![[personal profile]](/makepkg/implementing-https-www.dreamwidth.org/img/silk/identity/user.png){kind=small}
EMPIs and Patient Matching
Dec. 28th, 2017 02:12 pmA friend pointed me at https://siderea.dreamwidth.org/1377224.html - asking if I could provide some context and response.
In HIT-land, this article from the ONC best describes what the term “patient matching” means:
“For the uninitiated, we use “patient matching” in health IT as shorthand to describe the techniques used to match the data about you held by one health care provider with the data about you held by another (or many others). In practice, patient matching is the process of comparing different demographic elements from different health information technology (health IT) systems to determine if they refer to the same patient.”
I’d like to add in here, that you can have patient matching requirements within the same provider organizations (for example, if you have different systems to help each department do their work better. Pathology, may have a different system from Radiology, may have a different system from Oncology, etc…In all cases, it’s important that each system be able to communicate and identify each patient as the correct unique patient. Where you have a source of all truth registration system, and a requirement that each departmental system uses the same medical record number (MRN) from said registration system, this is relatively easy – but even the registration system will have a percentage of duplicates or near matches. I’ll get to that in a bit.
When you have different organizational systems needing to access each others services, the problem becomes more complex. For example, as is common in healthcare, a hospital system may have radiology and lab services that local providers can refer to. These providers may or may not be in the same parent system – and usually have an EHR different from the hospital system’s EHR (why? Because different EHRs do different things well – and an EHR that is optimized for inpatient use is going to have different features and requirements than one optimized for ambulatory care (like when you go to your PCP). Because these are individual entities and because we do not have a national patient identifier, medical record numbers are going to be different between the two systems.
So how do you resolve whether or not a patient presenting at a lab with an order from their primary physician is the same patient as someone already in the hospital system’s EHR? Do you create a new medical record number (sometimes yes), do you attempt to match a patient with an existing patient? What if the person taking patient information at registration misspells a name? If a new record is created based on a misspelling, that’s a duplicate. What if SSN or DOB information is transposed – there’s more duplication. What if two people with the same name and DOB show up to the same facility (not as uncommon as you might think)? There’s the SSN, but what if they have similar SSNs – or what if you’re still not sure? So duplicates and near matches are endemic to the registration system, and the issue is only compounded when you are trying to identify patients between systems.
That’s where patient matching software comes in (EMPI is a term you might hear with regards to patient matching). There are a number of software vendors out there that specialize in this: Initiate, NextGate, 4Medica, and many many more – this on top of about a bazillion home grown solutions that allow for matching within various custom software and propriety solutions. Each have their own flavor of how to do patient matching. All have a concept of “scoring” a match and thresholds. All of them share similar problems with regards to scoring matches and defining thresholds. And in every EMPI-adjacent project I’ve ever worked on, I *usually* hear something along the lines of “You’re the subject matter experts, so how can you help us [the healthcare system] figure this out” but the true subject matter experts with regards to patient matching is the provider organization itself.
What is this threshold and matching business? Well, EMPIs (and other patient matching system) work to match certain demographic characteristics of a patient. Most organizations use Name, DOB, often SSN, and sometimes gender as a bare minimum to establish identity. But in this data set alone you have have misspellings, phonetic spellings, nicknames, marriages, divorces, name changes, gender changes, transpositions in numbers between month/day and day/month systems depending on who is entering or giving information, bad data entry. Some of this can be mitigated if the patient has a rule for themselves to only give their full legal name (like I do) but some systems include a middle initial, some don’t…any time there’s uncertainty, you have an opportunity to create a near match or a duplicate. So EMPIs have created ways to allow other demographic data to be included in patient match – maybe a driver’s license can be scanned in, maybe a street address can be included, maybe a parent or spouse name can be included. But each of these have varying levels of reliability, so you may not want a street address, for example, to have equal weight to an SSN. Or maybe your provider organization has opted to do away with using SSNs as identifiers because of high rates of identity theft in your area, so they’re not reliable indicators of identify and you want to de-emphasize the role an SSN plays in identifying a match. This is weighting. Once all of the demographic indicators are calculated and weighted, you come up with a score. A threshold is a score that defines two things: 1) a score over which you are certain enough of a patient’s identity based on match that you automatically link your action (be it an encounter, a clinical note, a lab, a registration, an imaging study, etc) to a patient record ; and 2) a score below which everything should be considered absolutely not a match so don’t link it. This defines a range between which a score means manual review is necessary to resolve any matches/duplicates/etc. This is point at which software says, “We found a probable match to your patient, but need you to review to take next steps”. There are some organizations that have an upper threshold of 100% (ie., all demographics match exactly) and a lower threshold of something like 20% with very large manual review range – other organizations have a much narrower range for manual review.
As healthcare systems grow and change to respond to market dynamics in order to continue to provide services – more and more EHRs and other clinical systems need to be able to talk to each other and to be able to accurately and efficiently communicate about the same unique patient across systems. This also introduces more uncertainty – and again, more opportunities to create duplicate records within an entity’s system. So back to the original question – yes 20% seems about right for duplicate records, and yes, the problem is expected to grow.
![[livejournal.com profile]](/makepkg/implementing-https-www.dreamwidth.org/img/external/lj-userinfo.gif){kind=small}
