Sociodemographic and Health-Related Characteristics of a Safety-Net Patient Population

Background Safety-net health systems are an important source of healthcare for underserved or vulnerable individuals, but definitions of safety-net institutions are largely based on patient characteristics. Some definitions may not accurately identify such institutions. Therefore, we aimed to describe the characteristics of urban safety-net patients in Texas and compare the distribution of morbidities between safety-net and general population patients. Methods We used hospital claims data from the Dallas-Fort Worth Hospital Council Foundation to create a cross-sectional cohort. Eligible patients were aged ≥18 years and Tarrant County residents in 2018. Patients were divided into two groups for comparison. The first group represented patients with hospital claims from JPS Health Network (i.e. safety-net population). The second group represented all patients with hospital claims in Tarrant County (i.e. general population). We estimated frequencies of patient characteristics. In addition, we estimated overall and payor-stratified standardized morbidity ratios (SMRs) adjusted for age, gender, and race/ethnicity to compare the prevalence of common chronic diseases between safety-net patients and patients in the general population. Results Our study population comprised 459,827 patients, of whom 74,323 (16%) were safety-net patients. Patients aged ≥65 years comprised 23% of the general population and 11% of the safety-net population. Non-Hispanic Whites comprised 52% of the general population and 29% of safety-net patients. A larger proportion of safety-net patients were uninsured compared with general population patients (safety-net: 54%; general population: 25%), but Medicaid distribution was less discrepant (safety-net: 9%; general population: 7%). Medicare was the primary payor for 24% of general population patients and 14% of safety-net patients. Safety-net patients had relative excesses of mental health

and chronic conditions ranging between 5% and 230% for all selected conditions except dementia/Alzheimer's. The patterns for payor-stratified SMRs were consistent with the overall results.
Conclusions We observed considerable sociodemographic diversity and a high burden of mental health and chronic conditions among safety-net patients, which may support understanding the healthcare needs of safety-net populations. Our findings raise questions about definitions of safety-net institutions based on Medicaid distribution alone and the transportability of findings from studies in which safety-net populations are unrepresented.

Background
The Affordable Care Act (ACA) intended to expand Medicaid coverage for low-income individuals in the United States (U.S.), but states were given the option to decline expansion. As of August 2019, 14 states including Texas, have not expanded Medicaid coverage, which has created a coverage gap that affects an estimated 2.5 million adults [1,2]. Texas has the highest prevalence of uninsured adults in the U.S. [3] and accounts for 31% of people in the coverage gap [2]. Safety-net health systems offer service regardless of an individual's ability to pay and are an important source of healthcare for these uninsured individuals [4,5], but safety-net populations are understudied and limited information is available about their healthcare needs [6,7].
Individuals who rely on safety-net institutions are often broadly described as underserved or medically vulnerable populations and include not only the uninsured but also underinsured, racial/ethnic minorities, unemployed, immigrants, chronically ill (physical or mental), disabled, veterans, homeless, or incarcerated [5]. Prior reports involving safetynet populations have largely been about specific conditions [8,9] or conducted at the hospital-level using aggregate definitions of patient characteristics [10][11][12][13][14]. Reports of specific conditions have limited generalizability to the broader safety-net population and hospital-level (i.e. group-level) have a different unit of analysis [15]. In addition, grouplevel comparisons may obscure important differences in patient case mix [16]. A systematic individual-level assessment of people receiving care in a safety-net health system may thus contribute to a multilevel understanding [17] about underserved and vulnerable individuals. Therefore, we aimed to describe the characteristics of an urban safety-net population in Texas and compare the distribution of morbidities between safety-net patients and patients in the general population.

Study setting
JPS Health Network (JPS) is an urban integrated safety-net health system and a member of America's Essential Hospitals [18]. The network has a well-defined catchment area that encompasses Tarrant County, Texas, which is part of the Dallas-Fort Worth metropolitan area. Tarrant County is the third largest County in Texas with a population of 2.1 million people and an overall poverty of 13.5% [19,20]. The network includes a 573-bed academic teaching hospital in Fort Worth, Texas and over 40 satellite community health and school-based clinics distributed across the county. JPS provides comprehensive health services including inpatient, outpatient, and specialty services. Specialty services include Level 1 services for trauma and stroke, an accredited comprehensive Community Cancer Program, and a comprehensive HIV clinic (Healing Wings Clinic) supported by the Ryan White HIV/AIDS Program.

Data source and eligibility criteria
We used administrative claims data from the Dallas Fort Worth Hospital Council Foundation (DFWHC) regional database for this study. The DFWHC, which includes 90 member hospitals in the North Texas region, houses all administrative claims data reported from about 95% of hospitals in the region [21]. This comprehensive database has information on patient charges, emergency department (ED) use, and sociodemographic characteristics of patients from the North Texas region [21].
Our eligibility criteria included patients aged ≥18 years who were Tarrant County residents at any point in 2018 and had a hospital encounter at any DFWHC member hospital in Tarrant County during that year. Patients without a Tarrant County zip code during 2018 were ineligible. We defined our safety-net population as adults who generated either at least one claim for services rendered during a non-ED hospital encounter or who had claims that exclusively originated from the ED at JPS between January 1, 2018 and  Table S1).. Multi-morbidity burden was defined as the presence of more than one selected chronic illness per hospital encounter and classified from zero to greater than or equal to three (0, 1, 2, or ≥3).

