Computer-implemented detection and statistical analysis of errors by healthcare providers

1. A computer-implemented method comprising:

accessing a plurality of digitally stored, unstructured medical diagnostic data;

digitally displaying a first subset of the medical diagnostic data, the first subset of the medical diagnostic data including at least a first set of diagnostic reports, using a computer display device, concurrently with digitally displaying one or more quality control checklists that are specific to a medical discipline represented in the first set of diagnostic reports;

receiving digital input specifying one or more errors in the first set of diagnostic reports and digitally storing the digital input in association with the first subset of medical diagnostic data;

training a hierarchical Bayesian machine learning model in one or more training phases using the digital input as labels for training data comprising at least the first subset of medical diagnostic data, wherein the hierarchical Bayesian machine learning model:

is trained on provider hierarchy metadata; and

integrates a population-wide priors model, an inter-feature correlation model and an inter-reviewer variability model;

after completing the one or more training phases, performing an evaluation phase using the hierarchical Bayesian machine learning model to generate one or more provider error rate data, wherein the hierarchical Bayesian machine learning model generates the one or more provider error rate data based on an input comprising a second subset of the medical diagnostic data, the second subset being different from the first subset;

applying a grading algorithm to the one or more provider error rate data to yield one or more output provider quality score values;

wherein the method is performed by one or more computing devices.

2. The method of claim 1 , further comprising training the hierarchical Bayesian machine learning model at least in part using provider feature data and patient feature data.

3. The method of claim 1 , further comprising applying the grading algorithm to the one or more provider error rate data to yield one or more initial provider quality score values, modifying the initial provider quality score values by applying a provider weight value associated with a particular healthcare provider, again modifying the initial provider quality score values by applying a patient complexity adjustment value, to transform the initial provider quality score values into the one or more output provider quality score values.

4. The method of claim 3 , the provider weight value comprising a provider-specific clinical impact weighting factor.

5. The method of claim 1 , the unstructured medical diagnostic data comprising a plurality of Digital Imaging and Communications in Medicine (DICOM) digital images of patient anatomy and a corresponding plurality of radiology diagnostic reports that are based on the images.

6. The method of claim 1 , the accessing comprising determining a provider identity value that uniquely identifies a particular healthcare provider; based on the provider identity value, retrieving a sampling factor for the particular healthcare provider from a database;

sampling the first subset of the medical diagnostic data from all the medical diagnostic data based upon the sampling factor.

7. A computer-implemented method comprising:

from among a plurality of digitally stored records of diagnostic imaging examinations, selecting a first subset of records of diagnostic imaging examinations that are associated with a set of healthcare providers under assessment for structured exam quality review;

accessing a first subset of digitally stored, unstructured medical diagnostic data associated with the first subset of records, the first subset of medical diagnostic data including at least a first set of diagnostic reports;

conducting the structured exam quality review by digitally displaying the first subset of medical diagnostic data including at least the first set of diagnostic reports, using a computer display device, concurrently with digitally displaying one or more structured quality control checklists that are specific to a medical discipline and representing diagnostic dimensions or parameters of quality represented in the first set of diagnostic reports;

receiving digital input, via the one or more checklists, specifying one or more error attribute values representing attributes of errors in the first set of diagnostic reports and digitally storing the digital input in association with the first subset of medical diagnostic data and one or more identifiers corresponding to the one or more checklists, wherein each error attribute value represents a level of agreement or disagreement between an interpretation of a diagnostic imaging examination of the first subset of records of diagnostic imaging examinations by a healthcare provider of the set of healthcare providers under assessment and an interpretation of the diagnostic imaging examination of the first subset of records of diagnostic imaging examinations by a reviewer;

training a hierarchical Bayesian machine learning model in one or more training phases using the digital input as labels for training data comprising at least the first subset of medical diagnostic data for estimating rates of diagnostic errors committed by each healthcare provider under evaluation, the healthcare providers each comprising an individual radiologist or group of radiologists practicing together at a facility, facilities, or affiliated with a common provider organization;

after completing the one or more training phases, performing an evaluation phase using the hierarchical Bayesian machine learning model to generate one or more provider error rate data, wherein the hierarchical Bayesian machine learning model generates the one or more provider error rate data based on an input comprising a second subset of the medical diagnostic data, the second subset of medical diagnostic data including at least a second set of diagnostic reports different from the first set of diagnostic reports included in the first subset of medical diagnostic data;

applying a grading algorithm to the one or more provider error rate data to yield one or more output provider quality score values;

wherein the method is performed by one or more computing devices.

