Measure your identity data management maturity. Take our 5-minute maturity assessment now

New York HIE Goes All In on Patient Matching for the Homeless

Featured in Healthcare Innovation Group

As of April 2020, there were more than 60,000 homeless people in New York City, and in recent years the city’s homeless numbers have reached their highest levels since the Great Depression.

Healthix, the largest public health information exchange (HIE) in the nation, bringing together over 1,200 healthcare organizations at thousands of locations across New York City and Long Island, has set out to improve the health outcomes of people without stable housing through a coordinated care approach. Leaders of the HIE, which provides access to clinical data of more than 20 million patients, identified from the onset a foundational challenge: the homeless lack addresses, which often prohibits matching the right records to the right person.

Undoubtedly, one of the major challenges with caring for the homeless is being able to first identify them properly and then follow them across the care continuum. These core barriers have led to missed opportunities to collaborate with public health agencies like the New York City Department of Social Services to effectively address this critical social determinant of health (SDOH). Healthix’s senior vice president of innovation, Thomas Moore, recently spoke with Managing Editor Rajiv Leventhal, sharing key details on how the HIE has identified proxies for homelessness and used a master patient index (MPI) from patient matching company Verato to reconcile millions of medical records, identify the homeless, and alert ED staff. Below are excerpts of that interview.

Can you describe the decision-making process and impetus behind coordinating care for those without stable housing?

Healthix is a private not-for-profit [HIE], certified by the state as part of Department of Health, and to a great extent we service a public function. In healthcare, a lot of that is serving the underserved and improving the outcomes of the most vulnerable. You can’t pick a more vulnerable group than the homeless; these are individuals who count for a minority of the population but disproportionately use the healthcare system, and also have a much higher rate of mental health, substance abuse and criminal justice issues, and chronic disease. If we’re trying to get healthcare under control and improve the lives of New Yorkers, this is a group we really need to focus on. That’s obvious, and that was the impetus for us taking this [work] on.

Patient matching can certainly improve across the U.S., and for the homeless population specifically, it must be incredibly challenging, right?

The first thing we wanted to do from a technical standpoint is just find out who is homeless. That’s very rarely codified; there is such a thing as a diagnosis for homelessness, but it’s not reliable. So we first scoured the public records for the addresses of homeless shelters, places of worship, and government buildings that might be places where the homeless could provide as an address. When you have a medical record, you have an address, and it may not be a very good address, but it’s something that comes across the feed into the HIE.

We also went out to our participants; we have a number of [members] who focus primarily, or exclusively, on housing and homelessness, along with healthcare, and they helped us identify the addresses of these places. One of them, notably, was the HIV/AIDS Services Administration (HASA) in New York City. We also work closely with the New York City Department of Homeless Services (DHS).

Finally, we did analysis of our MPI, the technology engine used to link patient records across the 20 million people registered with Healthix. We did a frequency analysis; in any given address, you’d expect to find somewhere between 1 and 10 people, at the most, living at a single address. When we found an address that came back with a frequency of 100 people or more living there, we knew it wasn’t a home, and typically the addresses given were peoples’ home addresses. So we were able to significantly augment our list of addresses that we call “proxies for homelessness.” We put hundreds of addresses into this from various sources, and if any patient in Healthix registered at any one of our facilities with one of those addresses, we tagged that as “homeless”—meaning that encounter is from a person that was homeless at that time.

What have been your biggest lessons learned?

We worked with Verato for the MPI to do various analyses on these addresses and patients that were identified as homeless, so that we could do some cross-referencing. One of the key value propositions of the Verato engine is that they maintain several large public databases that give addresses on individuals, including credit cards, utilities, and public government records so that you can cross reference a person’s address with these databases. The reference database is a single curated database that includes over 300 million identifiers from a variety of data partners and is curated for accuracy confidence by Verato. The enterprise master patient index is unique to each customer and interacts with the reference data and their private data to augment records for improved matching.

So if you have a person who goes to two or three different healthcare facilities and gives a different address for each one, if we can find them all tied to a common individual across all those databases, then we can do a match. So the Verato system did a pretty good job matching those patients using that data source.

We’re also in discussions with Verato to use the proxy addresses to further augment their matching. That’s not part of the algorithm they use since it’s very specialized, but our goal with them is that in the future, we will be to take two or more addresses that are both proxies for homelessness, and assuming all the other demographic data matches for the patient—such as date-of-birth, name, phone number, and Social Security Number—and the addresses are both proxies for homelessness, we believe that could strengthen the algorithm even more. That takes a lot of analysis, because one of the things we don’t tolerate in patient matching is a false positive. If you have a false positive you can blend health records together for two people, and that can have adverse effects for their care. You need to be at the 99.999 percent accuracy [level].

How has this work evolved over time?

One thing we have done is establish an advisory council. We have quite a few organizations who already have been very involved with homeless services and issues for a long time. One takeaway has been that the status of homelessness is not exactly black and white; you can be homeless today and housed tomorrow, so you have to distinguish between current homelessness, a history of homelessness, and also “unstably housed” which [usually] means having more than three addresses in a 12-month period. All three of those are important pieces of information that healthcare providers need to know about.

One thing I have learned is that it’s very difficult for people who are chronically homeless to get them into steady housing, even when you give them an apartment. And that’s because when you give a homeless person [an apartment], they might be required to pay the utilities, agree to some monthly review, or come in and see a counselor. People who are homeless oftentimes don’t want to be part of any kind of structure. It’s a complicated problem—you might think homeless people want a place to stay, but sometimes the tradeoffs are such that they don’t actually want that.

How will Healthix plan to measure success for this initiative?

We have developed an alert, and alerting is our strongest services we provide in the community—essentially letting our providers know when their patients are going to the hospital, get released, or maybe they [end up] in jail or a nursing home, or even that they died. With COVID-19, we are letting providers know when patients are tested and what the results were.

Now that we have a flag for homelessness, we can alert people when one of their patients has become homeless, is living in a shelter, or that his or her status has changed. We have the opportunity to be more nuanced and also tell them when a person has had history of homelessness or is unstably housed. So the first way to measure [success], and the most straightforward way, is to implement the homeless alert with as many Healthix participants as possible, and then we can easily quantify how many alerts are being sent. Then we can interview the people who receive them and get a qualitative analysis of what they have done with it and how it’s benefitted their work.

Homelessness is clearly complex, and the [stakeholders] involved are very diverse—including social services, health services, mental health services, substance abuse services, and the criminal justice system. In order to manage the situation, we need to bring all those [entities] together, and Healthix is in a unique position to do that.