Some fundamental concepts concerning the analysis of longitudinal data will be covered in the thesis. These data are the outcome of repeated assessments of specific clinical indicators or physiological signals, often known as biomarkers, on the same individual. Some crucial theoretical aspects and common strategies for analyzing longitudinal data (in particular Linear Mixed-effects models, Extended Cox model, Joint models) and frequent issues such as missing data, which are typical in the context of dynamic data, will be covered. As a doctorate student, I have mostly focused on longitudinal data analyses from the Intensive Care Unit’s (ICU) clinical context, from an applied standpoint. ICU patients are subject to close and ongoing surveillance because of their medical condition, which usually starts on the day of admission and lasts for the following weeks. These people exhibit abrupt changes in their vital functions over time, which can be measured using particular markers. This calls for a constant observation and assessment of numerous clinical indications of interest throughout time. This often results in a significant amount of data that may be examined. Starting from this context, the results from the ORANGE study will be presented. The ORANGE (Outcome pRognostication of Acute brain injury with the NeuroloGical pupil indEx) study is an international, multicenter, prospective, observational trial that enrolled 514 adult patients admitted to the ICU, requiring intubation and mechanical ventilation, after Traumatic Brain Injury, Subarachnoid Haemorrhage, or Intracranial Haemorrhage, at 13 hospitals in eight countries in Europe and the United States. The study aimed to evaluate the association between abnormal pupillary function, assessed by the Neurological Pupil index (NPi) and long-term outcomes in patients with Acute Brain Injury (ABI). The co-primary outcomes of the study were functional neurological outcome assessed with the extended Glasgow Outcome Scale -GOSE- and mortality at 6 months. Overall, the results strongly suggest that repeatedly abnormal NPi values (NPi < 3), including the most extreme values of zero, in the first week after ABI predict poor outcome. An increase in the number of abnormal NPi measurements over time was associated with a higher probability of poor neurological outcome. Two consecutive NPi measurements equal to zero, or deterioration of NPi to a value of zero, were associated with an increased mortality risk. By contrast, the mortality risk was not increased when an NPi value of zero recovered to a higher value. These findings indicate the importance of the trajectories of NPi. Afterwards the results of a simulation work designed to evaluate the robustness of Extended Cox model (ECM) and Joint models (JM) when the longitudinal process is affected by different missing mechanisms will be presented. Starting from simulated “complete case” longitudinal data, considering two different functional forms for the population-averaged trajectories, i.e. linear and quadratic, we branched out into two possible scenarios. In the first case, after the missing process, obtained with a probabilistic approach, is applied, the biomarker profiles loose some observations, resulting in “intermittent” patterns. In the second case, instead, we took into account a scenario in which as soon as a visit is missed, that is after the first missing value in the biomarker profile, the following values are no longer observed. In both cases, the state and time at the patient's event are observed and unaltered. In other words, the missingness of some values in the individual markers profiles does not imply the drop-out of the individual.

Nella tesi verranno trattati alcuni concetti fondamentali relativi all'analisi dei dati longitudinali. Questi sono il risultato di misurazioni ripetute di specifici indicatori clinici o segnali fisiologici, noti come biomarcatori, sullo stesso individuo. Saranno trattati alcuni aspetti teorici cruciali e strategie comuni per l'analisi dei dati longitudinali (in particolare modelli lineari a effetti misti, modello di Cox esteso (ECM), modelli congiunti (JM)) e problemi frequenti come i dati mancanti, tipici nel contesto dei dati dinamici. Come attività di dottorato, mi sono concentrato, da un punto di vista applicativo ,sull'analisi di dati longitudinali nel contesto clinico dell'Unità di Terapia Intensiva (UTI). I pazienti di UTI sono soggetti a una stretta e continua sorveglianza a causa delle loro condizioni mediche, che di solito inizia il giorno del ricovero e perdura per le settimane successive. Tali soggetti presentano brusche variazioni delle funzioni vitali nel corso del tempo, che possono essere misurate con particolari marcatori. Ciò richiede una costante osservazione e valutazione di numerose indicazioni cliniche di interesse nel corso del tempo. Grazie a questo monitoraggio si puo’ ottenere una notevole quantità di dati che possono essere esaminati. Partendo da questo contesto, verranno presentati i risultati dello studio ORANGE. Lo studio ORANGE (Outcome pRognostication of Acute brain injury with the NeuroloGical pupil indEx) è uno studio internazionale, multicentrico, prospettico e osservazionale che ha arruolato 514 pazienti adulti ricoverati in terapia intensiva, che necessitavano di intubazione e ventilazione meccanica, dopo una lesione cerebrale traumatica, un'emorragia subaracnoidea o un'emorragia intracranica, in 13 ospedali di otto Paesi in Europa e Stati Uniti. Lo studio mirava a valutare l'associazione tra la funzione pupillare anormale, valutata mediante il Neurological Pupil index (NPi), e gli outcomes a lungo termine nei pazienti con lesione cerebrale acuta (ABI). Gli endpoints co-primari dello studio erano l'outcome funzionale neurologico valutato con la Glasgow Outcome Scale estesa -GOSE- e la mortalità a 6 mesi. Nel complesso, i risultati suggeriscono fortemente che valori NPi ripetutamente anormali (NPi < 3), compresi i valori più estremi di zero, nella prima settimana dopo l'ABI predicono un outcome negativo. Un aumento del numero di misurazioni NPi anomale nel tempo è stato associato a una maggiore probabilità di esito neurologico sfavorevole. Due misurazioni consecutive di NPi pari a zero, o il deterioramento di NPi fino a un valore pari a zero, erano associati a un aumento del rischio di mortalità. Al contrario, il rischio di mortalità non aumentava quando un valore NPi pari a zero ritornava a un valore poitivo. Questi risultati indicano l'importanza delle traiettorie di NPi. Dopodichè verranno presentati i risultati di un lavoro di simulazione volto a valutare la robustezza del modello di Cox esteso (ECM) e dei modelli congiunti (JM) quando il processo longitudinale è affetto da diversi meccanismi di missing. Partendo da dati longitudinali simulati del "caso completo" e considerando due diverse forme funzionali per le traiettorie medie della popolazione, ossia lineare e quadratica, abbiamo delineato due scenari possibili. Nel primo caso, dopo l'applicazione del processo di missing, ottenuto con un approccio probabilistico, i profili dei biomarcatori perdono alcune osservazioni, dando luogo a pattern "intermittenti". Nel secondo caso, invece, abbiamo preso in considerazione uno scenario in cui non appena si perde una misurazione, cioè dopo il primo valore mancante nel profilo del biomarcatore, i valori successivi non vengono più osservati. In entrambi i casi, lo stato e il tempo dell'evento del paziente sono osservati e inalterati. In altre parole, la mancanza di alcuni valori nei profili dei singoli marcatori non implica il drop-out dallo studio da parte dell'individuo.

