Reshaping free-text radiology notes into structured reports with generative question answering transformers.

BACKGROUND: Radiology reports are typically written in a free-text format, making clinical information difficult to extract and use. Recently, the adoption of structured reporting (SR) has been recommended by various medical societies thanks to the advantages it offers, e.g. standardization, completeness, and information retrieval. We propose a pipeline to extract information from Italian free-text radiology reports that fits with the items of the reference SR registry proposed by a national society of interventional and medical radiology, focusing on CT staging of patients with lymphoma. METHODS: Our work aims to leverage the potential of Natural Language Processing and Transformer-based models to deal with automatic SR registry filling. With the availability of 174 Italian radiology reports, we investigate a rule-free generative Question Answering approach based on the Italian-specific version of T5: IT5. To address information content discrepancies, we focus on the six most frequently filled items in the annotations made on the reports: three categorical (multichoice), one free-text (free-text), and two continuous numerical (factual). In the preprocessing phase, we encode also information that is not supposed to be entered. Two strategies (batch-truncation and ex-post combination) are implemented to comply with the IT5 context length limitations. Performance is evaluated in terms of strict accuracy, f1, and format accuracy, and compared with the widely used GPT-3.5 Large Language Model. Unlike multichoice and factual, free-text answers do not have 1-to-1 correspondence with their reference annotations. For this reason, we collect human-expert feedback on the similarity between medical annotations and generated free-text answers, using a 5-point Likert scale questionnaire (evaluating the criteria of correctness and completeness). RESULTS: The combination of fine-tuning and batch splitting allows IT5 ex-post combination to achieve notable results in terms of information extraction of different types of structured data, performing on par with GPT-3.5. Human-based assessment scores of free-text answers show a high correlation with the AI performance metrics f1 (Spearman's correlation coefficients>0.5, p-values<0.001) for both IT5 ex-post combination and GPT-3.5. The latter is better at generating plausible human-like statements, even if it systematically provides answers even when they are not supposed to be given. CONCLUSIONS: In our experimental setting, a fine-tuned Transformer-based model with a modest number of parameters (i.e., IT5, 220 M) performs well as a clinical information extraction system for automatic SR registry filling task. It can extract information from more than one place in the report, elaborating it in a manner that complies with the response specifications provided by the SR registry (for multichoice and factual items), or that closely approximates the work of a human-expert (free-text items); with the ability to discern when an answer is supposed to be given or not to a user query.

Advancing Italian biomedical information extraction with transformers-based models: Methodological insights and multicenter practical application.

The introduction of computerized medical records in hospitals has reduced burdensome activities like manual writing and information fetching. However, the data contained in medical records are still far underutilized, primarily because extracting data from unstructured textual medical records takes time and effort. Information Extraction, a subfield of Natural Language Processing, can help clinical practitioners overcome this limitation by using automated text-mining pipelines. In this work, we created the first Italian neuropsychiatric Named Entity Recognition dataset, PsyNIT, and used it to develop a Transformers-based model. Moreover, we collected and leveraged three external independent datasets to implement an effective multicenter model, with overall F1-score 84.77 %, Precision 83.16 %, Recall 86.44 %. The lessons learned are: (i) the crucial role of a consistent annotation process and (ii) a fine-tuning strategy that combines classical methods with a "low-resource" approach. This allowed us to establish methodological guidelines that pave the way for Natural Language Processing studies in less-resourced languages.

An eHealth App (CAPABLE) Providing Symptom Monitoring, Well-Being Interventions, and Educational Material for Patients With Melanoma Treated With Immune Checkpoint Inhibitors: Protocol for an Exploratory Intervention Trial.

