Unbiased treatment effect estimates by modeling the disease process of multiple sclerosis

• Published in: Journal of the neurological sciences Vol. 278; no. 1; pp. 54 - 59
• Main Authors: Healy, Brian C, Hayden, Douglas L, Sampat, Mehul P, Bakshi, Rohit, Guttmann, Charles R.G
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 15.03.2009; Elsevier
• Subjects: Active/inactive disease process; Biological and medical sciences; Brain - physiopathology; Clinical trials; Clinical Trials, Phase I as Topic; Clinical Trials, Phase II as Topic; Computer Simulation; Gadolinium - metabolism; Humans; Medical sciences; Models, Biological; Models, Statistical; MRI; Multiple sclerosis; Multiple sclerosis and variants. Guillain barré syndrome and other inflammatory polyneuropathies. Leukoencephalitis; Multiple Sclerosis, Relapsing-Remitting - physiopathology; Multiple Sclerosis, Relapsing-Remitting - therapy; Neurology; Normal Distribution; Poisson Distribution; Regression to the mean; Treatment Outcome; Clinical trials; Active/inactive disease process; Multiple sclerosis; Regression to the mean; MRI; Nervous system diseases; Treatment; Central nervous system disease; Clinical trial; Nuclear magnetic resonance imaging; Modeling; Inflammatory disease
• ISSN: 0022-510X; 1878-5883
• DOI: 10.1016/j.jns.2008.11.013

Abstract Gadolinium-enhancing lesions in the brain are commonly used as a primary outcome measure of disease activity in phase I/II clinical trials in multiple sclerosis (MS). The advent of effective therapy and the cost of clinical trials have led some researchers to adopt a one-arm study design with selection towards patients showing MRI activity. Regression to the mean is recognized as an important consideration in these trials, but the additional confounding effect of alternating active and inactive phases of disease has not been considered. Simulated data were generated from Poisson and normal distributions to mimic outcomes from phase I/II clinical trials of patients with relapsing–remitting MS under a constant or changing disease process model. In all cases, conventional comparison of pretreatment to on-treatment measurements overestimated the treatment effect. Although correction for regression to the mean provided unbiased estimates of the treatment effect under a constant disease process model, this correction also overestimated the treatment effect when disease activity changed over time. Conversely, unbiased estimates of the treatment effect under an alternating (active/inactive) disease process were obtained by correctly accounting for regression to the mean and the disease process. The implications of these results are discussed in terms of efficacy and safety.