Systems and Methods for Predicting Treatment-Regimen-Related Outcomes

Abstract

Systems and methods are provided for predicting treatment-regimen-related outcomes (e.g., risks of regimen-related toxicities). A predictive model is determined for predicting treatment-regimen-related outcomes and applied to a plurality of datasets. An ensemble algorithm is applied on result data generated from the application of the predictive model. Treatment-regimen-related outcomes are predicted using the predictive model. A combination of machine learning prediction and patient preference assessment is provided for enabling informed consent and precise treatment decisions.

It is claimed: 
     
         1 . A processor-implemented method for predicting regimen-related outcomes, the method comprising:
 generating, using the one or more data processors, one or more training datasets and one or more testing datasets based at least in part on clinical data or gene feature data of a plurality of patients;   determining, using one or more data processors, one or more initial predictive models using one or more machine learning algorithms based at least in part on the one or more training datasets;   applying, using the one or more data processors, the one or more initial predictive models on the one or more training datasets to generate result data;   performing, using the one or more data processors, an ensemble algorithm on the result data to generate ensemble data;   determining, using the one or more data processors, one or more final predictive models based at least in part on the ensemble data;   evaluating, using the one or more data processors, performance of the one or more final predictive models based at least in part on the one or more test datasets; and   predicting, using the one or more data processors, regimen-related outcomes using the one or more final predictive models.   
     
     
         2 . The method of  claim 1 , wherein the gene feature data includes data related to one or more single-nucleotide polymorphisms (SNPs). 
     
     
         3 . The method of  claim 2 , wherein generating one or more training datasets and one or more testing datasets based at least in part on clinical data or gene feature data of a plurality of patients includes:
 determining the gene feature data based at least in part on one or more predetermined SNPs.   
     
     
         4 . The method of  claim 2 , wherein generating one or more training datasets and one or more testing datasets based at least in part on clinical data or gene feature data of a plurality of patients includes:
 determining a plurality of SNPs;   filtering the plurality of SNPs to determine one or more filtered SNPs; and   determining the gene feature data based at least in part on the one or more filtered SNPs.   
     
     
         5 . The method of  claim 4 , wherein filtering the plurality of SNPs to determine the one or more filtered SNPs includes:
 removing a number of SNPs based on missing data from the plurality of SNPs.   
     
     
         6 . The method of  claim 4 , wherein filtering the plurality of SNPs to determine the one or more filtered SNPs includes:
 removing one or more SNPs that are associated from the plurality of SNPs.   
     
     
         7 . The method of  claim 4 , wherein filtering the plurality of SNPs to determine the one or more filtered SNPs includes:
 performing recursive partitioning for filtering the plurality of SNPs.   
     
     
         8 . The method of  claim 7 , wherein performing recursive partitioning for filtering the plurality of SNPs includes:
 dividing a gene feature dataset related to the plurality of SNPs into a plurality of sub-datasets;   selecting one or more first sub-datasets from the plurality of sub-datasets;   developing a first recursive partitioning model based at least in part on the one or more first sub-datasets; and   determining one or more first predictive SNPs based at least in part on the first recursive partitioning model, wherein the one or more first predictive SNPs are included into the one or more filtered SNPs.   
     
     
         9 . The method of  claim 8 , wherein performing recursive partitioning for filtering the plurality of SNPs further includes:
 selecting one or more second sub-datasets from the plurality of sub-datasets;   developing a second recursive partitioning model based at least in part on the one or more second sub-datasets; and   determining one or more second predictive SNPs based at least in part on the second recursive partitioning model, wherein the one or more second predictive SNPs are included into the one or more filtered SNPs.   
     
     
         10 . The method of  claim 1 , wherein the one or more machine learning algorithms correspond to one or more of the following: a penalized logistic regression algorithm, a random forests algorithm, and a C5.0 algorithm. 
     
     
         11 . The method of  claim 1 , wherein generating one or more training datasets and one or more testing datasets based at least in part on clinical data or gene feature data of a plurality of patients includes:
 generating one or more clinical predictor datasets based at least in part on the clinical data; and   generating one or more gene feature datasets based at least in part on the gene feature data.   
     
     
         12 . The method of  claim 11 , wherein applying the one or more initial predictive models on the one or more training datasets to generate result data includes:
 applying the initial predictive models on the one or more clinical predictor datasets to generate clinical result data; and   applying the initial predictive models on the one or more gene feature datasets to generate gene feature result data.   
     
     
         13 . The method of  claim 1 , wherein the ensemble algorithm corresponds to an average calculation or a logistic regression algorithm. 
     