BACKGROUND: Since treatment with immune checkpoint inhibitors (ICIs) is becoming standard therapy for patients with high-risk and advanced melanoma, an increasing number of patients experience treatment-related adverse events such as fatigue. Until now, studies have demonstrated the benefits of using eHealth tools to provide either symptom monitoring or interventions to reduce treatment-related symptoms such as fatigue. However, an eHealth tool that facilitates the combination of both symptom monitoring and symptom management in patients with melanoma treated with ICIs is still needed. OBJECTIVE: In this pilot study, we will explore the use of the CAPABLE (Cancer Patients Better Life Experience) app in providing symptom monitoring, education, and well-being interventions on health-related quality of life (HRQoL) outcomes such as fatigue and physical functioning, as well as patients' acceptance and usability of using CAPABLE. METHODS: This prospective, exploratory pilot study will examine changes in fatigue over time in 36 patients with stage III or IV melanoma during treatment with ICI using CAPABLE (a smartphone app and multisensory smartwatch). This cohort will be compared to a prospectively collected cohort of patients with melanoma treated with standard ICI therapy. CAPABLE will be used for a minimum of 3 and a maximum of 6 months. The primary endpoint in this study is the change in fatigue between baseline and 3 and 6 months after the start of treatment. Secondary end points include HRQoL outcomes, usability, and feasibility parameters. RESULTS: Study inclusion started in April 2023 and is currently ongoing. CONCLUSIONS: This pilot study will explore the effect, usability, and feasibility of CAPABLE in patients with melanoma during treatment with ICI. Adding the CAPABLE system to active treatment is hypothesized to decrease fatigue in patients with high-risk and advanced melanoma during treatment with ICIs compared to a control group receiving standard care. The Medical Ethics Committee NedMec (Amsterdam, The Netherlands) granted ethical approval for this study (reference number 22-981/NL81970.000.22). TRIAL REGISTRATION: ClinicalTrials.gov NCT05827289; https://clinicaltrials.gov/study/NCT05827289. INTERNATIONAL REGISTERED REPORT IDENTIFIER (IRRID): DERR1-10.2196/49252.

A comparative analysis of the effects of containment policies on the epidemiological manifestation of the COVID-19 pandemic across nine European countries.

The COVID-19 pandemic has been a catastrophic event that has seriously endangered the world's population. Governments have largely been unprepared to deal with such an unprecedented calamity, partially due to the lack of sufficient or adequately fine-grained data necessary for forecasting the pandemic's evolution. To fill this gap, researchers worldwide have been collecting data about different aspects of COVID-19's evolution and government responses to them so as to provide the foundation for informative models and tools that can be used to mitigate the current pandemic and possibly prevent future ones. Indeed, since the early stages of the pandemic, a number of research initiatives were launched with this goal, including the PERISCOPE (Pan-European Response to the ImpactS of COVID-19 and future Pandemics and Epidemics) Project, funded by the European Commission. PERISCOPE aims to investigate the broad socio-economic and behavioral impacts of the COVID-19 pandemic, with the goal of making Europe more resilient and prepared for future large-scale risks. The purpose of this study, carried out as part of the PERISCOPE project, is to provide a first European-level analysis of the effect of government policies on the spread of the virus. To do so, we assessed the relationship between a novel index, the Policy Intensity Index, and four epidemiological variables collected by the European Centre for Disease Control and Prevention, and then applied a comprehensive Pan-European population model based on Multilevel Vector Autoregression. This model aims at identifying effects that are common to some European countries while treating country-specific policies as covariates, explaining the different evolution of the pandemic in nine selected countries due to data availability: Spain, France, Netherlands, Latvia, Slovenia, Greece, Ireland, Cyprus, Estonia. Results show that specific policies' effectiveness tend to vary consistently within the different countries, although in general policies related to Health Monitoring and Health Resources are the most effective for all countries.

Localizing in-domain adaptation of transformer-based biomedical language models.

