System and method for infectious disease notification

11798,691

17/016563

September 10, 2020

A system is for infectious disease notification. The system includes at least one processor, configured to use a machine learning monitoring algorithm, trained on a large number of EMR datasets of patients, to calculate a probability for an infectious disease from a provided EMR dataset and compare the probability of the provided EMR dataset calculated with a known value. In training of the monitoring algorithm, the value represents whether there was an onset of an infectious disease or not and the monitoring algorithm is designed to adjust parameters of the monitoring algorithm. And in evaluating a notification, the value is a threshold value and the system is designed to output a notification upon the probability being greater than the threshold value.

Siemens Healthcare Diagnostics Inc. (Tarrytown, NY, US)

Siemens Healthcare Diagnostics Inc. (Tarrytown, NY, US)

1. A system for infectious disease notification comprising: at least one processor, configured to use a machine learning monitoring algorithm, trained on a large number of electronic medical record (EMR) datasets of previous cases of patients, to apply data imputation methods on a part of a provided EMR dataset where there are missing values, to bring the data into a proper representation, calculate a probability value for an infectious disease from the provided EMR dataset, compare the probability value of the provided EMR dataset calculated with a known value, output a notification upon the probability value lying above the known value, train the monitoring algorithm, wherein in training of the monitoring algorithm, the known value represents whether there was an onset of an infectious disease or not and the monitoring algorithm iteratively adjusts parameters of the monitoring algorithm, and evaluate a notification based on the adjusted parameters, wherein in evaluating the notification, the known value is a threshold value and the system is designed to output the notification upon the probability value being greater than the threshold value, wherein the monitoring algorithm is trained on the previous cases of patients and a current epidemiology profile of a hospital, the epidemiology profile of the hospital comprising data pertaining to a group of medication, including antibiotics, used within the hospital, and wherein the monitoring algorithm is a regression model with parameters θ that are, in a course of training with a number of n training EMR datasets x of patients where a ground truth z is known whether the patients developed an infection in a given time period or not, while calculating the probability (PV) y=f(x) of an infection in a given time period, wherein the parameters θ of a regression model to minimize error between the probability calculated y and the ground truth outcome z are calculated with θ=argmin θ′ Σ(dist(f (x ;θ′), z)), wherein a sum Σ runs over the n EMR datasets and wherein the function dist is an absolute or squared difference between the two values f(x) and z.

2. The system of claim 1 , wherein the monitoring algorithm is a regression model.

3. The system of claim 1 , wherein the monitoring algorithm is designed to process at least one of EMR data of a group vital signs, lab results, point of care test results, patient care related procedures, co-morbidities, other diseases and clinical care data.

4. The system of claim 1 , further comprising: a data connection to more than one medical institution, the system being designed to receive EMR datasets from the more than one medical institution, wherein the at least one processor trains the monitoring algorithm using the received EMR datasets.

5. The system of claim 1 , further comprising: an input interface for a data network or an input device for receiving marking data.

6. A method for infectious disease notification with a system, using a machine learning monitoring algorithm, trained on a large number of electronic medical records (EMR) datasets of previous cases of patients, from an EMR dataset of a patient provided to the system, the method comprising: applying data imputation methods on a part of an EMR dataset where there are missing values, to bring the data into a proper representation; providing a machine learning monitoring algorithm; calculating a probability for an infectious disease from the EMR dataset with the machine learning monitoring algorithm of the system; comparing the probability of the EMR dataset calculated, with a threshold value; training the machine learning monitoring algorithm based on the previous cases of patients and a current epidemiology profile of a hospital, the epidemiology profile of the hospital comprising data pertaining to a group of medication, including antibiotics, used within the hospital, wherein in training of the monitoring algorithm, the threshold value represents whether there was an onset of an infectious disease or not and the monitoring algorithm iteratively adjusts parameters of the monitoring algorithm; evaluating a notification based on the adjusted parameters; and outputting a notification upon the probability lying above the threshold value, and wherein the monitoring algorithm is a regression model with parameters θ that are, in a course of training with a number of n training EMR datasets x of patients where a ground truth z is known whether the patients developed an infection in a given time period or not, while calculating the probability (PV) y=f(x) of an infection in a given time period, wherein the parameters θ of a regression model to minimize error between the probability calculated y and the ground truth outcome z are calculated with θ=argmin θ′ Σ(dist(f (x ;θ′), z)), wherein a sum Σ runs over the n EMR datasets and wherein the function dist is an absolute or squared difference between the two values f(x) and z.

