How Should We Measure Social Deprivation in Orthopaedic Patients?

Abstract

Background

Social deprivation negatively affects a myriad of physical and behavioral health outcomes. Several measures of social deprivation exist, but it is unclear which measure is best suited to describe patients with orthopaedic conditions.

Questions/purposes

(1) Which measure of social deprivation, defined as “limited access to society’s resources due to poverty, discrimination, or other disadvantage,” is most strongly and consistently correlated with patient-reported physical and behavioral health in patients with orthopaedic conditions? (2) Compared with the use of a single measure alone, how much more variability in patient-reported health does the simultaneous use of multiple social deprivation measures capture?

Methods

Between 2015 and 2017, a total of 79,818 new patient evaluations occurred within the orthopaedic department of a single, large, urban, tertiary-care academic center. Over that period, standardized collection of patient-reported health measures (as described by the Patient-reported Outcomes Measurement Information System [PROMIS]) was implemented in a staged fashion throughout the department. We excluded the 25% (19,926) of patient encounters that did not have associated PROMIS measures reported, which left 75% (59,892) of patient encounters available for analysis in this cross-sectional study of existing medical records. Five markers of social deprivation were collected for each patient: national and state Area Deprivation Index, Medically Underserved Area Status, Rural-Urban Commuting Area code, and insurance classification (private, Medicare, Medicaid, or other). Patient-reported physical and behavioral health was measured via PROMIS computer adaptive test domains, which patients completed as part of standard care before being evaluated by a provider. Adults completed the PROMIS Physical Function version 1.2 or version 2.0, Pain Interference version 1.1, Anxiety version 1.0, and Depression version 1.0. Children ages 5 to 17 years completed the PROMIS Pediatric Mobility version 1.0 or version 2.0, Pain Interference version 1.0 or version 2.0, Upper Extremity version 1.0, and Peer Relationships version 1.0. Age-adjusted partial Pearson correlation coefficients were determined for each social deprivation measure and PROMIS domain. Coefficients of at least 0.1 were considered clinically meaningful for this purpose. Additionally, to determine the percentage of PROMIS score variability that could be attributed to each social deprivation measure, an age-adjusted hierarchical regression analysis was performed for each PROMIS domain, in which social deprivation measures were sequentially added as independent variables. The model coefficients of determination (r²) were compared as social deprivation measures were incrementally added. Improvement of the r² by at least 10% was considered clinically meaningful.

Results

Insurance classification was the social deprivation measure with the largest (absolute value) age-adjusted correlation coefficient for all adult and pediatric PROMIS physical and behavioral health domains (adults: correlation coefficient 0.40 to 0.43 [95% CI 0.39 to 0.44]; pediatrics: correlation coefficient 0.10 to 0.19 [95% CI 0.08 to 0.21]), followed by national Area Deprivation Index (adults: correlation coefficient 0.18 to 0.22 [95% CI 0.17 to 0.23]; pediatrics: correlation coefficient 0.08 to 0.15 [95% CI 0.06 to 0.17]), followed closely by state Area Deprivation Index. The Medically Underserved Area Status and Rural-Urban Commuting Area code each had correlation coefficients of 0.1 or larger for some PROMIS domains but neither had consistently stronger correlation coefficients than the other. Except for the PROMIS Pediatric Upper Extremity domain, consideration of insurance classification and the national Area Deprivation Index together explained more of the variation in age-adjusted PROMIS scores than the use of insurance classification alone (adults: r² improvement 32% to 189% [95% CI 0.02 to 0.04]; pediatrics: r² improvement 56% to 110% [95% CI 0.01 to 0.02]). The addition of the Medically Underserved Area Status, Rural-Urban Commuting Area code, and/or state Area Deprivation Index did not further improve the r² for any of the PROMIS domains.

Conclusion

To capture the most variability due to social deprivation in orthopaedic patients’ self-reported physical and behavioral health, insurance classification (categorized as private, Medicare, Medicaid, or other) and national Area Deprivation Index should be included in statistical analyses. If only one measure of social deprivation is preferred, insurance classification or national Area Deprivation Index are reasonable options. Insurance classification may be more readily available, but the national Area Deprivation Index stratifies patients across a wider distribution of values. When conducting clinical outcomes research with social deprivation as a relevant covariate, we encourage researchers to consider accounting for insurance classification and/or national Area Deprivation Index, both of which are freely available and can be obtained from data that are typically collected during routine clinical care.

Level of Evidence

Level III, therapeutic study.

