Decision support tools for reducing readmissions of individuals with acute myocardial infarction
System, methods and computer storage media are disclosed for providing a decision support tool for reducing readmissions of AMI patients through early prediction of readmission. An AMI patient may be identified using a working diagnosis and/or an elevated troponin level. One or more machine learning models may be utilized to predict readmission at a time prior to discharge. Based on the prediction, an intervening action may be automatically initiated. Further embodiments include training machine learning model(s) to predict readmission of an AMI patient. In some embodiments, a first model may be trained using reference patient data as it existed at a predetermined time following the patient's admission (e.g., 12 hours after admission), and a second model may be trained using reference patient data as it existed at a later time (e.g., discharge). Readmission risk scores from each model may be combined to determine an overall risk for an AMI patient.
1 . One or more non-transitory media having instructions that, when executed by one or more hardware processors, cause the one or more hardware processors to facilitate a plurality of operations, the operations comprising:
receiving an indication, associated with a current encounter, that a patient is an acute myocardial infarction (AMI) patient who is having or has had an AMI;
identifying (a) a first time interval and a second time interval in relation to patient treatment associated with AMI and (b) the patient as an AMI patient, wherein the first time interval and the second time interval differ;
collecting, via the one or more hardware processors, electronic encoded information specifying content associated with an AMI patient identifier, the patient, and AMI biometric patient data, wherein:
the one or more hardware processors are associated with an electronic memory at a medical records computer system,
the AMI biometric patient data is generated via a sensor device configured to indicate a measurement metric of an AMI condition corresponding to the patient, and
the AMI biometric patient data is associated with the AMI patient identifier and the electronic memory at the medical records computer system;
in response to the receiving of the indication, automatically querying an electronic health record for the patient for a first plurality of feature values corresponding to the first time interval and a second plurality of feature values corresponding to the second time interval, at least one feature value of the first plurality of feature values (a) differing from any feature value of the second plurality of feature values and (b) being associated with the AMI biometric patient data;
applying a first machine-learning electronic model to information associated with the first plurality of feature values to generate a first prediction, the first machine-learning electronic model trained on a first set of AMI patient data, associated with the first time interval, based on a first set of feature data, the first set of feature data generated by utilizing a feature selection algorithm comprising:
(i) stratifying and/or segregating reference data to produce results containing a set of continuous features and a set of categorical features; and
(ii) using the results with an extreme gradient boosting algorithm and an adaptive elastic-net algorithm to produce the first set of feature data;
applying a second machine-learning electronic model to information associated with the second plurality of feature values to generate a second prediction, the second machine-learning electronic model trained on a second set of AMI patient data, associated with the second time interval, based on a second set of feature data generated by utilizing the feature selection algorithm;
based on the first prediction and the second prediction, generating a readmission prediction of the patient being readmitted to a healthcare facility for AMI within a future time interval; and
electronically writing, via the one or more hardware processors, electronic encoded data to the electronic memory at the medical records computer system, wherein the electronic encoded data indicates the readmission prediction.
2 . The one or more non-transitory media of claim 1 , wherein the readmission prediction comprises a risk score, wherein the operations further comprise determining that the risk score satisfies a risk level threshold, and wherein the operations further comprise initiating one or more intervening actions for the patient based on the risk score satisfying the risk level threshold.
3 . The one or more non-transitory media of claim 1 , wherein the indication includes one or both of a diagnosis received from the electronic health record and a troponin lab value.
4 . The one or more non-transitory media of claim 3 , wherein the indication further includes a diagnosis code associated with AMI, the diagnosis code being entered into the electronic health record prior to discharge.
5 . The one or more non-transitory media of claim 1 , wherein the AMI biometric patient data indicates a troponin lab value of the patient, and wherein the indication that the patient is an AMI patient includes a determination that the troponin lab value of the patient satisfies an elevated threshold amount that is determined by an upper reference limit.
6 . The one or more non-transitory media of claim 1 , wherein the first machine-learning electronic model is trained on patient data associated with a time within 12 hours of admission and the second machine-learning electronic model is trained on patient data associated with a time of discharge, and wherein determining the readmission prediction comprises generating an overall prediction of readmission utilizing the first machine-learning electronic model and the second machine-learning electronic model.
