New York City has the largest homeless population in the country, an issue the city has tackled through prevention, affordable housing, and health care initiatives. At the safety-net health system for the city, New York City Health + Hospitals, which serves more than one million people a year, identifying and helping to care for homeless New Yorkers is a part of our transformation efforts. Using data science to identify and “phenotype” our homeless patients helps us tailor their care and match them to the right hospital and community-based supports—ultimately including housing itself.
Ideally, health systems should identify homeless patients at the point of care, since homelessness can affect a person’s ability to access services, take medications, and administer self-management (for example, monitoring blood glucose while taking insulin). However, in many big systems like ours, capture of this information is inconsistent and often in unstructured fields across fragmented data systems. Most of New York City’s homeless people reside in city shelters, some live on the streets, and some are temporarily doubled up with family members or couch surfing with friends. Fearing stigma, homeless patients may not disclose their homelessness to clinic staff or may convey it discreetly, for example by providing a proxy address such as a shelter as their home address.
To better characterize the homeless patients we see throughout NYC Health + Hospitals, we talked to our colleagues – including social workers, emergency room doctors, and homeless service providers – about proxy indicators for homelessness and where this information is documented in our vast and disparate information systems. They told us it could be found in registration documents (basic information on the patient such as name, address, and insurance), electronic medical records, and insurance claims. To capture as many patients as possible, we created a composite definition of homelessness by:
- Matching addresses for each of our one million patients to homeless shelters and our own hospitals
- Searching for the words “homeless,” “undomiciled,” or “shelter” anywhere in the address fields
- Flagging patients whose home zip code changed 10 or more times in one year
- Pulling records with a “homeless” flag from registration at the few facilities that record this
- Searching for the diagnosis code for homelessness on the clinical problem list, other diagnostic assessments, or in the billing data
While this search misses unstructured data like narrative social-work assessments, mining homeless status from our data systems flexibly across multiple domains captured more than 20,000 adult homeless patients served within one year. This enabled us to understand our homeless patients as a population, across clinical diagnoses, utilization, demographics and other factors. They come to us disproportionately through the emergency department, have longer and more frequent hospital stays, and experience higher rates of behavioral health issues such as substance use, mood disorders, and schizophrenia. Specifically, we found that they were nine times more likely to visit the ER or be admitted that our average patient. This finding was so substantial that we incorporated the homeless flag into our overall score predicting risk of high hospital use, which we are deploying to front line providers via the electronic medical record.
Categorizing our homeless patients also revealed important sub-groups, reminding us that homelessness is not a homogeneous experience. For instance, those living in family shelters had lower rates of chronic disease and were more likely to use primary care than other homeless patients, suggesting that a tailored version of our current care management programs for this group could be effective. Our presumed street homeless patients, on the other hand, were three times as likely as those in family shelters to have a behavioral health diagnosis and over eight times as likely to be diagnosed with schizophrenia, requiring more intensive services.
At NYC Health + Hospitals/Bellevue, which has the largest number of homeless patients of the hospitals in the system, we opened an outpatient complex-care clinic that offers flexible walk-in appointments and longer visits, which are required to better address overlapping health and social needs. The clinic also has dedicated staff knowledgeable about the issues homeless patients struggle with, such as substance use disorder and the need for assistance navigating the public housing system.
At NYC Health + Hospitals/Lincoln in the Bronx, one of the busiest hospitals in our system, the adult medicine clinic started a pilot program that identifies patients who are homeless or at risk of eviction and connects them to resources including food benefit programs, legal help, health insurance navigation, public shelters, and local housing organizations. Addressing housing instability places patients in a better position to engage with their clinical team and avoid preventable emergency room use and hospital stays. Since the cost of one night in the hospital is more than a typical month’s rent for a one-bedroom apartment in New York, investing in services for a group known to need a lot of inpatient care makes financial as well as clinical sense.
As housing is an essential component of healthcare for these patients, we use our data to compile lists of patients who meet eligibility criteria for local supportive housing complexes, such as people living with HIV/AIDS or with behavioral health diagnoses, and reach out to assist with housing applications. We also work with sister New York City agencies to match our high-risk homeless patients to a city housing agency database to verify approved housing applications and identify needed to be followed up. We also convey profiles of homeless patients shared with managed care plans such as Metroplus, New York City’s “public option,” so that they can be active partners in care management and support discharge planning for inpatients.
Given the links between homelessness and health, all health systems should work to address the underlying causes of illness, including lack of housing. The first step is identifying homeless patients and understanding how they intersect with our system; data science helps us meet patients where they are and connect them with health care and social supports on their own terms.