Introduction

According to the WHO, social determinants of health are the “conditions in which people are born, grow, work, live, and age, and the wider set of forces and systems [that shape] the conditions of daily life” [75]. The US government’s initiative Healthy People 2020 identifies several examples of social determinants, such as access to safe housing, reliable transportation, quality education, job training, nutritious food, recreational activities, and healthcare services [43]. These variables contribute to a multitude of health outcomes including physical functioning, mental well-being, the presence of pain, and hospital readmission rates, and they are associated with up to an 18-year difference in life expectancy [30, 41, 48, 53, 55, 68, 72]. Despite a growing awareness of the importance of social determinants, social deprivation is still a major source of health inequality in the United States [62]. Defined by the American Psychological Association as “limited access to society’s resources due to poverty, discrimination, or other disadvantage,” the importance of accounting for social deprivation when evaluating the effectiveness of a clinical intervention has become clear [6, 12, 45, 77]. Even when social deprivation is not a primary exposure or outcome of interest, it can exaggerate or obscure a study’s results if it is not appropriately considered during analysis.

To study the effects of social deprivation, it must be quantitatively defined. Available measures of social deprivation range from simple, single-domain markers such as income level to multidomain indices that account for 10 or more variables [3, 9, 19, 61, 66, 69]. To our knowledge, there is no consensus measure of social deprivation [6, 63], and the associations among various social deprivation markers and relevant indicators of health have not been compared systematically. Identifying the social deprivation measures that most thoroughly capture potential effects on health would promote standardization and improve comparability across studies. Furthermore, an ideal measure would be freely available and could be determined from data that are routinely captured during standard clinical care. The national and state Area Deprivation Indices, Medically Underserved Area classification, Rural-Urban Continuum Area codes, and patients’ type of health insurance are related to social determinants of health and fit these criteria [19, 23, 27, 66].

We therefore asked: (1) Which measure of social deprivation, defined as “limited access to society’s resources due to poverty, discrimination, or other disadvantage,” is most strongly and consistently correlated with patient-reported physical and behavioral health in patients with orthopaedic conditions? (2) Compared with the use of a single measure alone, how much more variability in patient-reported health does the simultaneous use of multiple social deprivation measures capture?

Patients and Methods

Study Design and Setting

This was a cross-sectional study of existing medical records from a single tertiary-care academic institution in the Midwestern United States. The institution has a multistate catchment area, including urban and rural regions.

Participants

Eligible participants included new patients who presented to an orthopaedic department between June 22, 2015 and November 1, 2017 for evaluation of an orthopaedic condition. On June 22, 2015, the department began collecting Patient-reported Outcomes Measurement Information System (PROMIS) measures as standard care in a staged fashion over several months. Patients were evaluated at one of five clinic locations throughout urban and suburban areas by an attending physician, resident or fellow trainee physician, or midlevel provider. Patients presenting to all service and clinic types were eligible, including surgical and nonsurgical, adult and pediatric, scheduled and walk-in, and our “safety net” clinics staffed by resident and fellow trainees. Patients who did not complete any PROMIS domains at their clinic encounter were excluded, which most often occurred due to the staged nature of implementation of PROMIS collection throughout the department.

Study Flow and Patients’ Descriptive Data

Of 202,951 patient encounters that occurred during the study period, 59,892 were new patient encounters that met the study criteria (40,603 adults and 19,289 children) (Fig. 1). Most patients self-reported white race (86% [34,774 of 40,603] of adults; 78% [15,112 of 19,289] of children) (Table 1). Patients were evenly distributed across the national Area Deprivation Index quartiles, but few patients lived in a designated Medically Underserved Area (8% [3130 of 40,603] of adults; 7% [1322 of 19,289] of children), and most patients lived in an urban metropolitan Rural-Urban Community Area classification (89% [35,221 of 39,764] of adults; 83% [15,361 of 18,480] of children). The mean scores for all PROMIS measures varied based on insurance classification for both adults (Table 2) and children (Table 3), with patients enrolled in private insurance consistently reporting better scores than those with government-sponsored Medicare or Medicaid/Medicaid replacement insurance.

Fig. 1.

This flowchart depicts reasons for patient inclusion and exclusion from the study.

Table 1.

Patient characteristics (n = 59,892)

	Adults (n = 40,603)		Children (n = 19,289)
	Number	Median (P25, P75) or % (n)	Number	Median (P25, P75) or % (n)
Age in years	40,603	55 (41, 65)	19,289	12 (8, 15)
Sex (female)	40,603	57 (23,225)	19,289	50 (9609)
Race	40,401		19,106
White/Caucasian		86 (34,774)		78 (15,112)
Black/African American		11 (4298)		16 (3081)
Asian		2 (625)		1 (244)
Other		2 (704)		4 (669)
Ethnicity (Hispanic/Latino origin)	40,538	1 (572)	19,238	4 (761)
Social deprivation measures
National ADI	33,862		16,744
Quartile 1 (least deprived)		26 (8956)		26 (4293)
Quartile 2		26 (8784)		22 (3736)
Quartile 3		25 (8362)		23 (3890)
Quartile 4 (most deprived)		23 (7760)		29 (4825)
State ADI	33,862		16,744
Decile		3 (1, 5)		3 (1, 6)
MUA	40,603		19,289
Underserved		8 (3130)		7 (1322)
Not underserved		92 (37,473)		93 (17,967)
RUCA	39,764		18,480
Urban metropolitan (1-3)		89 (35,221)		83 (15,361)
Urban micropolitan (4-6)		5 (1999)		7 (1374)
Rural city/town (7-9)		5 (1814)		6 (1167)
Rural area (10)		2 (730)		3 (578)
Insurance classification	40,603		19,289
Private		67 (27,282)		79 (15,228)
PPO / POS		61 (24,700)		65 (12,576)
HMO		6 (2582)		14 (2652)
Medicare		30 (12,240)		-
Medicaid/Medicaid replacement		2 (687)		19 (3654)
Other		1 (394)		2 (407)
PROMIS Scores
Anxiety	25,186	53.5 (45.7, 59.8]		-
Depression	32,821	47.5 (38.9, 54.3)		-
Pain Interference	33,071	61.6 (56.2, 66.9)	14,511	51.1 (44.7, 57.3)
Physical Function	33,062	40.0 (34.5, 46.9)		-
Peer Relationships		-	14,003	54.3 (46.4, 61.3)
Upper Extremity		-	14,481	47.3 (35.9, 57.2)
Mobility		-	14,541	41.3 (36.3, 49.1)

