1 Background

Social determinants of health (SDOH) are factors falling into five categories, including healthcare access, environment, social, financial, and education, while socioeconomic status (SES) is a theoretical composite measure that encompasses all of these categories.¹ In orthopaedics, socioeconomic factors have been linked to patient welfare in multiple clinical scenarios, including postoperative outcome.^2,3 However, the effect of SES on outcomes varies across studies and depends on the clinical environment (nonoperative vs. perioperative patients), subspecialty, type of surgery, measure of SES, and outcome measure.^4,5 There is no question that SES is an important moderator of orthopaedic outcomes. However, important questions remain regarding how SES should be incorporated into outcomes research.

There are several methods of quantifying SES, including measures of income, education, employment, wealth, and insurance status. Recently, geography-based indices based on census data have gained popularity. These indices provide quantitative scores based on the patient's home address or zip-code, allowing for estimation of SES at the level of individual neighborhoods.^3,4 Geography based estimators of SES are resistant to many forms of bias, inexpensive to collect, reproducible, and are available for retrospective studies. The Area Deprivation Index (ADI) is a metric of SES based on the patient's home address that was specifically designed to assess health outcomes. It considers 17 socioeconomic indicators obtained from United States census data and aggregated at the level of street address, weighted to estimate mortality rates.⁶ A growing number of studies demonstrate association between ADI and function in orthopaedic patients.^5,7–12

There are three important questions we need to answer as a community: (1) what are the important SES-related predictive factors, (2) which outcomes are most related to SES, and (3) how should this information be collected and integrated into future research to most effectively represent the effects of SES factors in the context of orthopaedic intervention?

The purpose of this study was to analyze how national ADI score based on home address is associated with 2-year postoperative patient-reported outcomes (PROs) in patients undergoing outpatient orthopaedic surgery, as measured by frequently used patient-reported outcome measures (PROMs). This study is exploratory in nature; statistical tests were selected comparing observed data to random chance to differentiate between a process where ADI is associated with outcome versus one where ADI and outcome are independent.

2 Methods

This study was a retrospective analysis of prospectively collected data from patients enrolled in an orthopaedic registry who underwent outpatient extremity orthopaedic surgery at an urban tertiary referral center between June 2015 and November 2018. Patients had to be 12 years or older, English-speaking, have an email address, and not be institutionalized to be eligible for enrollment, and all patients provided informed consent to participate. Patients completed electronic surveys within 1 week of presentation for surgery and at 2-year follow-up.¹³ This study was approved by the local institutional review board.

Demographic, socioeconomic, patient-specific health, and operative factors as enumerated in Table 1 were self-reported by the patients. American Society of Anesthesiologists (ASA) score and additional relevant operative characteristics were supplemented with a review of the electronic medical record. The Charlson Comorbidity Index (CCI) was calculated to measure comorbidity burden.¹⁴

Each survey included PROMs in conceptual clusters including 1) Patient-Reported Outcomes Measurement Information System (PROMIS) domains, 2) joint-specific PROMs, 3) pain, 4) activity measures, and 5) expectations/satisfaction. Computer-adaptive PROMIS domains included Physical Function, Pain Interference, Depression, Anxiety, Fatigue, and Social Satisfaction.¹⁵ Higher scores in the Physical Function and Social Satisfaction domains indicated better function, while higher scores in the Pain Interference, Depression, Anxiety, and Fatigue domains indicated worse function. Joint-specific PROMs included the American Shoulder and Elbow Surgeons (ASES) Score,¹⁶ the International Knee Documentation Committee (IKDC) Score,¹⁷ or the Brief Michigan Hand Questionnaire (BMHQ) Score,¹⁸ selected as appropriate for the operative body region. Activity measures included the Tegner Activity Scale (TAS),¹⁹ and the Marx Activity Rating Scale (MARS).^20,21 Patients’ preoperative expectations and met expectations were assessed using the Musculoskeletal Outcomes Data Evaluation and Management System (MODEMS) Expectations domains at baseline and 2 years after surgery, respectively.²² Postoperative satisfaction was assessed via the Surgical Satisfaction Questionnaire (SSQ-8).²³ Patients additionally completed a numeric pain scale (NPS) for operative site and whole body pain, as well as an anchoring question at the end of the 2-year postoperative survey asking: “Is the condition for which you underwent surgery completely better now?” with the option answer either “Yes” or “No”.

