A New Measure of Quantified Social Health Is Associated With Levels of Discomfort, Capability, and Mental and General Health Among Patients Seeking Musculoskeletal Specialty Care

Abstract

Background

A better understanding of the correlation between social health and mindsets, comfort, and capability could aid the design of individualized care models. However, currently available social health checklists are relatively lengthy, burdensome, and designed for descriptive screening purposes rather than quantitative assessment for clinical research, patient monitoring, or quality improvement. Alternatives such as area deprivation index are prone to overgeneralization, lack depth in regard to personal circumstances, and evolve rapidly with gentrification. To fill this void, we aimed to identify the underlying themes of social health and develop a new, personalized and quantitative social health measure.

Questions/purposes

(1) What underlying themes of social health (factors) among a subset of items derived from available legacy checklists and questionnaires can be identified and quantified using a brief social health measure? (2) How much of the variation in levels of discomfort, capability, general health, feelings of distress, and unhelpful thoughts regarding symptoms is accounted for by quantified social health?

Methods

In this two-stage, cross-sectional study among people seeking musculoskeletal specialty care in an urban area in the United States, all English and Spanish literate adults (ages 18 to 89 years) were invited to participate in two separate cohorts to help develop a provisional new measure of quantified social health. In a first stage (December 2021 to August 2022), 291 patients rated a subset of items derived from commonly used social health checklists and questionnaires (Tool for Health and Resilience in Vulnerable Environments [THRIVE]; Protocol for Responding to and Assessing Patient Assets, Risks and Experiences [PRAPARE]; and Accountable Health Communities Health-Related Social Needs Screening Tool [HRSN]), of whom 95% (275 of 291; 57% women; mean ± SD age 49 ± 16 years; 51% White, 33% Hispanic; 21% Spanish speaking; 38% completed high school or less) completed all items required to perform factor analysis and were included. Given that so few patients decline participation (estimated at < 5%), we did not track them. We then randomly parsed participants into (1) a learning cohort (69% [189 of 275]) used to identify underlying themes of social health and develop a new measure of quantified social health using exploratory and confirmatory factor analysis (CFA), and (2) a validation cohort (31% [86 of 275]) used to test and internally validate the findings on data not used in its development. During the validation process, we found inconsistencies in the correlations of quantified social health with levels of discomfort and capability between the learning and validation cohort that could not be resolved or explained despite various sensitivity analyses. We therefore identified an additional cohort of 356 eligible patients (February 2023 to June 2023) to complete a new extended subset of items directed at financial security and social support (5 items from the initial stage and 11 new items derived from the Interpersonal Support Evaluation List, Financial Well-Being Scale, Multidimensional Scale of Perceived Social Support, Medical Outcomes Study Social Support Survey, and 6-item Social Support Questionnaire, and “I have to work multiple jobs in order to finance my life” was self-created), of whom 95% (338 of 356; 53% women; mean ± SD age 48 ± 16 years; 38% White, 48% Hispanic; 31% Spanish speaking; 47% completed high school or less) completed all items required to perform factor analysis and were included. We repeated factor analysis to identify the underlying themes of social health and then applied item response theory–based graded response modeling to identify the items that were best able to measure differences in social health (high item discrimination) with the lowest possible floor and ceiling effects (proportion of participants with lowest or highest possible score, respectively; a range of different item difficulties). We also assessed the CFA factor loadings (correlation of an individual item with the identified factor) and modification indices (parameters that suggest whether specific changes to the model would improve model fit appreciably). We then iteratively removed items based on low factor loadings (< 0.4, generally regarded as threshold for items to be considered stable) and high modification indices until model fit in CFA was acceptable (root mean square of error approximation [RMSEA] < 0.05). We then assessed local dependencies among the remaining items (strong relationships between items unrelated to the underlying factor) using Yen Q3 and aimed to combine only items with local dependencies of < 0.25. Because we exhausted our set of items, we were not able to address all local dependencies. Among the remaining items, we then repeated CFA to assess model fit (RMSEA) and used Cronbach alpha to assess internal consistency (the extent to which different subsets of the included items would provide the same measurement outcomes). We performed a differential item functioning analysis to assess whether certain items are rated discordantly based on differences in self-reported age, gender, race, or level of education, which can introduce bias. Last, we assessed the correlations of the new quantified social health measure with various self-reported sociodemographic characteristics (external validity) as well as level of discomfort, capability, general health, and mental health (clinical relevance) using bivariate and multivariable linear regression analyses.

Results

We identified two factors representing financial security (11 items) and social support (5 items). After removing problematic items based on our prespecified protocol, we selected 5 items to address financial security (including “I am concerned that the money I have or will save won’t last”) and 4 items to address social support (including “There is a special person who is around when I am in need”). The selected items of the new quantified social health measure (Social Health Scale [SHS]) displayed good model fit in CFA (RMSEA 0.046, confirming adequate factor structure) and good internal consistency (Cronbach α = 0.80 to 0.84), although there were some remaining local dependencies that could not be resolved by removing items because we exhausted our set of items. We found that more disadvantaged quantitative social health was moderately associated with various sociodemographic characteristics (self-reported Black race [regression coefficient (RC) 2.6 (95% confidence interval [CI] 0.29 to 4.9)], divorced [RC 2.5 (95% CI 0.23 to 4.8)], unemployed [RC 1.7 (95% CI 0.023 to 3.4)], uninsured [RC 3.5 (95% CI 0.33 to 6.7)], and earning less than USD 75,000 per year [RC 2.7 (95% CI 0.020 to 5.4) to 6.8 (95% CI 4.3 to 9.3)]), slightly with higher levels of discomfort (RC 0.055 [95% CI 0.16 to 0.093]), slightly with lower levels of capability (RC -0.19 [95% CI -0.34 to -0.035]), slightly with worse general health (RC 0.13 [95% CI 0.069 to 0.18]), moderately with higher levels of unhelpful thoughts (RC 0.17 [95% CI 0.13 to 0.22]), and moderately with greater feelings of distress (RC 0.23 [95% CI 0.19 to 0.28]).

Conclusion

A quantitative measure of social health with domains of financial security and social support had acceptable psychometric properties and seems clinically relevant given the associations with levels of discomfort, capability, and general health. It is important to mention that people with disadvantaged social health should not be further disadvantaged by using a quantitative measure of social health to screen or cherry pick in contexts of incentivized or mandated reporting, which could worsen inequities in access and care. Rather, one should consider disadvantaged social health and its associated stressors as one of several previously less considered and potentially modifiable aspects of comprehensive musculoskeletal health.