(2024). The use of Joint and Cox models to assess the association between a longitudinal marker and a time-to-event: a simulation study under different missing mechanisms and applications in iCU setting. (Tesi di dottorato, Università degli Studi di Milano-Bicocca, 2024).

The use of Joint and Cox models to assess the association between a longitudinal marker and a time-to-event: a simulation study under different missing mechanisms and applications in iCU setting

PETROSINO, MATTEO
2024

Abstract

Some fundamental concepts concerning the analysis of longitudinal data will be covered in the thesis. These data are the outcome of repeated assessments of specific clinical indicators or physiological signals, often known as biomarkers, on the same individual. Some crucial theoretical aspects and common strategies for analyzing longitudinal data (in particular Linear Mixed-effects models, Extended Cox model, Joint models) and frequent issues such as missing data, which are typical in the context of dynamic data, will be covered. As a doctorate student, I have mostly focused on longitudinal data analyses from the Intensive Care Unit’s (ICU) clinical context, from an applied standpoint. ICU patients are subject to close and ongoing surveillance because of their medical condition, which usually starts on the day of admission and lasts for the following weeks. These people exhibit abrupt changes in their vital functions over time, which can be measured using particular markers. This calls for a constant observation and assessment of numerous clinical indications of interest throughout time. This often results in a significant amount of data that may be examined. Starting from this context, the results from the ORANGE study will be presented. The ORANGE (Outcome pRognostication of Acute brain injury with the NeuroloGical pupil indEx) study is an international, multicenter, prospective, observational trial that enrolled 514 adult patients admitted to the ICU, requiring intubation and mechanical ventilation, after Traumatic Brain Injury, Subarachnoid Haemorrhage, or Intracranial Haemorrhage, at 13 hospitals in eight countries in Europe and the United States. The study aimed to evaluate the association between abnormal pupillary function, assessed by the Neurological Pupil index (NPi) and long-term outcomes in patients with Acute Brain Injury (ABI). The co-primary outcomes of the study were functional neurological outcome assessed with the extended Glasgow Outcome Scale -GOSE- and mortality at 6 months. Overall, the results strongly suggest that repeatedly abnormal NPi values (NPi < 3), including the most extreme values of zero, in the first week after ABI predict poor outcome. An increase in the number of abnormal NPi measurements over time was associated with a higher probability of poor neurological outcome. Two consecutive NPi measurements equal to zero, or deterioration of NPi to a value of zero, were associated with an increased mortality risk. By contrast, the mortality risk was not increased when an NPi value of zero recovered to a higher value. These findings indicate the importance of the trajectories of NPi. Afterwards the results of a simulation work designed to evaluate the robustness of Extended Cox model (ECM) and Joint models (JM) when the longitudinal process is affected by different missing mechanisms will be presented. Starting from simulated “complete case” longitudinal data, considering two different functional forms for the population-averaged trajectories, i.e. linear and quadratic, we branched out into two possible scenarios. In the first case, after the missing process, obtained with a probabilistic approach, is applied, the biomarker profiles loose some observations, resulting in “intermittent” patterns. In the second case, instead, we took into account a scenario in which as soon as a visit is missed, that is after the first missing value in the biomarker profile, the following values are no longer observed. In both cases, the state and time at the patient's event are observed and unaltered. In other words, the missingness of some values in the individual markers profiles does not imply the drop-out of the individual.
REBORA, PAOLA
Dati longitudinali; Terapia Intensiva; simulazioni; Modello Cox esteso; modelli congiunti
Longitudinal data; ICU; simulations; Extended Cox model; Joint models
MED/01 - STATISTICA MEDICA
Italian
27-feb-2024
36
2022/2023
open
(2024). The use of Joint and Cox models to assess the association between a longitudinal marker and a time-to-event: a simulation study under different missing mechanisms and applications in iCU setting. (Tesi di dottorato, Università degli Studi di Milano-Bicocca, 2024).
File in questo prodotto:
File Dimensione Formato  
phd_unimib_876320.pdf

accesso aperto

Descrizione: THE USE OF JOINT AND COX MODELS TO ASSESS THE ASSOCIATION BETWEEN A LONGITUDINAL MARKER AND A TIME-TO-EVENT: A SIMULATION STUDY UNDER DIFFERENT MISSING MECHANISMS AND APPLICATIONS IN ICU SETTING
Tipologia di allegato: Doctoral thesis
Dimensione 4.82 MB
Formato Adobe PDF
4.82 MB Adobe PDF Visualizza/Apri

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/10281/465018
Citazioni
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
Social impact