Automatic coding of patient outcomes

• US 8,504,392 B2
• Filed: 11/11/2011
• Issued: 08/06/2013
• Est. Priority Date: 11/11/2010
• Status: Active Grant

• Current Assignee: Board of Trustees of the Leland Stanford Junior University (Stanford Management Co.)
• Original Assignee: Board of Trustees of the Leland Stanford Junior University (Stanford Management Co.)
• Inventors: Saria, Suchi, McElvain, Gayle, Rajani, Anand K., Penn, Anna A., Koller, Daphne L.
• Primary Examiner(s): Holcomb, Mark
• Application Number:
  US13/294,959
• Publication Number:
  US 20120290319A1
• Time in Patent Office: 634 Days
• Field of Search: 705 2- 4, 706/12, 706/25, 706 45- 61
• US Class Current: 705/3
• CPC Class Codes: G06Q 10/10 Office automation; Time management
• CPC Class Codes: G16H 10/60 for patient-specific data, e.g. for electronic patient records
• CPC Class Codes: G16H 15/00 ICT specially adapted for medical reports, e.g. generation or transmission thereof
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: G16H 50/30 for calculating health indices; for individual health risk assessment
• CPC Class Codes: G16H 70/60 relating to pathologies

Abstract

Systems and methods can mine structured clinical event data in an electronic health record (EHR) system to determine patient outcomes. Mining the structured clinical event data instead of or in addition to mining discharge summaries can increase the accuracy of patient outcome identification. Sophisticated language models can be used to extract outcomes from discharge summaries while also inferring outcomes from cues or hints contained in the structured clinical event data. For example, the clinical event data can include information regarding treatments and medications prescribed by clinicians to specifically manage patient complications; thus, presence or absence of relevant treatments in the clinical event data can provide independent indicators to disambiguate cases where current language processing approaches fail.

Claims

1. A system for classifying a health condition of a patient, the system comprising:

a model creation engine configured to create a medical classification model by at least;

receiving an identification of a clinical feature that is to be associated with a health condition, the identification being provided by one or more of an automated analysis of an electronic medical reference and a manual expert identification of the clinical feature, the clinical feature comprising one or more of the following features;

an identified medication, a clinical event, a microbial culture feature, and a radiology feature,

creating a rule that maps the clinical feature to the health condition in a model data repository comprising physical computer storage, wherein the rule reflects a relationship between the clinical feature and the health condition,

automatically learning a weight to apply to the rule with a supervised machine learning algorithm by at least analyzing the clinical feature with respect to pre-identified outcomes in a training data set, the training data set comprising first structured clinical event data, the rule reflecting a strength of the relationship between the clinical feature and the health condition, and

storing, in the model data repository, the learned weight associated with the rule for subsequent usage in identifying a patient health condition; and

an outcome identification module comprising computer hardware, the outcome identification module configured to at least;

access patient data corresponding to a patient, the patient data comprising second structured clinical event data stored in an electronic health record (EHR) database,

analyze the second structured clinical event data to determine whether the clinical feature exists in the clinical event data,

apply the rule and the weight of the medical classification model to the clinical feature to infer a possible health condition of the patient by at least matching the rule with a selected clinical feature in the second structured clinical event data corresponding to the patient, and

provide one or more billing codes configured to be processed by a medical billing system, said one or more billing codes being based at least in part on the possible health condition of the patient.

2. The system of claim 1, wherein the outcome identification module is further configured to apply the rule to infer the possible health condition of the patient by at least combining a count of occurrences of the clinical feature in the second structured clinical event data with the learned weight in a probabilistic function.

3. The system of claim 2, wherein the probabilistic function comprises a weighting function comprising one or more parameters derived from optimizing an objective on the training data.

4. The system of claim 3, wherein the weighting function comprises a logistic regression function.

5. The system of claim 1, wherein the model creation engine is further configured to incorporate a language-based rule and an associated weight in the model data repository.

6. The system of claim 5, wherein the outcome identification module is further configured to identify a language feature in clinician notes and to apply the language-based rule to the identified language feature as part of inferring the possible health condition of the patient.

7. The system of claim 1, wherein the outcome identification module infers the possible health condition of the patient by directly identifying the one or more billing codes.

8. The system of claim 1, wherein the machine learning algorithm comprises a supervised algorithm or a semi-supervised algorithm.

9. The system of claim 8, wherein the machine learning algorithm comprises one or more of the following:

a maximum likelihood estimation algorithm, a support vector machine, a boosting algorithm, and a neural network algorithm.

10. The system of claim 1, wherein the electronic medical reference comprises an online medical dictionary.