For state ADI decile, 1 is least deprived and 10 is most deprived; number represents the available sample size for each variable; ADI = Area Deprivation Index; MUA = Medically Underserved Area status; RUCA = Rural-Urban Commuting Area code; PPO = preferred provider organization; POS = point of service; HMO = health maintenance organization; PROMIS = Patient-reported Outcomes Measurement Information System.

Table 2.

Mean adult PROMIS scores by insurance classification

Insurance classification
PROMIS measure	Private (n = 27,282)		Medicare (n = 12,240)	Medicaid/Medicaid replacement (n = 687)	p value
	PPO / POS (n = 24,700)	HMO (n = 2582)
Anxiety	52 ± 10		55 ± 11	59 ± 11	< 0.001
	51 ± 10	53 ± 10
Depression	46 ± 10		49 ± 11	54 ± 12	< 0.001
	46 ± 9	47 ± 11
Pain Interference	61 ± 8		63 ± 8	66 ± 8	< 0.001
	61 ± 8	62 ± 8
Physical Function	42 ± 9		37 ± 9	36 ± 10	< 0.001
	42 ± 9	41 ± 9

Data presented as mean ± SD; minimum clinically important differences are approximately 2 to 3 points for PROMIS Anxiety and Depression and approximately 4 points for PROMIS Pain Interference and Physical Function [17, 32, 35, 54, 64]; p values represent comparison between private, Medicare, and Medicaid/Medicaid replacement classifications; PROMIS = Patient-reported Outcomes Measurement Information System; PPO = preferred provider organization; POS = point of service; HMO = health maintenance organization.

Table 3.

Mean pediatric PROMIS scores by insurance classification

PROMIS measure	Private (n = 15,228)
	PPO / POS (n = 12,576)	HMO (n = 2652)	Medicaid/Medicaid replacement (n = 3654)		p value
Peer Relationships	54 ± 10		50 ± 10		< 0.001
	54 ± 11	51 ± 10
Pain Interference	50 ± 9		53 ± 10		< 0.001
	50 ± 9	52 ± 10
Upper Extremity	46 ± 12		42 ± 12		< 0.001
	46 ± 12	43 ± 12
Mobility	44 ± 10		42 ± 10		< 0.001
	44 ± 10	43 ± 10

Data presented as mean ± SD; the minimum clinically important difference is approximately 3 points for these PROMIS measures [65]; p values represent comparison between private and Medicaid/Medicaid replacement classifications; PROMIS = Patient-reported Outcomes Measurement Information System; PPO = preferred provider organization; POS = point of service; HMO = health maintenance organization.

Variables

The definition of social deprivation for this study was taken to be “limited access to society’s resources due to poverty, discrimination, or other disadvantage,” as defined by the American Psychological Association. We chose this definition because it was developed by an established scientific organization, is descriptive yet broad, and is independent of any particular measure of the construct. In this study, we compared five measures of social deprivation. All domains are free of cost, can be determined from basic demographic information that is routinely collected as part of standard care, and have been used extensively in medical studies. The 2015 Area Deprivation Index was developed by Singh [61] and Kind et al. [27] to rank community-level social deprivation based on 17 variables, including residents’ housing quality, employment, and education [67]. Nine-digit ZIP code regions are ranked by national Area Deprivation Index percentile (with 1 indicating the least deprivation and 100 indicating the most deprivation) and by state Area Deprivation Index decile (with 1 indicating the least deprivation and 10 indicating the most). For this study, the national and state Area Deprivation Indices were assessed. Patients’ medically underserved area status is a binary determination based on their residential street address [19-22]. Medically Underserved Areas are designated by the US Health Resources and Services Administration based on four variables: a shortage of primary care providers, high infant mortality, high poverty, and/or a large elderly population. The 2013 Rural-Urban Commuting Area code is a designation from the US Department of Agriculture’s Economic Research Service based on the metropolitan population size and the “degree of urbanization and adjacency to metro areas” of a person’s residential county [66]. Using patients’ five-digit residential ZIP codes, patients are assigned to one of 10 Rural-Urban Commuting Area codes. Then, the codes are grouped into established categories (codes 1-3: urban metropolitan, codes 4-6: urban micropolitan, codes 7-9: rural city or town, and code 10: rural area). Finally, patients’ insurance classification was assessed as a fifth potential marker of social deprivation because it is reflective of income and employment history and has been correlated with orthopaedic patients’ symptom severity and outcomes [26, 71]. Patients were categorized into private (PPO/POS or HMO), Medicare (for adults only), or Medicaid/Medicaid replacement insurance. Because of the very low proportion of patients who presented with no insurance or some other type of insurance (such as, workers compensation or Medicare for children) and the heterogeneity of reasons why these patients had no/other insurance, these patients were excluded from the insurance-specific analyses. Of note, occasionally patients’ addresses could not be mapped to a corresponding Area Deprivation Index, Medically Underserved Area, or Rural-Urban Commuting Area, especially if a patient only had a post office box mailing address on file.

