Pediatric Emergency Department Visits for Homelessness After Shelter Eligibility Policy Change

This is a retrospective study of homeless children presenting to the ED of an urban tertiary-care children’s hospital in Massachusetts. We included children and adolescents ages 0 to 18 years old presenting to the ED from March 2010 to February 2016 with a chief complaint of homelessness or with a disposition affected by homelessness. For purposes of this study, being homeless was defined as having nowhere to stay on the night of presentation and requiring emergency shelter in the ED. Patients were included in this study even if they presented with a medical complaint but, during the ED visit, were identified as being homeless. We excluded patients who were housed in a shelter at the time of ED presentation because the goal was to identify only children whose primary reason for the visit was homelessness. This study was approved by our hospital’s institutional review board.

A natural language processing (NLP) tool, DrT,¹⁵ was used in several phases to search the ED physician electronic medical record (EMR) documentation during the study period to identify patients who met inclusion criteria. An NLP is a tool that allows for searching text by keywords¹⁶ and also includes an iterative algorithm used to identify similar text that does not include the keywords but has other shared words or phrases.^17,18 Because the International Classification of Diseases, Ninth Revision, Clinical Modification and the International Classification of Diseases, 10th Revision, Clinical Modification codes for homelessness (V60.0 and Z59.0, respectively) were not consistently used, diagnosis code alone would not have adequately identified this patient population.

In the first phase of record identification, the NLP tool was used to search all the ED EMRs from March 2010 to February 2016 to identify records with the following key terms: “homeless,” “had been living/staying,” “DHCD,” “do not have housing,” “evict,” “housing office/authority,” “housing placement,” “no place/nowhere to stay/sleep,” “ICD-9 v60.0,” and “ICD-10 z59.0.” These records were manually reviewed by the initial reviewer and categorized as “included” or “excluded” for the study definition of homelessness. Any equivocal cases were reviewed by a second reviewer and only included in the study if both reviewers agreed on inclusion.

In the next phase of EMR patient identification, the NLP tool was further refined by comparing the included and excluded records. The tool was then used to search the remainder of the ED records (ie, those not containing key search terms) from March 2010 to February 2016 and assign a score to each EMR on the basis of how each individual record was found to be similar to the manually characterized “included” charts in the program. These EMRs were then manually reviewed to determine a score low enough to ensure no cases were likely to be missed. This iterative process was repeated until no new cases were identified by using the NLP program. Finally, the records identified from the EMR were compared with social work paper records (when available) to ensure that all known homeless patients during the study period were identified. In total, 336 228 records were electronically reviewed by using the NLP program, 1168 were manually review by the study team, and 312 records met inclusion criteria. This process had an accuracy of 95.1% (95% confidence interval [CI]: 92.9 to 97.4) per the NLP program statistical analysis.

The primary outcome was the number of homeless children presenting to the ED before and after the regulation. Social work and physician documentation were reviewed to extract the following variables: date of ED visit, ED length of stay (LOS), age, sex, race, ethnicity, type of insurance, if siblings were present, which family member was present with the patient, if this was a repeat visit for homelessness, whether the family returned to the ED for any reason within a month of their visit for homelessness, whether the family had previously been seen at the Massachusetts housing office (DHCD), whether a medical complaint was present, chronic medical conditions, reason for homelessness, and where the family was staying before the ED visit. Facility and professional payments for the ED visit were also extracted. These variables were extracted into a Research Electronic Data Capture database compliant with the Health Insurance Portability and Accountability Act for analysis.

To compare the number of ED visits for homelessness to the number of homeless children in Massachusetts, the Department of Education McKinney-Vento annual survey was used to determine the number of homeless school-aged children in Massachusetts.¹⁹ To extrapolate this to the total number of homeless children in Massachusetts, we multiplied the number of school-aged children by 2.04 (100 divided by 49) to match a commonly cited methodology that is based on the most recent estimate that 49% of homeless children are ≥6 years old.³ Given that McKinney-Vento numbers are reported by school year, we compared them to ED visits by school year (July to June), rather than calendar year, and excluded years in which we did not have full school-year data on ED visits.

Frequencies of demographic and ED visit variables were calculated. Patient hours were calculated by adding the ED LOS for each included patient. Rate ratios (RRs) with 95% CIs were analyzed for changes in specific characteristics of children presenting to the ED for homelessness over the study period. Rate differences with 95% CIs were calculated to compare circumstances preceding homelessness. Because the DHCD regulation change was implemented on September 17, 2012, patients seen from March 2010 (date of EMR storage format change) to September 17, 2012, were analyzed as the prepolicy group, and those seen from September 18, 2012, to February 2016 were analyzed as the postpolicy group. The interrupted time series incidence RR with 95% CI was calculated to examine visits over the study period. All analyses were conducted by using Stata version 13.0 (Stata Corp, College Station, TX).