Clinical Relevance

A personalized, quantitative measure of social health would be useful to better capture and understand the role of social health in comprehensive musculoskeletal specialty care. The SHS can be used to measure the distinct contribution of social health to various aspects of musculoskeletal health to inform development of personalized, whole-person care pathways. Clinicians may also use the SHS to identify and monitor patients with disadvantaged social circumstances. This line of inquiry may benefit from additional research including a larger number of items focused on a broader range of social health to further develop the SHS.

Introduction

Social health (or circumstances) is a combination of role and identity (such as socioeconomic status, level of education, employment type; one’s position in society), security and stability (such as housing stability, food security, job security, access to transportation), and social relationships (such as friends, family, cohesive communities; social support) [2, 22]. There is evidence that certain aspects of social health are associated with variations in levels of comfort and capability [12, 26, 29, 34], less healthy mindsets (such as feelings of distress or unhelpful thoughts; mental health) [16, 20, 26, 30, 36, 48], personal health agency [14], and adverse events and urgent and emergency care visits after surgery [23, 24].

A better understanding of the relationship between social, mental, and physical health could aid the design of more comprehensive, individualized care strategies. However, the social health questionnaires currently available, such as the Tool for Health and Resilience in Vulnerable Environments (THRIVE) [44]; the Protocol for Responding to and Assessing Patient Assets, Risks, and Experiences (PRAPARE) [32]; and the Accountable Health Communities Health-Related Social Needs Screening Tool (HRSN) [5], are relatively long (10 to 20 items), burdensome, and designed for descriptive screening purposes rather than quantification. These instruments use checklists for identification of specific social health needs (such as problems with mold or leakage related to housing) and are not designed to quantify social health for research, quality improvement, or patient monitoring. Although area deprivation indices are quantitative and less burdensome, such metrics are prone to overgeneralization, lack depth in regard to personal circumstances, and evolve rapidly with gentrification. A personalized, quantitative measure of social health could facilitate analysis of the relative impact of social health and mental health to levels of discomfort and incapability. Uncovering these relationships using practical, user-friendly measures would provide important building blocks for more personalized, whole-person musculoskeletal care. To fill this void, we aimed to develop a brief, personalized social health measure that quantifies the most impactful aspects of a patient’s social circumstances among patients seeking musculoskeletal specialty care. Subsequently, we also assessed the clinical relevance of such a measure by seeking the correlation between quantified social health with various aspects of musculoskeletal health. Therefore, we asked: (1) What underlying themes of social health (factors) among a subset of items derived from available legacy checklists and questionnaires can be identified and quantified using a brief social health measure? (2) How much of the variation in levels of discomfort, capability, general health, feelings of distress, and unhelpful thoughts regarding symptoms is accounted for by quantified social health?

Patients and Methods

Study Design and Setting

In this cross-sectional study, all new and returning patients who presented to one of two musculoskeletal specialty offices in an urban area in the United States were invited to participate by one of several research assistants who were not involved in their care. Both offices have a relatively high percentage of patients referred by regional federally qualified health centers and covered under county safety net insurance. We intentionally limited enrollment to these offices to have a wide spectrum of social health and good representation of people who are socially disadvantaged. All patients provided verbal consent, and our institutional review board accepted completion of the questionnaires as confirmation of informed consent. English- or Spanish-speaking patients ages 18 years or older were deemed eligible, and patients with cognitive dysfunction or who could not read English or Spanish were excluded. All participants completed surveys before or after their clinic visit with one of the surgeons, residents, and/or physician assistants using a tablet with Health Insurance and Portability Accounting Act–compliant software (Research Electronic Data Capture). The questionnaires were translated to Spanish by two independent native-speaking researchers. One of the researchers translated the survey to Spanish, and the second translated it back to English to validate the translation, a process that led to a few minor edits, confirming adequate translation.

We conducted this study in two stages because of observed inconsistencies in the external consistency of the social health measure developed in the first stage (Supplemental Table 1; http://links.lww.com/CORR/B382). We found that social health was correlated with levels of discomfort and capability in the cohort used to develop the measure but not in the cohort used to validate the findings (Supplemental Table 2; http://links.lww.com/CORR/B382). We performed multiple sensitivity analyses and omitted various outlying observations, but we were not able to address or explain these inconsistencies. We therefore decided to enroll an additional cohort incorporating several modifications to ensure the development of a consistent and reliable measure. First, we enrolled a larger sample size and only approached patients in clinics who were more likely to have greater variation in social health, including individuals with safety-net insurance and lower socioeconomic status. Second, we used the findings of the initial factor analysis to include a more extensive set of items focused on financial security, alongside additional items focused on social support, given the evidence underscoring the importance of emotional and instrumental social support as important drivers of comfort and capability [33, 38, 40, 49]. Third, we included a greater number of explanatory variables to account for more potential confounders in case we found inconsistencies again. Last, we assessed a larger number of potentially relevant parameters derived from a more elaborate selection of statistical analyses.

Participants

In the first stage (December 2021 to August 2022), we enrolled 291 patients, of whom 95% (275 of 291) completed all social health items required to perform a factor analysis and were included. Given that so few patients decline participation (estimated at < 5%), we did not track the number of approached patients. The participants were parsed into a learning cohort (69% [189 of 275]) used to develop a new social health measure and a validation cohort (31% [86 of 275]) used to test and internally validate the findings from the learning cohort on new data not used in its development (Supplemental Table 1; http://links.lww.com/CORR/B382).

For the second stage (February 2023 to June 2023), we enrolled an additional 356 patients, of whom 95% (338 of 356) completed all items directed at social health required to perform factor analysis and were included.

Descriptive Data

In the initial stage, 79% (217 of 275) were English speakers, 33% (91 of 275) identified as Hispanic, 57% (158 of 275) identified as women, and the mean age was 49 ± 16 years (Supplemental Table 1; http://links.lww.com/CORR/B382). Thirty-eight percent (104 of 275) reported high school or less as their highest level of education.

