OBJECTIVES:

To evaluate the association between caregiver-reported social determinants of health (SDOH) and emergency department (ED) visits and hospitalizations by children with chronic disease.

METHODS:

This was a nested retrospective cohort study (December 2015 to May 2017) of children (0–18 years) receiving Supplemental Security Income and Medicaid enrolled in a case management program. Caregiver assessments were coded for 4 SDOH: food insecurity, housing insecurity, caregiver health concerns, and safety concerns. Multivariable hurdle Poisson regression was used to assess the association between SDOH with ED and hospital use for 1 year, adjusting for age, sex, and race and ethnicity. ED use was also adjusted for medical complexity.

RESULTS:

A total of 226 children were included. Patients were 9.1 years old (SD: 4.9), 60% male, and 30% Hispanic. At least 1 SDOH was reported by 59% of caregivers, including food insecurity (37%), housing insecurity (23%), caregiver health concerns (18%), and safety concerns (11%). Half of patients had an ED visit (55%) (mean: 1.5 per year [SD: 2.4]), and 20% were hospitalized (mean: 0.4 per year [SD: 1.1]). Previously unaddressed food insecurity was associated with increased ED use in the subsequent year (odds ratio: 3.43 [1.17–10.05]). Among those who had ≥1 ED visit, the annualized ED rate was higher in patients with a previously unaddressed housing insecurity (rate ratio: 1.55 [1.14–2.09]) or a safety concern (rate ratio: 2.04 [1.41–2.96]).

CONCLUSIONS:

Over half of caregivers of children with chronic disease enrolled in a case management program reported an SDOH insecurity or concern. Patients with previously unaddressed food insecurity had higher ED rates but not hospitalization rates.

Children with chronic disease (CCD) are pediatric patients experiencing persistent medical conditions that last beyond 1 year and are usually lifelong.1  CCD with progressive disease or the presence of ≥2 significant chronic conditions that require ongoing treatment are labeled as having “complex” chronic disease, whereas CCD who may have periods of good health (eg, type 1 diabetes) are labeled as having “noncomplex” chronic disease. As a population, CCD, especially those with medical complexity, consume a disproportionate amount of acute health care, such as emergency department (ED) visits and hospitalizations.24  Understanding the factors that might mitigate use in this population remains a challenge for health care systems, particularly how best to keep children out of the hospital and at home.5 

Social determinants of health (SDOH), such as income, language, and race and ethnicity, are linked to health outcomes of children.610  In previous research, it is suggested that clinical care may contribute only as much as 20% to patient health care outcomes compared with 40% contributed by SDOH.11  SDOH-related problems, such as adverse food insecurity and housing insecurity, are increasingly being recognized as affecting medical health care use, including hospitalization.1218  In previous research, it is also suggested that systemic social disparities create barriers to health care access for CCD.19 

Case management (CM) is 1 approach to improving care and decreasing acute health care use for CCD. CM is a process performed by a health professional to track and help patients get needed services, including social services.20  Screening for SDOH-related problems may help to identify CCD who would benefit from more-intensive CM. Given that CCD are more likely than their healthier peers to use health care, identification and mitigation of SDOH may have an even larger impact in the CCD population.

Pediatric Partners in Care is a program created to provide community-based CM to CCD living in a major metropolitan area in the Pacific Northwest, coordinated through the local freestanding children’s hospital. We recognized an opportunity with this program to conduct this study to evaluate the association between exposure to SDOH-related problems (eg, home and food insecurity) and subsequent ED and hospital use within the program’s population.

The program’s objective is to improve the health outcomes for CCD while also mitigating their costly acute health care use. Enrollment criteria for the program included the following: (1) residence in either 1 of 2 counties within the study’s metropolitan area; (2) receiving Supplemental Security Income and Medicaid (ie, as proxy for medical complexity or disability); either a (3a) Predictive Risk Intelligence System21  score of >1.0 (a Medicaid decision tool to select patients for care coordination) or (3b) provider request, typically because of social concerns (eg, low health literacy, multiple households); and (4) age of 18 years or younger at time of enrollment.

Enrolled patients and families were assigned a paired registered nurse case manager and nonclinical care coordinator, who had undergone program-specific training. Case managers had backgrounds in primary care, mental health, and/or public health, and coordinators had backgrounds in parenting support, health education, school health, and/or community resource navigation. Three coordinators were certified in Spanish proficiency. Each complementary case manager and coordinator pair was assigned on the basis of family-identified need. For example, a family needing help with a school-individualized education plan was assigned to a coordinator with school-based experience, whereas another family with low health literacy was paired with a coordinator who had a background in health education.

A case manager completed the intake assessment during patient enrollment. The intake assessment was a nonvalidated survey that included open-ended questions about communication preferences and SDOH thought to impact the patient’s care or family functioning; for consistency, coordinators were trained in the use of standardized prompts.

If identified in the assessment, food insecurity was addressed by providing families with information and application assistance with programs, such as the Women, Infant, and Children Program or Supplemental Nutrition Assessment Program. Housing insecurity also was addressed by providing help applying for housing assistance programs. CM for caregiver health or a safety concern was tailored on the basis of the specific elicited concern. For example, family caregivers who indicated emotional strain would receive mental health information. A caregiver with intimate partner violence was referred to safe harbor resources.

This was a nested retrospective cohort study of children enrolled in the CM program. Data for the study were collected from the following: (1) Medicaid claims data; (2) hospital demographic and clinical data; and (3) primary caregiver intake assessments created at patient enrollment.

The hospital’s institutional review board approved all study procedures.

Within the CM program population, we applied additional inclusion and exclusion criteria. Patients had to have a completed caregiver care coordination intake assessment created for the program from December 15, 2015, to May 27, 2016, and for whom insurance claims and eligibility files were collected for at least 10 months over the course of a 12-month frame after enrollment. Any months with missing claims data were censored. Intake assessment questions before and after December 15, 2015, to May 27, 2016, substantially changed because of programmatic changes, and therefore answers were not comparable. Patients with incomplete or duplicated intake assessments were excluded. In families with multiple children enrolled, only the child with the greatest medical complexity, determined by 2-physician chart review, was included to reduce oversampling.22 

State Medicaid claims data were used to identify ED and hospital use for each patient. Usage counts started the month during which the intake assessment was completed for each patient (from December 1, 2015, to May 31, 2016) and ended May 31, 2017, 1 year postenrollment. ED and hospital use were evaluated by how often an ED or hospitalization occurred during the study period (ie, annualized rate), defined as the rate of ED visits and/or hospitalizations per patient per year. Denominators for each patient were censored by months (fractions of a year) if either claims or eligibility data were not available.

