N Mounier1, C Ferme, H Flechtner, M Henzy-Amar, E Lepage. 1. Groupe d'Etude des Lymphomes de l'Adulte (GELA), Institut d'Hématologie, Hôpital Saint Louis, AP-HP, Paris, France. nicolas.mounier@sls.ap-hop-paris.fr
Abstract
BACKGROUND: The standard Q-TWiST approach defines a series of health states and weights each state's duration according to its quality of life (QOL) to calculate quality-adjusted lifetimes. However, a fixed weight may not adequately reflect time variations in QOL. METHODS: To account for measurements derived from irregular visits and informative missing data, the authors estimated the mean QOL profile using a mixed-effect growth curve model for the response, combined with a logistic regression model for the drop-out process. RESULTS: Using data from a clinical study of lymphoma patients, the authors demonstrated better readaptation to normal life for patients younger than 30. Sensitivity analyses and computer simulations demonstrated that modeling the drop-out probability as a function of the QOL measurements is necessary if conditioning by health state is not possible. CONCLUSION: Our model-based approach is useful to analyze studies with incomplete QOL data, especially when approximate QOL assessment by health state is not possible.
BACKGROUND: The standard Q-TWiST approach defines a series of health states and weights each state's duration according to its quality of life (QOL) to calculate quality-adjusted lifetimes. However, a fixed weight may not adequately reflect time variations in QOL. METHODS: To account for measurements derived from irregular visits and informative missing data, the authors estimated the mean QOL profile using a mixed-effect growth curve model for the response, combined with a logistic regression model for the drop-out process. RESULTS: Using data from a clinical study of lymphomapatients, the authors demonstrated better readaptation to normal life for patients younger than 30. Sensitivity analyses and computer simulations demonstrated that modeling the drop-out probability as a function of the QOL measurements is necessary if conditioning by health state is not possible. CONCLUSION: Our model-based approach is useful to analyze studies with incomplete QOL data, especially when approximate QOL assessment by health state is not possible.
Authors: Jeff A Sloan; Daniel J Sargent; Paul J Novotny; Paul A Decker; Randolph S Marks; Heidi Nelson Journal: J Pain Symptom Manage Date: 2013-11-15 Impact factor: 3.612