Data analysis
We estimated relative frequencies of sociodemographic characteristics in both safety-net patients and general population patients. In addition, we computed overall-and payorstratified (private insurance, Medicaid, Medicare, and uninsured) standardized morbidity ratios (SMRs) and corresponding 95% confidence limits (CL) to compare the observed number of cases for a selected chronic condition among safety-net patients with the expected number of cases if patients in the general population had the same age-, gender-, and racial/ethnic distribution as safety-net patients. An SMR greater than one thus indicates the relative excess number of cases for a particular condition among safety-net patients compared with patients in the general population.

Results
Our study population comprised 459,827 adult Tarrant County residents with an administrative claim during 2018, of whom 74,323 (16%) were safety-net patients ( Figure   1). Table 1 summarizes the sociodemographic characteristics of our study population by safety-net or general population status. Patients aged ≥65 years comprised 23% of the general population and 11% of the safety-net population. The majority of general population patients were non-Hispanic White (52%) and female (65%), whereas gender and race/ethnicity were more evenly distributed among safety-net patients. A larger proportion of safety-net patients were uninsured compared with general population patients (safety-net: 54%; general population: 25%). In contrast, Medicare was the primary payor for about 24% of general population patients but only 14% of patients used the safety-net health system. We identified two prior studies with an individual-level analysis comparing safety-net populations with patients in other settings. Our results are generally consistent with the study by Balasubramanian et al. [29], who reported a greater proportion of safety-net patients had higher comorbidity scores compared with patients at non-safety-net institutions. Nevertheless, the population for this study comprised patients aged 50-63 years and the data were derived from both administrative claims and clinical sources including electronic health records. Our study population included all patients aged ≥18 years and was based on administrative claims data, but the observed higher morbidity among safety-net patients is consistent between studies. The study by Whitaker et al. [16] is not directly comparable with our study considering the authors focused on lower postoperative complications after colorectal cancer surgery and not the broader comparisons pursued in our study.

Health-related characteristics
We observed sizable differences in the chronic disease burden between safety-net patients and general population patients by payor status. In particular, our results suggest substantially higher relative excess of mental health and chronic conditions even when comparing safety-net patients with patients in the general population by private insurance, Medicaid, and Medicare. These differences may be partially attributable to being underinsured [30], which prompted care at a safety-net health system. This finding, combined with the low frequency of Medicare beneficiaries in our population, adds further questions about standard definitions of disproportionate share hospitals that are used to identify safety-net institutions.
Our findings should be interpreted in the context of certain limitations. Our SMR estimates may be sensitive to outcome misclassification because of errors in coding claims and varying coding practices across participating hospitals. For example, a prior study [31] reported that administrative claims data have 61% sensitivity and 94% specificity for classifying depression when electronic health records are the reference standard. If outcome misclassification is precisely nondifferential (i.e. no variation in misclassification between safety-net patients and patients in the general population) given additional assumptions including independent errors [32], then our SMRs may underestimate the true magnitude of relative excess of depression or other conditions. In contrast, if outcome misclassification is differential, then our SMR estimates may underestimate or overestimate the true magnitude of relative excess. In addition, our SMR estimates may be sensitive to selection bias if individuals in our safety-net hospital also received care at other hospitals. This bias is related to an initial misclassification of hospital status but manifests as selection bias because the individual could "participate" in both the safetynet and general population of patients. The consequence may bias SMRs toward the null, which would underestimate the magnitude of health conditions among safety-net patients.
Lastly, claims data do not include services provided through outpatient labs or hospital-based outpatient services which limits the generalizability of these findings to those settings.
In summary, if the limitations of our study did not materially affect our results, then our findings provide useful evidence about safety-net populations. We observed considerable sociodemographic diversity and a high burden of mental health and chronic conditions among safety-net patients, which may support understanding the healthcare needs and considerations may be a greater concern, but these issues are not insurmountable with proper planning [6,44]. In the absence of adequate representation in research used to generate evidence for interventions and guidelines, greater attention should be given to evaluating interventions and guidelines for benefits and harms in safety-net settings.

Declarations
Ethics approval and consent to participate Not applicable Consent for publication: Not applicable

Availability of data and materials
The data that support the findings of this study are available from Dallas Fort Worth Hospital Council Foundation but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of Dallas Fort Worth Hospital Council Foundation.

Competing interests
The authors declare that they have no competing interests.

Supplementary Files
This is a list of supplementary files associated with the primary manuscript. Click to download. Supplementary.docx