Pedro Pereira Rodrigues

Abstract
Lack of timeliness and capacity are seen as fundamental problems that jeopardise healthcare delivery systems everywhere. Many believe the shortage of medical providers is causing this timeliness problem. This action research presents how one doctor implemented the theory of constraints (TOC) to improve the throughput (quantity of patients treated) of his ophthalmology imaging practice by 64% in a few weeks with little to no expense. The five focusing steps (5FS) guided the TOC implementation - which included the drum-buffer-rope scheduling and buffer management - and occurred in a matter of days. The implementation provided significant bottom-line results almost immediately. This article explains each step of the 5FS in general terms followed by specific applications to healthcare services, as well as the detailed use in this action research. Although TOC successfully addressed the practice problems, this implementation was not sustained after the TOC champion left the organisation. However, this drawback provided valuable knowledge. The article provides insightful knowledge to help readers implement TOC in their environments to provide immediate and significant results at little to no expense.

Abstract
PurposeTo develop, implement and assess the results of psychoeducation to improve the QoL of parents with CHD newborns.MethodsParticipants were parents of inpatient newborns with the diagnosis of non-syndromic CHD. We conducted a parallel RCT with an allocation ratio of 1:1 (intervention vs. control), considering the newborns, using mixed methods research. The intervention group received psychoeducation (Parental Psychoeducation in CHD [PPeCHD]) and the usual routines, and the control group received just the regular practices. The allocation concealment was assured. PI was involved in enrolling participants, developing and implementing the intervention, data collection and data analysis. We followed the Consolidated Standards of Reporting Trials (CONSORT) guidelines.ResultsParents of eight newborns were allocated to the intervention group (n = 15 parents) and eight to the control group (n = 13 parents). It was performed as an intention-to-treat (ITT) analysis. In M2 (4 weeks), the intervention group presented better QoL levels in the physical, psychological, and environmental domains of World Health Organization Quality of Life instrument (WHOQOL-Bref). In M3 (16 weeks), scores in physical and psychological domains maintained a statistically significant difference between the groups.ConclusionsThe PPeCHD, the psychoeducational intervention we developed, positively impacted parental QoL. These results support the initial hypothesis. This study is a fundamental milestone in this research field, adding new essential information to the literature.

Abstract
Several studies demonstrate that the structure of the brain increases in hierarchical complexity throughout development. We tested if the structure of artificial neural networks also increases in hierarchical complexity while learning a developing task, called the balance beam problem. Previous simulations of this developmental task do not reflect a necessary premise underlying development: a more complex structure can be built out of less complex ones, while ensuring that the more complex structure does not replace the less complex one. In order to address this necessity, we segregated the input set by subsets of increasing Orders of Hierarchical Complexity. This is a complexity measure that has been extensively shown to underlie the complexity behavior and hypothesized to underlie the complexity of the neural structure of the brain. After segregating the input set, minimal neural network models were trained separately for each input subset, and adjacent complexity models were analyzed sequentially to observe whether there was a structural progression. Results show that three different network structural progressions were found, performing with similar accuracy, pointing towards self-organization. Also, more complex structures could be built out of less complex ones without substituting them, successfully addressing catastrophic forgetting and leveraging performance of previous models in the literature. Furthermore, the model structures trained on the two highest complexity subsets performed better than simulations of the balance beam present in the literature. As a major contribution, this work was successful in addressing hierarchical complexity structural growth in neural networks, and is the first that segregates inputs by Order of Hierarchical Complexity. Since this measure can be applied to all domains of data, the present method can be applied to future simulations, systematizing the simulation of developmental and evolutionary structural growth in neural networks.

Abstract
Missing data is an issue that can negatively impact any task performed with the available data and it is often found in real -world domains such as healthcare. One of the most common strategies to address this issue is to perform imputation, where the missing values are replaced by estimates. Several approaches based on statistics and machine learning techniques have been proposed for this purpose, including deep learning architectures such as generative adversarial networks and autoencoders. In this work, we propose a novel siamese neural network suitable for missing data imputation, which we call Siamese Autoencoder-based Approach for Imputation (SAEI). Besides having a deep autoencoder architecture, SAEI also has a custom loss function and triplet mining strategy that are tailored for the missing data issue. The proposed SAEI approach is compared to seven state-of-the-art imputation methods in an experimental setup that comprises 14 heterogeneous datasets of the healthcare domain injected with Missing Not At Random values at a rate between 10% and 60%. The results show that SAEI significantly outperforms all the remaining imputation methods for all experimented settings, achieving an average improvement of 35%. This work is an extension of the article Siamese Autoencoder-Based Approach for Missing Data Imputation [1] presented at the International Conference on Computational Science 2023. It includes new experiments focused on runtime, generalization capabilities, and the impact of the imputation in classification tasks, where the results show that SAEI is the imputation method that induces the best classification results, improving the F1 scores for 50% of the used datasets.

Abstract
Experimental assessment of different missing data imputation methods often compute error rates between the original values and the estimated ones. This experimental setup relies on complete datasets that are injected with missing values. The injection process is straightforward for the Missing Completely At Random and Missing At Random mechanisms; however, the Missing Not At Random mechanism poses a major challenge, since the available artificial generation strategies are limited. Furthermore, the studies focused on this latter mechanism tend to disregard a comprehensive baseline of state-of-the-art imputation methods. In this work, both challenges are addressed: four new Missing Not At Random generation strategies are introduced and a benchmark study is conducted to compare six imputation methods in an experimental setup that covers 10 datasets and five missingness levels (10% to 80%). The overall findings are that, for most missing rates and datasets, the best imputation method to deal with Missing Not At Random values is the Multiple Imputation by Chained Equations, whereas for higher missingness rates autoencoders show promising results.