In the second stage, 69% (232 of 338) of the cohort were English speakers, 48% (162 of 338) identified as Hispanic, 53% (179 of 338) identified as women, and the mean ± SD age was 48 ± 16 years (Table 1). Forty-seven percent (159 of 338) reported high school or less as their highest level of education, 33% (111 of 338) were unemployed, and 48% (162 of 338) had an annual household income of below USD 24,000. Spanish speakers had a considerably larger proportion of incomplete responses compared with English speakers (10% [11 of 106] versus 2% [4 of 232]). We suggest that this might be because of lower literacy among this population as the completion time seemed to be longer for Spanish participants as well, and this is supported by available evidence [4, 7].

Table 1.

Descriptive data on the participants enrolled in the additional cohort (Stage 2) that were used to address the inconsistencies found in the initial stage and help develop a new measure of quantified social health

Variable	Total cohort (n = 338)
Age in years	48 ± 16
Gender
Women	53 (179)
Men	47 (158)
Nonbinary	0.30 (1)
Race/ethnicity
White	38 (128)
Hispanic	48 (162)
Black	10 (33)
Asian	3 (9)
Other	2 (6)
Marital status
Married	49 (167)
Single	37 (125)
Divorced	9 (29)
Widowed	5 (17)
Level of education
High school or less	47 (159)
2-year college	16 (53)
4-year college	21 (70)
Postcollege	17 (56)
Survey language
Spanish	31 (106)
English	69 (232)
Injury	38 (130)
Anatomical region
Upper extremity	28 (93)
Lower extremity	54 (183)
Spine	18 (62)
Adverse event	6 (19)
Employment status
Employed	45 (152)
Retired	12 (40)
Unemployed	33 (111)
Other (such as student, homemaker)	10 (35)
Annual household income in USD
< 24,000	48 (162)
24,000 to 46,000	16 (55)
46,001 to 75,000	8 (28)
75,001 to 121,000	10 (33)
> 121,000	18 (60)
Insurance type
Private	25 (84)
County safety net	42 (143)
Medicare	14 (46)
Medicaid	6 (19)
Other	7 (25)
Uninsured	6 (21)
Social health	13 ± 6.9
Financial security (subdomain)	8.5 ± 4.8
Social support (subdomain)	4.7 ± 3.6
PROMIS PF CAT score	41 ± 9.6
NRS pain intensity	5 (3-7)
EQ-5D-5L score	11 ± 4.0
EQ-5D VAS score	60 ± 24
Feelings of distress	8 (6-11)
Unhelpful thoughts	8 (6-10)

Data presented as mean ± SD, % (n), or median (IQR).

Included Items for Measurement Development

In the first stage, we selected items from the THRIVE [44], PRAPARE [32], and HRSN [5] screening tools to include in the development of the new social health measure, accounting for overlap between questions (Supplemental Table 3; http://links.lww.com/CORR/B382).

In the second stage, we selected the five best-performing questions from the initial factor analysis and added six new items directed at financial security (five items derived from the Interpersonal Support Evaluation List [8] and Financial Well-Being Scale [9], and one item was self-created [“I have to work multiple jobs in order to finance my life”]) and five new items directed at social support (derived from the Multidimensional Scale of Perceived Social Support [51], Medical Outcomes Study Social Support Survey [41], and 6-item Social Support Questionnaire [39]). We selected these items based on the findings of the first stage and the suspicion that social support was underrepresented in the initial factor analysis given the evidence that social relationships are an important component of social health [2, 22].

Development of New Social Health Measure

In the first stage, we used the learning subset to perform an exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) to identify groupings of items that were strongly correlated and seemed to resemble the same underlying theme. We also assessed which items were most strongly correlated with the underlying themes (high factor loadings). After selecting a subset of items to represent social health (Supplemental Table 1; http://links.lww.com/CORR/B382), we tested external consistency using Spearman rank correlation with nonparametric bootstrapping (with replacement n = 1000) and multivariable linear regression by assessing the correlation of the new social health measures with pain intensity and level of capability. We subsequently compared the findings between the learning and validation cohorts and found that social health was correlated with pain intensity and level of capability in the learning subset but not in the validation subset. The inconsistencies remained after various sensitivity analyses, including analyses to identify and omit outlying observations and resampling the learning/validation cohort (Supplemental Table 2; http://links.lww.com/CORR/B382). Because we were not able to address or explain these inconsistencies, we decided to enroll an entirely new cohort (Stage 2) with various modifications in an attempt to develop a more consistent and reliable measure.

In the second stage, we first used a parallel analysis to identify the optimal number of factors to retain in EFA. We then performed EFA and CFA to identify groupings of items that measure the same underlying construct. We assessed factor loadings (correlation of an individual item with the identified factor), R² (proportion of variation in the identified factor explained by an individual item), and modification indices (parameters that suggest whether specific changes to the model would improve model fit appreciably). We omitted items with factor loading of < 0.4 (given the evidence that such items are more likely to be unstable [18]) and addressed items with high modification indices until acceptable model fit in CFA was accomplished (root mean square of error approximation [RMSEA] < 0.05). The remaining items were moved to an item response theory (IRT)-based graded response model to assess item discrimination and item difficulty parameters. Item discrimination represents the extent to which an item is able to identify differences in social health among patients, in which higher values indicate a greater ability to differentiate levels of social health. Item difficulty represents the overall level of social health required to achieve a 50/50 chance of answering an item in a more positive or negative way. It essentially indicates where on the spectrum an item can identify differences in social health (higher or lower segment) and thus can be helpful to identify items with the highest potential to reduce floor and/or ceiling effects (proportion of people with the lowest or highest score possible, respectively). We also assessed local dependencies among items using the Yen Q3 approach [6, 50]. Local dependencies are strong relationships between items that exist outside the scope of the underlying factor, meaning that they remain after the contribution of the underlying factor is accounted for (a separate domain within a factor). Local dependencies can threaten the unidimensionality of a measure and can lead to biased parameter estimations and are thus best accounted for. We aimed to combine only items with values < 0.25 but were not able to account for all local dependencies because we exhausted our set of items. We also performed a differential item functioning (DIF) analysis to identify whether patients from certain demographic subgroups (including age, gender, race, level of education) have a different probability of answering an item in a certain way while the true underlying social health would otherwise be the same. For example, imagine that a student requires both knowledge about anticoagulation as well as unfamiliar English vocabulary to answer a question about anticoagulation. Given that English vocabulary is not required to understand the mechanisms of anticoagulation, such a question can lead to biased estimations of the student’s knowledge of anticoagulation. Considering that including items with differential item functioning can lead to less inclusive and biased measures, it is better to account for them. The remaining items were then moved to CFA to confirm adequate item selection (RMSEA < 0.05), and we used Cronbach alpha to assess internal consistency (the extent to which different subsets of included items would provide the same measurement results). We aimed for a Cronbach α > 0.80 and < 0.95, recognizing that high values may be an indicator of redundancy among the selected items [43]. Last, we tested external consistency by assessing the correlation of various self-reported sociodemographic variables using multiple-choice questions (including age, gender, race, marital status, level of education, employment status, annual household income, and insurance type) with the new quantified social health measure using multivariable linear regression.