Patient-reported health was defined using the PROMIS computer adaptive test measures [11], which have been used widely in orthopaedic studies [2, 4, 5, 8]. At the study institution, these measures are collected from all orthopaedic patients on computer tablets (iPad mini; Apple) as standard care before their evaluation by an orthopaedic provider. In adults, physical health was assessed using the PROMIS Physical Function version 1.2 or version 2.0 and Pain Interference version 1.1. Behavioral health was quantified with the PROMIS Anxiety version 1.0 and Depression version 1.0. In children younger than 18 years, physical health was assessed using the PROMIS Pediatric Mobility version 1.0 or version 2.0, Upper Extremity version 1.0, and Pain Interference version 1.0 or version 2.0. To avoid the potential for provoking psychological distress by asking children to answer questions about depression or anxiety when presenting for an orthopaedic problem, the PROMIS Pediatric Peer Relationships version 1.0 is used at our institution as a proxy for social and behavioral functioning in children. For PROMIS domains in which the computer adaptive test version was updated midway through the study, scores on both versions were combined for analysis because the upgraded PROMIS computer adaptive test versions have performed equivalently to previous versions [46, 47]. The Adult PROMIS Anxiety domain was implemented 10 months after the other domains, so these data are systematically missing for some patients but without the expectation of a directional bias.

All PROMIS domains used in this study are normalized to the general US population with a mean score of 50 and SD of 10. Higher scores represent more of the domain, such that a score of 60 on the PROMIS Physical Function represents more (that is, better) functioning than average, but a score of 60 on the PROMIS Pain Interference represents more (that is, worse) pain interference than average. In the orthopaedic evidence, the minimum clinically important differences (MCIDs) for these PROMIS domains have generally been cited as approximately 4 points for Adult PROMIS Physical Function and Pain Interference and 2 to 3 points for Adult PROMIS Anxiety and Depression [17, 32, 35, 54, 64]. The MCIDs in Pediatric PROMIS measures have not specifically been identified in the orthopaedic population but are thought to be approximately 3 points for a medical pediatric population [65].

Primary and Secondary Study Outcomes

Our primary study goal was to identify the measure which most thoroughly captures potential effects from social deprivation on patient-reported health in patients with orthopaedic conditions. To achieve this, we determined which measure of social deprivation most strongly and consistently correlated with patient-reported physical and behavioral health, as measured by age-adjusted partial correlation coefficients between each social deprivation measure and each PROMIS domain.

Our secondary study goal was to determine whether simultaneous use of multiple social deprivation measures would meaningfully capture more variability in patient-reported health when compared with the use of a single social deprivation measure at a time. To achieve this, we calculated the percentage of PROMIS score variability that could be attributed to the addition of each social deprivation measure one by one by using an age-adjusted hierarchical regression analysis.

Ethical Approval

Ethical approval for this study was obtained from the Washington University School of Medicine (IRB ID #: 201801039).

