Date d'introduction 30 09 2007 entreprise (nom, site internet

Malali

Description

30/09/2007 Date d'introduction Entreprise (Nom, site internet) IDDI, http://www.iddi.com Logo Nom, Prénom du contact, [e-mail] téléphone, adresse e-mail du linda.danielson@iddi.

Sujets

• Employabilité, métiers et formations

Informations

• Publié par: englue
• Nombre de lectures: 140
• Langue: Français

Extrait

________________________________________________________________________

Formulaire de propositions de sujets de TFE- Action de formation BioW in

1

Date d'introduction

30/09/2007

Entreprise

(Nom, site internet)

IDDI,

http://www.iddi.com

Logo

Nom, Prénom du contact,

téléphone, adresse e-mail du

contact*

[e-mail]

[e-mail]

Lieu

Avenue Provinciale, 30

1340 Ottignies / Louvain-La-Neuve - Belgium

Département de l'entreprise

biostatistique

Niveau d'études concerné

1

universitaire

Type de stage

1

Mémoire

Thématique

Use of Cox regression model in randomized clinical trials

This project will consist of

1. extracting from the medical literature a large number of

papers (50 to 100) describing the results of randomized trials

in which the Cox model was used to adjust the treatment

comparison;

2. comparing the significance of the adjusted and the

unadjusted treatment comparisons, using a correlation

approach and concordance statistics;

3. exploring (through simulations) the impact of non

proportional hazards on the outcome of early breast cancer

trials, using either a Cox regression model, a stratified or an

unstratified logrank test;

4. writing a paper with the findings.

Background

Sir David Cox introduced his famed regression model in

1972. His initial exposition of the model remains the most

frequently cited scientific paper of all times. Although the

model may be extremely useful in epidemiology to account

simultaneously for many confounders, it does make the

assumption of proportional hazards, which may or may not

be valid. In the context of randomized trials, use of the Cox

model is unnecessary at best (a stratified logrank is

appropriate to account for a few truly important prognostic

factors) and miseleading at worst (because of strong non

proportionality of the hazards and/or multicollinearity of the

prognostic factors included in the model).

It would therefore be of interest to investigate use of the Cox

model in randomized clinical trials from two perspectives:

(a) is there empirical evidence that published results of past

trials are unaffected by use of the Cox model? and

(b) is there empirical evidence from current trials that strong

departures from proportional hazards and/or correlations

between the baseline covariates spuriously affect the

statistical significance of treatment comparisons?

Formations visées

Statistique, mathématique, biostatistique

Techniques abordées

Compétences visées

Compétences requises

A good knowledge of survival analysis is needed for the

project (logrank test, Cox regression model). The student

should additionally have a broad interest in clinical trial

methodology. Good proficiency in English is a must.

Référénces

REFERENCES

1. Cox DR, J R Stat Soc (B) 1972;34187.