The evidence regarding minimum sample size for graded response modeling is ambiguous, with claims ranging from 250 to 1000 observations. We aimed to enroll 350 patients, of which 338 were included because of missing data, based on the evidence that a sample of 250 to 400 observations can be sufficient when at least 15 items are included [11, 13, 17, 25].

Variations in Levels of Discomfort, Capability, Mental Health and General Health Accounted for by Quantified Social Health

To assess the clinical relevance of our new social health measure, we ran multiple multivariable linear regression models to seek the correlation of quantified social health with levels of discomfort, capability, and general health while accounting for potential confounders such as feelings of distress and unhelpful thoughts. We also assessed the correlation of social health with feelings of distress and unhelpful thoughts regarding symptoms as secondary outcomes. The level of capability was measured using the Patient-Reported Outcome Measurement Information System Physical Function Computerized Adaptive Test (PROMIS PF CAT) [10, 21, 37]. The PROMIS questionnaires are based on the general US population, in which a score of 50 represents the population mean, and every 10 points above or below an SD. Higher scores indicate a greater level of capability. The level of discomfort was measured using an 11-point ordinal scale from 0 (representing no pain) to 10 (the worst pain imaginable) [47]. General health was measured using the 5-level EuroQol 5 Dimension (EQ-5D-5L) scale, which includes the following themes rated on a 5-point Likert scale: mobility, self-care, usual activities (such as, work, study, housework, family or leisure activities), pain/discomfort, and anxiety/depression [15]. Higher scores indicate worse general health. Patients also completed a VAS of their overall health status, with higher scores indicating better overall health.

Explanatory Variables Accounted for in the Multivariable Regression Models

We measured various explanatory variables to account for in the multivariable regression models, including self-reported sociodemographic information (including race, level of education, employment status), self-reported mental health (including three items about feelings of distress and three items about unhelpful thoughts regarding symptoms [45] rated on a 5-point Likert scale, for which higher scores indicate greater levels of distress and unhelpful thoughts, respectively), and clinical information derived from the clinician or electronic health record (including trauma, anatomic region, adverse events) (Table 1). Feelings of distress and unhelpful thoughts regarding symptoms were also used as a secondary outcome in separate multivariable linear regression models. Potential confounders were added using a forward selection method, only including explanatory variables that improved the model fit (decrease in Akaike information criterion). We decided to exclude sociodemographic variables from the multivariable regression analysis because of suspected collinearity with quantified social health [46].

Ethical Approval

The University of Texas at Austin Institutional Review Board approved the study protocol.

Results

What Underlying Themes of Social Health Can Be Identified and Quantified Using a Brief Social Health Measure?

In the second stage, we identified two factors: one that seems to represent financial security (factor 1) and one that seems to represent social support (factor 2).

Factor 1 “financial security”: Most of the items (11 of 16) were loaded onto this factor (Table 2). Six items were omitted based on low factor loadings, high modification indices, low item discrimination and item difficulty parameters, and local dependencies according to the prespecified criteria. The remaining five items were included to quantify social health related to financial security: (1) In the past 12 months, lack of reliable transportation kept you from medical appointments, meetings, work, or from getting things needed for daily living; (2) I am concerned that the money I have or will save won’t last; (3) I have money left over at the end of the month; (4) Giving a gift for a wedding, birthday, or other occasion would put a strain on my finances for the month; and (5) I have to work multiple jobs in order to finance my life. The five items measuring financial security displayed good model fit in one-factor CFA (RMSEA 0.041) and good internal consistency (Cronbach α = 0.80), although there were multiple local dependencies that could not be resolved because we exhausted our set of items (Supplemental Table 4; http://links.lww.com/CORR/B382). Two items were flagged for differential item functioning based on level of education (item representing lack of reliable transportation) and self-reported race (item representing being concerned that money will not last), which were not deemed clinically relevant. There were no items flagged for differential item functioning based on age or gender. The financial security score displayed a floor effect of 7.4% (Table 3).

Table 2.

Psychometric properties of the identified factors among items reflecting social health

Social health items	Origin of question	Description	CFA factor loadinga	R2b	Item discriminationc	Item difficulty (1)d	Item difficulty (2)d	Item difficulty (3)d	Item difficulty (4)d
Factor 1: Financial security
1. I am worried about losing my housing or have already lost my housing	PRAPARE	Omitted based on CFA, IRT, and clinical evaluation
2. Within the last 12 months, you were worried that your food would run out before you got money to buy more	HRSN	Omitted based on local dependencies and item difficulty
3. I have trouble paying for my heating or electricity bill	THRIVE	Omitted based on local dependencies and item difficulty
4. In the past 12 months, lack of reliable transportation kept you from medical appointments, meetings, work, or from getting things needed for daily livinge	HRSN	DIF education	0.67	0.45	1.8	-0.17	0.86	1.4	2.1
5. I could handle a major unexpected expense	FWBS	Omitted based on high modification indices and low factor loading
6. I am concerned that the money I have or will save won’t laste	FWBS	DIF race (nonuniform)	0.78	0.61	2.2	-1.3	-0.63	0.13	1.2
7. I have money left over at the end of the monthe	FWBS		0.69	0.47	1.8	-1.6	-0.44	0.39	1.2
8. Giving a gift for a wedding, birthday, or other occasion would put a strain on my finances for the monthe	FWBS		0.78	0.60	2.5	-0.81	-0.10	0.55	1.4
9. I have to work multiple jobs in order to finance my lifee	Self-created		0.73	0.54	2.1	-0.62	0.22	0.93	2.2
10. I often feel lonely or isolated from those around me	HRSN	Omitted based on high modification indices and clinical evaluation
11. If I needed an emergency loan of $100, there is someone (friend, relative, or acquaintance) I could get it from	ISEL	Omitted based on high modification indices
Factor 2: Social support
12. There is a special person who is around when I am in neede	MSPSS		0.80	0.64	2.5	-0.46	0.73	1.3	1.9
13. I have someone to prepare my meals if I am unable to do it myselfe	MOS-SS		0.73	0.53	2.0	-0.81	0.47	0.92	1.7
14. I have someone who understands my problemse	MOS-SS		0.91	0.83	4.5	-0.50	0.66	1.2	2.0
15. I have someone who accepts me totally, including both my worst and my best pointse	SSQ-6		0.84	0.71	3.2	-0.34	0.88	1.6	2.2
16. I have someone I can really count on to help me feel better when I am feeling down or upset	SSQ-6	Omitted based on local dependencies