Statistical Analysis

Descriptive statistics for all study variables were computed separately for adults and children. The primary analysis consisted of age-adjusted partial Pearson correlations between each of the five social deprivation measures and each of the eight PROMIS measures. Age (in years) was treated as a continuous variable. We chose to adjust for age because age is recorded in a standard manner in orthopaedic clinical research (in years), it influences a wide range of health domains, and clinical analyses in orthopaedics very typically adjust for this parameter. Partial correlations were computed by correlating residuals from pairs of linear regressions. The first regression of each pair predicted a social deprivation measure based on age, and the second regression predicted a PROMIS measure based on age. Partial correlations with categorical predictors such as insurance classification were estimated via the square root of delta r² from two-step hierarchical regressions that predicted age-adjusted PROMIS scores in the first step and the categorical variable in the second step. The national percentile and state decile Area Deprivation Indices were treated as continuous variables, Medically Underserved Area status was binary, and Rural-Urban Commuting Area codes and insurance classification were treated as categorical variables. For patients with multiple insurance plans, only the primary insurance plan was considered (for example, a patient with Medicare and Medicaid insurance was categorized as having Medicare). The social deprivation measure exhibiting the largest partial correlation in absolute value with a given PROMIS measure was interpreted as the measure that accounts for the most variability in patient-reported health based on social deprivation. Given that the purpose of this study was to identify relevant covariates and confounders (rather than primary independent variables), correlation coefficients of 0.1 or greater were considered clinically meaningful [36]. As a subanalysis to assess whether the patterns of correlation were consistent across the demographic spectrum, partial correlations were calculated for the following subgroups: male sex, female sex, white race, and nonwhite race. Additionally, to assess whether consistent performance of a social deprivation measure across PROMIS measures was the result of truly consistent performance as opposed to interdependence of PROMIS measures, we created two correlation matrices among PROMIS measures: one for adults and one for children. To address the second research question, we performed a hierarchical linear regression analysis for each PROMIS measure. Starting with age as the only independent variable, each social deprivation measure was sequentially added to the linear regression model in the order of the largest to smallest partial correlation in absolute value since, based on the first research question of the study, the measure(s) with the largest partial correlation would generally be recommended to be incorporated into clinical outcomes research as a marker of social deprivation. We compared each model’s coefficients of determination (r²) as social deprivation measures were incrementally added. We considered improvement in the total r² by more than 10% as clinically meaningful. Given the robust sample size, data imputation was unnecessary, and patients with missing data were included whenever possible. All tests were two-sided, and a significance level of p < 0.05 was used for all analyses. SAS Base version 9.4 (SAS Institute Inc) and R (version 4.0.2) were used for all calculations and to determine patients’ Area Deprivation Index, Medically Underserved Area, and Rural-Urban Commuting Area statuses by geocoding their residential addresses to recover the corresponding Federal Information Processing Standards codes.

Results

Comparison of Social Deprivation Measures

Of the social deprivation measures we compared, insurance classification had the largest and most consistent relationship with patients’ physical and behavioral health measures (Fig. 2). That is, insurance classification was the social deprivation measure with the largest (absolute value) age-adjusted correlation coefficient for all adult (Table 4) and pediatric (Table 5) PROMIS health domains. All age-adjusted correlation coefficients between insurance classification and PROMIS measures met the predetermined threshold for clinical meaningfulness as well (adults: correlation coefficient 0.40 to 0.43 [95% CI 0.39 to 0.44]; pediatrics: correlation coefficient 0.10 to 0.19 [95% CI 0.08 to 0.21]; all p < 0.001). Consistently, the social deprivation measure with the second largest age-adjusted correlation coefficient for all assessed domains was the national Area Deprivation Index (adults: correlation coefficient 0.18 to 0.22 [95% CI 0.17 to 0.23]; pediatrics: correlation coefficient 0.08 to 0.15 [95% CI 0.06 to 0.17), followed closely by state Area Deprivation Index. PROMIS Pediatric Mobility was the only health domain whose correlation coefficient with the national and state Area Deprivation Index did not meet the predetermined threshold to be considered clinically meaningful. The Medically Underserved Area classification and Rural-Urban Commuting Area codes each had correlation coefficients of 0.1 or larger for some PROMIS domains, but neither had consistently stronger coefficients than the other. The correlations between patient-reported health and social deprivation were generally weaker in the pediatric population than in adults. The correlation between patient-reported health and social deprivation for the populations subgrouped by sex (male and female) and race (white and nonwhite) revealed patterns similar to the analysis of the entire population together (Supplemental Digital Content 1; http://links.lww.com/CORR/A660). Correlation matrices of PROMIS measures revealed coefficient absolute values ranging from 0.39 to 0.76 between PROMIS adult measures and from 0.05 to 0.58 between PROMIS pediatric measures (Supplemental Digital Content 2; http://links.lww.com/CORR/A661).

Fig. 2.

A-B These plots depict the age-adjusted correlations (y-axis) between social deprivation measures and self-reported health on PROMIS measures (x-axis) in (A) adults and (B) children. Error bars represent 95% CIs. A color image accompanies the online version of this article.

Table 4.

Absolute values of age-adjusted correlation coefficients between social deprivation measures and adult PROMIS measures