In the era of digital healthcare, the huge volumes of textual information generated every day in hospitals constitute an essential but underused asset that could be exploited with task-specific, fine-tuned biomedical language representation models, improving patient care and management. For such specialized domains, previous research has shown that fine-tuning models stemming from broad-coverage checkpoints can largely benefit additional training rounds over large-scale in-domain resources. However, these resources are often unreachable for less-resourced languages like Italian, preventing local medical institutions to employ in-domain adaptation. In order to reduce this gap, our work investigates two accessible approaches to derive biomedical language models in languages other than English, taking Italian as a concrete use-case: one based on neural machine translation of English resources, favoring quantity over quality; the other based on a high-grade, narrow-scoped corpus natively written in Italian, thus preferring quality over quantity. Our study shows that data quantity is a harder constraint than data quality for biomedical adaptation, but the concatenation of high-quality data can improve model performance even when dealing with relatively size-limited corpora. The models published from our investigations have the potential to unlock important research opportunities for Italian hospitals and academia. Finally, the set of lessons learned from the study constitutes valuable insights towards a solution to build biomedical language models that are generalizable to other less-resourced languages and different domain settings.

The Case Manager: An Agent Controlling the Activation of Knowledge Sources in a FHIR-Based Distributed Reasoning Environment.

BACKGROUND: Within the CAPABLE project the authors developed a multi-agent system that relies on a distributed architecture. The system provides cancer patients with coaching advice and supports their clinicians with suitable decisions based on clinical guidelines. OBJECTIVES: As in many multi-agent systems we needed to coordinate the activities of all agents involved. Moreover, since the agents share a common blackboard where all patients' data are stored, we also needed to implement a mechanism for the prompt notification of each agent upon addition of new information potentially triggering its activation. METHODS: The communication needs have been investigated and modeled using the HL7-FHIR (Health Level 7-Fast Healthcare Interoperability Resources) standard to ensure proper semantic interoperability among agents. Then a syntax rooted in the FHIR search framework has been defined for representing the conditions to be monitored on the system blackboard for activating each agent. RESULTS: The Case Manager (CM) has been implemented as a dedicated component playing the role of an orchestrator directing the behavior of all agents involved. Agents dynamically inform the CM about the conditions to be monitored on the blackboard, using the syntax we developed. The CM then notifies each agent whenever any condition of interest occurs. The functionalities of the CM and other actors have been validated using simulated scenarios mimicking the ones that will be faced during pilot studies and in production. CONCLUSION: The CM proved to be a key facilitator for properly achieving the required behavior of our multi-agent system. The proposed architecture may also be leveraged in many clinical contexts for integrating separate legacy services, turning them into a consistent telemedicine framework and enabling application reusability.

A synthetic dataset of liver disorder patients.

The data in this article include 10,000 synthetic patients with liver disorders, characterized by 70 different variables, including clinical features, and patient outcomes, such as hospital admission or surgery. Patient data are generated, simulating as close as possible real patient data, using a publicly available Bayesian network describing a casual model for liver disorders. By varying the network parameters, we also generated an additional set of 500 patients with characteristics that deviated from the initial patient population. We provide an overview of the synthetic data generation process and the associated scripts for generating the cohorts. This dataset can be useful for the machine learning models training and validation, especially under the effect of dataset shift between training and testing sets.

Why did AI get this one wrong? - Tree-based explanations of machine learning model predictions.

Increasingly complex learning methods such as boosting, bagging and deep learning have made ML models more accurate, but harder to interpret and explain, culminating in black-box machine learning models. Model developers and users alike are often presented with a trade-off between performance and intelligibility, especially in high-stakes applications like medicine. In the present article we propose a novel methodological approach for generating explanations for the predictions of a generic machine learning model, given a specific instance for which the prediction has been made. The method, named AraucanaXAI, is based on surrogate, locally-fitted classification and regression trees that are used to provide post-hoc explanations of the prediction of a generic machine learning model. Advantages of the proposed XAI approach include superior fidelity to the original model, ability to deal with non-linear decision boundaries, and native support to both classification and regression problems. We provide a packaged, open-source implementation of the AraucanaXAI method and evaluate its behaviour in a number of different settings that are commonly encountered in medical applications of AI. These include potential disagreement between the model prediction and physician's expert opinion and low reliability of the prediction due to data scarcity.

A New Interactive Tool to Visualize and Analyze COVID-19 Data: The PERISCOPE Atlas.

