Abstract
Numerous diagnostic decisions are made every day by healthcare professionals. Bayesian networks can provide a useful aid to the process, but learning their structure from data generally requires the absence of missing data, a common problem in medical data. We have studied missing data imputation using a step-wise nearest neighbors' algorithm, which we recommended given its limited impact on the assessed validity of structure learning Bayesian network classifiers for Obstructive Sleep Apnea diagnosis.

Abstract
Obstructive sleep apnea (OSA) is a significant sleep problem with various clinical presentations that have not been formally characterized. This poses critical challenges for its recognition, resulting in missed or delayed diagnosis. Recently, cluster analysis has been used in different clinical domains, particularly within numeric variables. We applied an extension of k-means to be used in categorical variables: k-modes, to identify groups of OSA patients. Demographic, physical examination, clinical history, and comorbidities characterization variables (n=46) were collected from 318 patients; missing values were all imputed with k-nearest neighbors (k-NN). Feature selection, through Chi-square test, was executed and 17 variables were inserted in cluster analysis, resulting in three clusters. Cluster 1 having an age between 65 and 90 years (54%), 78% of males, with the presence of diabetes and gastroesophageal reflux, and high OSA prevalence; Cluster 2 presented a lower percentage of OSA (46%), with middle-aged women without comorbidities, but with gastroesophageal reflux; and Cluster 3 was very similar to cluster 1, only differing in age (45-64) and comorbidities were not present. Our results suggest that there are different groups of OSA patients, creating the need to rethink the baseline characteristics of these patients before being sent to perform polysomnography (gold standard exam for diagnosis).

Abstract
The varied phenotypes of obstructive sleep apnea (OSA) poses critical challenges, resulting in missed or delayed diagnosis. In this work, we applied k-modes, aiming to identify groups of OSA patients, based on demographic, physical examination, clinical history, and comorbidities characterization variables (n=41) collected from 318 patients. Missing values were imputed with k-nearest neighbours (k-NN) and chi-square test was held. Thirteen variables were inserted in cluster analysis, resulting in three clusters. Cluster 1 were middle-aged men, while Cluster 3 were the oldest men and Cluster 2 mainly middle-aged women. Cluster 3 weighted the most, whereas Cluster 1 weighted the least. The same effect was described in increased neck circumference. The percentages of variables driving sleepiness, congestive heart failure, arrhythmias and pulmonary hypertension were very low (<20%) and OSA severity was more common in mild level. Our results suggest that it is possible to phenotype OSA patients in an objective way, as also, different (although not considered innovative) visualizations improve the recognition of this common sleep pathology.

Abstract
In pharmacovigilance, reported cases are considered suspected adverse drug reactions (ADR). Health authorities have thus adopted structured causality assessment methods, allowing the evaluation of the likelihood that a drug was the causal agent of an adverse reaction. The aim of this work was to develop and validate a new causality assessment support system used in a regional pharmacovigilance centre. A Bayesian network was developed, for which the structure was defined by experts while the parameters were learnt from 593 completely filled ADR reports evaluated by the Portuguese Northern Pharmacovigilance Centre medical expert between 2000 and 2012. Precision, recall and time to causality assessment (TTA) was evaluated, according to the WHO causality assessment guidelines, in a retrospective cohort of 466 reports (April-September 2014) and a prospective cohort of 1041 reports (January-December 2015). Additionally, a simplified assessment matrix was derived from the model, enabling its preliminary direct use by notifiers. Results show that the network was able to easily identify the higher levels of causality (recall above 80%), although struggling to assess reports with a lower level of causality. Nonetheless, the median (Q1:Q3) ITA was 4 (2:8) days using the network and 8 (5:14) days using global introspection, meaning the network allowed a faster time to assessment, which has a procedural deadline of 30 days, improving daily activities in the centre. The matrix expressed similar validity, allowing an immediate feedback to the notifiers, which may result in better future engagement of patients and health professionals in the pharmacovigilance system.

Abstract
The early classification of university students according to their potential academic performance can be a useful strategy to mitigate failure, to promote the achievement of better results and to better manage resources in higher education institutions. This paper proposes a two-stage model, supported by data mining techniques, that uses the information available at the end of the first year of students' academic career (path) to predict their overall academic performance. Unlike most literature on educational data mining, academic success is inferred from both the average grade achieved and the time taken to conclude the degree. Furthermore, this study proposes to segment students based on the dichotomy between the evidence of failure or high performance at the beginning of the degree program, and the students' performance levels predicted by the model. A data set of 2459 students, spanning the years from 2003 to 2015, from a European Engineering School of a public research University, is used to validate the proposed methodology. The empirical results demonstrate the ability of the proposed model to predict the students' performance level with an accuracy above 95%, in an early stage of the students' academic path. It is found that random forests are superior to the other classification techniques that were considered (decision trees, support vector machines, naive Bayes, bagged trees and boosted trees). Together with the prediction model, the suggested segmentation framework represents a useful tool to delineate the optimum strategies to apply, in order to promote higher performance levels and mitigate academic failure, overall increasing the quality of the academic experience provided by a higher education institution.

Abstract
Doctoral programmes are facing several challenges in modern societies. The societal role of the University, funded by the state, requires it to: a) increase the offer and admission of third cycle students; b) to reach industry/companies expectations; c) to ensure reasonable employability prospects for the PhD candidates. With the current demography, most candidates can only find a job in industry/companies. Therefore, significant pressure is being put on doctoral programmes to include transferable skills in their curriculum. This paper presents a course "Fit for Industry?" aiming at filling this need. The course design methodology is presented in detail. It includes: a) the involvement of industry since its inception; b) the joint identification of a small number of key competencies to be addressed; c) the inclusion of assessment and feedback mechanisms in its design; d) an immersive and international dimension. It was found that the course had a profound impact on the candidates' perceptions of industry and valued by industry participants. Other stakeholders, such as PhD supervisors, also had a positive perception. The paper concludes with recommendations for those willing to replicate the course locally.