A rigorous stepwise coding was conducted to categorize the SDOH intake information for this study. First, the primary investigator (C.F.) performed open coding (n = 226) to classify the assessment responses into SDOH-related problem categories hypothesized to be related to use, including housing concerns, living arrangements, food support, parental health and wellness concerns, and caregiver or parent safety concerns. Coding was blinded to caregiver and patient demographics, such as sex, age, and race and ethnicity. The questions regarding housing concerns and living arrangements were combined because they were documented interchangeably across case managers and then recoded under “housing insecurity” generally. For questions related to both housing insecurity and food insecurity, open coding revealed case managers documented both concerns that had been previously addressed (eg, food insecurity had occurred but has already received sufficient aid) and previously unaddressed (eg, ongoing food insecurity). The remaining 2 questions regarding caregiver health and patient or caregiver safety were coded to indicate whether a concern was documented as present or not, as expressed by the family caregiver.

Once open coding was complete, the research team agreed on the final 4 SDOH-related problems to assess: housing insecurity (previously addressed and unaddressed), food insecurity (previously addressed and unaddressed), caregiver health concerns, and safety concerns. Two trained research assistants each coded 20 patients and met with the primary investigator to review the codebook to clarify definitions and refine the codebook. The research assistants then recoded the original 20 patients with the refined codebook and confirmed their coding as in agreement. The research assistants double coded a random 15% sample with the finalized codebook, and interrater reliability was confirmed with a κ statistic of ≥0.8. The research assistants then coded a random half of the remaining intake assessments. Any ambiguous answers were reconciled between the research assistants with the primary investigator.

Covariates for the regression included patient age, sex, and race and ethnicity and were identified from hospital data. The level of medical complexity was determined using the Pediatric Medical Complexity Algorithm (PMCA), which distinguishes whether patients have complex chronic disease or noncomplex chronic disease by using a combination of International Classification of Diseases, Ninth Revision codes and frequency of use for the conditions coded to determine categorization.1  PMCA categorization was used as the medical complexity coefficient in regression analyses. Nursing intake assessments were used to identify medical technology dependence (eg, gastrostomy tube).

Descriptive analyses of the frequencies, means, and variance for covariates and exposure variables were conducted. Statistical software R (version 3.4.4) was used for modeling. Because of the relatively large amount of nonevents (0’s) in the outcomes, we fit Poisson hurdle models to evaluate the association between SDOH-related problems with annualized ED visit frequency and hospitalization frequency in the targeted population, adjusting for age, sex, race and ethnicity, and medical complexity. A Poisson hurdle model has 2 components, 1 component to model the probability of having a nonzero outcome, usually using logistic regression, and the other to model the count data among those having nonzero outcome values using a truncated Poisson. Food and housing insecurity were examined as “not present,” “previously addressed,” or “previously unaddressed,” whereas health and safety concerns were examined as either “not present” or “present.” A sensitivity analysis was conducted to assess whether the number of SDOH-related problems (ie, dosage effect) was associated with use. In the model for each outcome, we include all 4 social risk factors as well as age, sex, race and ethnicity, and PMCA condition. Limited English proficiency was excluded from multivariate analysis because of co-linearity with race and ethnicity. A P value of <.05 was considered statistically significant.

Of 813 patients enrolled, 226 were eligible for this study (Fig 1). Patients were on average 9.1 years old (SD: 4.9), mostly male (n = 136 [60%]), and evenly distributed across Hispanic (n = 69 [30%]), white non-Hispanic (n = 54 [24%]), and African American (n = 47 [21%]) races and ethnicities (Table 1). Patients were mostly English-speaking (n = 156 [69%]), and approximately one-third had limited English proficiency. A large majority (n = 191 [85%]) of patients had complex chronic disease. Mental health conditions (n = 160 [71%]), progressive chronic illnesses deemed to be associated with shortened adult life expectancy (n = 106 [47%]), and neurologic diagnoses (n = 101 [45%]) were most common; however, patients experienced a wide range of medical diagnoses (Table 1). Approximately 1 in 5 patients (n = 38 [17%]) had some form of medical technology dependence.

FIGURE 1

Identification of patients with medical complexity who had a completed caregiver care coordination intake assessment. a211 patients (93%) had all 12 months of data, and 15 patients (7%) had 10 to 11 months of data. CCC, caregiver care coordination; PPIC, Pediatric Partners in Care.

FIGURE 1

Identification of patients with medical complexity who had a completed caregiver care coordination intake assessment. a211 patients (93%) had all 12 months of data, and 15 patients (7%) had 10 to 11 months of data. CCC, caregiver care coordination; PPIC, Pediatric Partners in Care.

