System and method for combining hetergeneous predictors with an application to survival anaylsis
A method, a system, and a computer-readable medium for predicting a risk in a survival analysis for a plurality of individuals characterized by at least one predictor are disclosed. A method for estimating risk order of an individual, given information about a set of individuals, characterized by one or many predictors, and provided that direction of association between each predictor and the risk order is known, comprising the step of comparing the individual with each individual within the set of individuals, and estimating risk of individual based on set comparisons.
1 . A method for estimating risk order of an individual, given information about a set of individuals, characterized by one or many predictors, and provided that a direction of association between each predictor and the risk order is known, comprising the steps of:
comparing the individual with each individual within the set of individuals; and,
estimating a risk of an individual based on the comparison.
2 . The method according to claim 1 , further comprising
performing a Univariate Shift Procedure, wherein a measure of distance between each pair of individuals by a single predictor is calculated;
performing a Multi-dimensional Order Procedure, wherein a pair of individuals, based on univariate shifts, is ordered; and,
performing a Scoring Procedure, wherein a score for an individual is calculated.
3 . The method according to claim 2 , wherein said Univarite Shift Procedure further comprises calculating a “trimmed distance”:
d
i
(
a
,
b
)
=
{
f
i
′
(
a
,
b
)
,
if
f
i
′
(
a
,
b
)
>
μ
0
,
otherwise
}
,
wherein μ is a sensitivity threshold for a given predictor.
4 . The method according to the claim 3 , wherein the function ƒ i ′(a,b) on the individuals a, b can be calculated as ƒ i (a)−ƒ i (b), where ƒ i (a) and ƒ i (b) are values of the predictor i for the individual a and b, respectively, and provided that all predictors have a positive association with the risk order.
5 . The method according to claim 2 , wherein the Multi-dimensional Order Procedure further comprises:
determining an aggregate order of the pair of individuals using all predictors using:
q ( a,b )=sign( U ( d 1 ( a,b ), . . . , d m ( a,b ))),
wherein U is an aggregation function, and m is a number of predictors in the dataset.
6 . The method according to claim 2 , wherein the Scoring Procedure further comprises, for an individual a in the plurality of individuals, given a set n of individuals I characterized by predictors {ƒ i }, wherein i=1, . . . , m, define the scoring function as:
s
(
a
)
=
∑
x
ε
I
q
(
a
,
x
)
n
-
1
wherein q(a, x) is an order of individuals a and x obtained by application of the Multi-dimensional Order Procedure; and wherein the sum is taken over all individuals in the dataset of individuals I.
7 . The method according to claim 5 , wherein the aggregated order function is selected from a group including, but not limited to mean, maximum, minimum, sum, median, voting, and consensus.
8 . Computer-readable medium encoded with computer program instructions for performing a method for predicting a risk in a survival analysis for a plurality of individuals characterized by at least one predictor, the method comprising:
performing the following procedures:
a Univariate Shift Procedure, wherein a measure of distance between each pair of individuals by a single predictor is calculated;
a Multi-dimensional Order Procedure, wherein a pair of individuals is ordered, based on univariate shifts;
a Scoring Procedure, wherein a score for an individual is calculated.