Portal de Periódicos da CAPES

Abstracts from ASENT 2004 Annual Meeting March 11–13, 2004

2004; Springer Nature; Volume: 1; Issue: 4 Linguagem: Inglês

10.1602/neurorx.1.4.506

1545-5351

Yuyan Duan, Ilya Lipkovich, Saeeduddin Ahmed, Jonna Ahl, Thomas A. Hardy, Diane Haldane, Robert L. Baker, Mauricio Tohen, Hong Liu‐Seifert, Kristine Healey, Bruce J. Kinon, Saeed Ahmed, Ilya Lipkovich, Mauricio Tohen, Vicki Poole Hoffmann, Dong Ding, Ellen Frank, Lizheng Shi, Janey Shin, Diego Novick, Paul B. van den Berg, Haya Ascher‐Svanum, Josep María Haro, I. Gasquet, Spyridon Tziveleskis, Fabio Blandini, Marie‐Thérèse Armentero, Roberto Fancellu, Giuseppe Nappi, David G. White, Mark A. Jensen, Barry G.W. Arnason, Samuel Frank, Karl Kieburtz, Robert G. Holloway, Renée Wilson, Carol Zimmerman, Scott Y. H. Kim, Jordan J. Elm, Barbara C. Tilley, Amy Yu, Paulo Guimarães, Christopher Goetz, Bernard Ravina, Karl Keiburtz, Steven M. Leventer, Karen Raudibaugh, John C. Keogh, Robert F. Kucharik, Deirdre O’Hara, Naidong Ye, Kimm Galbraith, Brian Speicher, Kevin L. Keim, Alireza Atri, Matthew L. LoPresti, Seth Sherman, Haline E. Schendan, Michael E. Hasselmo, Chantal E. Stern, Joseph Jankovic, Christine Hunter, Kevin Dat Vuong, R. Horowski, Heike Beneš, Dirk Woitalla, H. Przuntek, Jan Tack, George R. Uhl, James P. Bennett, Lorelie Villarete, C. P. Liu, Howard L. Weiner, M. J. Tong, Arash Rassoulpour, Hsi‐Yang Wu, P. Guidetti, Helen E. Scharfman, Guy M. McKhann, Robert Goodman, Edward H. Bertram, Robert Schwarcz, Francesco Bibbiani, Aiste Kielaite, Lauren C. Costantini, Thomas N. Chase, Irene Avila, Justin D. Oh, Edward Castañeda, Christopher P. Smith, Thomas N. Chase, Xiaoxia Wang, Gerda Andringa, William Bara‐Jimenez, Emory Encarnacio, Michael Morris, A.M. Bridgeman, Catherine Bennett, Madhavi Thomas, Tetsuo Ashizawa, Thomas W. Weickert, Terry E. Goldberg, Aaron L. Mishara, José Apud, Bhaskar Kolachana, Michael Egan, Daniel R. Weinberger,

Background: Analyses of categorical repeated measures of clinical data using conventional methods can give biased estimates of treatment effects and associated SEs when dropouts are not completely at random (depending on observed clinical outcomes).We test the utility of multiple imputation (MI) analysis in reducing these biases.Methods: We used simulation to compare performance of MI versus conventional methods, including restricted pseudolikelihood methods and generalized estimating equations, in five typical clinical profiles for 1) estimating overall treatment effects, and 2) treatment differences at last scheduled visit.Results: The power to detect treatment differences with MI is consistently higher than with conventional methods.Type I error rates (estimated from scenarios in which no treatment difference existed) were consistently smaller with MI than with conventional methods.However, MI tended to overestimate variability of treatment differences at endpoint.Among tested profiles, the advantage of MI over conventional methods in terms of power to detect overall treatment differences was greatest when treatments separated from each other early, then converged later.Conclusion: Compared to conventional techniques, MI may lead to less biased estimates of treatment differences in categorical analyses of continuous data, especially in clinical trials with a high (40-60%) proportion of dropouts.However, MI did not perform well when dropouts were (partially) driven by clinical outcomes that were also not observed.Of course, this conclusion is limited by the specifics of the simulation scenarios tested and as such, does not constitute theoretical proof.