TABLE 1

Patient Demographics and Clinical Characteristics

Characteristicn (%)
Age  
 Mean age in y (SD) 9.1 (4.9) 
Sex  
 Male 136 (60) 
Race and ethnicity  
 Hispanic 69 (30) 
 White non-Hispanic 54 (24) 
 African American 47 (21) 
 Other race or ethnicity 43 (19) 
 Unknown 13 (6) 
Language  
 English 156 (69) 
 Spanish 41 (18) 
 Other 27 (12) 
 Unknown 2 (1) 
Medical complexity  
 Complex chronic disease 191 (85) 
 Noncomplex chronic disease 28 (12) 
 Without chronic disease 4 (2) 
aNonclassifiable 3 (1) 
bMedical diagnoses  
 Mental health 160 (71) 
 Progressive 106 (47) 
 Neurologic 101 (45) 
 Pulmonary or respiratory 94 (42) 
 Gastrointestinal 85 (38) 
 Genitourinary 68 (30) 
 Cardiac 65 (29) 
 Musculoskeletal 65 (29) 
 Ophthalmic 46 (20) 
 Genetic 38 (17) 
 Metabolic 33 (15) 
 Endocrine 31 (14) 
 Hematologic 22 (10) 
 Malignancy 16 (7) 
 Immunologic 15 (7) 
 Renal 12 (5) 
 Otologic 9 (4) 
 Dermatologic 7 (3) 
 Craniofacial 6 (3) 
Medical technology use  
cAny medical technology use 38 (17) 
 Gastrostomy tube 27 (12) 
 Positive pressure (CPAP or BiPAP) 6 (3) 
 Central line 6 (3) 
 Nasogastric or nasoduodenal tube 4 (2) 
 Oxygen (only) 2 (1) 
 Tracheostomy with or without ventilator 2 (1) 
Characteristicn (%)
Age  
 Mean age in y (SD) 9.1 (4.9) 
Sex  
 Male 136 (60) 
Race and ethnicity  
 Hispanic 69 (30) 
 White non-Hispanic 54 (24) 
 African American 47 (21) 
 Other race or ethnicity 43 (19) 
 Unknown 13 (6) 
Language  
 English 156 (69) 
 Spanish 41 (18) 
 Other 27 (12) 
 Unknown 2 (1) 
Medical complexity  
 Complex chronic disease 191 (85) 
 Noncomplex chronic disease 28 (12) 
 Without chronic disease 4 (2) 
aNonclassifiable 3 (1) 
bMedical diagnoses  
 Mental health 160 (71) 
 Progressive 106 (47) 
 Neurologic 101 (45) 
 Pulmonary or respiratory 94 (42) 
 Gastrointestinal 85 (38) 
 Genitourinary 68 (30) 
 Cardiac 65 (29) 
 Musculoskeletal 65 (29) 
 Ophthalmic 46 (20) 
 Genetic 38 (17) 
 Metabolic 33 (15) 
 Endocrine 31 (14) 
 Hematologic 22 (10) 
 Malignancy 16 (7) 
 Immunologic 15 (7) 
 Renal 12 (5) 
 Otologic 9 (4) 
 Dermatologic 7 (3) 
 Craniofacial 6 (3) 
Medical technology use  
cAny medical technology use 38 (17) 
 Gastrostomy tube 27 (12) 
 Positive pressure (CPAP or BiPAP) 6 (3) 
 Central line 6 (3) 
 Nasogastric or nasoduodenal tube 4 (2) 
 Oxygen (only) 2 (1) 
 Tracheostomy with or without ventilator 2 (1) 

Demographics and clinical characteristics of enrolled patients with a completed caregiver care coordination intake assessment and at least 12 mo of insurance claims and eligibility are shown (N = 226). BiPAP, bilevel positive airway pressure; CPAP, continuous positive airway pressure.

a

The PMCA’s accuracy of identifying a child as having complex chronic disease is dependent on having ≥2 chronic diseases as well as ≥2 usage claims in which these diagnoses were coded during the study period. The algorithm could not classify n = 3 patients over a retrospective period of <3 y.

b

Patients could have >1 medical diagnosis.

c

Missing data for technology use was varied for different technologies ranging from 6 to 14 patients.

Of the 226 studied patients, 133 (59%) had at least 1 social determinant concern of any type. Details of the frequency of caregiver-reported SDOH-related problems are shown in Table 2. Twenty-seven caregivers (12%) reported previously unaddressed food insecurity, such as needing help identifying ways to access food aid, dealing with aid ineligibility, and accessing additional resources when aid was insufficient. For the 33 caregivers (15%) who had a previously unaddressed housing insecurity, most were due to financial reasons (n = 27); however, a few caregivers (n = 5) reported their children’s medical conditions (eg, wheelchair accessibility) as barriers to housing. Of the 40 caregivers (18%) who had a health concern for themselves, most were due to physical conditions (n = 21), but emotional strain (n = 8) and mental health conditions (n = 5) also were reported. Lastly, caregiver-reported safety concerns were most often related to their children’s own behavior, including aggressive or self-injurious behavior (n = 13 [6%]).

TABLE 2

Frequency and Categorization of Caregiver-Reported Social Determinant Concern

CharacteristicN = 226, n (%)
aFrequency of caregiver-reported social determinant concerns  
 Caregivers with ≥1 social determinant concern (any) 133 (59) 
  Caregivers with 1 social determinant concern 84 (37) 
  Caregivers with 2 social determinants concern 36 (16) 
  Caregivers with ≥3 social determinants concern 13 (6) 
Categorization of caregiver-reported social determinant concerns  
 Food insecurity (eg, food stamps) (any) 81 (36) 
 Previously unaddressed food need 27 (12) 
  Requests or needs help accessing food aid (eg, caregiver wants aid referral) 10 (4) 
  Applied for aid but determined ineligible (eg, income ineligible) 9 (4) 
  Current aid insufficient to meet needs (eg, ran out of stamps, stamps insufficient) 4 (2) 
  Other not specified 4 (2) 
Housing insecurity (eg, government housing voucher) (any) 51 (23) 
 Previously unaddressed housing need 33 (15) 
  Financial reasons (eg, homelessness, needs rent assistance) 27 (12) 
  Medically problematic arrangement (eg, wheelchair accessibility) 5 (2) 
  Socially problematic arrangement (eg, high density of family members) 4 (2) 
  Other not specified 3 (1) 
 Caregiver health or wellness concern (any) 40 (18) 
  Physical medical condition (eg, surgery, back pain, etc) 21 (9) 
  Emotional strain (eg, “a lot going on”) 8 (4) 
  Mental health disorder (eg, depression or psychotic disorder) 5 (2) 
  Other not specified 6 (3) 
 Caregiver or patient safety concern (any) 25 (11) 
  Patient’s behavior (eg, aggressive and/or self-injurious behavior) 13 (6) 
  Patient’s medical state or physical care (eg, patient at risk for fall) 6 (3) 
  Intimate partner or family violence (eg, restraining order against father) 4 (2) 
  Other not specified 4 (2) 
CharacteristicN = 226, n (%)
aFrequency of caregiver-reported social determinant concerns  
 Caregivers with ≥1 social determinant concern (any) 133 (59) 
  Caregivers with 1 social determinant concern 84 (37) 
  Caregivers with 2 social determinants concern 36 (16) 
  Caregivers with ≥3 social determinants concern 13 (6) 
Categorization of caregiver-reported social determinant concerns  
 Food insecurity (eg, food stamps) (any) 81 (36) 
 Previously unaddressed food need 27 (12) 
  Requests or needs help accessing food aid (eg, caregiver wants aid referral) 10 (4) 
  Applied for aid but determined ineligible (eg, income ineligible) 9 (4) 
  Current aid insufficient to meet needs (eg, ran out of stamps, stamps insufficient) 4 (2) 
  Other not specified 4 (2) 
Housing insecurity (eg, government housing voucher) (any) 51 (23) 
 Previously unaddressed housing need 33 (15) 
  Financial reasons (eg, homelessness, needs rent assistance) 27 (12) 
  Medically problematic arrangement (eg, wheelchair accessibility) 5 (2) 
  Socially problematic arrangement (eg, high density of family members) 4 (2) 
  Other not specified 3 (1) 
 Caregiver health or wellness concern (any) 40 (18) 
  Physical medical condition (eg, surgery, back pain, etc) 21 (9) 
  Emotional strain (eg, “a lot going on”) 8 (4) 
  Mental health disorder (eg, depression or psychotic disorder) 5 (2) 
  Other not specified 6 (3) 
 Caregiver or patient safety concern (any) 25 (11) 
  Patient’s behavior (eg, aggressive and/or self-injurious behavior) 13 (6) 
  Patient’s medical state or physical care (eg, patient at risk for fall) 6 (3) 
  Intimate partner or family violence (eg, restraining order against father) 4 (2) 
  Other not specified 4 (2) 