FWBS = Financial Well-Being Scale; ISEL = Interpersonal Support Evaluation List; MSPSS = Multidimensional Scale of Perceived Social Support; MOSS-SS = Medical Outcomes Study Social Support Survey; SSQ-6 = 6-item Social Support Questionnaire.

^aFactor loading in CFA resembles the correlation between an individual item and the identified factor (in this case: “financial security” and “social support”). Model fit statistics for the two-factor CFA: RMSEA = 0.046.

^bR² resembles the percentage of variance of each item that is explained by the identified factor (in this case: “financial security” and “social support”), in which a higher R² indicates that the item measures a larger proportion of the identified factor.

^cItem discrimination resembles the ability of an item to identify differences in social health (more specifically, related to “financial security” or “social support”). Higher item discrimination indicates that relatively small differences in the level of financial security or social support lead to relatively large changes in the probability of answering a different response (that is: strongly disagree, disagree, neutral, agree, strongly agree).

^dItem difficulty resembles the level of social health (more specifically, related to “financial security” or “social support”) at which the participant has a 50/50 chance of answering an item in a more positive or negative manner; that is, there are four item difficulty parameters that each represent the transition to a higher probability (> 50%) of answering a more positive or negative option, for example: “Strongly disagree” or “Disagree” (1), “Disagree” or “Neutral” (2), “Neutral” or “Agree” (3), and “Agree” or “Strongly agree” (4). One can consider these thresholds as stepping stones, in which each item difficulty parameter (1-4) represents a point at which the probability of choosing an answer option that represents worse social health increases. The further these 1, 2, 3, and 4 values are apart, the more informative an item can be considered given there is more distinction between the levels of social health required for someone to have a higher probability to select different options.

^eThese items were chosen for the final measure.

Table 3.

Descriptive statistics of quantified social health including the financial security and social support subdomains

	Range	Median (IQR)	Ceiling effect, %	Floor effect, %
Social health	0-34	14 (8-18)	0	3.6
Financial security	0-20	9 (5-12)	0.3	7.4
Social support	0-16	4 (2-7)	1.5	17

Ceiling effect = proportion of patients with the highest possible score. Floor effect = proportion of patients with the lowest possible score.

Factor 2 “social support”: Only 5 of 16 items were loaded onto this factor (Table 2). We identified multiple local dependencies and removed one item because of this but were unable to address them completely because we exhausted our set of items (Supplemental Table 4; http://links.lww.com/CORR/B382). The remaining four items were selected to quantify social health related to social support: (1) There is a special person who is around when I am in need, (2) I have someone to prepare my meals if I am unable to do it myself, (3) I have someone who understands my problems, and (4) I have someone who accepts me totally, including both my worst and my best points. We found a marginal model fit in one-factor CFA (RMSEA 0.091) and good internal consistency (Cronbach α = 0.84). There were no items flagged for differential item functioning based on age, gender, self-reported race, or level of education for any of the included items. The social support score had a floor effect of 17% (Table 3).

The financial security and social support scores function as subdomains of a combined measure of quantified social health (Social Health Scale [SHS]) (Fig. 1). The combined measure displayed good model fit in two-factor CFA (RMSEA 0.046). The floor effect of the combined social health measure was 3.6% (Table 3). The SHS displayed a relatively normal distribution with some skewness to the left (Fig. 2).

Fig. 1.

The scoring manual for the new SHS measure. The patient is asked to self-administer this survey, in which all statements are answered on a 5-point Likert scale: Strongly disagree, Disagree, Neutral, Agree, Strongly agree. The answers are then scored from 0 to 4, dependent on the framing of the statements: Negatively framed statements (items 1, 2, 4, 5) = strongly disagree (0) to strongly agree (4). Positively framed statements (items 3, 6, 7, 8, 9) = strongly disagree (4) to strongly agree (0). The values of all nine statements are then summed to generate the total score on the 9-item SHS, providing a score ranging between 0 and 36. The financial security subdomain consists of items 1 to 5, and the social support subdomain consists of items 6 to 9. Higher scores indicate more disadvantaged social health.

Fig. 2.

Graph displaying the distribution of the SHS scores of all participants in the second stage.

We found that a more disadvantaged social health score was associated with self-reported Black race compared with self-reported White race (regression coefficient [RC] 2.6 [95% confidence interval (CI) 0.29 to 4.9]), being divorced compared with having a partner (RC 2.5 [95% CI 0.23 to 4.8]), being unemployed compared with being employed (RC 1.7 [95% CI 0.023 to 3.4]), being uninsured compared with private insurance (RC 3.5 [95% CI 0.33 to 6.7]), and earning less than USD 75,000 per year (RC 2.7 [95% CI 0.020 to 5.4] to 6.8 [95% CI 4.3 to 9.3]) (Table 4). The regression coefficients indicate how much the outcome variable is expected to be different in the group of interest compared with the reference group while accounting for the effects of other variables.

Table 4.

