Neural network drug dosage estimation

• US 6,658,396 B1
• Filed: 11/29/1999
• Issued: 12/02/2003
• Est. Priority Date: 11/29/1999
• Status: Expired due to Fees

• Current Assignee: Prediction Sciences LLC
• Original Assignee: Prediction Sciences LLC
• Inventors: Diamond, Cornelius, Arouh, Scott, Tang, Sharon S.
• Primary Examiner(s): Starks, Jr., Willbert L.
• Application Number:
  US09/451,249
• Time in Patent Office: 1,464 Days
• Field of Search: 706/17, 600/300, 600/544
• US Class Current: 706/17
• CPC Class Codes: C12Q 2600/156 Polymorphic or mutational markers
• CPC Class Codes: G01N 33/5091 for testing the pathological state of an organism
• CPC Class Codes: G16H 10/60 for patient-specific data, e.g. for electronic patient records
• CPC Class Codes: G16H 20/10 relating to drugs or medications, e.g. for ensuring correct administration to patients
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: Y02A 90/10 Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Abstract

Neural networks are constructed (programmed), trained on historical data, and used to predict any of (1) optimal patient dosage of a single drug, (2) optimal patient dosage of one drug in respect of the patient'"'"'s concurrent usage of another drug, (3a) optimal patient drug dosage in respect of diverse patient characteristics, (3b) sensitivity of recommended patient drug dosage to the patient characteristics, (4a) expected outcome versus patient drug dosage, (4b) sensitivity of the expected outcome to variant drug dosage(s), (5) expected outcome(s) from drug dosage(s) other than the projected optimal dosage. Both human and economic costs of both optimal and sub-optimal drug therapies may be extrapolated from the exercise of various optimized and trained neural networks. Heretofore little recognized sensitivities—such as, for example, patient race in the administration of psychotropic drugs—are made manifest. Individual prescribing physicians employing deviant patterns of drug therapy may be recognized. Although not intended to prescribe drugs, nor even to set prescription drug dosage, the neural networks are very sophisticated and authoritative “helps” to physicians, and to physician reviewers, in answering “what if” questions.

Claims

1. A method of predicting an optimal dosage of a particular drug for a particular patient comprising:

receiving a database of the characteristics and optimal drug dosage of the particular drug for a multiplicity of patients;

receiving data on the characteristics of a particular patient corresponding at least in part to those characteristics that are within the database; and

interpolating in a computer neural network the optimal drug dosage for the particular patient in consideration of the characteristics and optimal drug dosage database.

2. The method according to claim 1 wherein the interpolating is in a computer neural network optimized to relate optimal drug dosage to the patient characteristics of the database.

3. A computerized method of predicting an optimal dosage of a particular drug for a Particular Patient in consideration of previously determined optimal dosages of the drug for members of a patient population, the computerized optimal drug dosage prediction method comprising:

programming a plurality of neural networks each having an architecture of one or more slabs collectively relating input data to output data,

wherein said input data includes at least a selected three (3) of a person'"'"'s traits drawn from at least two (2) of the following three (3) groups consisting of

Group 1 overt indications of

(1a) age,

(1b) gender,

(1c) race,

(1d) ethnicity,

(1e) diet type,

(1f) height,

(1g) weight, and

(1h) body surface area,

Group 2 medical diagnostic indications of

(2a) blood pressure,

(2b) use of a drug other than the particular drug at the same time as use of the particular drug,

(2c) fitness,

(2d) peptide levels, and

(2e) genetic predisposition to a particular disease, and

Group 3 pharmacological indications of

(3a) pharmacokinetic parameters,

(3b) pharmacodynamic parameters, and

wherein said output data includes clinically-determined optimal drug dosage for the same person;

training each of the plurality of programmed neural networks with a training data set drawn from a multiplicity of historical medical records of a multiplicity of persons historically administered the particular drug, the records relating the selected input data to the output data;

selecting one of plurality of neural networks that performs best on the training data set to be a selected trained neural network;

using the selected trained neural network to predict an optimal dosage of the particular drug for the Particular Patient, the using transpiring by inputting the selected input data, which input data is at least a selected three (3) of The Particular Patient'"'"'s traits drawn from at least two (2) of the three (3) groups consisting of Group 1 overt indications and Group 2 medical diagnostic indications and Group 3 pharmacological indications, to ascertain as (2) output of the trained neural network the output data, which output data is the predicted optimal dosage for the Particular Patient.

4. The computerized drug dosage prediction method according to claim 3

wherein the architectures of the plurality of neural networks are established by a same human who does the programming of the neural network, one human thus acting as both neural network architect and neural network programmer.

5. The computerized drug dosage prediction method according to claim 3 wherein the selecting comprises:

choosing one of the plurality of neural networks by a genetic algorithm, the genetic algorithm acting to select the one of the plurality of neural networks that performs best on the training data set; and

wherein the training of each of the plurality of neural architectures permits, along with the choosing, not only a selection a single one of the plurality of neural networks that performs best on the training data set, but the training of this selected one of the plurality of neural networks to optimally relate the input data to the output data.