     
         14 . The method of  claim 1 , wherein generating one or more training datasets and one or more testing datasets based at least in part on clinical data or gene feature data of a plurality of patients includes:
 generating one or more clinical predictor datasets by generating binary predictor data based at least in part on the clinical data.   
     
     
         15 . The method of  claim 1 , further comprising:
 performing 10-fold cross-validation on the one or more training datasets to determine one or more tuning parameters for the initial predictive models.   
     
     
         16 . The method of  claim 1 , wherein the clinical data includes diagnosis data, cancer-stage data, regimen related data, and neuropathy related data. 
     
     
         17 . A processor-implemented method for determining a treatment regimen for a patient, the method comprising:
 generating, using the one or more data processors, one or more training datasets and one or more testing datasets based at least in part on sample clinical data or sample gene feature data;   determining, using one or more data processors, one or more initial predictive models using one or more machine learning algorithms based at least in part on the one or more training datasets;   applying, using the one or more data processors, the one or more initial predictive models on the one or more training datasets to generate result data;   performing, using the one or more data processors, an ensemble algorithm on the result data to generate ensemble data;   determining, using the one or more data processors, one or more final predictive models based at least in part on the ensemble data;   evaluating, using the one or more data processors, performance of the one or more final predictive models based at least in part on the one or more test datasets;   predicting, using the one or more data processors, regimen-related outcomes using the one or more final predictive models based at least in part on clinical data or gene feature data of a patient;   assessing patient preferences of the patient to generate patient preference data; and   determining a treatment regimen for the patient based on the regimen-related outcomes and the patient preference data.   
     
     
         18 . A processor-implemented method for building a predictive model for predicting regimen-related outcomes, the method comprising:
 dividing, using one or more data processors, a training dataset into a plurality of sub-datasets;   selecting, using the one or more data processors, one or more first training sub-datasets from the plurality of sub-datasets;   determining, using the one or more data processors, a first predictive model using one or more machine learning algorithms based at least in part on the one or more first training sub-datasets;   evaluating, using the one or more data processors, the performance of the first predictive model using the plurality of sub-datasets excluding the one or more first training sub-datasets; and   determining, using the one or more data processors, a final predictive model based at least in part on the performance evaluation of the first predictive model.   
     
     
         19 . The method of  claim 18 , further comprising:
 selecting one or more second training sub-datasets from the plurality of sub-datasets;   determining a second predictive model using the one or more machine learning algorithms based at least in part on the one or more second training sub-datasets; and   evaluating the performance of the second predictive model using the plurality of sub-datasets excluding the one or more second training sub-datasets.   
     
     
         20 . The method of  claim 19 , wherein the final predictive model is determined based at least in part on the comparison of the performance of the first predictive model and the performance of the second predictive model. 
     
     
         21 . The method of  claim 19 , further comprising:
 selecting one or more third training sub-datasets from the plurality of sub-datasets;   determining a third predictive model using the one or more machine learning algorithms based at least in part on the one or more third training sub-datasets; and   evaluating the performance of the third predictive model using the plurality of sub-datasets excluding the one or more third training sub-datasets.   
     
     
         22 . The method of  claim 21 , wherein the final predictive model is determined based at least in part on the comparison of the performance of the first predictive model, the performance of the second predictive model, and the performance of the third predictive model. 
     
     
         23 . The method of  claim 18 , further comprising:
 performing cross-validation on the plurality of sub-datasets to determine one or more tuning parameters of the final predictive model.   
     
     
         24 . The method of  claim 18 , wherein the one or more machine learning algorithms correspond to one or more of the following: a penalized logistic regression algorithm, a random forests algorithm, and a C5.0 algorithm. 
     
     
         25 . A processor-implemented system for predicting regimen-related outcomes, the system comprising:
 one or more processors configured to:
 generate one or more training datasets and one or more testing datasets based at least in part on clinical data or gene feature data of a plurality of patients; 
 determine one or more initial predictive models using one or more machine learning algorithms based at least in part on the one or more training datasets; 
 apply the one or more initial predictive models on the one or more training datasets to generate result data; 
 perform an ensemble algorithm on the result data to generate ensemble data; 
 determine one or more final predictive models based at least in part on the ensemble data; 
 evaluate performance of the one or more final predictive models based at least in part on the one or more test datasets; and 
 predict regimen-related outcomes using the one or more final predictive models; 
   one or more non-transitory machine-readable storage media for storing a computer database having a database schema that includes and interrelates clinical data fields, gene feature data fields, result data fields, ensemble data fields and predictive model data fields,   the clinical data fields storing the clinical data,   the gene feature data fields storing the gene feature data,   the result data fields storing the result data,   the ensemble data fields storing the ensemble data, and   the predictive model data fields storing parameter data of the initial predictive models and the final predictive models.   
     