Since the start of the 21st century, the world has not confronted a more serious threat to global public health than the COVID-19 pandemic. While governments initially took radical actions in response to the pandemic to avoid catastrophic collapse of their health care systems, government policies have also had numerous knock-on socioeconomic, political, behavioral and economic effects. Researchers, thus, have a unique opportunity to forward our collective understanding of the modern world and to respond to the emergency situation in a way that optimizes resources and maximizes results. The PERISCOPE project, funded by the European Commission, brings together a large number of research institutions to collect data and carry out research to understand all the impacts of the pandemic, and create predictive models that can be used to optimize intervention strategies and better face possible future health emergencies. One of the main tangible outcomes of this project is the PERISCOPE Atlas: an interactive tool that allows to visualize and analyze COVID-19-related health, economic and sociopolitical data, featuring a WebGIS and several dashboards. This paper describes the first release of the Atlas, listing the data sources used, the main functionalities and the future development.

Differential impact of the COVID-19 pandemic on primary care utilization related to common mental disorders in four European countries: A retrospective observational study.

Background: The COVID-19 pandemic is commonly believed to have increased common mental disorders (CMD, i.e., depression and anxiety), either directly due to COVID-19 contractions (death of near ones or residual conditions), or indirectly by increasing stress, economic uncertainty, and disruptions in daily life resulting from containment measure. Whereas studies reporting on initial changes in self-reported data frequently have reported increases in CMD, pandemic related changes in CMD related to primary care utilization are less well known. Analyzing time series of routinely and continuously sampled primary healthcare data from Sweden, Norway, Netherlands, and Latvia, we aimed to characterize the impact of the pandemic on CMD recorded prevalence in primary care. Furthermore, by relating these changes to country specific time-trajectories of two classes of containment measures, we evaluated the differential impact of containment strategies on CMD rates. Specifically, we wanted to test whether school restrictions would preferentially affect age groups corresponding to those of school children or their parents. Methods: For the four investigated countries, we collected time-series of monthly counts of unique CMD patients in primary healthcare from the year 2015 (or 2017) until 2021. Using pre-pandemic timepoints to train seasonal Auto Regressive Integrated Moving Average (ARIMA) models, we predicted healthcare utilization during the pandemic. Discrepancies between observed and expected time series were quantified to infer pandemic related changes. To evaluate the effects of COVID-19 measures on CMD related primary care utilization, the predicted time series were related to country specific time series of levels of social distancing and school restrictions. Results: In all countries except Latvia there was an initial (April 2020) decrease in CMD care prevalence, where largest drops were found in Sweden (Prevalence Ratio, PR = 0.85; 95% CI 0.81-0.90), followed by Netherlands (0.86; 95% CI 0.76-1.02) and Norway (0.90; 95% CI 0.83-0.98). Latvia on the other hand experienced increased rates (1.25; 95% CI 1.08-1.49). Whereas PRs in Norway and Netherlands normalized during the latter half of 2020, PRs stayed low in Sweden and elevated in Latvia. The overall changes in PR during the pandemic year 2020 was significantly changed only for Sweden (0.91; 95% CI 0.90-0.93) and Latvia (1.20; 95% CI 1.14-1.26). Overall, the relationship between containment measures and CMD care prevalence were weak and non-significant. In particular, we could not observe any relationship of school restriction to CMD care prevalence for the age groups best corresponding to school children or their parents. Conclusion: Common mental disorders prevalence in primary care decreased during the initial phase of the COVID-19 pandemic in all countries except from Latvia, but normalized in Norway and Netherlands by the latter half of 2020. The onset of the pandemic and the containment strategies were highly correlated within each country, limiting strong conclusions on whether restriction policy had any effects on mental health. Specifically, we found no evidence of associations between school restrictions and CMD care prevalence. Overall, current results lend no support to the common belief that the pandemic severely impacted the mental health of the general population as indicated by healthcare utilization, apart from in Latvia. However, since healthcare utilization is affected by multiple factors in addition to actual need, future studies should combine complementary types of data to better understand the mental health impacts of the pandemic.