7 . The one or more non-transitory media of claim 1 , wherein the first machine-learning electronic model uses a first subset of feature values to determine a first risk level and the second machine-learning electronic model uses a second subset of feature values to determine a second risk level, the first risk level and the second risk level differing and being combined to generate the readmission prediction.
8 . The one or more non-transitory media of claim 7 , wherein the first risk level and the second risk level are combined by computing a weighted average of the first risk level and the second risk level, the weights being based on time since admission.
9 . The one or more non-transitory media of claim 1 , wherein the operations further comprise electronically communicating, to a caregiver of the patient, a notification comprising an indication of the readmission prediction or a recommendation based on the readmission prediction.
10 . A system having one or more hardware processors configured to facilitate a plurality of operations, the operations comprising:
receiving an indication, associated with a current encounter, that a patient is an acute myocardial infarction (AMI) patient who is having or has had an AMI;
identifying (a) a first time interval and a second time interval in relation to patient treatment associated with AMI and (b) the patient as an AMI patient, wherein the first time interval and the second time interval differ;
collecting, via the one or more hardware processors, electronic encoded information specifying content associated with an AMI patient identifier, the patient, and AMI biometric patient data, wherein:
the one or more hardware processors are associated with an electronic memory at a medical records computer system,
the AMI biometric patient data is generated via a sensor device configured to indicate a measurement metric of an AMI condition corresponding to the patient, and
the AMI biometric patient data is associated with the AMI patient identifier and the electronic memory at the medical records computer system;
in response to the receiving of the indication, automatically querying an electronic health record for the patient for a first plurality of feature values corresponding to the first time interval and a second plurality of feature values corresponding to the second time interval, at least one feature value of the first plurality of feature values (a) differing from any feature value of the second plurality of feature values and (b) being associated with the AMI biometric patient data;
applying a first machine-learning electronic model to information associated with the first plurality of feature values to generate a first prediction, the first machine-learning electronic model trained on a first set of AMI patient data, associated with the first time interval, based on a first set of feature data, the first set of feature data generated by utilizing a feature selection algorithm comprising:
(i) stratifying and/or segregating reference data to produce results containing a set of continuous features and a set of categorical features; and
(ii) using the results with an extreme gradient boosting algorithm and an adaptive elastic-net algorithm to produce the first set of feature data;
applying a second machine-learning electronic model to information associated with the second plurality of feature values to generate a second prediction, the second machine-learning electronic model trained on a second set of AMI patient data, associated with the second time interval, based on a second set of feature data generated by utilizing the feature selection algorithm;
based on the first prediction and the second prediction, generating a readmission prediction of the patient being readmitted to a healthcare facility for AMI within a future time interval; and
electronically writing, via the one or more hardware processors, electronic encoded data to the electronic memory at the medical records computer system, wherein the electronic encoded data indicates the readmission prediction.
11 . The system of claim 10 , wherein the readmission prediction comprises a risk score, wherein the operations further comprise determining that the risk score satisfies a risk level threshold and initiating one or more intervening actions, and wherein the one or more intervening actions are based on the risk score satisfying the risk level threshold.
12 . The system of claim 10 , wherein the operations further comprise electronically communicating, to a caregiver of the patient, a notification comprising an indication of the readmission prediction or a recommendation based on the prediction.
