Patient Record Matching: An Identity Resolution Problem

The vision of the Office of the National Coordinator for Health IT is an interoperable health IT ecosystem where health care providers can share patient health information and match their patients’ Electronic Health Records to other providers’. Such an interoperable system would reduce medical test duplication, medical errors, and delays in treatment. Additionally, the resulting more complete records would lead to improved medical decisions.

A critical part of this interoperability vision is accurate patient record matching. Errors in patient record matching can compromise patient safety and privacy and lead to adverse events, increased health care costs (e.g., repeat tests), and liability.

Why is Patient Record Matching Hard?

Patient health records typically consist of several attributes including the patient’s name, date of birth, gender, social security number, mail address, email address, and phone number.

Unfortunately, none of these attributes is unique or sufficient by itself to be the only basis for patient record matching. For instance, social security numbers in patient health records are not always filled or reliable. Some patients are not willing to provide their social security numbers and/or may provide a fake one. In other cases, records for different patients share the same social security number, for instance, because a parent may provide his/her own social security number when checking a child into a hospital.

Names present their own unique challenges and require special handling. Names are not only not unique but also the name of a patient may differ in two records due to misspellings, nicknames (e.g., “Nick” for “Nicholas”), word order differences, or missing components.

Finally, all attributes in a record are susceptible to data entry errors and some may come in different formats (e.g., date of birth).

The Solution: Identity Resolution Software

Performing patient record matching requires assessing the similarities in multiple key attributes and quantifying the combined evidence. This type of problem is best handled by Identity Resolution software, also known as Entity Matching software. It is important that the Identity Resolution software:

1. perform well in terms of both precision (number of extraneous matches) and recall (number of missing matches)
2. be scalable for Big Data
3. be customizable

NetOwl’s probabilistic machine learning-based entity matching achieves highly accurate identity resolution using probabilistic rules derived from large-scale, real-world, multi-ethnicity name and other attribute data. It performs identity resolution against tens of millions or more entity records with sub-second response times. It is customizable, allowing parameter tuning as well as addition of custom rules and dictionaries.

For more information about NetOwl’s Identity Resolution software, contact us today.