Homelessness is a critical social determinant of health, driving disparities in healthcare utilization, morbidity and mortality. In 2018, the Canadian Institute for Health Information (CIHI) mandated the coding of homelessness in hospital administrative data, which more than doubled case identification. However, 25 % of cases remain undetected, and two-thirds of flagged patients were not currently homeless, though they have a documented history of homelessness. We summarize recent evidence and present opportunities for CIHI and health systems to further improve the accuracy of homelessness coding in Canadian hospital administrative data, which would enhance its utility for health research, policy making and health system planning.

Introduction

Homelessness is a critical social determinant of health with profound implications for morbidity and mortality. In Ontario, nearly 30% of people experiencing homelessness fall within the top 10% of most costly healthcare patients (Wiens et al. 2021a), and per-person costs are six times higher than for housed individuals of the same age and sex, even after adjustment for morbidity, mental health and substance use (Richard et al. 2024a). Yet, Canadian hospital administrative databases have historically identified only a very limited proportion of this population (Richard et al. 2019).

Two interrelated factors contribute to this gap. First, whether homelessness is known and documented in patient records depends heavily on individual clinician documentation practices. Second, the International Statistical Classification of Diseases and Related Health Problems, 10th Revision, Canada (ICD-10-CA) code for homelessness (Z59.0), like many Z-codes, typically does not meet the Canadian Institute for Health Information's (CIHI) criteria for mandatory coding (being the most responsible diagnosis or being a documented contributor to increased length of stay or intensity of care) (CIHI 2004). This left the coding of homelessness largely optional and at the discretion of health information management (HIM) professionals. Therefore, homelessness health research has historically had to rely on high-quality but costly studies that recruited representative samples of homeless individuals from the community (Hwang et al. 2011; Richard et al. 2022).

In 2018, in response to recommendations from healthcare and policy experts, the CIHI mandated the coding of ICD-10-CA code Z59.0 wherever present on the patient chart (CIHI 2023). This change led to a dramatic increase in the identification of patients experiencing homelessness across Canada (De Prophetis et al. 2023). However, data quality following the mandate was unknown.

This paper describes results of recent validation work assessing the impact of this coding mandate on homelessness coding quality in Ontario, summarizes persistent data quality issues identified and highlights opportunities to further improve the utility of these data for research and policy making.

Methods

As this paper highlights published data, detailed methods are described elsewhere (Richard et al. 2024b).

Briefly, we linked prospective cohort data from a randomly recruited representative group of people experiencing homelessness in 2021 and 2022 in Toronto to ICES data holdings (Schull et al. 2020). Full sampling and recruitment details for this cohort are detailed in the cohort study protocol (Richard et al. 2022). We compared periods of known housing status (classified as homeless or housed) to encounters in databases with homelessness indicators, including the Discharge Abstract Database (DAD), National Ambulatory Care Reporting System (NACRS) and Ontario Mental Health Reporting System (OMHRS). We developed homelessness case definitions combining various datasets, indicators (including but not exclusive to Z59.0) and grace periods between the housing episode and healthcare encounter and compared these to true homelessness status during each housing period. Accuracy of each definition was measured by the percentage of true homelessness encounters identified (sensitivity), the percentage of true housed encounters identified (specificity) and the percentage of encounters flagged as homeless that were correctly identified (positive predictive value).

This study was approved by Unity Health Toronto's Research Ethics Board (REB# 20-272).

Results

The best performing definition (“any homeless indicator in DAD, NACRS or OMHRS during or within 180 days of a housing episode”) demonstrated a sensitivity of 52.9%, specificity of 99.5% and positive predictive value of 36.2% (Table 1). By contrast, when focusing on housing periods with at least one healthcare encounter (which is necessary for identification to be possible, i.e., “patients experiencing homelessness”), the same definition had a sensitivity of 75.1%, specificity of 98.5% and positive predictive value of 35.9%.

In comparison to the equivalent case definition assessed pre-mandate (Richard et al. 2019), which considered the same set of homelessness indicators, the sensitivity of the case definition more than doubled from 25% to 53%, while positive predictive value decreased from 51% to 36% (Figure 1). About 75% of the false positives were previously homeless, suggesting indicators may not distinguish between current and past history of homelessness.