Frequency and type of caregiver-reported SDOH concern documented by a case manager during a caregiver care coordination intake assessment.

a

Categories are not mutually exclusive (ie, patients could have >1) because some caregivers expressed multiple concerns.

A total of 124 (55%) patients had >1 ED visit, with an annualized mean of 1.5 (SD: 2.4) ED visits per patient (range: 0–18). A total of 45 (20%) patients had ≥1 hospitalization, with an annualized mean of 0.4 (SD: 1.1) hospitalizations per patient (range: 0–8).

Because of a relatively large number of 0 events in the ED visit and hospitalization outcomes, we considered both 0-inflated and hurdle models. Poisson hurdle models fit the data better than 0-inflated Poisson models, therefore we only present results from the univariate and multivariate Poisson hurdle models (Tables 3 through 6). The medical complexity coefficient had a large SE, indicating poor model fitting, so it was excluded from adjustment for the hospital visit outcome.

TABLE 3

Estimated Probability of Having >0 ED or Hospital Visits, by Risk Factor

Caregiver-Reported Social DeterminantED VisitsHospitalizations
Probability95% CIProbability95% CI
Housing insecurity     
 Not present 0.51 0.44–0.59 0.17 0.12–0.24 
 Previously addressed need 0.69 0.49–0.96 0.12 0.03–0.46 
 Previously unaddressed need 0.65 0.50–0.84 0.32 0.19–0.54 
Food insecurity     
 Not present 0.50 0.43–0.60 0.18 0.13–0.26 
 Previously addressed need 0.52 0.40–0.69 0.17 0.09–0.33 
 Previously unaddressed need 0.80 0.66–0.97 0.28 0.15–0.52 
Caregiver health concern     
 Not present 0.54 0.47–0.62 0.19 0.14–0.26 
 Present 0.55 0.42–0.74 0.18 0.09–0.36 
Safety concern     
 Not present 0.55 0.48–0.62 0.20 0.15–0.27 
 Present 0.52 0.35–0.79 0.10 0.03–0.36 
Caregiver-Reported Social DeterminantED VisitsHospitalizations
Probability95% CIProbability95% CI
Housing insecurity     
 Not present 0.51 0.44–0.59 0.17 0.12–0.24 
 Previously addressed need 0.69 0.49–0.96 0.12 0.03–0.46 
 Previously unaddressed need 0.65 0.50–0.84 0.32 0.19–0.54 
Food insecurity     
 Not present 0.50 0.43–0.60 0.18 0.13–0.26 
 Previously addressed need 0.52 0.40–0.69 0.17 0.09–0.33 
 Previously unaddressed need 0.80 0.66–0.97 0.28 0.15–0.52 
Caregiver health concern     
 Not present 0.54 0.47–0.62 0.19 0.14–0.26 
 Present 0.55 0.42–0.74 0.18 0.09–0.36 
Safety concern     
 Not present 0.55 0.48–0.62 0.20 0.15–0.27 
 Present 0.52 0.35–0.79 0.10 0.03–0.36 

Univariate analysis of Poisson hurdle models showing the association between caregiver-reported SDOH and subsequent ED visits and hospitalizations. N = 226. CI, confidence interval.

TABLE 4

Estimated Rates of ED or Hospital Visits Among Those Who Had >0 ED or Hospital Visits, by Risk Factor

Caregiver-Reported Social DeterminantED VisitsHospitalizations
Rate95% CIRate95% CI
Housing insecurity    
 Not present 2.22 1.90–2.60 1.81 1.32–2.46 
 Previously addressed need 2.16 1.42–3.28 3.94 2.01–7.73 
 Previously unaddressed need 3.08 2.42–3.92 1.13 0.59–2.14 
Food insecurity    
 Not present 2.52 2.15–2.94 2.09 1.56–2.82 
 Previously addressed need 1.90 1.40–2.58 1.25 0.63–2.48 
 Previously unaddressed need 2.57 1.94–3.41 1.12 0.50–2.53 
Caregiver health concern    
 Not present 2.41 2.09–2.77 1.80 1.35–2.38 
 Present 2.32 1.74–3.09 1.57 0.84–2.96 
Safety concern    
 Not present 2.22 1.93–2.55 1.71 1.31–2.24 
 Present 3.75 2.79–5.04 2.42 1.00–5.84 
Caregiver-Reported Social DeterminantED VisitsHospitalizations
Rate95% CIRate95% CI
Housing insecurity    
 Not present 2.22 1.90–2.60 1.81 1.32–2.46 
 Previously addressed need 2.16 1.42–3.28 3.94 2.01–7.73 
 Previously unaddressed need 3.08 2.42–3.92 1.13 0.59–2.14 
Food insecurity    
 Not present 2.52 2.15–2.94 2.09 1.56–2.82 
 Previously addressed need 1.90 1.40–2.58 1.25 0.63–2.48 
 Previously unaddressed need 2.57 1.94–3.41 1.12 0.50–2.53 
Caregiver health concern    
 Not present 2.41 2.09–2.77 1.80 1.35–2.38 
 Present 2.32 1.74–3.09 1.57 0.84–2.96 
Safety concern    
 Not present 2.22 1.93–2.55 1.71 1.31–2.24 
 Present 3.75 2.79–5.04 2.42 1.00–5.84 