Sociodemographic characteristics associated with quantified social health

Variable	Regression coefficient (95% CI)	Semipartial R2	p value
Age	-0.0016 (-0.052 to 0.049)	0.021	0.95
Language		< 0.001
English	Reference value
Spanish	-0.73 (-2.7 to 1.3)		0.47
Gendera		0.0011
Women	Reference value
Men	-0.27 (-1.6 to 1.0)		0.68
Race/ethnicity		0.0049
White	Reference value
Hispanic	0.79 (-1.1 to 2.7)		0.40
Black	2.6 (0.29 to 4.9)		0.03
Asian	0.55 (-3.3 to 4.4)		0.78
Other	1.3 (-3.4 to 5.9)		0.59
Marital status		0.016
Married	Reference value
Single	1.4 (-0.12 to 2.9)		0.07
Divorced	2.5 (0.23 to 4.8)		0.03
Widowed	2.3 (-0.62 to 5.3)		0.12
Level of education		< 0.001
High school or less	Reference value
2-year college	-0.95 (-2.7 to 0.84)		0.30
4-year college	-0.16 (-2.0 to 1.7)		0.86
Postcollege	0.18 (-1.9 to 2.3)		0.87
Employment status		0.010
Employed	Reference value
Retired	-2.6 (-5.0 to -0.081)		0.04
Unemployed	1.7 (0.023 to 3.4)		0.047
Other (such as student, homemaker)	1.5 (-0.66 to 3.7)		0.17
Annual household income in USD		0.10
< 24,000	Reference value
24,000 to 46,000	-1.3 (-3.1 to 0.64)		0.19
46,001 to 75,000	-2.0 (-4.7 to 0.61)		0.13
75,001 to 121,000	-2.7 (-5.4 to -0.020)		0.048
> 121,000	-6.8 (-9.3 to -4.3)		< 0.01
Insurance type		0.0058
Private	Reference value
MAP	1.9 (-0.48 to 4.3)		0.12
Medicare	-0.69 (-3.2 to 1.9)		0.60
Medicaid	1.8 (-1.4 to 5.0)		0.26
Other	2.5 (-0.14 to 5.1)		0.06
Uninsured	3.5 (0.33 to 6.7)		0.03

Model fit parameters: Akaike information criterion = 2121, R² = 0.42, adjusted R² = 0.37. The regression coefficients represent how much the outcome variable is expected to change for each 1-unit increase (or for categorical variables: difference relative to the reference value) in the explanatory variable of interest while accounting for the effects of other variables included in the model. The linear regression model captures relationships in the context of a specific model with the assumption that the correlations with the outcome variable are linear. Semipartial R² represents the proportion of the variation in the outcome variable that is accounted for by the explanatory variable of interest.

^aOne nonbinary individual was pooled with men based on random assignment.

How Much of the Variation in Levels of Discomfort, Capability, General Health, Feelings of Distress, and Unhelpful Thoughts Regarding Symptoms Is Accounted for by Quantified Social Health?

Using bivariate analyses, we found that more disadvantaged social health measured with the SHS was moderately associated with greater levels of discomfort (ρ = 0.30 [95% CI 0.20 to 0.41]), moderately with lower levels of capability (ρ = -0.20 [95% CI -0.31 to -0.11]), moderately with worse general health (ρ = 0.34 [95% CI 0.25 to 0.44]), and moderately with worse mental health (ρ = 0.36 [95% CI 0.27 to 0.45] to 0.44 [95% CI 0.35 to 0.54]) (Table 5). The ρ coefficients indicate the strength and direction of the correlation between two independent variables but do not quantify the effect size or account for the effects of other explanatory variables.

Table 5.

The correlations of quantified social health with various outcomes assessed using Spearman rank correlation with nonparametric bootstrapping (replacement n = 1000)

Variable	Social health	p value	Financial security (subdomain)	p value	Social support (subdomain)	p value
Social support	NA	NA	0.33 (0.22 to 0.43)	< 0.01	NA	NA
Level of capability	-0.20 (-0.31 to -0.11)	< 0.01	-0.19 (-0.29 to -0.085)	< 0.01	-0.13 (-0.24 to -0.024)	0.02
Level of discomfort	0.30 (0.20 to 0.41)	< 0.01	0.32 (0.22 to 0.43)	< 0.01	0.20 (0.092 to 0.30)	< 0.01
EQ-5D-5L	0.34 (0.25 to 0.44)	< 0.01	0.35 (0.25 to 0.45)	< 0.01	0.21 (0.10 to 0.32)	< 0.01
EQ-5D VAS	-0.30 (-0.40 to -0.20)	< 0.01	-0.29 (-0.40 to -0.19)	< 0.01	-0.20 (-0.30 to -0.099)	< 0.01
Feelings of distress	0.44 (0.35 to 0.54)	< 0.01	0.48 (0.39 to 0.57)	< 0.01	0.29 (0.18 to 0.39)	< 0.01
Unhelpful thoughts	0.36 (0.27 to 0.45)	< 0.01	0.41 (0.32 to 0.50)	< 0.01	0.21 (0.10 to 0.31)	< 0.01

Data presented as ρ coefficient (95% CI). The ρ coefficient indicates the strength and direction of a monotonic correlation between two independent variables (ranging between –1 and 1) but does not quantify the size of the effect. The correlation strength can be interpreted as negligible correlation (ρ < 0.1), weak correlation (ρ = 0.1 to 0.3), moderate correlation (ρ = 0.3 to 0.5), strong correlation (ρ = 0.5 to 0.7), and very strong correlation (ρ > 0.7). NA = not applicable (we did not perform these analyses).

While accounting for the effects of other explanatory variables such as age, affected area, and mental health in multivariable linear regression models, we found that more disadvantaged social health was associated with slightly greater levels of discomfort (RC 0.055 [95% CI 0.16 to 0.093]), slightly lower levels of capability (RC -0.19 [95% CI -0.34 to -0.035]), and slightly worse general health (RC 0.13 [95% CI 0.069 to 0.18]) (Table 6). We also found that more disadvantaged social health was associated with moderately higher levels of distress (RC 0.23 [95% CI 0.19 to 0.28]) and moderately higher levels of unhelpful thoughts (RC 0.17 [95% CI 0.13 to 0.22]) (Supplemental Tables 5 and 6; http://links.lww.com/CORR/B382). The regression coefficients indicate how much the outcome variable is expected to change for each one-unit increase in the explanatory variable of interest (quantified social health) while accounting for the effects of other variables.

Table 6.

