Subtype classification and heterogeneous prognosis model construction in precision medicine

Quasi-linear Cox proportional hazards model with cross- L1 penalty

Background

To accurately predict the response to treatment, we need a stable and effective risk score that can be calculated from patient characteristics. When we evaluate such risks from time-to-event data with right-censoring, Cox’s proportional hazards model is the most popular for estimating the linear risk score. However, the intrinsic heterogeneity of patients may prevent us from obtaining a valid score. It is therefore insufficient to consider the regression problem with a single linear predictor.

Methods

we propose the model with a quasi-linear predictor that combines several linear predictors. This provides a natural extension of Cox model that leads to a mixture hazards model. We investigate the property of the maximum likelihood estimator for the proposed model. Moreover, we propose two strategies for getting the interpretable estimates. The first is to restrict the model structure in advance, based on unsupervised learning or prior information, and the second is to obtain as parsimonious an expression as possible in the parameter estimation strategy with cross- L₁ penalty. The performance of the proposed method are evaluated by simulation and application studies.

Results

We showed that the maximum likelihood estimator has consistency and asymptotic normality, and the cross- L₁-regularized estimator has root-n consistency. Simulation studies show these properties empirically, and application studies show that the proposed model improves predictive ability relative to Cox model.

Conclusions

It is essential to capture the intrinsic heterogeneity of patients for getting more stable and effective risk score. The proposed hazard model can capture such heterogeneity and achieve better performance than the ordinary linear Cox proportional hazards model.

Data mining to understand health status preceding traumatic brain injury

The use of precision medicine is poised to increase in complex injuries such as traumatic brain injury (TBI), whose multifaceted comorbidities and personal circumstances create significant challenges in the domains of surveillance, management, and environmental mapping. Population-wide health administrative data remains a rather unexplored, but accessible data source for identifying clinical associations and environmental patterns that could lead to a better understanding of TBIs. However, the amount of data structured and coded by the International Classification of Disease poses a challenge to its successful interpretation. The emerging field of data mining can be instrumental in helping to meet the daunting challenges faced by the TBI community. The report outlines novel areas for data mining relevant to TBI, and offers insight into how the above approach can be applied to solve pressing healthcare problems. Future work should focus on confirmatory analyses, which subsequently can guide precision medicine and preventive frameworks.

A Potential Prognostic Gene Signature for Predicting Survival for Glioblastoma Patients

OBJECTIVE

This study aimed to screen prognostic gene signature of glioblastoma (GBM) to construct prognostic model.

METHODS

Based on the GBM information in the Cancer Genome Atlas (TCGA, training set), prognostic genes (Set X) were screened by Cox regression. Then, the optimized prognostic gene signature (Set Y) was further screened by the Cox-Proportional Hazards (Cox-PH). Next, two prognostic models were constructed: model A was based on the Set Y; model B was based on part of the Set X. The samples were divided into low- and high-risk groups according to the median prognosis index (PI). GBM datasets in Gene Expression Ominous (GEO, GSE13041) and Chinese Glioma Genome Atlas (CGGA) were used as the testing datasets to confirm the prognostic models constructed based on TCGA.

RESULTS

We identified that the prognostic 14-gene signature was significantly associated with the overall survival (OS) in the TCGA. In model A, patients in high- and low-risk groups showed the significantly different OS (P = 7.47 × 10-9, area under curve (AUC) 0.995) and the prognostic ability were also confirmed in testing sets (P=0.0098 and 0.037). The model B in training set was significant but failed in testing sets.

CONCLUSION

The prognostic model which was constructed based on the prognostic 14-gene signature presented a high predictive ability for GBM. The 14-gene signature may have clinical implications in the subclassification of GBM.