Univariate analysis of Poisson hurdle models showing the association between caregiver-reported SDOH and subsequent ED visits and hospitalizations. N = 226. CI, confidence interval.

TABLE 5

Raw Proportion and Adjusted Annualized Odds Ratios of Having >0 ED or Hospital Visits Comparing Patients Whose Caregivers Reported a Social Determinant With Patients Whose Caregivers Did Not

Caregiver-Reported Social DeterminantED VisitsHospitalizations
Raw ProportionAdjusted Odds Ratio95% CIRaw ProportionAdjusted Odds Ratio95% CI
Housing insecurity       
 Not present 0.53 Reference — 0.18 Reference — 
 Previously addressed need 0.61 2.88 0.85–9.77 0.17 1.03 0.19–5.49 
 Previously unaddressed need 0.61 1.42 0.62–3.29 0.30 2.18 0.88–5.40 
Food insecurity       
 Not present 0.49 Reference — 0.18 Reference — 
 Previously addressed need 0.54 1.02 0.50–2.04 0.20 0.91 0.36–2.3 
 Previously unaddressed need 0.81 3.43 1.17–10.05 0.26 1.44 0.51–4.08 
Caregiver health concern       
 Not present 0.54 Reference — 0.20 Reference — 
 Present 0.55 0.82 0.37–1.84 0.18 1.08 0.40–2.92 
Safety concern       
 Not present 0.55 Reference — 0.20 Reference — 
 Present 0.52 1.03 0.38–2.80 0.16 0.51 0.10–2.64 
Caregiver-Reported Social DeterminantED VisitsHospitalizations
Raw ProportionAdjusted Odds Ratio95% CIRaw ProportionAdjusted Odds Ratio95% CI
Housing insecurity       
 Not present 0.53 Reference — 0.18 Reference — 
 Previously addressed need 0.61 2.88 0.85–9.77 0.17 1.03 0.19–5.49 
 Previously unaddressed need 0.61 1.42 0.62–3.29 0.30 2.18 0.88–5.40 
Food insecurity       
 Not present 0.49 Reference — 0.18 Reference — 
 Previously addressed need 0.54 1.02 0.50–2.04 0.20 0.91 0.36–2.3 
 Previously unaddressed need 0.81 3.43 1.17–10.05 0.26 1.44 0.51–4.08 
Caregiver health concern       
 Not present 0.54 Reference — 0.20 Reference — 
 Present 0.55 0.82 0.37–1.84 0.18 1.08 0.40–2.92 
Safety concern       
 Not present 0.55 Reference — 0.20 Reference — 
 Present 0.52 1.03 0.38–2.80 0.16 0.51 0.10–2.64 

Adjusted multivariate Poisson hurdle regression modeling association between caregiver-reported SDOH and the number of ED visits and hospitalizations. Estimates are adjusted for age, sex, race and ethnicity, and medical complexity for the ED outcome and all except medical complexity for the hospital visits outcome (which caused problematic model fitting because of large SE). N = 226. CI, confidence interval; —, not applicable.

TABLE 6

Raw Rate and Adjusted Rate Ratio of Annualized ED Visits and Hospital Visits Comparing Patients Whose Caregivers Reported a Social Determinant With Patients Whose Caregivers Did Not, Among Those Patients Who Had >0 ED Visits or Hospitalizations

Caregiver-Reported Social DeterminantED VisitsHospitalizations
Raw RateAdjusted Rate Ratio95% CIRaw RateAdjusted Rate Ratio95% CI
Housing insecurity       
 Not present 2.61 Reference — 2.17 Reference — 
 Previously addressed need 2.42 0.77 0.47–1.24 3.36 3.12 0.94–10.37 
 Previously unaddressed need 3.26 1.55 1.14–2.09 1.55 0.46 0.21–0.98 
Food insecurity       
 Not present 2.93 Reference — 2.41 Reference — 
 Previously addressed need 2.34 0.69 0.48–0.99 1.88 0.86 0.37–2.00 
 Previously unaddressed need 2.60 0.91 0.64–1.28 1.68 0.49 0.19–1.29 
Caregiver health concern       
 Not present 2.67 Reference — 2.14 Reference — 
 Present 2.86 1.04 0.73–1.46 1.93 0.92 0.43–1.95 
Safety concern       
 Not present 2.56 Reference — 2.07 Reference — 
 Present 3.67 2.04 1.41–2.96 2.57 2.66 0.83–8.50 
Caregiver-Reported Social DeterminantED VisitsHospitalizations
Raw RateAdjusted Rate Ratio95% CIRaw RateAdjusted Rate Ratio95% CI
Housing insecurity       
 Not present 2.61 Reference — 2.17 Reference — 
 Previously addressed need 2.42 0.77 0.47–1.24 3.36 3.12 0.94–10.37 
 Previously unaddressed need 3.26 1.55 1.14–2.09 1.55 0.46 0.21–0.98 
Food insecurity       
 Not present 2.93 Reference — 2.41 Reference — 
 Previously addressed need 2.34 0.69 0.48–0.99 1.88 0.86 0.37–2.00 
 Previously unaddressed need 2.60 0.91 0.64–1.28 1.68 0.49 0.19–1.29 
Caregiver health concern       
 Not present 2.67 Reference — 2.14 Reference — 
 Present 2.86 1.04 0.73–1.46 1.93 0.92 0.43–1.95 
Safety concern       
 Not present 2.56 Reference — 2.07 Reference — 
 Present 3.67 2.04 1.41–2.96 2.57 2.66 0.83–8.50 

Adjusted multivariate Poisson hurdle regression modeling association between caregiver-reported SDOH and the number of ED visits and hospitalizations. Estimates are adjusted similarly to Table 5. N = 226. CI, confidence interval; —, not applicable.