Results of multiple separate multivariable linear regression models to seek the associations of quantified social health with levels of comfort, capability, general health, and overall health while accounting for the effects of other variables

Variable	Level of discomfort (NRS pain intensity)			Level of capability (PROMIS PF CAT)			General health (EQ-5D-5L)			Overall health (EQ-5D-VAS)
	Regression coefficient (95% CI)	Semipartial R2	p value	Regression coefficient (95% CI)	Semipartial R2	p value	Regression coefficient (95% CI)	Semipartial R2	p value	Regression coefficient (95% CI)	Semipartial R2	p value
Age	0.016 (0.0021 to 0.031)	0.012	0.02	-0.14 (-0.20 to -0.086)	0.064	< 0.01	0.040 (0.019 to 0.061)	0.030	< 0.01	-0.078 (-0.22 to 0.064)	0.0030	0.28
Affected area		0.0014			0.012			0.0072			0.0001
Lower extremity	Reference value			Reference value			Reference value			Reference value
Upper extremity	0.31 (-0.23 to 0.85)		0.26	1.7 (-0.40 to 3.8)		0.11	-0.53 (-1.3 to 0.28)		0.20	-0.63 (-6.0 to 4.8)		0.68
Spine	-0.64 (-1.3 to -0.0022)		0.049	7.0 (4.5 to 9.5)		< 0.01	-2.6 (-3.6 to -1.7)		< 0.01	1.3 (-5.0 to 7.7)		0.82
Feelings of distress	0.30 (0.20 to 0.40)	0.070	< 0.01	-0.64 (-1.0 to -0.24)	0.017	< 0.01	0.49 (0.34 to 0.64)	0.069	< 0.01	-2.3 (-3.3 to -1.3)	0.047	< 0.01
Unhelpful thoughts	0.087 (-0.022 to 0.20)	0.0051	0.09	-0.42 (-0.85 to -0.015)	0.0077	0.06	0.12 (-0.047 to 0.28)	0.0034	0.16	-0.78 (-1.9 to 0.31)	0.0046	0.16
Social health	0.055 (0.16 to 0.093)	0.014	< 0.01	-0.19 (-0.34 to -0.035)	0.0094	0.02	0.13 (0.069 to 0.18)	0.028	< 0.01	-0.52 (-0.90 to -0.13)	0.016	< 0.01

Akaike information criterion and R² of all models: discomfort model = 1443 and 0.31, capability model = 2357 and 0.26, general health model = 1721 and 0.39, and overall health model = 2984 and 0.22. The regression coefficients represent how much the outcome variable is expected to change for each 1-unit increase (or for categorical variables: difference relative to the reference value) in the explanatory variable of interest while accounting for the effects of other variables included in the model. The linear regression model captures relationships in the context of a specific model with the assumption that the correlations with the outcome variable are linear. Semipartial R² represents the proportion of the variation in the outcome variable that is accounted for by the explanatory variable of interest. Note that these models may be distorted somewhat by the known collinearity between thoughts and feelings regarding symptoms and their relationship to social health. This can be seen in the relatively lower contributions of unhelpful thoughts to the model and the relatively low semipartial R² relative to the R² of the overall model.

Discussion

A better understanding of the relative contributions of social health, mental health, and pathophysiology to variations in levels of discomfort and incapability among patients seeking musculoskeletal specialty care could aid the design of personalized whole-patient care strategies. There is a relative lack of brief, quantitative measures of social health as most available assessments of social health are checklists designed for descriptive screening purposes rather than quantification for research, patient monitoring, or quality improvement. To fill this void, we aimed to develop a brief, quantitative social health measure for patients seeking musculoskeletal specialty care. In this two-stage study, we identified a brief number of items that quantify two domains of social health representing financial security and social support. The new measure of quantified social health, the SHS (Fig. 1), displayed acceptable psychometric properties with adequate external validity (correlation with sociodemographic characteristics), although there were some remaining local dependencies that could not be resolved because we exhausted our set of items. There were no discordantly rated items (differential item functioning) based on differences in age, gender, self-reported race, and level of education that were deemed clinically relevant. The clinical relevance of the SHS was evident as underlined by the relatively independent associations with levels of discomfort, capability, general health, and mental health.

Limitations

This study can be viewed in the light of a number of limitations. First, we initially did not account for local dependencies, modification indices, and differential item functioning, which potentially could have helped address the observed inconsistencies. We advise others to account for these parameters in addition to factor loadings, R² (proportion of variation in the latent factor explained by an item), and item-rest correlations to explore underlying constructs in more detail. Second, we used a relatively small subset of items in both our first and second factor analysis including only 16 items derived from commonly used social health checklists in the second stage. This left only limited room to omit redundant or problematic items and resulted in the inability to account for all local dependencies because we exhausted our set of items. This especially seems relevant considering that we used the results of the initial factor analysis to select a broader set of items directed at financial security and social support. It is possible that we were not sufficiently broad at the outset, and that there are other aspects of social health that were not sufficiently represented in this measure (such as, role and identity). On the other side, one can argue whether including a larger number of similar items holds sufficient additional value. Regardless, the selected items have face validity and seem clinically relevant. Further development of quantified social health measure may benefit from including more social health items based on qualitative research, particularly examining social health domains in diverse populations with greater variation and scope than can currently be found in legacy checklists.

In the same light, one can argue that our sample size was relatively low for IRT based on the ambiguous evidence on minimum sample size (ranging between 250 and 1000 observations). There is evidence that our sample was sufficient for factor analysis [31], and that samples as low as 250 observations can lead to adequate parameter estimations [35]. Given that our item selection mostly relied on local dependencies and parameters from the sufficiently powered factor analysis, in which item discrimination and item difficulty were only considered and not used to calculate T-scores, we believe that our sample was sufficient. Future studies incorporating a greater number of items may increase their sample size to 500 observations to ensure sufficient power relatively independent from data-related conditions.

Third, we combined race and ethnicity with a limited number of self-reported categories that do not capture potentially important nuances. These limited categories, and the nonrepresentative numbers (especially for Black and Asian race), are not able to address the degree to which these social constructions are potentially related to variations in social health and associated forms of disadvantage. Considering that the main purpose of this study was to develop a new social health measure, and not to identify specific correlations, we believe that these variables were sufficient for assessment of external validity related to measurement development. After all, the item selection process was not dependent on the sociodemographic characteristics measured in this study. Future studies may further elaborate on the correlation of social health with more nuanced variables of race, ethnicity, culture, and other social constructions among a representative subset of patients with sufficient sample sizes in the considered groups.

Fourth, one can argue that area deprivation indices pose as a useful alternative to quantify social health with little survey burden. However, area deprivation indices tend to be imprecise because of overgeneralization, lack of depth about personal social circumstances, and rapid evolvement with gentrification. We therefore did not consider or compare our new measure to these metrics as we aimed to develop a brief, more personalized, quantitative measure of social health.