Social deprivation measures	Anxiety			Depression			Pain Interference			Physical Function
	Number	r (95% CI)	p value	Number	r (95% CI)	p value	Number	r (95% CI)	p value	Number	r (95% CI)	p value
MUA	24,941	0.04 (0.02-0.05)	< 0.001	32,515	0.02 (0.01-0.03)	< 0.001	32,761	0.02 (0.01-0.03)	0.002	32,753	0.05 (0.04-0.06)	< 0.0 001
RUCA	24,433	0.11 (0.10-0.13)	< 0.001	31,854	0.08 (0.06-0.09)	< 0.001	32,097	0.11 (0.10-0.12)	< 0.001	32,088	0.04 (0.03-0.05)	< 0.001
National ADI	20,916	0.22 (0.20-0.23)	< 0.001	27,141	0.18 (0.17-0.20)	< 0.001	27,355	0.21 (0.20-0.22)	< 0.001	27,349	0.19 (0.18-0.20)	< 0.001
State ADI	20,916	0.21 (0.19-0.22)	< 0.001	27,141	0.17 (0.16-0.18)	< 0.001	27,355	0.19 (0.18-0.21)	< 0.001	27,349	0.18 (0.16-0.19)	< 0.001
Insurance classification	24,941	0.40 (0.39-0.41)	< 0.001	32,515	0.43 (0.42-0.44)	< 0.001	32,761	0.40 (0.39-0.41)	< 0.001	32,753	0.43 (0.42-0.44)	< 0.001

PROMIS = Patient-reported Outcomes Measurement Information System; MUA = Medically Underserved Area status; RUCA = Rural-Urban Commuting Area code; ADI = Area Deprivation Index.

Table 5.

Absolute values of age-adjusted correlation coefficients between social deprivation measures and pediatric PROMIS measures

Social deprivation measures	Peer Relationships			Pain Interference			Upper Extremity			Mobility
	Number	r (95% CI)	p value	Number	r (95% CI)	p value	Number	r (95% CI)	p value	Number	r (95% CI)	p value
MUA	13,839	0.03 (0.01-0.05)	< 0.001	14,334	0.03 (0.02-0.05)	< 0.001	14,304	0.11 (0.10-0.13)	< 0.001	14,363	0.00 (-0.02-0.02)	0.90
RUCA	13,355	0.01 (-0.01-0.03]	0.42	13,841	0.06 (0.04-0.08)	< 0.001	13,812	0.02 (0.00-0.04)	0.014	13,869	0.04 (0.02-0.06)	< 0.001
National ADI	12,258	0.15 (0.13-0.17)	< 0.001	12,700	0.14 (0.12-0.16)	< 0.001	12,671	0.12 (0.10-0.14)	< 0.001	12,726	0.08 (0.06-0.10)	< 0.001
State ADI	12,258	0.12 (0.11-0.14)	< 0.001	12,700	0.13 (0.11-0.15)	< 0.001	12,671	0.12 (0.10-0.14)	< 0.001	12,726	0.08 (0.06-0.10)	< 0.001
Insurance classification	13,839	0.18 (0.16-0.20)	< 0.001	14,334	0.19 (0.17-0.21)	< 0.001	14,304	0.15 (0.14-0.17)	< 0.001	14,363	0.10 (0.08-0.12)	< 0.001

PROMIS = Patient-reported Outcomes Measurement Information System; MUA = Medically Underserved Area status; RUCA = Rural-Urban Commuting Area code; ADI = Area Deprivation Index.

Simultaneous Use of Multiple Social Deprivation Measures

Overall, insurance classification combined with the national Area Deprivation Index captured more variability in patient-reported health than use of either measure in isolation. That is, insurance classification and national Area Deprivation Index together explained more of the variation in age-adjusted PROMIS scores than did the use of insurance classification alone for all adult PROMIS assessments (Table 6) and all pediatric PROMIS assessments except for the Upper Extremity domain (Table 7). The addition of the national Area Deprivation Index to the hierarchical regression analysis increased the r² by 32% to 189% (95% CI 0.02 to 0.04) for the PROMIS adult measures and by 56% to 110% (95% CI 0.01 to 0.02) for the PROMIS pediatric measures. Addition of the Medically Underserved Area status, Rural-Urban Commuting Area code, and/or state Area Deprivation Index did not further improve the r² for any of the PROMIS domains to a meaningful degree.

Table 6.

Hierarchical linear regression analyses that include age and social deprivation measures as independent variables and adult PROMIS scores as dependent variables

Predictors in model	PROMIS measure
	Anxiety	Depression	Pain Interference	Physical Function
Age only	0.01	0.01	0.01	0.04
Insurance classification	0.03	0.03	0.02	0.07
	366%	327%	227%	58%
Insurance classification + national ADI	0.06	0.05	0.05	0.09
	140%	90%	189%	32%
Insurance classification + national ADI + RUCA	0.06	0.05	0.05	0.09
	0%	1%	1%	1%
Insurance classification + national ADI + RUCA + MUA	0.06	0.05	0.05	0.09
	0%	0%	1%	1%
Insurance classification + national ADI + RUCA + MUA + state ADI	0.06	0.05	0.05	0.09
	0%	0%	0%	0%

Top row: R² of a hierarchical linear regression model that includes age and the social deprivation measures listed; bottom row: interval percent improvement in the R² by adding another social deprivation measure to the model; PROMIS = Patient-reported Outcomes Measurement Information System; ADI = Area Deprivation Index; RUCA = Rural-Urban Commuting Area code; MUA = Medically Underserved Area status.

Table 7.