Table 3 shows the estimated probability of patients having at least 1 ED visit or hospitalization during the study period. Patients whose caregivers reported previously unaddressed food insecurity had an 80% chance (66%–97%) of having at least 1 ED visit compared with a 50% chance (43%–60%) for patients whose caregivers reported no food insecurity and a 52% chance (40%–69%) for patients whose caregivers reported a previously addressed food insecurity.

Table 4 shows the estimated rates of ED visits or hospitalization among those who had at least 1 ED or hospital visit respectively. Among those patients with at least 1 ED visit, patients whose caregivers reported a safety concern had an annualized mean of 3.75 (2.79–5.04) ED visits, compared with an annualized mean of 2.22 (1.93–2.55) ED visits for patients whose caregivers reported no safety concerns.

Table 5 shows the adjusted odds ratios of having >0 ED or hospital visits, comparing patients whose caregivers reported a social determinant compared with patients whose caregivers did not. Patients whose caregivers reported a previously unaddressed food insecurity had a higher chance of having at least 1 ED visit (odds ratio: 3.43 [1.17–10.05]) compared with patients whose caregivers did not report any food insecurity. Remaining results were not statistically significant.

Table 6 shows the adjusted rate ratio of annualized ED visits and hospitalizations comparing patients whose caregivers reported a social determinant compared with patients whose caregivers did not, among those patients who had >0 ED or hospital visits. Patients with a caregiver-reported safety concern had twice the annualized ED use (rate ratio: 2.04 [1.41–2.96]) compared with those who did not. Patients whose caregivers reported a previously unaddressed housing insecurity had a 55% higher risk of annualized ED use compared with patients whose caregivers reported no housing insecurity (rate ratio: 1.55 [1.14–2.09]). However, this same group had a reduced risk of hospitalization (rate ratio: 0.46 [0.21–0.98]). Previously addressed food insecurity was associated with a comparative lower rate of ED use (rate ratio: 0.69 [0.48–0.99]).

No association was found in the sensitivity analysis evaluating the number of SDOH-related problems with use (data not shown).

In this study, we evaluated the association between caregiver-reported SDOH-related problems and ED and hospital use for a CCD cohort insured by Medicaid and enrolled in an intensive CM program. We found that as many as 6 out of 10 caregivers of CCD endorsed at least 1 SDOH insecurities or concerns. Previously unaddressed food insecurity was associated with increased ED use in the subsequent year, regardless of referrals placed by the CM program. Among those who had ≥1 ED or hospital visits, safety concerns and previously unaddressed housing insecurity were associated with increased ED use in the subsequent year, despite referrals to housing resources at the time of enrollment.

Whereas the use of screening for SDOH-related problems in the pediatric population has been more broadly described, with this study, we add to our understanding of how SDOH-related problems may influence ED and hospital use for CCD.2326  Specifically, identifying previously unaddressed food insecurity may be predictive of CCD who will visit the ED in the subsequent year. Among those who do have ≥1 ED or hospital visits, safety concerns or previously unaddressed housing insecurity may be predictive of how often a CCD will visit the ED within the following year. Notably, our study reinforces the importance of eliciting not just if a social problem has existed for a family but also whether that need is being fully addressed by ongoing services given that we found no association of ED or hospital use with addressed housing insecurity and a protective effect with addressed food insecurity.

Although SDOH-related insecurities or concerns mostly were not associated with hospitalization, this may be because hospitalization may be more medically necessary and unavoidable for this population than ED visits, given the additional medical threshold for admission. Why patients with previously unaddressed housing insecurity were less likely to be hospitalized may reflect a more-frequent use of the ED, but the actual reason is unknown. Future work in which researchers examine reasons for hospitalization may better elucidate these relationships.

With the increasing focus on value-based payment, improved integration between “traditional” hospital systems and social services, including governmental or other nonprofit aide groups, may be encouraged as hospital systems attempt to mitigate their financial risk and improve patient health outcomes.27  Evidence-based examples of such innovative models to reduce medical care spending and improve health outcomes are increasingly common.24,28,29  In our study, providing general information and help applying for previously unaddressed housing and food aid was not, in and of itself, sufficient to address all of the risk associated with these insecurities given that the effect persisted within the year after its identification. Without a control group, it is unknown to what degree some of that risk was reduced with the informational referrals; however, with this study, it is suggested that there may be a need for a more-intensive intervention to mitigate the impact of housing and food insecurities for the CCD population.

As such, the relationship between housing and food insecurity and CCD health care use requires further investigation. The interplay between caring for a child’s medical conditions and SDOH is likely bidirectional and multifactorial. Poor living conditions may affect CCD health directly, or caregivers may be more likely to use the ED for temporary shelter when caring for their child’s medical issues. For example, in our coding, a caregiver-reported housing concern was directly related to the child’s medical problem (eg, housing need for a wheelchair-accessible unit). In turn, a child’s health care cost may exacerbate a family’s financial strain and hinder caregiver employment opportunities, a key driver toward housing insecurity.3033 

This was a single-state observational study in which we evaluated only select Medicaid plans within a CM population, which limits its generalizability. However, we were able to examine health plan rather than single-institution usage data. Coding qualitative responses to the intake assessment involves subjective interpretation and is subject to reporting bias, misclassification, and missing data (ie, may not have captured all relevant SDOH); however, the κ achieved was excellent (≥0.8), and a rigorous approach was used. Given information was not collected from a control group nor was the exact timing of the addressed needs captured, we may have incorrectly estimated the impact of SDOH-related concerns on subsequent use. We have no available data to verify whether services offered were accepted or accessed by families or if need reduction actually occurred and when.

