Grid multi-category response logistic models

Abstract

BackgroundMulti-category response models are very important complements to binary logistic models in medical decision-making. Decomposing model construction by aggregating computation developed at different sites is necessary when data cannot be moved outside institutions due to privacy or other concerns. Such decomposition makes it possible to conduct grid computing to protect the privacy of individual observations.MethodsThis paper proposes two grid multi-category response models for ordinal and multinomial logistic regressions. Grid computation to test model assumptions is also developed for these two types of models. In addition, we present grid methods for goodness-of-fit assessment and for classification performance evaluation.ResultsSimulation results show that the grid models produce the same results as those obtained from corresponding centralized models, demonstrating that it is possible to build models using multi-center data without losing accuracy or transmitting observation-level data. Two real data sets are used to evaluate the performance of our proposed grid models.ConclusionsThe grid fitting method offers a practical solution for resolving privacy and other issues caused by pooling all data in a central site. The proposed method is applicable for various likelihood estimation problems, including other generalized linear models.Electronic supplementary materialThe online version of this article (doi:10.1186/s12911-015-0133-y) contains supplementary material, which is available to authorized users.

Privacy-Preserving Logistic Regression with Distributed Data Sources via Homomorphic Encryption

“…Logistic regression with data privacy protection receives significant attentions from researchers in biomedical informatics [16], [29], [32], considering different settings of integrating and sharing data. These works do not use the model of [25] and hence are different from ours.…”

Clinical Research Informatics for Big Data and Precision Medicine

“…With massive data sets that are far too large or too complex to analyze using traditional local computational methods, new approaches for performing analytics using distributed techniques that "bring analytics to the data" rather than "submit data to the analytics" are a new area of active CRI research [70][71][72][73][74]. These have also been adapted to enable distributed analytics of sensitive clinical data without requiring data partners to release any patient-level data, offering a new approach for reducing concerns about patient re-identification.…”

Embracing study heterogeneity for finding genetic interactions in large‐scale research consortia

“…Such data sharing can not only achieve larger sample sizes but also reduce bias caused by nonrandom selection (Wolfson et al, 2010). These include logistic regression-based models (Ji, Jiang, Wang, Xiong, & Ohno-Machado, 2014;Jiang et al, 2013;Wang et al, 2013;Wu, Jiang, Kim, & Ohno-Machado, 2012;Wu et al, 2015), splitting regression analyses (Wolfson et al, 2010), cryptographic techniques for participant-level genomic data (Kamm, Bogdanov, Laur, & Vilo, 2013), and genetic association meta-analyses (Xie et al, 2014), among others. A number of methods have been developed to address the challenge of protecting patient-level information in multiple GWAS/genome databases.…”