7. The method of claim 6 , wherein the monitoring algorithm of the system is further trained when a new EMS dataset of a patient is added to the system, the further training comprising: calculating a probability for an infectious disease from the EMR dataset with the system; comparing the probability of the EMR dataset calculated with a value representing whether there was an onset of an infectious disease or not; and adjusting parameters of the monitoring algorithm based upon the comparing.

8. The method of claim 6 , wherein the monitoring algorithm of the system is further trained when a feedback of a specialist is added to the system, the further training comprising: calculating a probability for an infectious disease from the EMR dataset connected with the feedback with the system; comparing the probability of the EMR dataset calculated with a value representing whether there was an onset of an infectious disease or not based on the feedback; and adjusting parameters of the monitoring algorithm based upon the comparing.

9. The method of claim 6 , wherein an EMR dataset or identification data of a patient is marked.

10. The method of claim 6 , further comprising: a preprocessing step, wherein data of the EMR dataset is normalized to always be represented in a same way, required for the regression model input.

11. A network service system, comprising a system for infectious disease notification including: at least one processor, configured to use a machine learning monitoring algorithm, trained on a large number of electronic medical record (EMR) datasets of previous cases of patients, to apply data imputation methods on a part of a provided EMR dataset where there are missing values, to bring the data into a proper representation, calculate a probability value for an infectious disease from the provided EMR dataset, compare the probability value of the provided EMR dataset calculated with a known value, output a notification upon the probability value lying above the known value, and train the monitoring algorithm, wherein in training of the monitoring algorithm, the known value represents whether there was an onset of an infectious disease or not and the monitoring algorithm iteratively adjusts parameters of the monitoring algorithm, and evaluate a notification based on the adjusted parameters, wherein in evaluating the notification, the known value is a threshold value and the system is designed to output the notification upon the probability value being greater than the threshold value, wherein the monitoring algorithm is trained on the previous cases of patients and a current epidemiology profile of a hospital, the epidemiology profile of the hospital comprising data pertaining to a group of medication, including antibiotics, used within the hospital, and wherein the monitoring algorithm is a regression model with parameters θ that are, in a course of training with a number of n training EMR datasets x of patients where a ground truth z is known whether the patients developed an infection in a given time period or not, while calculating the probability (PV) y=f(x) of an infection in a given time period, wherein the parameters θ of a regression model to minimize error between the probability calculated y and the ground truth outcome z are calculated with θ=argmin θ′ Σ(dist(f (x ;θ′), z)), wherein a sum Σ runs over the n EMR datasets and wherein the function dist is an absolute or squared difference between the two values f(x) and z.

12. A non-transitory computer program product storing a computer program, directly loadable into a computing device, including program elements for performing the method of claim 6 when the computer program is executed by the computing device.

13. A non-transitory computer-readable medium storing program elements, readable and executable by a computer unit, to perform the method of claim 6 when the program elements are executed by the computer unit.

14. The system of claim 2 , wherein the monitoring algorithm is a regression model that is a continuously learning regression model.

15. The system of claim 14 , wherein the continuously learning regression model is based on regularized regression models, random forest, support vector machines or deep neural networks.

16. The system of claim 5 , wherein the system is designed to mark the EMR dataset or identification data of the patient with flags indicating that there is the risk of an infection, that the patient needs further monitoring or that the patient is wrongly flagged.

17. The system of claim 1 , wherein the monitoring algorithm is designed to be trained on data pertaining to the group of medication used, outcomes, biochemistry and vital signs.

18. The method of claim 7 , wherein the monitoring algorithm of the system is further trained when a feedback of a specialist is added to the system, the further training comprising: calculating a probability for an infectious disease from the EMR dataset connected with the feedback with the system; comparing the probability of the EMR dataset calculated with a value representing whether there was an onset of an infectious disease or not based on the feedback; and adjusting the parameters of the monitoring algorithm based upon the comparing.

19. The method of claim 9 , wherein an EMR dataset or identification data of a patient is marked with flags indicating that the patient needs further monitoring or that a patient is wrongly flagged.

8090590 January 2012 Fotsch
8359223 January 2013 Chi
9955869 May 2018 Meltzer
10249389 April 2019 Athey
10327697 June 2019 Stein
20050182303 August 2005 Yancey, Jr.
20180107926 April 2018 Choi
20180315494 November 2018 Kolde
20190214138 July 2019 Aoyagi
20200243163 July 2020 Van Aggelen
WO-2019025901 February 2019
WO 2019025901 February 2019
WO 2019165004 August 2019

Extended European Search Report for European Application No. 19199183.5 dated Apr. 1, 2020. cited by applicant

edspgr.11798691