Sample size and the infrequency of hospitalization may have limited our power to examine associations with SDOH-related concerns and may also reflect that the claims data used to identify “high-using” patients for the program were not adequate predictors for future use. PMCA’s sensitivity for correctly classifying children with medical complexity is imperfect.1  The classification of 12% of the children in the study population as having noncomplex chronic or no chronic disease may reflect a lack of use where their chronic diseases were coded in Medicaid claims data. The accuracy of PMCA in correctly identifying a child with medical complexity is dependent on the child having ≥2 chronic diseases as well as ≥2 usage claims where these diagnoses were coded during the study period. In this study, PMCA category may reflect level of severity or stability of a child’s chronic conditions because it relies on health care use for correct classification. Our study population did have a relatively high percentage of patients with mental health diagnoses and lower percentages of patients dependent on medical technology compared with other published studies of CCD,2,3,22  which may account, in part, for the lower rates of observed ED and inpatient use. This difference is important to consider. For children with less complex chronic disease, decreasing ED and inpatient use may not be the right outcome to assess CM program effectiveness.

We found that 6 out of 10 caregivers of CCD enrolled in a Medicaid CM program endorsed food insecurity, housing insecurity, a caregiver health concern, and/or a safety concern. Previously unaddressed food insecurity was associated with increased ED use in the subsequent year. Among those who had ≥1 ED or hospital visits, previously unaddressed housing insecurity was associated with increased ED use in the subsequent year, despite referral for housing and safe haven assistance. Among those who had ≥1 ED or hospital visits, previously addressed food insecurity and previously unaddressed housing insecurity were associated with a reduced rate of ED visits and hospitalization, respectively.

Given these mixed findings, in future work, researchers should investigate the interplay between the social and medical needs of CCD and their caregivers if we wish to better meet the needs of this vulnerable population. Future investigators should consider whether it is more efficacious to screen for and aggressively target insecurities within a specific pediatric CM program or as part of a larger community initiative in which these challenges are the primary focus (eg, Housing First initiatives).3436  Lastly, studies in which researchers evaluate how continuity of care across the outpatient and acute care settings (ie, ED and hospital) interplay to impact health care use in CCD remains an important consideration both in addressing SDOH-related concerns and unplanned health care use.

Dr Foster conceptualized and designed the study, developed and led coding and the analysis, and drafted and revised the manuscript; Dr Simon participated in study conceptualization and design and aided with data analysis and interpretation and manuscript writing; Dr Qu aided in study design, conducted data analysis and interpretation, and reviewed the final manuscript as submitted; Ms Holmes aided in data collection and data interpretation and reviewed the final manuscript as submitted; Mr Chang and Ms Ramos aided in data collection, data management and cleaning, and data analysis and reviewed the final manuscript as submitted; Ms Koutlas aided in data collection and coding analysis and interpretation and reviewed the final manuscript as submitted; Dr Rivara helped secure funding and participated in study design and conceptualization, data collection, and manuscript writing; Dr Melzer was primary investigator of the Pediatric Partners in Care program, participated in study design and conceptualization and data collection, and reviewed the final manuscript as submitted; and Dr Mangione-Smith participated in study conceptualization and design, data collection, data analysis and interpretation, and manuscript writing; and all authors approved the final manuscript as submitted.

FUNDING: Supported by grant 1C1CMS331341 from the Department of Health and Human Services, Centers for Medicare and Medicaid Services. The contents of this document are solely the responsibility of the authors and do not necessarily represent the official views of the US Department of Health and Human Services or any of its agencies.

Dr Mangione-Smith’s current affiliations are Kaiser Permanente Washington Health Research Institute, Seattle, WA, and Department of Health Systems Science at Kaiser Permanente School of Medicine, Pasadena, CA.

Dr Foster’s current affiliations are Division of Academic General Pediatrics and Primary Care, Department of Pediatrics, Feinberg School of Medicine, Northwestern University and Mary Ann and J. Milburn Smith Child Health Research, Advocacy, and Outreach Center, Ann and Robert H. Lurie Children’s Hospital of Chicago, Chicago, IL.