6. The computerized drug dosage prediction method according to claim 3 extended and expanded to account for interaction between at least two, a first and a second, drugs taken concurrently by the Particular Patient, the extended method comprising:

performing the programming, the training, the selecting and the using in respect of a first drug to predict in a first selected trained neural network an optimal dosage of the first drug for the Particular Patient;

performing the programming, the training, the selecting and the using in respect of a second drug to predict in a second selected trained neural network an optimal dosage of the second drug for the Particular Patient;

programming a plurality of drug interaction neural networks each having an architecture of one or more slabs collectively relating (3) second data inputs drawn from the group consisting of

(3a) the first drug optimal dosage;

(3b) the second drug optimal dosage; and

(3c) the first data inputs for the first trained neural network, and

(3d) the first data inputs for the second trained neural network, and having as an (4) output

(4a) the clinically-determined optimal drug dosage for the first drug, and

(4b) the clinically-determined optimal drug dosage for the second drug; and

training each of the plurality of drug interaction neural networks with a training data set drawn from a multiplicity of historical medical records of a multiplicity of persons historically administered the two particular drugs, these records relating the input data to the output data, to produce a plurality of trained drug interaction neural networks;

selecting a one of plurality of trained drug interaction neural networks that performs best on the training data set to be a selected trained drug interaction neural network;

using the selected trained drug interaction neural network to predict the optimal dosage of both the first and the second drug for the Particular Patient;

wherein each drug'"'"'s optimal dosage is in respect of the other drug'"'"'s optimal dosage.

7. The computerized drug dosage prediction method according to claim 6 wherein the selecting comprises:

choosing a one of the plurality of drug interaction neural networks by a genetic algorithm, the genetic algorithm acting to select the one of the plurality of drug interaction neural networks that performs best on the training data set;

wherein the training of the drug interaction neural network is of each of the plurality of neural network architectures to produce a single optimal drug interaction neural network architecture as well as to train this single optimal drug interaction neural network architecture, when identified, to relate the optimal dosage of a first drug in respect of a patient'"'"'s simultaneous usage of a second drug.

8. A computerized method of predicting an optimal dosage of a particular drug for a Particular Patient, the method comprising:

programming a neural net that relates drug dosage to drug efficacy and to drug side effects;

training the neural network in consideration of both (i) efficacy and (ii) side effect measures from usages of the drug at determined dosages on members of a population to produce a trained neural network;

using the trained neural network to predict a drug dosage for an individual patient that (i) delivers adequate measures of efficacy while (ii) minimizing adverse side effects.

9. The computerized method of predicting an optimal dosage of a particular drug according to claim 8 wherein the using comprises:

repeatedly exercising the neural network in respect of various drug dosages to assess the (i) measures of efficacy and (ii) adverse side effects in respect of each dosage, ultimately selecting the drug dosage for the Particular Patient that (i) delivers the adequate measures of efficacy while (ii) minimizing the adverse side effects.

10. A method of creating a neural network both (i) optimized and (ii) trained to predict the optimal dosage of a particular drug for a particular patient, the optimal trained neural network creation method comprising:

acquiring neural network training data as

raw data from patient medical records, the raw data being in the form of recorded patient responses to recorded patient drug dosages in respect of patient characteristics,

the patient medical records being categorized for relevance, so as to produce

relevant categorized patient drug response data in respect of patient characteristics;

designing and programming a plurality of neural networks each of which may suitably operate upon the neural network training data by

programming a framework for each of the plurality of neural networks;

programming a genetic algorithm for framework for each of the plurality of neural networks, and

interfacing the programmed framework with the programmed genetic algorithm;

then

training each of the plurality of neural networks with the neural network training data until, by operation of the genetic algorithm, a single optimal, trained neural network is selected;

wherein the selected single optimal, trained neural network is then both (i) optimized, meaning selected as the best of many, and (ii) trained, meaning proven on data of relevance, to predict the optimal dosage of a particular drug for a particular patient.

11. The method according to claim 10 wherein the training is by the back propagation method.

12. The method according to claim 10 wherein the training is by the Levenberg-Marquardt optimization method.

13. An optimal, trained, neural network created by the method of claim 10.

14. A computerized method of predicting an optimal dosage of a particular drug for a Particular Patient, the method comprising:

performing in a recurring, iterative, looped process each of the steps of

(1) programming a neural network that relates drug dosage to drug efficacy and to drug side effects, and

(2) training the neural network in consideration of a historical data set of both efficacy and side effect measures from usages of the drug at determined dosages on members of a population,

wherein the (1) programming transpires in consideration of the results of the (ii) training,

so long as, and until, an optimally programmed and trained neural network adequately accurately predicts both drug efficacy and drug side effects relative to drug dosage;

then

exercising the optimally programmed and trained neural network on a range of drug dosages in order to examine the predicted sensitivity of both drug efficacy and drug side effects relative to the range of drug dosages;

wherein a physician prescribing the drug can, by observation of the results of the exercising, make a better informed decision as to what patient dosages of the drug are likely to deliver both tolerable efficacy and tolerable side effects.

15. A computerized method of predicting the effect of one or both of dietary items consumed or exercises performed on measured physiological characteristics of a Particular Patient, the method comprising:

programming a neural net that relates any of selected dietary items consumed and exercises performed on measured physiological characteristics of patients as might be expected to be affected by the selected dietary items and exercises;

training the neural network in consideration of historical data on the impact of the dietary items consumed and exercises performed on the physiological characteristics evidenced by members of a population, therein to produce a trained neural network;

using the trained neural network to predict the change in physiological characteristics to be anticipated for an individual Particular Patient who consumes any of the selected dietary items and/or performs any of the selected exercises.

16. The computerized method of predicting the effect of one or both of dietary items consumed or exercises performed on measured physiological characteristics of a Particular Patient according to claim 15 wherein the using comprises:

repeatedly exercising the neural network in respect of various selected dietary items consumed and/or selected exercises performed to assess the (i) measures of efficacy and (ii) adverse side effects in respect of each selected dietary item consumed and/or selected exercise performed, ultimately selecting dietary items and/or exercises the Particular Patient that (i) delivers adequate measures of efficacy while (ii) maximizing acceptability and suitability to the Particular Patient'"'"'s preferences and conduct.