Method and apparatus for prediction of complications after surgery

Patent metadata

Field	Value
Publication number	US-12014832-B2
Application number	US-201816616534-A
Country	US
Kind code	B2
Filing date	Jun 1, 2018
Priority date	Jun 2, 2017
Publication date	Jun 18, 2024
Grant date	Jun 18, 2024

Abstract

Methods and systems disclosed herein utilize an automated analytics framework to implement a perioperative complication risk algorithm that uses existing clinical data in electronic health records to forecast patient-level probabilistic risk scores for eight major postoperative complications. An example method includes accessing health record data for a patient, normalizing the accessed health record data to generate a health record data set for the patient, transforming one or more features from the health record data set, selecting one or more transformed features from the health record data set, calculating risk probabilities for one or more complication risk categories based on the health record data set and the selected one or more features, calculating mortality risk probabilities for one or more mortality risk categories based on the calculated risk probabilities, and generating a personalized risk panel based on the calculated risk probabilities and the calculated mortality risk probabilities.

First claim

That which is claimed:

1. A surgery risk analytics system for surgical risk evaluation based at least in part on a surgery risk analytics platform service, the surgery risk analytics system comprising a processor and a memory having computer coded instructions therein when executed by the processor, cause the surgery risk analytics system to:
receive, using a data transformer module, health record data associated with monitoring of a patient for surgery risks via a client device, the health record data comprising a plurality of variables associated with a plurality of features, the plurality of features comprising a plurality of perioperative predictor features;
normalize, using the data transformer module, the received health record data by replacing missing values of one or more of the plurality of variables with replacement values based at least in part on variable type and by preprocessing the one or more of the plurality of variables based at least in part on respective one or more of a plurality of preprocessing types associated with the one or more of the plurality of variables;
transform, using the data transformer module, one or more of the plurality of features by optimizing the plurality of variables based at least in part on one or more conditional properties associated with a plurality of outcomes, thereby conforming the plurality of variables with inputs of a plurality of predictive models of a data analytics module based at least in part on the optimizing, wherein the plurality of variables are optimized by substituting the plurality of variables with probability ratios associated with ordered variables, wherein each of the plurality of predictive models is associated with a given complication risk and is configured to (i) generate one or more complication risk probabilities based at least in part on the one or more transformed features and (ii) execute independently from other ones of the plurality of predictive models;
select, using the data transformer module, the one or more transformed features based at least in part on variance factors of the one or more transformed features, the variance factors comprising a likelihood of excessive collinearity;
generate, using the plurality of predictive models, a plurality of complication risk probabilities for a plurality of respective complication risks associated with the plurality of predictive models based at least in part on the health record data and the selected one or more features;
generate, using one or more trained machine learning classifier models of the data analytics module, one or more mortality risk probabilities at given future times after a surgery event, wherein (i) the one or more trained machine learning classifier models comprises a plurality of random forest classifiers that are trained to generate a plurality of predictions associated with respective ones of the plurality of outcomes based at least in part on training data comprising the plurality of complication risk probabilities and (ii) the one or more mortality risk probabilities comprise an aggregation of the plurality of predictions;
generate, using an output generator, a user interface configured to render a personalized risk panel for the patient based at least in part on the plurality of complication risk probabilities and the one or more mortality risk probabilities; and
transmit the personalized risk panel from the output generator to the client device for graphical output on a client application executing on the client device, wherein (i) the personalized risk panel comprises results of the monitoring and a determination of level of care based at least in part on the plurality of complication risk probabilities and the one or more mortality risk probabilities and (ii) the personalized risk panel is associated with a surgery risk analytics platform that is configured for medical decision making associated with modifying surgical planning or staffing before or during surgery on the patient based at least in part on the personalized risk panel.

2. The surgery risk analytics system of claim 1 , wherein the computer coded instructions, when executed by the processor, further cause the data transformer module to:
remove one or more outliers from the health record data;
replace the one or more missing variables with replacement data; and
normalize the health record data to generate a health record data for the patient.

3. The surgery risk analytics system of claim 1 , wherein the plurality of features comprise one or more of perioperative demographic, socio-economic, administrative, clinical, pharmacy, or laboratory variables.

4. The surgery risk analytics system of claim 1 , wherein the computer coded instructions, when executed by the processor, further cause the data analytics module to:
generate the plurality of complication risk probabilities using a generalized additive model (GAM) with logistic link function; and
define one or more complication risk categories based at least in part on the GAM.

5. The surgery risk analytics system of claim 1 , wherein the computer coded instructions, when executed by the processor, further cause the data analytics module to:
apply a random forests classifier model of the one or more trained machine learning classifier models over the plurality of complication risk probabilities to identify probability of death at one, three, six, twelve, and twenty four months after surgery.

6. The surgery risk analytics system of claim 1 , wherein the personalized risk panel comprises a list of features contributing to the plurality of complication risk probabilities.

7. The surgery risk analytics system of claim 1 , wherein transforming, using the data transformer module, the plurality of features comprises remodeling raw features based at least in part on a plurality of predefined dictionaries for use in the plurality of predictive models.

8. A method for evaluating surgical risk and providing a surgery risk analytics platform service, the method comprising:
receiving, by a data transformer module, health record data associated with monitoring of a patient for surgery risks via a client device, the health record data comprising a plurality of variables associated with a plurality of features, the plurality of features comprising a plurality of perioperative predictor features;
normalizing, by the data transformer module, the health record data by replacing missing values of one or more of the plurality of variables with replacement values based at least in part on variable type and by preprocessing the one or more of the plurality of variables based at least in part on respective one or more of a plurality of preprocessing types associated with the one or more of the plurality of variables;
transforming, by the data transformer module, one or more of the plurality of features by optimizing the plurality of variables based at least in part on one or more conditional properties associated with a plurality of outcomes, thereby conforming the plurality of variables with inputs of a plurality of predictive models of a data analytics module based at least in part on the optimizing, wherein the plurality of variables are optimized by substituting the plurality of variables with probability ratios associated with ordered variables, wherein each of the plurality of predictive models is associated with a given complication risk and is configured to (i) generate one or more complication risk probabilities based at least in part on the one or more transformed features and (ii) execute independently from other ones of the plurality of predictive models;
selecting, by the data transformer module, the one or more transformed features based at least in part on variance factors of the one or more transfor

Assignees

• Univ Florida

Inventors

• Bihorac Azra
• Li Xiaolin
• Rashidi Parisa
• Pardalos Panagote
• Ozrazgat-Baslanti Tezcan
• Hogan Wiliam
• Wang Daisy Zhe
• Momcilovic Petar
• Lipori Gloria

Classifications

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G16H10/60

  for patient-specific data, e.g. for electronic patient records · CPC title

• G16H20/40

  relating to mechanical, radiation or invasive therapies, e.g. surgery, laser therapy, dialysis or acupuncture · CPC title

• G06N20/20

  Ensemble learning · CPC title

• G16H50/30 Primary

  for calculating health indices; for individual health risk assessment · CPC title