During the study period, there were 312 ED visits for homelessness; 95% (n = 297) of visits were after the 2012 policy (Fig 1). The demographic characteristics of the homeless children are presented in Table 1. The median LOS increased from 6 to 13 hours between the pre- and postperiod groups (difference: 7.0; 95% CI: 4.27 to 9.73), and the maximum LOS increased from 17 to 93 hours. This resulted in a total of 4834 ED patient hours for homeless children during the study period.

Sixty-five percent (194 of 297) of children presenting to the ED for homelessness after the policy had no medical complaint, compared with only 20% (3 of 15) of patients who presented before the policy change (RR: 3.27; 95% CI: 1.18 to 9.01). Nearly all the medical complaints were low acuity, with wheezing and/or bronchiolitis (2.2%) being the only condition likely requiring ED management. After the policy, only 26.3% (78 of 297) of the homeless children had any chronic medical condition, compared with 53.3% (8 of 15) of homeless children before the policy (RR: 0.49; 95% CI: 0.30 to 0.82). Twenty-nine children (9.1%) post-policy had a repeat ED visit for homelessness and 20 children (6.4%) had return visits within a month for medical complaints. Forty-eight percent of homeless children seen after the policy change had already been seen at the housing office within the 2 weeks before their ED visit, compared with 13% of homeless children before the policy. Families reported multiple reasons for becoming homeless (Table 2).

The overall rate of visits for homelessness per month increased >4 times from the pre- to the postpolicy period (incidence RR: 4.5; 95% CI: 1.33 to 15.23; Fig 2). This increase is not explained by an increase in ED volume. The number of homeless children in Massachusetts did increase over the study period; however, the number of children presenting to the ED for homelessness rose out of proportion to that increase. During the 2010–2011 school year, there were 14 247 homeless schoolchildren in Massachusetts, and during the 2014–2015 school year (the last year with full ED data), there were 19 515 homeless schoolchildren in Massachusetts.²⁰ Thus, over the study period, the number of homeless children in Massachusetts increased 1.37 times, whereas the number of homeless children in the ED increased 13.2 times (Fig 3).

During the study period, 90% of children were insured by state-based health insurance plans (Table 1). The total payments for ED visits for homelessness (adjusted to 2016 dollars) were $173 950 over the study period, with an average payment per visit of $557 (Table 3). Most payments ($159 269) occurred after the policy. A large portion (89%) was paid by state-based insurers ($155 149). In the prepolicy period, 10.7% of payments were for patients with no medical complaint; this increased to 61.2% of payments after the policy. Payments for visits with no medical complaint amounted to $99 092, with $97 514 occurring after the 2012 regulation. Although private insurance payments only increased 1.9 times in the postpolicy period compared with the prepolicy period, payments by MassHealth (Massachusetts state–based Medicaid and Children's Health Insurance Program health insurance) and state-based–managed care organizations increased 11.6 and 31.2 times, respectively.

A 2012 emergency shelter policy change drove many homeless families to spend a night in a place “not meant for human habitation” for eligibility, which resulted in a dramatic increase in the number of pediatric ED visits for homelessness. This was not accounted for by changes in ED volume or increases in homeless children in Massachusetts during this time. Additionally, the children seen for homelessness after the policy were more likely to have no chronic medical conditions or medical complaints during their visit, meaning they came to the ED only to seek shelter. Thus, these visits were for healthy children who would not otherwise be presenting to the ED, and their associated costs should be considered preventable. As a result of the policy, there was likely increasing awareness by families, pediatricians, and housing advocates of the option of families to stay in the ED for shelter eligibility. Since the policy was implemented, 2650 families have entered emergency assistance shelters by meeting this requirement.²¹ 

Massachusetts has a favorable climate for housing-insecure and homeless families, including the “right to shelter” policy for eligible homeless families,¹² but housing prices are high. In Massachusetts, 24% of households pay more than 50% of their income for rent, and a family must make >$24 per hour to afford rent for a 2-bedroom apartment.⁴ These factors likely contributed to the overall increase in emergency assistance shelter system funding from $143 to $193 million between 2013 and 2016.²² Thus, 1 goal of the 2012 “not meant for human habitation” regulation was reducing costs by making eligibility requirements more stringent.²³ 

