models utilize the observed events more effectively than other models, allowing them to keep a stable C-index in different censoring levels. In contrast, classical machine learning models deteriorate when the censoring level decreases due to their inability to improve on ranking the events versus other events. © 2024 The Author(s)

Background: In the European Union alone, more than 100 million people present to the emergency department (ED) each year, and this has increased steadily year-on-year by 2–3%. Better patient management decisions have the potential to reduce ED crowding, the number of diagnostic tests, the use of inpatient beds, and healthcare costs. Methods: We have established the Skåne Emergency Medicine (SEM) cohort for developing clinical decision support systems (CDSS) based on artificial intelligence or machine learning as well as traditional statistical methods. The SEM cohort consists of 325 539 unselected unique patients with 630 275 visits from January 1st, 2017 to December 31st, 2018 at eight EDs in the region Skåne in southern Sweden. Data on sociodemographics, previous diseases and current medication are available for each ED patient visit, as well as their chief complaint, test results, disposition and the outcome in the form of subsequent diagnoses, treatments, healthcare costs and mortality within a follow-up period of at least 30 days, and up to 3 years. Discussion: The SEM cohort provides a platform for CDSS research, and we welcome collaboration. In addition, SEM’s large amount of real-world patient data with almost complete short-term follow-up will allow research in epidemiology, patient management, diagnostics, prognostics, ED crowding, resource allocation, and social medicine. © The Author(s) 2024.

The development of black-box survival models has created a need for methods that explain their outputs, just as in the case of traditional machine learning methods. Survival models usually predict functions rather than point estimates. This special nature of their output makes it more difficult to explain their operation. We propose a method to generate plausible counterfactual explanations for survival models. The method supports two options that handle the special nature of survival models' output. One option relies on the Survival Scores, which are based on the area under the survival function, which is more suitable for proportional hazard models. The other one relies on Survival Patterns in the predictions of the survival model, which represent groups that are significantly different from the survival perspective. This guarantees an intuitive well-defined change from one risk group (Survival Pattern) to another and can handle more realistic cases where the proportional hazard assumption does not hold. The method uses a Particle Swarm Optimization algorithm to optimize a loss function to achieve four objectives: the desired change in the target, proximity to the explained example, likelihood, and the actionability of the counterfactual example. Two predictive maintenance datasets and one medical dataset are used to illustrate the results in different settings. The results show that our method produces plausible counterfactuals, which increase the understanding of black-box survival models. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.

Background: Primary graft failure (PGF) remains the most common cause of short-term mortality after heart transplantation. The main objective was to develop and validate a risk model for prediction of short-term mortality due to PGF after heart transplantation using the ISHLT Heart Transplant Registry. Methods: We developed a non-linear artificial neural networks (ANN) model to evaluate the association between recipient-donor variables and post-transplant PGF. Patients in the ISHLT registry were randomly divided into derivation and an independent internal validation cohort. The primary endpoint was PGF defined as death within 30 days due to Graft failure or Cardiovascular causes or retransplant within 30 days for causes other than rejection. Results: Among 64,964 adult recipients transplanted between 1994 and 2013, mean age was 51 years and 22% were female. The incidence of PGF up to 30 days was 3.7%. The ANN model selected 33 of 77 risk variables as relevant for PGF prediction. The C-index in the test cohort was 0.70 (95% CI: 0.68-0.71). The risk variables which most influenced the PGF were underlying HF diagnosis, ischemia time and sex, while renal function had a lower influence. Conclusion: An ANN model to predict primary graft dysfunction was derived and independently validated. The good discrimination of the ANN model likely results from its flexibility to model potentially non-linear relationships and interactions. Whether this model with improved discrimination can assist in clinical decisions at the time of transplant should be tested. © 2023 The Authors

Using electronic health records data and machine learning to guide future decisions needs to address challenges, including 1) long/short-term dependencies and 2) interactions between diseases and interventions. Bidirectional transformers have effectively addressed the first challenge. Here we tackled the latter challenge by masking one source (e.g., ICD10 codes) and training the transformer to predict it using other sources (e.g., ATC codes). © 2023 European Federation for Medical Informatics (EFMI) and IOS Press.