System and method for record linkage

• US 10,249,385 B1
• Filed: 05/01/2013
• Issued: 04/02/2019
• Est. Priority Date: 05/01/2012
• Status: Active Grant

• Current Assignee: Cerner Innovation, Inc. (Oracle Corporation)
• Original Assignee: Cerner Innovation, Inc. (Oracle Corporation)
• Inventors: McNair, Douglas S., Kailasam, Kanakasabha, Murrish, John Christopher
• Primary Examiner(s): Tiedeman, Jason S
• Application Number:
  US13/874,961
• Time in Patent Office: 2,162 Days
• Field of Search: 705 2, 705 3
• CPC Class Codes: G16H 10/60 for patient-specific data, e.g. for electronic patient records
• CPC Class Codes: G16H 80/00 ICT specially adapted for facilitating communication between medical practitioners or patients, e.g. for collaborative diagnosis, therapy or health monitoring

Abstract

Methods, systems, and computer-readable media are provided for facilitating record matching and entity resolution and for enabling improvements in record linkage. A power-spectrum-based temporal pattern-specific weight may be incorporated into record linkage methods to enhance the record linkage accuracy and statistical performance. For example, in embodiments, a value-specific weight may be calculated from a population-based frequency of field-specific values and provides an opportunity to capture and measure the relative importance of specific values found in a field. A timeseries-derived Bayesian power spectrum weight may be calculated from the population-based frequency of temporal pattern-specific values in terms of intensities at various frequencies of the power spectrum computed from the timeseries, and further provides an opportunity to capture and measure the relative importance of specific sequences of care episodes.

Claims

1. One or more computer-readable storage devices having computer-usable instructions embodied thereon that, when executed, enable a given processor to perform a method of determining that a plurality of health records are related to the same human patient, the method comprising:

receiving a target record from a first health-records system, the target record comprising at least one target record blocking variable and a date-time variable associated with the target record;

receiving one or more candidate records from a second health-records system, each candidate record comprising at least one candidate record blocking variable and information of a plurality of episodes associated with a candidate patient, said information including a date-time value associated with each episode of the plurality of episodes, wherein the at least one candidate record blocking variable is the same as the at least one target record blocking variable and comprises a birth day, birth month, birth year, or an indication of a condition, treatment, diagnosis, or other context; and

for each candidate record;

(1) based on said date-time variable associated with each episode, determining a timeseries of time intervals representing the time between each episode and a time duration from the date-time variable associated with a last candidate record episode to the target record date-time variable;

(2) for each timeseries, determining a normalized power-spectrum likelihood weight (“

power spectra weight”

), based on the timeseries;

(3) determining a record linkage weight based on a measure of lexical similarity between the candidate record and target record;

(4) based on the determined record linkage weight and power spectra weight, determining a composite candidate record score using a root-mean-square transformation, a cosine transformation, or correlation coefficient;

(5) performing a comparison of the candidate record score to a threshold;

(6) based on the comparison, determining that the candidate record score satisfies the threshold, designating the candidate record as related to the target record, adding an indication of the target record to the candidate record thereby creating an updated candidate record, and storing the updated candidate record in the first or second health-records system.

2. The one or more computer-readable storage devices of claim 1, further comprising determining the threshold, and wherein the determined scores for the candidate records are rank-ordered;

wherein the threshold is determined based on a gap in successive rank-ordered scores; and

wherein for each candidate record, the threshold is determined as satisfied where the candidate score is greater than the threshold.

3. The one or more computer-readable storage devices of claim 1, wherein the time duration from the date-time value associated with a last candidate record episode comprises a date-time value associated with a most recent episode in the candidate record.

4. The one or more computer-readable storage devices of claim 1, wherein the first and second health-records systems comprise the same health-records system.

5. The one or more computer-readable storage devices of claim 1, wherein the first and second health-records systems comprise disparate records systems.

6. The one or more computer-readable storage devices of claim 1, wherein the candidate record designated as related to the target record is presented to a user.

