Chronic lung disease (CLD) in premature infants caused by bronchopulmonary dysplasia (BPD) is a growing problem worldwide due to the increase of the survival rate of premature infants at lower gestational age. Pulmonary edema is a frequently reported pathology impacting premature newborns affected by bronchopulmonary dysplasia (BPD), and also respiratory distress syndrome (RDS). Its etiology can be summarized as an excess of fluids in the alveolar or proto-alveolar epithelium caused by a disequilibrium between the blood flow in the pulmonary circulation and the clearance capacity of the lymphatics vessels. Several treatments have been proposed including both mechanical ventilation and pharmacological therapy. Diuretics are one of the most frequently used medications in Neonatal Intensive Care Unit (NICU). Despite the widespread likelihood of administering such drugs, their associated short and long-term effects are still unclear. This thesis aims to characterize the response to a routinely administered 3-days diuretic trial in a cohort of premature newborns. The proposed approach leverages the richness of the continuously recorded vital signs as well as the information abstracted from the electronic health records (EHR). Secondly, we aim to examine the concordance between the doctors’ recommendations (discontinuation versus continuation of the treatment) versus the novel unsupervised learning analysis illustrated in this work, which exploits vital signs, such as heart rate (HR), respiratory rate (Resp), and oxygen saturation (SpO2), and EHR-derived features. Lastly, we investigated the relationship between length of stay (LOS) (after the 3-days trial) and the development of BPD. Results derived from the proposed analytic pipeline reveal low to moderate agreement between clinical decisions and the proposed clustering, especially on associated clinical outcomes as LOS. Specifically, the vital signs average trends for standard deviation (SD) suggest a reconsideration of the traditional classification of responder versus not-responder. Vital signs trend-based clustering confirm our primary hypothesis that the dynamics of physiological signals offer additional insights on treatment response (compared to the solely utilization of EHR) and may be instrumental in developing a more reliable monitoring system.
La malattia polmonare cronica causata dalla displasia broncopolmonare è un problema crescente nel mondo per via dell’aumento del tasso di sopravvivenza dei neonati prematuri ad età gestazionali basse. L’edema polmonare è una frequente patologia che affligge i bambini affetti da sindrome da distress respiratorio o displasia broncopolmonare. La sua eziologia può essere riassunta come un eccesso di fluidi negli alveoli o nell’epitelio proto-alveolare a causa di un disequilibrio tra il flusso sanguigno nel circolo polmonare e la capacità di clearance dei vasi linfatici. Molti trattamenti sono stati proposti, dalla ventilazione meccanica ai trattamenti farmacologici. L’uso dei diuretici è uno dei più frequenti trattamenti in terapia intensiva neonatale. Nonostante la larga somministrazione, i loro effetti a breve e lungo termine sono ancora incerti. Questa tesi ha lo scopo di caratterizzare la risposta di un gruppo di neonati prematuri a una routine di somministrazione di tre giorni di diuretici. Questo approccio deve la sua ricchezza all’utilizzo di segnali vitali costantemente registrati, estratti dalla cartella clinica elettronica. Secondariamente, si analizzerà la concordanza tra la decisione medica (di continuare o interrompere il trattamento) e l’approccio non supervisionato illustrato che sfrutta i segnali vitali, come la frequenza cardiaca e respiratoria e la saturazione dell’ossigeno, e features estratte dalla cartella clinica elettronica Infine, si analizzerà la relazione di tali risultati con lo sviluppo di displasia broncopolmonare e la durata dell’ospedalizzazione a seguito dei tre giorni. I nostri risultati mostrano una moderata concordanza tra la decisione medica e il nostro approccio, specialmente per quanto riguarda il tempo trascorso in terapia intensiva dopo il trial. In particolare, la riduzione della deviazione standard per questi segnali suggerisce una riconsiderazione della classica divisione in rispondenti e non rispondenti. Il clustering basato sull’andamento dei segnali vitali ha confermato la nostra prima ipotesi secondo cui essi offrano maggiori informazioni rispetto al solo utilizzo della cartella clinica, dimostrando che potrebbe essere uno strumento per lo sviluppo di sistemi di monitoraggio più affidabile.
A vital signs-driven approach for clustering the responses to diuretics treatment in premature newborns
Asnaghi, Riccardo
2021/2022
Abstract
Chronic lung disease (CLD) in premature infants caused by bronchopulmonary dysplasia (BPD) is a growing problem worldwide due to the increase of the survival rate of premature infants at lower gestational age. Pulmonary edema is a frequently reported pathology impacting premature newborns affected by bronchopulmonary dysplasia (BPD), and also respiratory distress syndrome (RDS). Its etiology can be summarized as an excess of fluids in the alveolar or proto-alveolar epithelium caused by a disequilibrium between the blood flow in the pulmonary circulation and the clearance capacity of the lymphatics vessels. Several treatments have been proposed including both mechanical ventilation and pharmacological therapy. Diuretics are one of the most frequently used medications in Neonatal Intensive Care Unit (NICU). Despite the widespread likelihood of administering such drugs, their associated short and long-term effects are still unclear. This thesis aims to characterize the response to a routinely administered 3-days diuretic trial in a cohort of premature newborns. The proposed approach leverages the richness of the continuously recorded vital signs as well as the information abstracted from the electronic health records (EHR). Secondly, we aim to examine the concordance between the doctors’ recommendations (discontinuation versus continuation of the treatment) versus the novel unsupervised learning analysis illustrated in this work, which exploits vital signs, such as heart rate (HR), respiratory rate (Resp), and oxygen saturation (SpO2), and EHR-derived features. Lastly, we investigated the relationship between length of stay (LOS) (after the 3-days trial) and the development of BPD. Results derived from the proposed analytic pipeline reveal low to moderate agreement between clinical decisions and the proposed clustering, especially on associated clinical outcomes as LOS. Specifically, the vital signs average trends for standard deviation (SD) suggest a reconsideration of the traditional classification of responder versus not-responder. Vital signs trend-based clustering confirm our primary hypothesis that the dynamics of physiological signals offer additional insights on treatment response (compared to the solely utilization of EHR) and may be instrumental in developing a more reliable monitoring system.File | Dimensione | Formato | |
---|---|---|---|
ASNAGHI_Riccardo_Executive_Summary.pdf
solo utenti autorizzati dal 01/12/2023
Descrizione: Executive Summary
Dimensione
987.95 kB
Formato
Adobe PDF
|
987.95 kB | Adobe PDF | Visualizza/Apri |
ASNAGHI_Riccardo_Master_Thesis.pdf
solo utenti autorizzati dal 01/12/2023
Descrizione: Master Thesis
Dimensione
7.97 MB
Formato
Adobe PDF
|
7.97 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/197261