Quality of life and health-related utility after trans-oral surgery for head and neck cancers.

PURPOSE: The purpose of this study was to assess utility coefficients of health states following two minimally invasive surgical approaches for head and neck cancer, namely trans-oral robotic surgery and trans-oral laser microsurgery. Those utility coefficients will be later exploited in an economic evaluation study comparing the two approaches. METHODS: The above cited economic evaluation will be done from the Swiss healthcare system perspective and, as such, Swiss healthcare professionals were interviewed to elicit utility coefficients. Health states, ranging from remission to palliative care, were described using clinical vignettes. A computerized tool (UceWeb) implementing standard gamble and rating scale methods was used. RESULTS: Utility coefficients for 18 different health states were elicited with the two methods from 47 individuals, for a total of 1692 values. Elicited values varied from 0.980 to 0.213. Comparison with values elicited in previous studies show the need for population-specific elicitation, mainly for the worst health states. CONCLUSION: Herein we report health utility coefficients for the Swiss population for health states following minimally invasive trans-oral surgery. This study provides utility values that can be used not only for a specific cost-utility analysis, but also for future studies involving the same health states.

Cost-utility of two minimally-invasive surgical techniques for operable oropharyngeal cancer: transoral robotic surgery versus transoral laser microsurgery.

BACKGROUND: In the past few decades, a re-evaluation of treatment paradigms of head and neck cancers with a desire to spare patients the treatment-related toxicities of open surgery, has led to the development of new minimally invasive surgical techniques to improve outcomes. Besides Transoral Laser Microsurgery (TLM), a new robotic surgical technique namely Transoral Robotic Surgery (TORS) emerged for the first time as one of the two most prominent and widely used minimally invasive surgical approaches particularly for the treatment of oropharyngeal cancer, a sub-entity of head and neck cancers. Recent population-level data suggest equivalent tumor control, but different total costs, and need for adjuvant chemoradiation. A comparative analysis of these two techniques is therefore warranted from the cost-utility (C/U) point of view. METHODS: A cost-utility analysis for comparing TORS and TLM was performed using a decision-analytical model. The analyses adopted the perspective of a Swiss hospital. Two tertiary referral centers in Lausanne and Zurich provided data for model quantificantion. RESULTS: In the base case analysis TLM dominates TORS. This advantage remains robust, even if the costs for TORS reduce by up to 25%. TORS begins to dominate TLM, if less than 59,7% patients require adjuvant treatment, whereby in an interval between 55 and 62% cost effectiveness of TORS is sensitive to the prescription of adjuvant chemoradiation therapy (CRT). Exceeding 29% of TLM patients requiring a revision of surgical margins renders TORS more cost-effective. CONCLUSION: Non-robotic endoscopic surgery (TLM) is more cost-effective than robotic endoscopic surgery (TORS) for the treatment of oropharyngeal cancers. However, this advantage is sensitive to various parameters, i.e.to the number of re-operations and adjuvant treatment.

A Health eLearning Ontology and Procedural Reasoning Approach for Developing Personalized Courses to Teach Patients about Their Medical Condition and Treatment.

We propose a methodological framework to support the development of personalized courses that improve patients' understanding of their condition and prescribed treatment. Inspired by Intelligent Tutoring Systems (ITSs), the framework uses an eLearning ontology to express domain and learner models and to create a course. We combine the ontology with a procedural reasoning approach and precompiled plans to operationalize a design across disease conditions. The resulting courses generated by the framework are personalized across four patient axes-condition and treatment, comprehension level, learning style based on the VARK (Visual, Aural, Read/write, Kinesthetic) presentation model, and the level of understanding of specific course content according to Bloom's taxonomy. Customizing educational materials along these learning axes stimulates and sustains patients' attention when learning about their conditions or treatment options. Our proposed framework creates a personalized course that prepares patients for their meetings with specialists and educates them about their prescribed treatment. We posit that the improvement in patients' understanding of prescribed care will result in better outcomes and we validate that the constructs of our framework are appropriate for representing content and deriving personalized courses for two use cases: anticoagulation treatment of an atrial fibrillation patient and lower back pain management to treat a lumbar degenerative disc condition. We conduct a mostly qualitative study supported by a quantitative questionnaire to investigate the acceptability of the framework among the target patient population and medical practitioners.