7. The one or more computer-readable storage devices of claim 6, further comprising receiving a information from the user indicating a confirmation that the candidate record designated as related to the target record is related to the target record.

8. The one or more computer-readable storage devices of claim 1, wherein the target record date-time value comprises a near real-time date-time value.

9. The one or more computer-readable storage devices of claim 1, wherein determining a composite candidate record score comprises combining the power spectra weight and record linkage weight using the root-mean-square transformation.

10. The one or more computer-readable storage devices of claim 1, wherein the plurality of episodes comprises at least three episodes.

11. The one or more computer-readable storage devices of claim 1, wherein determining the power spectra weight for each candidate record comprises:

determining the power spectra weight based on at least a portion of the time series;

based on the determined power spectra, determining a likelihood value for each spectrum; and

normalizing the determined likelihood values.

12. The one or more computer-readable storage devices of claim 11, wherein the likelihood value of each spectrum is determined by iteratively permuting the spectrum and sampling the resulting permutations by Bayesian Markov Chain Monte Carlo simulation, wherein the median likelihood is retained as the likelihood value.

13. A computer-implemented method of determining related records, the method comprising:

receiving a reference record from a first records system, the reference record including a reference record date;

receiving one or more candidate records from a second records system, each candidate record associated with an entity and comprising information of a plurality of episodes, each episode associated with an episode date; and

for each candidate record;

(1) based on the episode date associated with each episode, determining a timeseries comprising time intervals, the time intervals representing the time between each episode date and the time between a last episode date of the candidate record and the reference record date;

(2) for each timeseries, determining a normalized power-spectrum likelihood weight (“

power spectra weight”

), based on the timeseries;

(3) based on the determined power spectra weight, determining a candidate record score;

(4) performing a comparison of the candidate record score to a threshold; and

(5) based on the comparison determining that the candidate record score satisfies the threshold, designating the candidate record as related to the reference record and adding an indication of the reference record to the candidate record thereby creating an updated candidate record.

14. The computer-implemented method of claim 13, further comprising determining the threshold, and wherein the determined scores for the candidate records are rank-ordered;

wherein the threshold is determined based on a gap in successive rank-ordered scores; and

wherein for each candidate record, the threshold is determined as satisfied where the candidate score is greater than the threshold.

15. The computer-implemented method of claim 13, wherein the entity comprises an object, person, or account.

16. The computer-implemented method of claim 13, wherein an episode comprises a change in a status associated with an object, a patient-care episode, or a transaction.

17. The computer-implemented method of claim 13, wherein for each candidate record, the method further comprises:

determining a record linkage weight based on a measure of lexical similarity between the candidate record and reference record; and

wherein determining a candidate record score is further based on the determined record linkage weight.

18. The computer-implemented method of claim 13, wherein the last episode-date of the candidate record comprises the episode-date associated with a most recent episode in the candidate record, and wherein the episode date comprises information representing a date, time, or date and time.

19. A computer-implemented method of determining related records, the method comprising:

receiving a plurality or records from at least one database, each record comprising at least one blocking variable and information of one or more episodes, wherein date-time information is associated with each episode, and wherein each blocking variable comprises a birth day, birth month, birth year, or indication of a condition, treatment, diagnosis, or other context;

for each record, determining a value of a linkage indicator variable, wherein the value of the linkage indicator variable indicates whether information in the record is associated with information in one or more other records in the at least one database;

based on the record linkage variable and blocking variable of each record, determining a subset of plausibly related records from the plurality of records;

determining a normalized power-spectrum likelihood weight (“

power spectra weight”

) for each record in the subset;

determining a record linkage weight for each record in the subset based on a measure of lexical similarity;

determining a composite score based on the determined power spectra weight and record linkage weight and using a root-mean-square transformation, a cosine transformation, or correlation coefficient;

comparing the composite score to a threshold value; and

if the composite score is greater than the threshold value, identifying a record as a related record and adding a related-record indication to the related record, but if the composite score is less than the threshold, identifying the record as not related.