NAGIOS: RODERIC FUNCIONANDO

Anemia management in end stage renal disease patients undergoing dialysis: a comprehensive approach through machine learning techniques and mathematical modeling

Repositori DSpace/Manakin

IMPORTANT: Aquest repositori està en una versió antiga des del 3/12/2023. La nova instal.lació está en https://roderic.uv.es/

Anemia management in end stage renal disease patients undergoing dialysis: a comprehensive approach through machine learning techniques and mathematical modeling

dc.contributor.advisor	Soria Olivas, Emilio
dc.contributor.advisor	Martín Guerrero, José David
dc.contributor.advisor	Canaud, Bernard
dc.contributor.author	Barbieri, Carlo
dc.contributor.other	Departament d'Enginyeria Electrònica	es_ES
dc.date.accessioned	2016-07-22T12:26:31Z
dc.date.available	2016-07-23T04:45:06Z
dc.date.issued	2016	es_ES
dc.date.submitted	18-07-2016	es_ES
dc.identifier.uri	http://hdl.handle.net/10550/54701
dc.description.abstract	Kidney impairment has global consequences in the organism homeostasis and a disorder like Chronic Kidney Disease (CKD) might eventually exacerbates into End Stage Renal Disease (ESRD) where a complete renal replacement therapy like dialysis is necessary. Dialysis partially reintegrates the blood ltration process; however, even when it is associated to a pharmacological therapy, this is not su fficient to completely replace the renal endocrine role and causes the development of common complications, like CKD secondary anemia (CKD-anemia) The availability of exogenous Erythropoiesis Stimulating Agents (ESA, synthetic molecules with similar structure and same mechanism of action as human erythropoietin) improved the treatment of CKD-anemia although the clinical outcomes are still not completely successful. In particular, for ERSD dialysis patients main di culties in the selection of an optimal therapy dosing derive from the high intra- and inter-individual response variability and the temporal discrepancy between the short ESA permanence in the blood (hours) and the long Red Blood Cells lifespan (months). The aim of this thesis has been to describe the development of the Anemia Control Model (ACM), a tool designed to support physicians in managing anemia for ESRD patines undergoing dialysis. Five main pillars constitute the foundation of this work: - Understanding the medical problem; - Availability of the data needed to derive the models; - Mathematical and Machine Learning modeling; - Development of a product usable at the point of care; - Medical device certi cation and clinical evaluation of the developed product. The understanding of the medical problem is fundamental for two reasons: firstly because the medical problem must be the driver of the product scope and consequently of its design; secondly because a good understanding of the medical problem is of fundamental importance to develop optimized models. In the case of anemia management the drug dosing is an important task where predictive models could support physicians to improve the treatment quality. In particular, considering that hemoglobin is the typical parameter used to measure anemia, our model were tailored to predict hemoglobin response to the two main drugs normally used to correct anemia, that is ESA and Iron. In a mathematical model based on di erential equations, like the one presented in this thesis, the knowledge of the main physiological processes related to anemia is the base to properly design the equations. A machine learning approach in principle can be built with no hypotesis, because it relays in learning from data, nevertheless knowledge of the domain helps to make better use of the available data. The medical problem has been discussed in Chapter 1. The availability of a huge database of very well structured data was basic for the development of models. Quality of the data is another important aspect. Chapter 2 gives the reader an overview of the available data.. The core of the ACM is the capability to predict for each patient the future hemoglobin concentrations as a function of past patient's clinical history and future drug prescription. By means of well performing and personalized predictive model it is possible to simulate how, for each specific c patient, di erent doses would a ffect hemoglobin trends. Mathematical and machine learning models present both advantages and limitations. Chapter 3 describes the mathematical model and analyzes its performances, while Chapter 4 is dedicated to the machine learning models. In our case the machine learning approach resulted more suitable for our scope, because its was well performing on the entire population, more stable and, once trained, very quick in elaborating the prediction. Once the predictive model was obtained, the next step was to wrap it into a service that could be consumed by a third party system (for example an app or a clinical system) where physicians could benefi t from the model prediction capability. To achieve that, firstly an algorithm for the dose selection was developed; secondly, a data structure for the communication with the third party system was defi ned; fi nally, the whole package was wrapped in a web service. These arguments have been discussed in the rst part of Chapter 5. Mistakes in ESA or Iron dosing might have serious consequences on patients' health, for this reason ACM intended use was limited to provide dose suggestions only; physicians must evaluate them and decide whether to accept or reject them. Nevertheless, such a tool could be considered as Medical Device under European Medical Device Directive (MDD); for this reason, to be on the safe side, it was decided to certify the ACM as medical device. A novel approach was developed to perform the risk assessment, the main idea being that ACM might generate risks when a dose suggestion is produced based on a wrong prediction. To assess this risk the model error distribution over the test set was utilized as estimation of the error distribution of the live system. Finally, a clinical evaluation of the ACM in three pilot clinics has been performed before deciding to roll-out the tool in more clinics. These arguments have been discussed in the second part of Chapter 5.	en_US
dc.format.extent	150 p.	es_ES
dc.language.iso	en	es_ES
dc.subject	artificial Intelligence	es_ES
dc.subject	predictive modeling	es_ES
dc.subject	dialysis	es_ES
dc.subject	anemia	es_ES
dc.subject	machine learning	es_ES
dc.subject	mathematical models	es_ES
dc.title	Anemia management in end stage renal disease patients undergoing dialysis: a comprehensive approach through machine learning techniques and mathematical modeling	es_ES
dc.type	doctoral thesis	es_ES
dc.subject.unesco	UNESCO::CIENCIAS TECNOLÓGICAS	es_ES
dc.embargo.terms	0 days	es_ES