Many families reported that they were staying in cars, on the streets, or in other places “not meant for human habitation” before their ED visit, and nearly half of children seen for homelessness after 2012 had already been seen in the housing office before their ED visit. These places should have fulfilled the eligibility requirement but may have been more difficult to prove and are physically less comfortable and potentially less safe than staying in the ED.¹³ These data reveal how more stringent eligibility requirements can lead to barriers in obtaining housing and create burdens on the health care system. In addition to affecting the health care system, unnecessary ED visits also put children at risk for nosocomial infections and injuries, lost parental wages, and exposure to the hectic environment of the ED.

The majority of ED visits for homelessness were paid for by state-based insurance companies. With an average payment of $557 per ED visit, these visits cost >4.8 times a night in an emergency assistance shelter, which averages $117 per night in Massachusetts.^23,24 Thus, these visits incur additional costs to the taxpayers, but they are counted as health care rather than housing spending in an era when controlling health care costs is critical. Given the increase in family homelessness in the United States,³ other states may face similar budgetary challenges while trying to care for these families and should take into account the potential unintended health care costs (especially for children without a medical need) when evaluating housing policies.

Not only did the number of children seen in the ED for homelessness increase but so did their ED LOS. ED boarders, such as patients requiring admission who are waiting for inpatient beds, are known to lead to treatment delays and increases in ED crowding and left without being seen rates.^25,26 Given that homeless children accounted for 4834 patient hours over the study period, they potentially affected patient flow and the management of other patients in the ED. To address this, when the volume of homeless children presenting to our ED began increasing, we developed processes to minimize the impact of the homeless families on the ED patient flow. This included rooming homeless children in the ED only late in the evening or overnight, once the ED volume had slowed.

Massachusetts is currently considering legislation that would seek to change the emergency shelter eligibility requirement to being “within 24 hours” of spending a night in a place not meant for human habitation.²⁷ This would allow families to become eligible for shelter because of an imminent risk of homelessness without the documentation of staying in a place “not meant for human habitation.” Critics of these proposals are concerned that this would cause a large increase in the number of families eligible for shelter at a time when the state is focusing on reducing the number of families in emergency assistance shelters.²³ In contrast, proposal advocates argue that it would bypass the requirement to stay in an unsafe location or in an ED while only adding 1 night to each family’s emergency assistance shelter stay.

Our study has some limitations. During the case identification phase, it is possible that we missed cases or that there was ascertainment bias; however, we attempted to reduce this possibility by running multiple iterations of the NLP program, comparing with social work paper records when available. It is also unlikely that there was any systematic difference in the number of missed cases in the pre- and postpolicy periods. Additionally, to compare the number of homeless children seen in the ED with the total number of homeless children in Massachusetts, the Department of Education McKinney-Vento survey data were used, which are limited to school-aged children. In the McKinney-Vento survey, children who live in homes where they are doubled up or moving frequently are also considered to be homeless, which varies from our inclusion criteria for homelessness. However, these data are considered the most reliable counts of homeless children by state, and because we only used these data to compare overall trends, this was the best approach to evaluate changes in the population of homeless children over time. We used a commonly cited calculation from the Horizons for Homeless Children’s “America’s Youngest Outcasts”³ to determine the total number of homeless children in Massachusetts for comparison on the basis of the best available data regarding the age distribution of homeless children in the United States. Generalizability is also a limitation because Massachusetts is the only state with a “not meant for human habitation” shelter policy at this time; however, another Massachusetts pediatric ED has seen similar increases in visits by homeless children, suggesting this issue is not limited to our ED.²⁸ 

There was a significant increase in the number of children presenting to the ED for homelessness over our study period after a regulation change in emergency assistance shelter eligibility. This resulted in considerable associated costs to the health care system, most of which were borne by state-based insurance plans. It is important to understand the magnitude of the problem of child homelessness, the characteristics of children being seen for homelessness, the reasons for families’ homelessness, and the unintended health care costs to enable legislators and policy makers to create more effective policies for these families in Massachusetts and throughout the United States.

CI

confidence interval

DHCD

Department of Housing and Community Development

ED

emergency department

EMR

electronic medical record

LOS

length of stay

NLP

natural language processing

RR

rate ratio

We thank Mai-Han Nguyen-Thanh and Christopher Shin for their assistance with the McKinney-Vento data and Dr Michael Monuteaux for his assistance with the statistical analyses.