Investigating the prediction ability of survival models based on both clinical and omics data: two case studies

Abstract

In biomedical literature numerous prediction models for clinical outcomes have been developed based either on clinical data or, more recently, on high-throughput molecular data (omics data). Prediction models based on both types of data, however, are less common, although some recent studies suggest that a suitable combination of clinical and molecular information may lead to models with better predictive abilities. This is probably due to the fact that it is not straightforward to combine data with different characteristics and dimensions (poorly characterized high dimensional omics data, well-investigated low dimensional clinical data). In this paper we analyze two publicly available datasets related to breast cancer and neuroblastoma, respectively, in order to show some possible ways to combine clinical and omics data into a prediction model of time-to-event outcome. Different strategies and statistical methods are exploited. The results are compared and discussed according to different criteria, including the discriminative ability of the models, computed on a validation dataset.

Boosting in Cox regression: a comparison between the likelihood-based and the model-based approaches with focus on the R-packages CoxBoost and mboost

“…In the last years, the importance of combining clinical and molecular data in a prediction model becomes clear in the biomedical field, and recent studies contrasted different solutions and methods to profitably use both kinds of data in the model building process (Boulesteix & Sauerbrei, 2011;De Bin et al, 2014b;Truntzer et al, 2014). The main issue related to the combination of clinical and molecular information is the different nature of the data, which belong to the low-and the high-dimensional world, respectively.…”

“…In Boulesteix & Sauerbrei (2011), these two strategies are called "favoring" and "residuals", respectively (De Bin et al (2014b) use the more opportune term "clinical offset" for the latter strategy). These two strategies have some theoretical differences which may influence the model building process and may result more adequate in specific situations.…”

“…those variables that for some reasons must be treated in a special way in the statistical analysis. Typical examples are clinical variables in biomedical studies also involving omics data: several papers (see, e.g., Binder & Schumacher, 2008;Boulesteix & Sauerbrei, 2011;De Bin et al, 2014b) show that by considering the clinical variables as mandatory it is possible to obtain better prediction models than by simply merging them with the omics data. We will see that the two R packages implement two different strategies to deal with this issue, and, as a novel contribution of the paper, we will explore the possibility of extending solutions currently only implemented in one package to the other.…”

IPF-LASSO: IntegrativeL1-Penalized Regression with Penalty Factors for Prediction Based on Multi-Omics Data

“…In this section, we discuss the connections of IPF-LASSO to these methods. In the scenario investigated by De Bin et al [8], we have two modalities ( M = 2). The first modality includes only a small number p 1 of clinical variables, such that a classical regression approach can be applied to this modality (the rule of thumb that the number of variables times 5 or 10 should not exceed the number of observations is typically satisfied).…”

“…The case of variables from one low-dimensional modality (typically, a few clinical variables relevant to the outcome to be predicted) and one high-dimensional modality (e.g., a microarray gene expression dataset) has been extensively investigated by De Bin et al [8], where they assess the “residual” two-step approach and the “favoring” approach (see Section 2.2 for more details).…”