ADI is a composite score that considers 17 socioeconomic factors derived from United States census data at the level of the census block group. A score of 1 indicates the lowest level of deprivation (highest SES), while a score of 100 indicates the highest level of deprivation (lowest SES).²⁴ Scores may be ranked at the state or national level. National ADI was calculated using a geocoding algorithm to determine the street-level geographic coordinates of each patient's residential address, based on their street name and number, and matching this with the appropriate census block.

Continuous variables were reported as means with standard deviation. The raw scores for the BMHQ, MARS, MODEMS, and SSQ-8 were each normalized to a 0–100 point scale to facilitate comparison between outcome measures. Categorical variables were reported as counts and percentages. Wilcoxon rank-sum tests and Kruskal-Wallis tests were conducted to analyze relationships between categorical variables and national ADI. Post-hoc Wilcoxon analysis was performed for categorical variables with more than two subgroups. Spearman's correlation coefficient was used to assess relationships between continuous variables and national ADI. Linear regression was conducted with the national ADI divided by 10 to estimate magnitude of effect relative to 10 point ADI change. All statistically significant patient characteristics from the bivariate analysis were used in a multivariable model to determine if ADI remained an independent predictor of 2-year outcomes after statistically controlling for other recorded SES factors. A multivariable logistic regression model using these same covariates was used to estimate likelihood of score change ≥ the minimal clinically important difference (MCID) (JMP, Version 16 SAS Institute Inc., Cary, NC). MCID values were estimated based on previous literature.^25–30

3 Results

4 Discussion

There is an obvious interaction between SES and patient function, with lower levels of SES being associated with greater symptom severity and functional impairment as measured by PROs. In a reimbursement environment headed toward outcomes-based compensation models, we need an improved understanding of these dynamics to avoid creating a healthcare system which systematically disincentivizes the delivery of elective care to patients with low SES, worsening the health gap in the United States.

Although there are still conflicting findings about the presence or absence of SES-based outcomes moderation in specific contexts, consensus is building that the question is not if SES is an important moderator of outcomes, but rather what features of SES moderate which outcomes. Recent systematic reviews in all seven foundational subspecialties of orthopaedics including arthroplasty,² sports,⁷ hand,⁸ spine,⁹ trauma,¹⁰ pediatrics,¹¹ and oncology,¹² have concluded that SES and other SDOH are drivers of outcomes. Despite the known importance of SES, there are persistently low levels of incorporation in orthopaedic studies.³¹ It is our responsibility as a research community to develop a thoughtful, comprehensive, reproducible, and relevant research strategy to integrate this important predictor of patient outcomes.

Many authors currently describe orthopaedics as being early on the learning curve regarding techniques for incorporating SES in outcomes research.^32,33 Developing a framework to account for SES and SDOH will require a more granular understanding of three areas: 1) the important predictive factors to consider, including discrete SDOH factors and continuous estimators of SES such as geography-based estimators, 2) an understanding of which outcomes are most susceptible to SES-based moderation, including clinical outcomes (infection, readmission, reoperation, and death) and PROs (such as the PROMIS domains, which report global function, or ASES, BMHQ, and IKDC, which report joint-specific function), 3) how data collection (self-reported vs. chart review) and modeling mechanics (single measurement vs. repeated measures designs) can most effectively be used to represent the effects of SES factors in the context of orthopaedic intervention.

This study improves our understanding of the scope of these interactions by demonstrating that not all PROs are equally susceptible to SES-based moderation, and that the ability of a study to identify SES-based changes in outcomes depends on the details of the data collection and analysis. PROMIS was developed to allow comparison between disease states that are otherwise incomparable for the purpose of prioritizing the distribution of resources. By design, these PROMs are sensitive to effects from non-orthopaedic life factors, including SES. Joint-specific or activity-specific PROMs may avoid some of the confounding effects inherent in PROMIS domains by only focusing on orthopaedic-specific function. PROMs’ sensitivity to SES moderation will also depend on evaluation of score at 2 years versus change in scores 2 years after surgery. SES has an effect on scores both preoperatively and postoperatively, but its effect on change between these two time-points may be specific to the PROM. Looking at only a single PROM collected postoperatively may systematically misrepresent the efficacy of surgical intervention based on SES.