13 . A computer-implemented method, comprising:
receiving an indication, associated with a current encounter, that a patient is an acute myocardial infarction (AMI) patient who is having or has had an AMI;
identifying (a) a first time interval and a second time interval in relation to patient treatment associated with AMI and (b) the patient as an AMI patient, wherein the first time interval and the second time interval differ;
collecting, via one or more hardware processors, electronic encoded information specifying content associated with an AMI patient identifier, the patient, and AMI biometric patient data, wherein:
the one or more hardware processors are associated with an electronic memory at a medical records computer system,
the AMI biometric patient data is generated via a sensor device configured to indicate a measurement metric of an AMI condition corresponding to the patient, and
the AMI biometric patient data is associated with the AMI patient identifier and the electronic memory at the medical records computer system;
in response to the receiving of the indication, automatically querying an electronic health record for the patient for a first plurality of feature values corresponding to the first time interval and a second plurality of feature values corresponding to the second time interval, at least one feature value of the first plurality of feature values (a) differing from any feature value of the second plurality of feature values and (b) being associated with the AMI biometric patient data;
applying a first machine-learning electronic model to information associated with the first plurality of feature values to generate a first prediction, the first machine-learning electronic model trained on a first set of AMI patient data, associated with the first time interval, based on a first set of feature data, the first set of feature data generated by utilizing a feature selection algorithm comprising:
(i) stratifying and/or segregating reference data to produce results containing a set of continuous features and a set of categorical features; and
(ii) using the results with an extreme gradient boosting algorithm and an adaptive elastic-net algorithm to produce the first set of feature data;
applying a second machine-learning electronic model to information associated with the second plurality of feature values to generate a second prediction, the second machine-learning electronic model trained on a second set of AMI patient data, associated with the second time interval, based on a second set of feature data generated by utilizing the feature selection algorithm;
based on the first prediction and the second prediction, generating a readmission prediction of the patient being readmitted to a healthcare facility for AMI within a future time interval; and
electronically writing, via the one or more hardware processors, electronic encoded data to the electronic memory at the medical records computer system, wherein the electronic encoded data indicates the readmission prediction.
14 . The computer-implemented method of claim 13 , wherein the first machine-learning electronic model and the second machine-learning electronic model are run at a same time to each predict a risk level of a particular AMI patient being readmitted.
15 . The one or more non-transitory media of claim 1 , wherein the readmission prediction is generated at a first time that is associated with the patient being admitted to the healthcare facility, and wherein the first machine-learning electronic model and the second machine-learning electronic model are applied, at a time within one day after admission of the patient to the healthcare facility, to generate the readmission prediction.
16 . The one or more non-transitory media of claim 1 , wherein the readmission prediction is generated at a time that is after admission of the patient to the healthcare facility and that is prior to discharge of the patient from the healthcare facility.
17 . The one or more non-transitory media of claim 1 , wherein the feature selection algorithm performs (i) the stratifying and/or segregating of the reference data before (ii) using a set of machine learning models with the results to identify the first set of feature data.
18 . The one or more non-transitory media of claim 1 , wherein the first plurality of feature values and the second plurality of feature values differ.
19 . The computer-implemented method of claim 13 , wherein the readmission prediction is generated at first time interval is associated with an admission of an AMI reference patient to a healthcare facility and at a second time interval associated with a discharge of the AMI reference patient from the healthcare facility.
20 . The one or more non-transitory media of claim 1 , wherein the operations further comprise: based on the readmission prediction, automatically initiating one or more intervening actions to reduce a likelihood of readmission for the patient.
21 . The one or more non-transitory media of claim 1 , wherein the first set of patient data and the second set of patient data differ.
22 . The one or more non-transitory media of claim 1 , wherein the feature selection algorithm consecutively stratifies and segregates the reference data to produce the results.
23 . The one or more non-transitory media of claim 1 , wherein the feature selection algorithm consecutively utilizes (a) differing machine learning models with (b) the results to identify the first set of feature data.
24 . The one or more non-transitory media of claim 1 , wherein the feature selection algorithm performs (i) the stratifying and/or segregating of the reference data before (ii) using a the set of machine learning models with the results to identify the first set of feature data.
25 . The one or more non-transitory media of claim 1 , wherein executing the instructions causes: initiating feature selection comprising: (i) identifying feature values from a set of data including the reference data and performing a first feature selection operation to generate a first set of features, and (ii) performing a second feature selection operation based on the first set of features to generate a second set of features.
26 . The one or more non-transitory media of claim 25 , wherein: the feature values are segregated, at a training component of a decision support tool associated with the one or more hardware processors, into the continuous features and the categorical features (i) after the identifying of the feature values and (ii) prior to the performing of the first feature selection operation.
27 . The one or more non-transitory media of claim 25 , wherein the operations further comprise utilizing the continuous features and the categorical features with an Extreme Gradient Boosting (XGboost) model to perform the first feature selection operation.
28 . The one or more non-transitory media of claim 1 , wherein the operations further comprise utilizing an adaptive elastic-net model with the first set of features to perform a second feature selection operation.