Published July 2020
Background of the Problem
Nationwide initiatives designed to improve the efficiency, safety, and quality of the delivery of healthcare are driving the adoption of interoperable health information exchange. The Office of the National Coordinator for Health Information Technology (ONC) released in 2014, its 10-year vision to achieve an interoperable health information technology (IT) infrastructure, which identified patient matching as part of its three-year agenda1. In its vision statement, the ONC stated that it wanted to address a multitude of issues plaguing data within the healthcare industry and patient matching was a key issue.
To be able to meet their vision of increasing the quality and quantity of information movement across disparate systems, vendors have pursued multiple ways to improve data provenance, data quality and reliability. Effective patient matching is no exception and requires accurate data for any patient matching methodology to work. Hospitals have pursued Master Patient Indexes (MPI) for years; however it is frequently accurate only within the system in which the hospital or health system operates and the acceptance of data from other sources like HIEs, other EMRs, and highly protected data like that from a Prescription Drug Management Program (PDMP) are frequently not incorporated. Short of having a nationwide patient data matching strategy that facilitates interoperable standardization of primary and secondary data elements, and the adoption of a uniform data capture methodology, CIOs and vendors must utilize the most sophisticated matching methodology possible.
How Should CIOs Evaluate A Patient Matching Methodology?
Most patient record linking approaches can be described as deterministic, probabilistic, referential or a blend of all three. A blend of all three should be the gating criteria for internal matching or utilizing a vendor. Additionally, CIOs should inquire with potential vendors about the magnitude of their searches and the estimated error rate.
Proper Patient Matching Explained
Utilizing the information in the patient record that identifies the individual, a deterministic matching approach looks for exact matches between multiple records. For example, if the patient’s name, date of birth, and social security number exactly match on two separate prescription records, those two records can be said to be for the same patient.
A probabilistic approach to patient linking is another method but it introduces a measure of uncertainty to the linking. Probabilistic matching should be used in select circumstances to link records where typos or name variants prohibit exact matching. This can be as simple as assuming that a patient named “Staci” and “Stacey” with the same last name and date of birth are the same person, or more complicated and potentially more risky, such as linking two names based on how rare they are for the area. If a record has the same name, date of birth and almost an exact match on a patient’s phone number except for one digit, probabilistic matching would say that the probability of there being a typo present in the phone number is higher than the probability that there are two different people with the same name and date of birth, but that also have a phone number that differs by only one digit, and support matching the two records.
Referential approaches to matching rely on external datasets that maintain lists of individuals or households, such as change of addresses databases, to be able to link records together. As more types of records from different sources are linked, the likelihood of connecting patient histories from different regions of the country or different types of medical records does go up, but this can also increase the chances of linking two patients inappropriately, as field reliability can vary among datasets. We recommend hospitals utilize in combination the above techniques of matching in a manner which balances the riskiness of a mismatch with ensuring all records belonging to an individual are captured.
Assessing patient linking accuracy is a difficult metric to measure. A true gold-standard method of assessing patient record linking would be to go over the entire linked record set with each individual to confirm its accuracy. Lacking the ability to do a massive manual review with each patient, multiple methods should be used to validate the patient linking. Therefore, the quantity of searches compared to the reported errors found is essential to confirming accuracy. This approximates how often a patient record is reviewed and can capture how often a patient is inappropriately linked to another individual with a current patient linking methodology. Another way to estimate error rate is to look for how many patient groups have ever had a manual edit to their linking structure, capturing historically how often corrections to patient linking have been requested. Providers review of patient linking is more likely to notice two records that should not be linked together than the lack of a record that should have been linked.
Conclusion
CIOs must continually monitor patient linking, looking for ways to tailor the filtering, data cleaning, and linking process to improve linking accuracy both within and across patient matching systems. If there is a combination of personal identifiers that is consistently incorrectly linking together two different patients, the algorithm must be adjusted to reduce errors. Additional fields also must be re-assessed for potential patient linking gains.
About Appriss Health
Appriss Health manages 43 state PDMPs with highly sensitive and state regulated data, but also maintains an inter-state PDMP communication platform (Interconnect) that allows for a patient’s controlled-substance records across multiple state PDMPs to be viewed by providers and pharmacists. Each year, Appriss Health receives over 360 million PDMP searches conducted by 1.4 million users and returns consolidated patient PDMP records with excellent accuracy. Our customer support team now fields about one call for every 770,000 PDMP patient searches related to patient linking (0.00013% call rate). Of the 176 million patient record groups across all PDMPs, only 1 out of every 2,997 groups (0.03%) have ever had a manual edit to the patient linking. The remaining 99.97% of patient groups have not had a manual change to the linking. To learn more about how to properly patient match, visit www.apprisshealth.com to download a free whitepaper on patient matching.
- Office of the National Coordinator for Health Information Technology. Connecting Health and Care for the Nation: A 10-Year Vision to Achieve an Interoperable Health IT Infrastructure. Available at http://healthit.gov/sites/default/files/ONC10yearInteroperabilityConceptPaper.pdf