Fifth, we excluded patients who were illiterate in Spanish or English for practical reasons, while one can argue that illiteracy may be an important aspect of disadvantaged social health. We found that people rarely declined participation or were rarely excluded because of English or Spanish illiteracy (estimated < 2%), and thus we believe that this had little effect on our findings. Besides, one can argue that having an independent researcher aid someone with survey completion has other limitations, including risk of bias attributed to social desirability or dishonesty related to feelings of shame regarding one’s social circumstances.

Sixth, the wording of the included items varied between first and third person considering that we derived them from various available instruments. We decided to stick to the origin of the questions, although we estimate that changing the wording has little to no effect on the item performance. Future studies may elaborate on whether nuances in wording have an effect on the performance of items.

Seventh, participants completed items either before or after their clinic visit, and one could argue that timing may have influenced the findings attributed to the Hawthorne [1] or other effects. The Hawthorne effect [1] refers to potential altercations in survey responses attributed to the awareness that one is being studied, which seems unavoidable in studies with patient-reported measures and is estimated to have little effect. Given that the clinic visit is unlikely to alter an individual’s social health, we believe that this likely had no effect and would otherwise have led to an underestimation of the found correlations.

Last, we did not track whether patients received surgery and whether they presented as new or returning patient to limit survey burden, while one may argue that this information seems relevant. Although surgery may be associated with social health (for instance, greater financial insecurity attributed to costs or greater dependence on social support), the aim of this study was to develop a new social health measure and assess its clinical relevance rather than to identify factors associated with disadvantaged social health. Future studies elaborating on the factors associated with social health may incorporate such explanatory variables.

What Underlying Themes of Social Health Can Be Identified and Quantified Using a Brief Social Health Measure?

The observation that we identified two factors representing financial security and social support is in line with the theoretical framework that social health consists of role and identify, security and stability, and social relationships [2, 22]. The included items displayed good psychometric properties with good internal consistency and good model fit in two-factor CFA. More research is warranted to further explore these domains and the items that best measure them, considering that there are still local dependencies that we were not able to address because we exhausted our set of items. These local dependencies raise several possibilities. First, we may need to include more items representing a greater breadth of social health, including more varied aspects of social relationships and security and stability, and items focused on role and identity. A second possibility is that social health items are highly correlated, and attempts to quantify social health with items that contribute usefully and independently to a score for social health may be infeasible. Perhaps the ability to quantify social health has inherent limits.

The observation that quantified social health was associated with various sociodemographic variables, including self-reported race, marital status, employment status, annual household income, and insurance type, has face validity and supports external validity of the measure, as these can all be associated with social disadvantage, but also merits caution and consideration [27, 28]. The social construction of race is responsible for long-standing systematic disadvantage. In this regard, the correlation of lower quantified social health with Black race in the US might help validate the measure. However, the people whose socioeconomic disadvantage is harming their musculoskeletal health should not be further disadvantaged by limiting their access to care as a way of gaming results of care measured using patient-reported outcomes. That would represent an unethical extension of historical disadvantage. Rather, one can understand stress from social disadvantage as one of many previously less considered and potentially modifiable aspects of comprehensive musculoskeletal health [28].

How Much of the Variation in Levels of Discomfort, Capability, General Health, Feelings of Distress, and Unhelpful Thoughts Regarding Symptoms Is Accounted for by Quantified Social Health?

The observation that social health was associated with levels of discomfort, capability, mental health, and general health affirms the relevance and importance of social health in musculoskeletal specialty care, which may evolve along with improvements in our ability to quantify social health. These findings are in line with available evidence showing the association of social health with various outcomes including level of capability, mental health, personal health agency, and adverse events [3, 12, 14, 16, 20, 26, 29, 30, 34, 36]. There also is evidence that having a good social support system can alleviate or limit symptoms of depression and anxiety [20]. This line of inquiry directs us to further develop a quantitative measure of social health to aid the design and implementation of comprehensive musculoskeletal specialty care. Such pathways may incorporate experts such as social workers or behavioral therapists who can address both mindset and circumstances. A recent systematic review [19] found that psychosocial interventions can lead to improved social support systems and reduced healthcare use, which seemed to reflect better health. There are also signs that integration of social workers can lead to lower costs and improved cost-effectiveness [42].

Practical Applications and Future Directions

The 9-item SHS (Fig. 1) is a patient-reported, quantitative social health measure with financial security and social support as subdomains that can be used to assess an individual’s social health in both clinical and research settings. One can use this measure to quantify social health in clinical research settings or quality improvement initiatives, but also to identify or monitor patients with disadvantaged social circumstances who may benefit from additional care or community resources. Future studies may further develop this measure using larger subsets of potentially relevant items derived from qualitative research and other legacy measures that may capture other, more specific aspects of social health. Such studies may also be able to eliminate remaining local dependencies, perform internal validation, and incorporate items focused on someone’s role and identity (more specifically, their content with sociodemographic characteristics such as race, level of education, or employment status), which may also be an important aspect of social health. Future studies can also elaborate on the correlation of quantified social health with levels of discomfort and capability as well as the development of comprehensive care pathways aimed to improve social health among patients seeking musculoskeletal care.

Conclusion

In this two-stage study, we identified two domains of social health representing financial security and social support and developed a quantitative social health measure with acceptable psychometric properties but considerable local dependencies. The development of a quantitative social health measure was accompanied by several difficulties, and more research is needed. Specifically, the relatively low number of social health items we included prohibited us from removing more problematic items because we exhausted our set of items, and thus future research should, if possible, incorporate a larger number of items covering a wider scope of social health. Qualitative research involving social workers may be useful. On the positive side, the selected items have face validity, performed relatively well in terms of internal consistency and external validity, and were independently associated with levels of discomfort, capability, and general health. These findings underline the clinical relevance of quantified social health and point to the importance of designing and implementing comprehensive whole-person musculoskeletal care strategies with attention to social circumstances. In particular, people with disadvantaged social health should not be further disadvantaged by using a quantitative measure of social health to screen or cherry pick in contexts of incentivized or mandated reporting, which could worsen inequities in access and care. Rather, one should consider disadvantaged social health and its associated stressors as one of several previously less considered and potentially modifiable aspects of comprehensive musculoskeletal health.