A mixed model approach for intent-to-treat analysis in longitudinal clinical trials with missing values

Abstract

Missing values and dropouts are common issues in longitudinal studies in all areas of medicine and public health. Intent-to-treat (ITT) analysis has become a widely accepted method for the analysis of controlled clinical trials. In most controlled clinical trials, some patients do not complete their intended followup according to the protocol for a variety of reasons; this problem generates missing values. Missing values lead to concern and confusion in identifying the ITT population, which makes the data analysis more complex and challenging. No adequate strategy exists for ITT analyses of longitudinal controlled clinical trial data with missing values. Several ad hoc strategies for dealing with missing values for an ITT analysis are common in the practice of controlled clinical trials. We performed a detailed investigation based on simulation studies to develop recommendations for this situation. We compared sizes (type I errors) and power between some popular ad hoc approaches and the linear mixed model approach under different missing value scenarios. Our results suggest that, for studies with a high percentage of missing values, the mixed model approach without any ad hoc imputation is more powerful than other options.

Eight Hours of Nightly Continuous Positive Airway Pressure Treatment of Obstructive Sleep Apnea Improves Glucose Metabolism in Patients with Prediabetes. A Randomized Controlled Trial

“…Regression coefficients and 95% confidence intervals are reported. The mixed model was used because it permits inclusion of all available data and is a valid and advantageous approach for analysis of randomized controlled trials with missing data (37,38). Further details on missing data are provided in the online supplement.…”

Individualized Low-Load Motor Control Exercises and Education Versus a High-Load Lifting Exercise and Education to Improve Activity, Pain Intensity, and Physical Performance in Patients With Low Back Pain: A Randomized Controlled Trial

“…Linear mixed-model analysis was chosen because it uses all the available information in data in a repeated-measures design and is robust in handling missing data. 5 Further, the baseline values were compared between the participants who attended the follow-ups and those who dropped out using the 2-sided Student t test or the chi-square test (APPENDIX B, available online at www.jospt.org). Also, the percentages of the participants who changed their scores by 30% or greater (minimal important change 26 ) for each outcome measure were calculated.…”

A Mixture Dropout Mechanism in a Longitudinal Study With Two Time Points: A Methadone Study

“…Furthermore, this approach was addressed by (Allison, 2001;Schafer and Graham, 2002;Durrant, 2005;Donders et al, 2006;Baraldi and Enders, 2010). On the other hand, linear mixed model is one of the advanced methods to deal with dropouts in longitudinal studies (Molenberghs and Kenward, 2007;Chakraborty and Gu, 2009;Atif et al, 2014).…”

“…A comprehensive detailed texts are provided by (Allison, 2001;Schafer and Graham, 2002;Durrant, 2005;Donders et al, 2006;Baraldi and Enders, 2010;Collins et al, 2001;Wei and Shih, 2001;Brick et al, 2004;Carpenter and Kenward, 2013). On the other hand, linear mixed effect model is a fascinating model for coping with dropouts (missing values) in longitudinal studies which applies maximum likelihood to estimate parameters (Chakraborty and Gu, 2009;Atif et al, 2014).…”