The Case Manager: Driving Medical Reasoning in a Distributed Environment for Home Patient Monitoring.

The CAPABLE project has been funded by the EU Horizon 2020 Programme over the years 2020-24 to support home care. A system is being designed and implemented supporting remote monitoring and virtual coaching for cancer patients. The system is based on a distributed modular architecture involving many components encapsulating various knowledge. The Case Manager has been designed as a separate component with the aim of coordinating the problem solving strategies. A first version of the Case Manager has been released and used by the components in a prototypical scenario shown at the first project review.

Eliciting and Exploiting Utility Coefficients in an Integrated Environment for Shared Decision-Making.

BACKGROUND: In shared decision-making, a key step is quantifying the patient's preferences in relation to all the possible outcomes of the compared clinical options. According to utility theory, this can be done by eliciting utility coefficients (UCs) from the patient. The obtained UCs are then used in decision models (e.g., decision trees). The elicitation process involves the choice of one or more elicitation methods, which is not easy for decision-makers who are unfamiliar with the theoretical framework. Moreover, to our knowledge there are no tools that integrate functionalities for UC elicitation with functionalities to run decision models that include the elicited values. OBJECTIVES: The first aim of this work is to provide decision support to the clinicians for the selection of the elicitation method. The second aim is to bridge the gap between UC elicitation and the exploitation of those UCs in shared decision-making. METHODS: Based on evidence from the utility theory literature, we developed a set of production rules that recommend the optimal elicitation method(s) according to the patient's profile and health state. We then complemented this decision support tool with a functionality for quantifying and running decision trees defined through the commercial software TreeAge. RESULTS: The result is an integrated framework for shared decision-making. Given the primary aim of this work, we focus for result evaluation on the elicitation tool. It was tested on 51 volunteers, who expressed UCs for four purposely selected health states. The insights on the collected UCs validated the rules included in the decision support system. The usability of the tool was assessed through the System Usability Scale, obtaining positive results. CONCLUSION: We developed an integrated environment to facilitate shared decision-making in the clinical practice. The next step is the validation of the entire framework and its use besides shared decision-making. As a matter of fact, it may also be exploited to target cost-utility analysis to a specific patient population.

Spatial Enablement to Support Environmental, Demographic, Socioeconomics, and Health Data Integration and Analysis for Big Cities: A Case Study With Asthma Hospitalizations in New York City.

The percentage of the world's population living in urban areas is projected to increase in the next decades. Big cities are heterogeneous environments in which socioeconomic and environmental differences among the neighborhoods are often very pronounced. Each individual, during his/her life, is constantly subject to a mix of exposures that have an effect on their phenotype but are frequently difficult to identify, especially in an urban environment. Studying how the combination of environmental and socioeconomic factors which the population is exposed to influences pathological outcomes can help transforming public health from a reactive to a predictive system. Thanks to the application of state-of-the-art spatially enabled methods, patients can be stratified according to their characteristics and the geographical context they live in, optimizing healthcare processes and the reducing its costs. Some public health studies focusing specifically on urban areas have been conducted, but they usually consider a coarse spatial subdivision, as a consequence of scarce availability of well-integrated data regarding health and environmental exposure at a sufficient level of granularity to enable meaningful statistical analyses. In this paper, we present an application of highly fine-grained spatial resolution methods to New York City data. We investigated the link between asthma hospitalizations and a combination of air pollution and other environmental and socioeconomic factors. We first performed an explorative analysis using spatial clustering methods that shows that asthma is related to numerous factors whose level of influence varies considerably among neighborhoods. We then performed a Geographically Weighted Regression with different covariates and determined which environmental and socioeconomic factors can predict hospitalizations and how they vary throughout the city. These methods showed to be promising both for visualization and analysis of demographic and epidemiological urban dynamics, that can be used to organize targeted intervention and treatment policies to address the single citizens considering the factors he/she is exposed to. We found a link between asthma and several factors such as PM2.5, age, health insurance coverage, race, poverty, obesity, industrial areas, and recycling. This study has been conducted within the PULSE project, funded by the European Commission, briefly presented in this paper.