Finally, Figure 2 presents characteristics associated with a significantly greater risk of misclassification as either a false positive or a false negative. Housed people under 30 years of age were more often misclassified as homeless (false positives). Conversely, people assigned female at birth, who identified as Black, without a mental health or substance use disorder, or who experienced a shorter duration of homelessness, were more often misclassified as housed (false negatives).

Implications and Opportunities

High-quality homelessness data from hospital administrative databases present major opportunities for cost-effective, high-impact research. Since the mandate, these data have been used to assess disparities in COVID-19 pandemic outcomes (Richard et al. 2021), diabetes care (Wiens et al. 2024), dementia (Booth et al. 2024) and cold exposure-related injuries (Richard et al. 2023), with many more studies forthcoming. Public health units, health system leaders and policy makers are also increasingly appreciating the value of these identifiers for improving patient care and public health planning.

However, although sensitivity has clearly improved, persistent issues in the quality of homelessness indicators remain, which may limit their utility. One in four patients experiencing homelessness continues to not be identified, disproportionately affecting people who are unhoused for shorter periods, women, people who identify as Black and those without mental health or substance use disorders. It is also concerning that two-thirds of patients who were identified as homeless were actually housed, although 75% of these were individuals with a history of homelessness.

Although administrative data are by definition imperfect, the following strategies have been identified in recent collaborations as opportunities to enhance data accuracy (and, by extension, utility).

Data integration across care settings

Integrating data across care settings within an episode of care can significantly enhance the identification of homelessness. At St. Michael's Hospital in Toronto, cross-referencing emergency department records against in-patient admission data revealed hundreds of additional cases of homelessness annually (L. Veta, written communication, May 2024). In response, this hospital's decision-support team has implemented quarterly reviews to add identifiers identified during care transitions. Such routine reviews are potentially feasible at most large hospitals and/or academic health science centres having a dedicated team or staff focused on data quality but may be difficult or impossible where such dedicated supports are unavailable. CIHI could also (or alternatively) enhance identifiers across episodes of care during routine data cleaning and standardization activities.

Incorporating no fixed address and shelter addresses as homelessness indicators

Address fields completed during registration or triage offer key indications of homelessness. For example, “no fixed address” or emergency shelter addresses are often used for patients experiencing homelessness (Hayes 2023). Despite this, analysts and decision support staff from various hospital networks (including St. Michael's Hospital, University Health Network, London Health Sciences Centre and William Osler Health System) have all observed that these address-based indicators do not consistently result in a Z59.0 code, because HIM professionals are not required to review these during coding (verbal and written communications, February to July 2024).

To address this gap, hospital networks could develop locally relevant lists of address-based indicators of homelessness (e.g., permanent emergency shelters) and develop procedures to flag such records in electronic health record (EHR) data for the benefit of both clinicians and HIM professionals. This, as with the prior recommendation, would be most easily implemented by hospitals or academic health science centres with dedicated data quality teams or staff. Unfortunately, CIHI does not receive patient addresses, making it impossible for them to assist with this work.

Routinely screening for housing status in EHR systems

Most EHRs (including the three most used EHRs in Canada: Meditech, Epic and Cerner) allow for customization, offering opportunities to introduce a structured field to routinely collect housing status. For example, Unity Health Toronto recently moved to Epic and began collecting social determinants of health data, such as gender and racial identity (Jingco 2024). In many other settings, such a collection is already routine (Richwine et al. 2025). Screening for housing status provides a vital opportunity to reduce missed cases (e.g., patients who are couch surfing and may not view their situation as homelessness) and update outdated information when patients become housed, particularly for groups disproportionately misclassified as homeless. This screening could replace or complement address-based identification.

