Causal effects of multiple sclerosis therapies in left-truncated registry data.
Left-truncation is an unrecorded interval between multiple sclerosis (MS) onset and initial data in observational studies. This delay may bias estimates of disease-modifying therapy (DMT) effectiveness, especially when determined by patient or disease characteristics.
To examine whether causal effect estimates of DMTs over the full disease course can be reliably derived from left-truncated registry data.
We analysed data from MSBase (144 centres, 41 countries) to assess the impact of left-truncation on causal treatment effect estimates. Cox marginal structural models (MSMs) estimated hazard ratios (HRs) for relapses, disability worsening and improvement, considering left-truncation at random and not-at-random. Fixed-time truncation and multivariable adjustment were applied to remediate bias.
The study included 5588 patients tracked from true MS onset. The null model, without left-truncation, estimated the DMT effect on relapse risk (HR = 0.64; 95% confidence interval (CI) = 0.54-0.77). Left-truncation inflated this estimate. Shorter random truncation (1 year) produced greater bias (HR = 0.34), decreasing with longer durations (3-year HR = 0.48). Truncation not-at-random biased relapse estimates (HR = 0.37). Disability outcomes were less sensitive.
MSMs can reliably estimate DMT effectiveness in left-truncated MS registry data, although accuracy depends on truncation mechanism and duration. Both random and not-at-random truncation impact relapse estimates. Disability outcomes appear less sensitive. Fixed-time truncation and covariate adjustment mitigated bias.
To examine whether causal effect estimates of DMTs over the full disease course can be reliably derived from left-truncated registry data.
We analysed data from MSBase (144 centres, 41 countries) to assess the impact of left-truncation on causal treatment effect estimates. Cox marginal structural models (MSMs) estimated hazard ratios (HRs) for relapses, disability worsening and improvement, considering left-truncation at random and not-at-random. Fixed-time truncation and multivariable adjustment were applied to remediate bias.
The study included 5588 patients tracked from true MS onset. The null model, without left-truncation, estimated the DMT effect on relapse risk (HR = 0.64; 95% confidence interval (CI) = 0.54-0.77). Left-truncation inflated this estimate. Shorter random truncation (1 year) produced greater bias (HR = 0.34), decreasing with longer durations (3-year HR = 0.48). Truncation not-at-random biased relapse estimates (HR = 0.37). Disability outcomes were less sensitive.
MSMs can reliably estimate DMT effectiveness in left-truncated MS registry data, although accuracy depends on truncation mechanism and duration. Both random and not-at-random truncation impact relapse estimates. Disability outcomes appear less sensitive. Fixed-time truncation and covariate adjustment mitigated bias.
Authors
Haile Haile, Diouf Diouf, Ozakbas Ozakbas, Horakova Horakova, Havrdova Havrdova, Patti Patti, Eichau Eichau, Alroughani Alroughani, Lugaresi Lugaresi, Tomassini Tomassini, Prat Prat, Girard Girard, Terzi Terzi, Yamout Yamout, Khoury Khoury, Grammond Grammond, Blanco Blanco, Shaygannejad Shaygannejad, Foschi Foschi, Surcinelli Surcinelli, Neri Neri, Weinstock-Guttman Weinstock-Guttman, Prevost Prevost, Amato Amato, Barnett Barnett, Gerlach Gerlach, John John, Kermode Kermode, Fabis-Pedrini Fabis-Pedrini, Carroll Carroll, van der Walt van der Walt, Butzkueven Butzkueven, van Pesch van Pesch, Soysal Soysal, Gouider Gouider, Mrabet Mrabet, Spitaleri Spitaleri, Cartechini Cartechini, Maimone Maimone, Ampapa Ampapa, Laureys Laureys, Ramo-Tello Ramo-Tello, Di Gregorio Di Gregorio, Lapointe Lapointe, Slee Slee, Karabudak Karabudak, Garber Garber, Altintas Altintas, Hodgkinson Hodgkinson, Sanchez-Menoyo Sanchez-Menoyo, Castillo-Triviño Castillo-Triviño, Habek Habek, Al-Asmi Al-Asmi, Al-Harbi Al-Harbi, Csepany Csepany, Cárdenas-Robledo Cárdenas-Robledo, Taylor Taylor, Foong Foong, Willekens Willekens, Shalaby Shalaby, Moore Moore, McGuigan McGuigan, Baghbanian Baghbanian, Massey Massey, Hardy Hardy, Ramanathan Ramanathan, Gross-Paju Gross-Paju, Gray Gray, Decoo Decoo, Shaw Shaw, Simu Simu, Rozsa Rozsa, Stuart Stuart, Sharmin Sharmin, Roos Roos, Kalincik Kalincik
View on Pubmed