Browse Prior Art Database

A Method for Clinical Risk Prediction across Heterogeneous Feature Types using Multi-linear Regression Models

IP.com Disclosure Number: IPCOM000238067D
Publication Date: 2014-Jul-30
Document File: 4 page(s) / 229K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method to use a tensor to represent the clinical observations from a patient; each mode of the tensor corresponds to a feature type. A logistic regression objective receives the tensors as input and predicts the risk for each patient.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 53% of the total text.

Page 01 of 4

A Method for Clinical Risk Prediction across Heterogeneous Feature Types using Multi-

-linear Regression Models

linear Regression Models

This article addresses the problem of predicting clinical risk based on heterogeneous feature types. Existing work ignores the heterogeneity and concatenates the features as a vector. This leads to high dimensionality, extreme sparsity, and loses the interaction between different feature types.

Existing risk prediction models represent the clinical observations of a patient as a vector. The vector based representation is not ideal when clinical features that are heterogeneous in nature are present. Examples of heterogeneous clinical features include, but are not limited to:

• Diagnosis + lab + medication + procedure + "

• Concatenating different feature types together , which leads to very
high dimensional feature space and over fitting in practice

Figure 1: Vector Based Representation of Patients

Normalization across different feature types is very challenging, if not impossible.

The proposed solution models the multi type clinical features as a tensor and shows how to solve the corresponding multi linear logistic regression problem. The novel method uses a tensor to represent the clinical observations from a patient . Each mode of the tensor corresponds to a feature type . A logistic regression objective receives the tensors as input and predicts the risk for each patient .

The solution comprises a number of analytic methods . A tensor based data model is used for heterogeneous clinical feature types. The data model captures the interaction between different feature types. A multi linear logistic regression algorithm uses the tensors as input and predicts the risk for each patient . Finally, the method implements an efficient optimization procedure....