Incorporating Laboratory Values Into a Machine Learning Model Improves In-Hospital Mortality Predictions After Rapid Response Team Call.

OBJECTIVES: Machine learning models have been used to predict mortality among patients requiring rapid response team activation. The goal of our study was to assess the impact of adding laboratory values into the model. DESIGN: A gradient boosted decision tree model was derived and internally validated to predict a primary outcome of in-hospital mortality. The base model was then augmented with laboratory values. SETTING: Two tertiary care hospitals within The Ottawa Hospital network. PATIENTS: Inpatients over the age of 18 years who experienced a rapid response team activation between January 1, 2015, and May 31, 2016. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: A total of 2,061 rapid response team activations occurred during the study period. The in-hospital mortality rate was 29.4%. Patients who died were older (median age, 72 vs 68 yr; p < 0.001), had a longer length of stay (length of stay) prior to rapid response team activation (4 vs 2 d; p < 0.001), and more often had respiratory distress (31% vs 22%; p < 0.001). Our base model without laboratory values performed with an area under the receiver operating curve of 0.71 (95% CI, 0.71-0.72). When the base model was augmented with laboratory values, the area under the receiver operating curve improved to 0.77 (95% CI, 0.77-0.78). Important mortality predictors in the base model were age, estimated ratio of Pao2 to Fio2 (calculated using oxygen saturation and estimated Fio2), length of stay prior to rapid response team activation, and systolic blood pressure. CONCLUSIONS: Machine learning models can identify rapid response team patients at a high risk of mortality and potentially supplement clinical decision making. Incorporating laboratory values into model development significantly improved predictive performance in this study.

How Do Spinal Surgeons Perceive The Impact of Factors Used in Post-Surgical Complication Risk Scores?

When deciding about surgical treatment options, an important aspect of the decision-making process is the potential risk of complications. A risk assessment performed by a spinal surgeon is based on their knowledge of the best available evidence and on their own clinical experience. The objective of this work is to demonstrate the differences in the way spine surgeons perceive the importance of attributes used to calculate risk of post-operative and quantify the differences by building individual formal models of risk perceptions. We employ a preference-learning method - ROR-UTADIS - to build surgeon-specific additive value functions for risk of complications. Comparing these functions enables the identification and discussion of differences among personal perceptions of risk factors. Our results show there exist differences in surgeons' perceived factors including primary diagnosis, type of surgery, patient's age, body mass index, or presence of comorbidities.

Ideating Mobile Health Behavioral Support for Compliance to Therapy for Patients with Chronic Disease: A Case Study of Atrial Fibrillation Management.

Poor patient compliance to therapy results in a worsening condition that often increases healthcare costs. In the MobiGuide project, we developed an evidence-based clinical decision-support system that delivered personalized reminders and recommendations to patients, helping to achieve higher therapy compliance. Yet compliance could still be improved and therefore building on the MobiGuide project experience, we designed a new component called the Motivational Patient Assistant (MPA) that is integrated within the MobiGuide architecture to further improve compliance. This component draws from psychological theories to provide behavioral support to improve patient engagement and thereby increasing patients' compliance. Behavior modification interventions are delivered via mobile technology at patients' home environments. Our approach was inspired by the IDEAS (Integrate, Design, Assess, and Share) framework for developing effective digital interventions to change health behavior; it goes beyond this approach by extending the Ideation phase' concepts into concrete backend architectural components and graphical user-interface designs that implement behavioral interventions. We describe in detail our ideation approach and how it was applied to design the user interface of MPA for anticoagulation therapy for the atrial fibrillation patients. We report results of a preliminary evaluation involving patients and care providers that shows the potential usefulness of the MPA for improving compliance to anticoagulation therapy.