Hierarchical linear regression analyses that include age and social deprivation measures as independent variables and pediatric PROMIS scores as dependent variables

Predictors in model	PROMIS measure
	Peer Relationships	Pain Interference	Upper Extremity	Mobility
Age only	0.01	0	0.1	0
Insurance classification	0.03	0.01	0.10	0.004
	120%	10,323%	8%	918%
Insurance classification + national ADI	0.04	0.03	0.11	0.01
	56%	83%	8%	110%
Insurance classification + national ADI + RUCA	0.04	0.03	0.11	0.01
	5%	1%	0%	9%
Insurance classification + national ADI + RUCA + MUA	0.04	0.03	0.11	0.01
	1%	0%	0%	0%
Insurance classification + national ADI + RUCA + MUA + state ADI	0.05	0.03	0.11	0.01
	3%	0%	0%	0%

Top row: R² of a hierarchical linear regression model that includes age and the social deprivation measures listed; bottom row: interval percent improvement in the R² by adding another social deprivation measure to the model; PROMIS = Patient-reported Outcomes Measurement Information System; ADI = Area Deprivation Index; RUCA = Rural-Urban Commuting Area code; MUA = Medically Underserved Area status.

Discussion

Social deprivation influences physical and behavioral health in the general population and particularly in patients with orthopaedic conditions [7, 13, 76]. To truly understand and improve clinical outcomes, social deprivation must be addressed and appropriately accounted for in clinical research [37]. To this end, we evaluated five markers of social deprivation to determine which measure best accounts for variability in patient-reported health. We found that insurance classification consistently correlated the strongest with age-adjusted adult and pediatric physical and behavioral health, followed by national Area Deprivation Index percentile. The state Area Deprivation Index decile performed almost as well as the national Area Deprivation Index percentile, and patients’ Medically Underserved Area status and Rural-Urban Commuting Area codes were minimally correlated with self-reported health. Correlations between social deprivation measures and patient-reported health were somewhat weaker in the pediatric population than in the adult population. For all adult health domains and most of the assessed pediatric health domains, including insurance classification and national Area Deprivation Index together meaningfully captured more variability in patients’ age-adjusted self-reported health than did the use of insurance classification alone. When conducting clinical outcomes research with social deprivation as a relevant covariate, we encourage researchers to include patients’ insurance classification and/or national Area Deprivation Index in the analysis, both of which are freely available and can be obtained from data that are typically collected during standard care. Insurance classification may be more readily available, but the national Area Deprivation Index stratifies patients across a wider distribution of values.

Limitations

This study has several limitations. First, it was not always possible to determine all measures of social deprivation for a given patient. For instance, some patients’ addresses could not be mapped to a corresponding Area Deprivation Index, Medically Underserved Area, and/or Rural-Urban Commuting Area, especially if a patient only had a post office box mailing address on file. These patients might have been more likely to live in a rural area or to have unstable housing because of social deprivation. Additionally, most study patients had at least some type of insurance. Therefore, uninsured patients were not included in the insurance-specific analyses, and these patients are clearly at risk of being in poor health because of social deprivation. Because of the low proportion of adults with workers compensation insurance and pediatric patients with Medicare insurance, these patients were also excluded from the insurance-related analyses. These limitations may have resulted in an underestimated correlation between social deprivation measures and patient-reported health because the full spectrum of social deprivation (including unstable housing or lack of insurance) could not be represented in this study. Nevertheless, the correlation and r² magnitudes calculated in this study are consistent with clinically meaningful associations between social deprivation and self-reported health. The magnitudes are not large enough to create a robust prediction model of patient-reported health based solely on social deprivation, and that was not the intention or expectation of this study. Misclassification of the address-based social deprivation measures could have also occurred because the measures rely on state and national surveys (such as the US Census), which are only administered every few years. However, this likely would have resulted in a random error. Finally, two statistical limitations are worth noting. First, some PROMIS data were missing. However, we do not anticipate this was a large source of systematic bias because once phased implementation of PROMIS measures was complete across our orthopaedic department, our study population had a 94% PROMIS completion rate. Second, because multiple social deprivation measures and multiple patient-reported health outcomes were included in the analysis, there is a risk of statistical Type I error (such as, false positive findings). Nevertheless, the robust sample size and small p values (almost all p < 0.001) of our analyses suggest this risk was low.

Regarding the generalizability of our study findings, the primary limitations are that our patient population was predominantly white, we had a low prevalence of patients with Medicaid/Medicaid replacement insurance, and all patients were evaluated at a single institution. Nevertheless, 22% of our patients resided in the worst national Area Deprivation Index quartile, our department has a multistate urban and rural catchment area, and patients were evaluated across five clinic sites in urban and suburban regions in the highest and lowest national Area Deprivation Index quartiles. Furthermore, our robust sample size included patients who presented to a variety of types of orthopaedic providers and clinic settings (physicians and midlevel providers, surgeons and nonsurgeons, and scheduled and walk-in appointment types). Because patient-reported health measures were recorded before evaluation by the provider, these findings might be generalized to nonorthopaedic patients as well. Nevertheless, our findings represent cross-sectional associations between social deprivation measures and a single set of patient-reported measures (PROMIS). Therefore, the results of this study should not be used to definitively assert which measure “best” captures social deprivation for all longitudinal clinical outcomes studies, especially when outcomes other than PROMIS measures are being considered.