     
         26 . A processor-implemented system for building a predictive model for predicting regimen-related outcomes, the system comprising:
 one or more processors configured to:
 divide a training dataset into a plurality of sub-datasets; 
 select one or more first training sub-datasets from the plurality of sub-datasets; 
 determine a first predictive model using one or more machine learning algorithms based at least in part on the one or more first training sub-datasets; 
 evaluate the performance of the first predictive model using the plurality of sub-datasets excluding the one or more first training sub-datasets; and 
 determine a final predictive model based at least in part on the performance evaluation of the first predictive model; 
   one or more non-transitory machine-readable storage media for storing a computer database having a database schema that includes and interrelates training data fields, first predictive model data fields, and final predictive model data fields,   the training data fields storing the training dataset,   the first predictive model data fields storing parameter data of the first predictive model, and   the final predictive model data fields storing parameter data of the final predictive model.   
     
     
         27 . A non-transitory computer-readable medium encoded with instructions for commanding one or more processors to execute operations of a method for predicting regimen-related outcomes, the method comprising:
 generating one or more training datasets and one or more testing datasets based at least in part on clinical data or gene feature data of a plurality of patients;   determining one or more initial predictive models using one or more machine learning algorithms based at least in part on the one or more training datasets;   applying the one or more initial predictive models on the one or more training datasets to generate result data;   performing an ensemble algorithm on the result data to generate ensemble data;   determining one or more final predictive models based at least in part on the ensemble data;   evaluating performance of the one or more final predictive models based at least in part on the one or more test datasets; and   predicting regimen-related outcomes using the one or more final predictive models.   
     
     
         28 . A non-transitory computer-readable medium encoded with instructions for commanding one or more processors to execute operations of a method for building a predictive model for predicting regimen-related outcomes, the method comprising:
 dividing a training dataset into a plurality of sub-datasets;   selecting one or more first training sub-datasets from the plurality of sub-datasets;   determining a first predictive model using one or more machine learning algorithms based at least in part on the one or more first training sub-datasets;   evaluating the performance of the first predictive model using the plurality of sub-datasets excluding the one or more first training sub-datasets; and   determining a final predictive model based at least in part on the performance evaluation of the first predictive model.   
     
     
         29 . A non-transitory computer-readable medium for storing data for access by an application program being executed on a data processing system, comprising:
 a data structure stored in said memory, said data structure including information, resident in a database used by said application program and including:
 one or more clinical data objects stored in said memory, the clinical data objects containing clinical data of a plurality of patients from said database; 
 one or more gene feature data objects stored in said memory, the gene feature data objects containing gene feature data of the plurality of patients from said database; 
 one or more training data objects stored in said memory, the training data objects containing one or more training datasets generated based at least in part on the clinical data or the gene feature data; 
 one or more initial predictive model data objects stored in said memory, the initial predictive model data objects containing parameters of one or more initial predictive models determined using one or more machine learning algorithms based at least in part on the one or more training datasets; 
 one or more result data objects stored in said memory, the result data objects containing result data generated by applying the initial predictive models on the one or more training datasets; 
 one or more ensemble data objects stored in said memory, the ensemble data objects containing ensemble data generated by performing an ensemble algorithm on the result data; and 
 one or more final predictive model data objects stored in said memory, the final predictive model data objects containing parameters of one or more final predictive models determined based at least in part on the ensemble data; 
   wherein the final predictive model data objects are used by said application program for predicting regimen-related outcomes.   
     
     
         30 . A non-transitory computer-readable medium for storing data for access by an application program being executed on a data processing system, comprising:
 a data structure stored in said memory, said data structure including information, resident in a database used by said application program and including:
 one or more training data objects stored in said memory, the training data objects containing a training dataset from said database, the training dataset including a plurality of sub-datasets; 
 one or more first predictive model data objects stored in said memory, the first predictive model data objects containing parameter data of a first predictive model determined using one or more machine learning algorithms based at least in part on one or more first training sub-datasets from the plurality of sub-datasets; 
 one or more final predictive model data objects stored in said memory, the final predictive model data objects containing parameter data of a final predictive model determined based at least in part on performance evaluation of the first predictive model; 
   wherein the final predictive model data objects are used by said application program for predicting regimen-related outcomes.