COOKIES USE We use necessary system cookies for the correct functioning of the website and optional Google Analytics cookies to obtain visit statistics. +info
Cookies config
Necessary
The necessary cookies are absolutely essential for the website to work properly. This category only includes cookies that guarantee basic website security and functionality. These cookies do not store any personal information.
Name
Proveedor
Propiedad
Finalidad
Caducidad
+info
_GRECAPTCHA
google.com
Own
provide antispam protection with the reCaptcha service
6 months
cc_cookie_accept
fidmag.org
Own
Usada per confirmar que l'usuari ha confirmat / refusat les cookies (i quins tipus accepta)
1 any
WEB_SESSION
fidmag.org
Own
Cookie técnica: cookie de sessió PHP. Guarda l'id de sessió d'usuari.
al acabar la sessió
Analisys
Analytical cookies are used to understand how visitors interact with the website. These cookies help to provide information on meters, the name of visitors, the percentage of bounces, the font of the traffic, etc.
Name
Proveedor
Propiedad
Finalidad
Caducidad
+info
_ga
Google Analytics
From third parties
Cookie d'anàlisi o mesurament: Identifica els usuaris i proporciona informació sobre com els usuaris troben la pàgina web i com la utilitzen per a realització d'Informes estadístics
2 anys
_gat_gtag_UA_141706552_1
Google Analytics
From third parties
Cookie d'anàlisi o mesurament: Tracking per part de google per google analytics
1 minut
_gid
Google Analytics
From third parties
Cookie d'anàlisi o mesurament: S'usa per limitar el percentatge de sol·licituds
Biased accuracy in multisite machine-learning studies due to incomplete removal of the effects of the site.
Solanes A, Palau P, Fortea L, Salvador R, González-Navarro L, Llach CD, Valentí M, Vieta E, Radua J
Limited service to collaborators of the network of Sisters Hospitalarias Centers. You will receive a message in your email with a link to download this article.
Abstract
Brain MRI researchers conducting multisite studies, such as within the ENIGMA Consortium, are very aware of the importance of controlling the effects of the site (EoS) in the statistical analysis. Conversely, authors of the novel machine-learning MRI studies may remove the EoS when training the machine-learning models but not control them when estimating the models' accuracy, potentially leading to severely biased estimates. We show examples from a toy simulation study and real MRI data in which we remove the EoS from both the "training set" and the "test set" during the training and application of the model. However, the accuracy is still inflated (or occasionally shrunk) unless we further control the EoS during the estimation of the accuracy. We also provide several methods for controlling the EoS during the estimation of the accuracy, and a simple R package ("multisite.accuracy") that smoothly does this task for several accuracy estimates (e.g., sensitivity/specificity, area under the curve, correlation, hazard ratio, etc.).