Estimating personalized drug responses from real world evidence

A computer program product comprising a non-transitory computer readable storage medium having a computer readable program stored therein, wherein the computer readable program, when executed on at least one processor of a data processing system, causes the data processing system to implement a drug response estimation engine, wherein the computer readable program causes the data processing system to:

• receive, by the drug response estimation engine, real-world evidence for a plurality of patients;
• build, by a patient similarity network builder component executing within the drug response estimation engine, a patient similarity network, wherein the patient similarity network is leveraged by a regression analysis component executing within the drug response estimation engine to ensure similar patients have similar drug responses to a given outcome of interest;
• build, by the regression analysis component, a statistical model using a network localized regression analysis approach, wherein building the statistical model comprises using the patient similarity network for regularizing a baseline regularization framework using a network localized LASSO approach to ensure model sparsity and to regularize drug response parameters of similar patients, wherein the statistical model learns parameters that predict drug responses for each patient and for each time point of the real-world evidence, wherein the statistical model optimizes the following objective function:
• arg
• ⁢
• ⁢
• min
• W
• ⁢
• 
• Dy
• -
• DZw
• 
• 2
• 2
• +
• λ
• 1
• ⁢
• ∑
• i
• =
• 1
• N
• ⁢
• 
• w
• i
• 
• 1
• 2
• +
• λ
• 2
• ⁢
• ∑
• i
• >
• i
• ′
• N
• ⁢
• ∑
• i
• ′
• =
• 1
• N
• -
• 1
• ⁢
• r
• ii
• ′
• ⁢
• 
• w
• i
• -
• w
• i
• ′
• 
• 2
• wherein D is a difference matrix, y is a column matrix of laboratory test measurements, Z is a block diagonal matrix of drug exposures, w is a drug response vector, wherein λ1 is a hyper-parameter that controls an exclusive lasso penalty and λ2 is a hyper-parameter that controls a network lasso penalty, wherein wi is a vector that denotes a drug response for each individual patient i, and wherein rii′ represents a patient similarity between the ith and i′th patients based on the similarity network;
• group, by a patient clustering component executing within the drug response estimation engine, patients based on demographics and comorbidities to form a plurality of patient groups, wherein the patient clustering component groups the patients using the learned parameters that predict drug responses such that each patient group within the plurality of patient groups represents patients that have similar drug responses and background information;
• estimate, by the drug response estimation engine, drug responses for a given patient based on the patient similarity network, the statistical model, and the plurality of patient groups;
• and
• output, by the drug response estimation engine, the drug responses for the given patient.

An apparatus comprising:

• a processor;
• and
• a memory coupled to the processor, wherein the memory comprises instructions which, when executed by the processor, cause the processor to implement a drug response estimation engine, wherein the instructions cause the processor to:
• receive, by the drug response estimation engine, real-world evidence for a plurality of patients;
• build, by a patient similarity network builder component executing within the drug response estimation engine, a patient similarity network, wherein the patient similarity network is leveraged by a regression analysis component executing within the drug response estimation engine to ensure similar patients have similar drug responses to a given outcome of interest;
• build, by the regression analysis component, a statistical model using a network localized regression analysis approach, wherein building the statistical model comprises using the patient similarity network for regularizing a baseline regularization framework using a network localized LASSO approach to ensure model sparsity and to regularize drug response parameters of similar patients, wherein the statistical model learns parameters that predict drug responses for each patient and for each time point of the real-world evidence, wherein the statistical model optimizes the following objective function:
• arg
• ⁢
• ⁢
• min
• W
• ⁢
• 
• Dy
• -
• DZw
• 
• 2
• 2
• +
• λ
• 1
• ⁢
• ∑
• i
• =
• 1
• N
• ⁢
• 
• w
• i
• 
• 1
• 2
• +
• λ
• 2
• ⁢
• ∑
• i
• >
• i
• ′
• N
• ⁢
• ∑
• i
• ′
• =
• 1
• N
• -
• 1
• ⁢
• r
• ii
• ′
• ⁢
• 
• w
• i
• -
• w
• i
• ′
• 
• 2
• wherein D is a difference matrix, y is a column matrix of laboratory test measurements, Z is a block diagonal matrix of drug exposures, w is a drug response vector, wherein ∥1 is a hyper-parameter that controls an exclusive lasso penalty and λ2 is a hyper-parameter that controls a network lasso penalty, wherein wi is a vector that denotes a drug response for each individual patient i, and wherein rii′ represents a patient similarity between the ith and i′th patients based on the similarity network;
• group, by a patient clustering component executing within the drug response estimation engine, patients based on demographics and comorbidities to form a plurality of patient groups, wherein the patient clustering component groups the patients using the learned parameters that predict drug responses such that each patient group within the plurality of patient groups represents patients that have similar drug responses and background information;
• estimate, by the drug response estimation engine, drug responses for a given patient based on the patient similarity network, the statistical model, and the plurality of patient groups;
• and
• output, by the drug response estimation engine, the drug responses for the given patient.