8. The method of claim 7 , further comprising determining one or more correlations and qualitative relationships between the error attribute values generated by the structured exam quality reviews, estimates of provider error rates, and additional provider attributes.

9. The method of claim 8 , further comprising using the one or more correlations and qualitative relationships to perform one of: generating measures of diagnostic error rates or weighted diagnostic error rates with improved accuracy, precision, or narrower confidence intervals; generating predicted measures of diagnostic error rates or weighted diagnostic error rates for radiology providers which have not had any of their imaging exams subjected to the structured exam quality reviews and for whom only supplementary data elements and additional provider attribute values are available.

10. The method of claim 7 , the rates of diagnostic errors being related to an aggregated set of diagnostic errors or subsets of diagnostic errors defined by one or more of a relationship to a common anatomical structure or structures.

11. The method of claim 7 , the rates of diagnostic errors being related to a type of diagnostic error selected from among false positive, false negative, overcall, or undercall.

12. The method of claim 7 , the rates of diagnostic errors being related to a level of severity of clinical impact to patients' care or outcomes.

13. The method of claim 7 , the rates of diagnostic errors being related to a level of egregiousness of a specific diagnostic error with respect to an estimated difficulty of making a correct diagnosis.

14. The method of claim 7 , further comprising generating one or more benchmarks and quality thresholds for use in assigning designations and categorizations to providers based on their assessed quality performance levels.

15. The method of claim 7 , further comprising generating one or more reports of the provider error rate data or provider quality score values.

16. The method of claim 7 , further comprising digitally displaying the one or more structured checklists with a plurality of checklist input fields that are programmed to receive one or more of: (1) a presence and severity of specific pathologies or anatomical abnormalities that are present in medical images, (2) an inclusion of specific diagnostic findings in medical reports, (3) a presence of specific technical quality issues in the medical images and reports, (4) a presence of specific diagnostic errors in the medical reports, (5) and assessment of a likelihood of present diagnostic errors and technical quality issues causing a clinically important impact on a patient.

17. The method of claim 7 , the selecting further comprising automatically determining a number and a set of types of diagnostic imaging exams to review from each provider.

18. The method of claim 7 , the selecting further comprising weighting the selecting based upon magnitudes of one or more provider attribute values selected from among provider practice size, patient mix, and provider training profiles.

19. The method of claim 7 , the selecting further comprising randomly sampling from among the plurality of digitally stored records of diagnostic imaging examinations to select the first subset of records of diagnostic imaging exams for the structured exam quality review.

20. The method of claim 7 , the selecting further comprising selecting a subset of exam records for multiple independent structured exam quality reviews, and in response, repeating, via two or more iterations, the step of receiving digital input, via the checklists, specifying one or more errors in the first set of diagnostic reports and digitally storing the digital input in association with the first subset of medical diagnostic data, in which each iteration comprises receiving different digital input from a different computer.

21. The method of claim 7 , the machine learning algorithm being programmed to generate estimated values of confidence intervals associated with each provider diagnostic error rate value.

22. The method of claim 17 , the machine learning algorithm being programmed to analyze data from exam records that have been subjected to multiple independent instances of the structured exam quality review to produce improved estimates of confidence intervals associated with each provider error rate value.

23. The method of claim 7 , the rates of diagnostic errors comprising weighted combinations of category rates of one or more of relationship to a common anatomical structure or structures, relationship to a type of diagnostic error selected from among false positive, false negative, overcall, or undercall, or relationship to a level of severity or egregiousness selected from among large overcall or small overcall.

24. The method of claim 23 , wherein weight values of the category rates represent likelihoods that specific diagnostic errors will have a clinically important impact on the patient.

25. The method of claim 24 , the weight values of the category rates being generated based on additional data elements during the structured exam quality reviews, stand-alone assessments by radiologists or other medical experts of a likely clinical impact of specific types of diagnostic errors, or analysis of historical medical records of patients in combination with diagnostic error data to estimate the correlation of specific diagnostic errors or providers with specific error rates and impacts to patients' treatment patterns, costs, and outcomes.

26. The method of claim 25 , further comprising adjusting the rates of diagnostic errors or the weight values based on an estimated measure of patient complexity of a patient population associated with each healthcare provider undergoing the structured exam quality review.

27. The method of claim 7 , further comprising automatically machine generating synthetic error attribute values based on one or more of medical images, medical reports, and/or combinations of medical images and reports, the machine generating being trained using the error attribute values that were generated as part of conducting the structured exam quality review.