One key barrier to screening, aside from the technical challenges of customizing the EHR to include housing question(s), is fostering a hospital culture that avoids inadvertently stigmatizing patients who disclose their homelessness. In many hospital settings, fear of being dismissed or receiving substandard care is an important factor that discourages patients from reporting their housing status (Reilly et al. 2022). Patient partners with lived experience of homelessness must be involved in the design and implementation of any screening question to ensure that appropriate language and procedure steps are used. If recommended by these experts, new or additional implicit bias and trauma- and violence-informed care training should also be embedded into hospital-wide education efforts.

Introducing a “history of homelessness” code

Finally, current evidence that most false positives have a history of homelessness suggests two additional gaps may be present, each having a similar coding-related solution. First, health professionals may be documenting past homelessness on patient records due to its relevance to current healthcare needs. As the mandate did not distinguish between current and past homelessness, this situation forces HIM professionals to code Z59.0 even if not fully appropriate. Second, beginning in 2022, CIHI directed HIM professionals to code for homelessness based on the Canadian Observatory on Homelessness definition, which includes individuals at risk of homelessness.

Advocacy following the most recent validation study and continued queries from HIM professionals has resulted in CIHI issuing a bulletin in March 2025 around the coding of homelessness. This bulletin, among other things, directs HIM professionals to not code Z59.0 when the patient is only at risk of homelessness or has a history of homelessness (CIHI 2025).

This is a very welcome step, which should reduce false positives from both aforementioned gaps. However, CIHI should also consider introducing an additional ICD-10-CA code for “history of homelessness.” This code would allow researchers and policy makers to better understand the lasting effects of homelessness on health, separately from the effects of current homelessness. Research shows that a history of homelessness impacts healthcare utilization long after a person has secured housing (Wiens et al. 2021b), and individuals who exit homelessness remain vulnerable to future episodes of homelessness (Espinoza and Randle 2025). Hospitals and researchers have a vested interest in identifying and quantifying this particular group of patients to better understand their contribution to local healthcare utilization and targeted interventions that may improve care and reduce costs. A suitable ICD-10-CA code for this purpose might be Z91.9, currently unassigned but available in the class of codes described as “Personal history of risk factors.”

Conclusions

CIHI's coding mandate demonstrated the transformative potential of policies aimed at improving the capturing and quality of homelessness data in Canadian hospital administrative databases. Major gaps in quality remain, however, and strategies such as data integration, address-based coding, routine screening and the introduction of more granular coding would build upon progress to maximize the value of these data for researchers, hospital administrators and policy makers.

Accurate, comprehensive and nuanced housing data are critical to addressing health disparities and improving the health of unhoused populations. By adopting these recommendations, CIHI and Canadian health systems can help elevate the visibility of this highly marginalized population and advance the broader goal of achieving health equity in Canada.

Declarations

Funding statement: No funding supported this submission.

Conflicts of interest: None to declare.

Ethics approval: Not applicable.

Consent to participate: Not applicable.

Consent for publication: Not applicable.

Availability of data and material: Not applicable.

Data Sources and Permissions

Data in this paper were drawn from previously published work: Richard et al. (2024b).

The lead author wrote this paper, and the broader study team permitted the reuse of these data.

Correspondence may be directed to Lucie Richard by e-mail at lucie.richard@unityhealth.to.

Possibilités d'améliorer la codification pour l'itinérance dans les données administratives des hôpitaux canadiens

Résumé

L'itinérance est un important déterminant social de la santé, qui entraîne des disparités en matière d'utilisation des soins, de morbidité et de mortalité. En 2018, l'Institut canadien d'information sur la santé (ICIS) rendait obligatoire la codification pour l'itinérance dans les données administratives des hôpitaux, ce qui a plus que doublé le repérage des cas. Cependant, 25% des cas ne sont pas détectés, et deux tiers des patients signalés ne sont pas actuellement sans abri, bien qu'ils aient un historique documenté en la matière. Nous résumons les données récentes et présentons les possibilités pour l'ICIS et les systèmes de santé d'améliorer davantage la précision de ce type de codage dans les données administratives des hôpitaux canadiens, ce qui en favoriserait l'utilité pour la recherche, pour l'élaboration de politiques et pour la planification des systèmes de santé.

Comments

Be the first to comment on this!