Comparison of Social Deprivation Measures

Of the social deprivation measures compared in this study, insurance classification had the largest and most consistent relationship with patients’ physical and behavioral health measures. Although we know of no other study that has directly compared multiple social deprivation measures in patients with orthopaedic conditions, our findings are relatively consistent with previous studies that used one or two of these measures in isolation. Specifically, compared with having private insurance, having Medicaid insurance is independently associated with lower levels of physical activity in children, a higher prevalence of untreated depression, and reduced feasibility of accessing pediatric and adult orthopaedic care and physical therapy for conditions ranging from back pain to fractures [1, 15, 28, 34, 39, 42, 52, 56, 59]. Furthermore, national and state Area Deprivation Indices are associated with numerous other negative health measures and outcomes, including a higher prevalence of obesity, less daily physical activity, increased opioid use and opioid-related harms, and worse pain, function, and psychosocial well-being in adult and pediatric patients with orthopaedic conditions such as fractures, carpal tunnel syndrome, shoulder osteoarthritis, and knee pain [12, 18, 24, 33, 40, 44, 50, 58, 60, 67, 77]. Although the state Area Deprivation Index deciles performed nearly as well as the national Area Deprivation Index percentiles in our study, we advocate for the use of the national Area Deprivation Index designation because it provides a more standard comparison across studies and within a single study that may have a catchment area that crosses state lines. Interestingly, rural communities are known to have reduced access to orthopaedic care [38, 73, 74], but studies that have evaluated disparities based on Rural-Urban Commuting Area codes have reported mixed results. Increased rurality measured by Rural-Urban Commuting Area codes is associated with an increased prevalence of obesity but it is not associated with differences in dietary intake or physical activity [14]. Similarly, a previous study involving the Medically Underserved Area designation found that living in a Medically Underserved Area might be associated with a reduced likelihood of receiving necessary medical care [25]. However, Medically Underserved Area status has not been found to correlate with most physical functioning, mental health, or pain measures [29], and there has been a longstanding call to revise the variables considered in the federal government’s designation of Medically Underserved Areas [49, 51, 78]. In our study and in the established evidence, we hypothesize that the Area Deprivation Index may be a more robust measure than the Rural-Urban Commuting Area codes or Medically Underserved Area classification because it considers more variables and scores residents across a wider scale. We suspect that the heterogeneity of the social deprivation measures (such as, the number/type of input variables, available score range, and individual versus neighborhood-level metrics) contributed to the differential associations we found between these measures and patients’ self-reported health.

Simultaneous Use of Multiple Social Deprivation Measures

Insurance classification combined with the national Area Deprivation Index captured more variability in patient-reported health than either measure used in isolation; however, consideration of additional social deprivation measures (the Medically Underserved Area classification, Rural-Urban Community Area codes, and/or state Area Deprivation Index) did not capture meaningfully more variability in patients’ health measures. In most cases, considering insurance class and national Area Deprivation Index together achieves the most efficient balance between creating a parsimonious model and capturing variability in patients’ age-adjusted health based on social deprivation. However, because there is some overlap but also some divergence in how each measure is related to social deprivation, some situations may warrant the use of fewer or more than these two variables. For instance, it may be appropriate to consider insurance classification in isolation when assessing the effect of differential insurance coverage or Medicaid expansion [10, 31, 70], but insurance classification would likely not be an ideal variable to use in a patient population that predominantly has the same type of insurance. As another example, in some situations, it might be appropriate to consider specific markers of both rurality (using Rural-Urban Commuting Area codes) and social deprivation (using insurance class and/or national Area Deprivation Index) because the social deprivation measures assessed in this study were not designed to capture the same constructs and are not truly interchangeable [16, 57].

Conclusion

Social deprivation, defined as “limited access to society’s resources due to poverty, discrimination, or other disadvantage,” influences patients’ clinical presentations and access to treatment. When conducting clinical research, it is essential to consider the potential influence of social deprivation on clinical outcomes. Our study findings suggest that simultaneous consideration of patients’ insurance classification (categorized as private, Medicare, Medicaid, or other) and national Area Deprivation Index percentile (determined by residential nine-digit ZIP code) captures the most variability in self-reported physical and behavioral health related to social deprivation. If only one measure of social deprivation is preferred, either insurance classification or national Area Deprivation Index would be a reasonable option. Insurance classification may be more readily available, but the national Area Deprivation Index stratifies patients across a wider distribution of values. When conducting clinical outcomes research with social deprivation as a relevant covariate, we encourage researchers to include patients’ insurance classification and/or national Area Deprivation Index in the analysis. Both variables are freely available and can be obtained from data that are typically collected during standard care.