1
Cohen
E
,
Kuo
DZ
,
Agrawal
R
, et al
.
Children with medical complexity: an emerging population for clinical and research initiatives
.
Pediatrics
.
2011
;
127
(
3
):
529
538
2
Simon
TD
,
Berry
J
,
Feudtner
C
, et al
.
Children with complex chronic conditions in inpatient hospital settings in the United States
.
Pediatrics
.
2010
;
126
(
4
):
647
655
3
O’Mahony
L
,
O’Mahony
DS
,
Simon
TD
,
Neff
J
,
Klein
EJ
,
Quan
L
.
Medical complexity and pediatric emergency department and inpatient utilization
.
Pediatrics
.
2013
;
131
(
2
).
4
Kuo
DZ
,
Hall
M
,
Agrawal
R
, et al
.
Comparison of health care spending and utilization among children with medicaid insurance
.
Pediatrics
.
2015
;
136
(
6
).
5
Coller
RJ
,
Nelson
BB
,
Sklansky
DJ
, et al
.
Preventing hospitalizations in children with medical complexity: a systematic review
.
Pediatrics
.
2014
;
134
(
6
).
6
Li
M
,
Riis
JL
,
Ghazarian
SR
,
Johnson
SB
.
Income, family context, and self-regulation in 5-year-old children
.
J Dev Behav Pediatr
.
2017
;
38
(
2
):
99
108
7
Yoshikawa
H
,
Aber
JL
,
Beardslee
WR
.
The effects of poverty on the mental, emotional, and behavioral health of children and youth: implications for prevention
.
Am Psychol
.
2012
;
67
(
4
):
272
284
8
Elgar
FJ
,
Gariépy
G
,
Torsheim
T
,
Currie
C
.
Early-life income inequality and adolescent health and well-being
.
Soc Sci Med
.
2017
;
174
:
197
208
9
Eneriz-Wiemer
M
,
Sanders
LM
,
Barr
DA
,
Mendoza
FS
.
Parental limited English proficiency and health outcomes for children with special health care needs: a systematic review
.
Acad Pediatr
.
2014
;
14
(
2
):
128
136
10
Khan
SQ
,
Berrington de Gonzalez
A
,
Best
AF
, et al
.
Infant and youth mortality trends by race/ethnicity and cause of death in the United States
.
JAMA Pediatr
.
2018
;
172
(
12
):
e183317
11
Hood
CM
,
Gennuso
KP
,
Swain
GR
,
Catlin
BB
.
County health rankings: relationships between determinant factors and health outcomes
.
Am J Prev Med
.
2016
;
50
(
2
):
129
135
12
Gitterman
BA
,
Chilton
LA
,
Cotton
WH
, et al
;
Council on Community Pediatrics
;
Committee on Nutrition
.
Promoting food security for all children
.
Pediatrics
.
2015
;
136
(
5
).
13
Gitterman
BA
,
Chilton
LA
,
Cotton
WH
, et al
;
Council on Community Pediatrics
.
Poverty and child health in the United States
.
Pediatrics
.
2016
;
137
(
4
):
e20160339
14
Lax
Y
,
Martinez
M
,
Brown
NM
.
Social determinants of health and hospital readmission
.
Pediatrics
.
2017
;
140
(
5
):
e20171427
15
Balistreri
KS
.
Food insufficiency and children with special healthcare needs
.
Public Health
.
2019
;
167
:
55
61
16
Flaherty
EG
,
Thompson
R
,
Dubowitz
H
, et al
.
Adverse childhood experiences and child health in early adolescence
.
JAMA Pediatr
.
2013
;
167
(
7
):
622
629
17
Sharfstein
J
,
Sandel
M
,
Kahn
R
,
Bauchner
H
.
Is child health at risk while families wait for housing vouchers?
Am J Public Health
.
2001
;
91
(
8
):
1191
1192
18
Sandel
M
,
Sheward
R
,
Ettinger de Cuba
S
, et al
.
Unstable housing and caregiver and child health in renter families
.
Pediatrics
.
2018
;
141
(
2
):
e20172199
19
Kuo
DZ
,
Goudie
A
,
Cohen
E
, et al
.
Inequities in health care needs for children with medical complexity
.
Health Aff (Millwood)
.
2014
;
33
(
12
):
2190
2198
20
Centers for Medicare and Medicaid Services
.
Glossary. Available at: https://www.cms.gov/apps/glossary/. Accessed April 6, 2017
21
Court
B
,
Mancuso
D
,
Zhu
C
,
Krupski
A
.
Predictive Risk Intelligence System (PRISM): A Decision-Support Tool for Coordinating Care for Complex Medicaid Clients
.
New York
:
John Wiley & Sons, Inc
;
2011
22
Simon
TD
,
Cawthon
ML
,
Stanford
S
, et al
;
Center of Excellence on Quality of Care Measures for Children with Complex Needs (COE4CCN) Medical Complexity Working Group
.
Pediatric medical complexity algorithm: a new method to stratify children by medical complexity
.
Pediatrics
.
2014
;
133
(
6
). Available at: www.pediatrics.org/cgi/content/full/133/6/e1647
23
Arthur
KC
,
Lucenko
BA
,
Sharkova
IV
,
Xing
J
,
Mangione-Smith
R
.
Using state administrative data to identify social complexity risk factors for children
.
Ann Fam Med
.
2018
;
16
(
1
):
62
69
24
Gottlieb
LM
,
Hessler
D
,
Long
D
, et al
.
Effects of social needs screening and in-person service navigation on child health: a randomized clinical trial
.
JAMA Pediatr
.
2016
;
170
(
11
):
e162521
25
Kerker
BD
,
Storfer-Isser
A
,
Szilagyi
M
, et al
.
Do pediatricians ask about adverse childhood experiences in pediatric primary care?
Acad Pediatr
.
2016
;
16
(
2
):
154
160
26
Chung
EK
,
Siegel
BS
,
Garg
A
, et al
.
Screening for social determinants of health among children and families living in poverty: a guide for clinicians
.
Curr Probl Pediatr Adolesc Health Care
.
2016
;
46
(
5
):
135
153
27
Garg
A
,
Homer
CJ
,
Dworkin
PH
.
Addressing social determinants of health: challenges and opportunities in a value-based model
.
Pediatrics
.
2019
;
143
(
4
):
e20182355
28
Sandberg
SF
,
Erikson
C
,
Owen
R
, et al
.
Hennepin Health: a safety-net accountable care organization for the expanded Medicaid population
.
Health Aff (Millwood)
.
2014
;
33
(
11
):
1975
1984
29
Kuehn
BM
.
Hospitals turn to housing to help homeless patients
.
JAMA
.
2019
;
321
(
9
):
822
824
30
Kish
AM
,
Newcombe
PA
,
Haslam
DM
.
Working and caring for a child with chronic illness: a review of current literature
.
Child Care Health Dev
.
2018
;
44
(
3
):
343
354
31
Lindly
OJ
,
Chavez
AE
,
Zuckerman
KE
.
Unmet health services needs among US children with developmental disabilities: associations with family impact and child functioning
.
J Dev Behav Pediatr
.
2016
;
37
(
9
):
712
723
32
McClung
N
,
Glidewell
J
,
Farr
SL
.
Financial burdens and mental health needs in families of children with congenital heart disease
.
Congenit Heart Dis
.
2018
;
13
(
4
):
554
562
33
Schaible
B
,
Colquitt
G
,
Caciula
MC
,
Carnes
A
,
Li
L
,
Moreau
N
.
Comparing impact on the family and insurance coverage in children with cerebral palsy and children with another special healthcare need
.
Child Care Health Dev
.
2018
;
44
(
3
):
370
377
34
Martinez
TE
,
Burt
MR
.
Impact of permanent supportive housing on the use of acute care health services by homeless adults
.
Psychiatr Serv
.
2006
;
57
(
7
):
992
999
35
Ly
A
,
Latimer
E
.
Housing first impact on costs and associated cost offsets: a review of the literature
.
Can J Psychiatry
.
2015
;
60
(
11
):
475
487
36
Larimer
ME
,
Malone
DK
,
Garner
MD
, et al
.
Health care and public service use and costs before and after provision of housing for chronically homeless persons with severe alcohol problems
.
JAMA
.
2009
;
301
(
13
):
1349
1357

Competing Interests

POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.

FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.