Clinical experience using registry data from our institution has suggested a moderating effect of ADI on baseline scores, with patients with low SES having greater dysfunction.³⁴Table 3 demonstrates that this baseline shift persists if measured by a single assessment at 2 years postoperatively, with lower SES once again being associated with greater dysfunction in all PROs except PROMIS Anxiety and Depression and the BMHQ. This study advances the conversation by looking at the change in PROs for each patient from perioperatively to postoperatively using a repeated measures study design. Table 3, columns 6 and 7 show that in univariable analysis, there is a statistically significant moderating effect of ADI on score improvement after 2 years on PROMIS Physical Function, Pain Interference, Social Satisfaction, and on the TAS, with higher ADIs lowering improvement. PROMIS Fatigue, Anxiety, and Depression show much lower score changes, which is likely because orthopaedic procedures do not directly modify these domains. In the context of the reduced magnitude of effect of orthopaedic intervention, the moderating effect of ADI on anxiety, depression, and fatigue is insignificant. The same analysis did not show associations with NPS operative site or whole body, the MARS upper or lower extremity activity scale, or the ASES or BMHQ scores, all of which are region-specific rather than holistic.

Finally, Table 3 shows multivariable linear regression parameter estimates demonstrating that ADI remains independently associated with only a subset of PROs as measured in single assessment at 2 years postoperatively. PROMIS Social Satisfaction, NPS whole body, met expectations, and satisfaction no longer are associated with ADI after accounting for other measures of SES. This suggests that there is variability in the data captured by ADI that is relevant to postoperatively collected PROs that is not otherwise captured by the panel of self-reported SES factors. When examining the effect of ADI on PRO improvement while controlling for other SES factors (Table 3), ADI remains independently associated with lower improvement on only the PROMIS Physical Function, Pain Interference, and IKDC PROs. The magnitude of effect after controlling for other factors appears to be below the threshold of clinical significance in the PROMIS domains. This is demonstrated by the results of the multivariable logistic regression regarding score change ≥ MCID, which shows that IKDC is the only PROM that remains associated with ADI specifically (Table 3).

The persistent sensitivity of IKDC to ADI despite controlling for other SES factors warrants discussion. This may be for reasons intrinsic to the items on the test, such that they pick up behaviors which SES may also disproportionately affect. However, it may also be secondary to peculiarities of the study population. Knee surgery is overrepresented in this cohort (representing 42 % of the total cohort), many of whom are student athletes. The rehabilitative resources available to these patients relative to their self-reported SES is not in line with the rest of the study population, so their inclusion may generate anomalous findings. This study is exploratory only, so additional interaction testing or subgroup analysis to further explore this persistent finding is outside the scope of this paper.

There are important limitations to this study. The follow-up rate in this study was 70 %. It is likely that nonresponses occur from a nonrandom and low SES portion of the population as we have previously described.^35,36 However, 70 % response rate is very high for postoperative satisfaction surveys, so the effect of nonresponse bias is likely to be small. The study uses self-reported data on sensitive issues (income, education, drug use, etc.), which are subject to reporting bias. However, deliberate misrepresentation by the patient is a source of error in all self-reported SES metrics. It is not unique to this study, and its effect is thought to be small. The use of ADI assigns an estimated SES value to each patient based on the population average of the area surrounding their home address. This may systematically misrepresent SES based on other known features of the patient, like procedure or history of trauma. For example, trauma patients from a given neighborhood may have lower SES than arthroplasty patients from the same neighborhood. This is mitigated by using the smallest available geographic area. The residual error introduced by this effect is not currently knowable, and should be the topic of future study. Lastly, it is worth emphasizing that although there is a statistically significant correlation between ADI and most PROs, the strength of correlation for each scale is relatively low (Rs ≤ 0.22), although some have suggested a threshold of Rs > 0.1 can be considered meaningful in behavioral sciences research, and this has been utilized in recent related orthopaedic research.³⁷ Additionally, we are making the statement that these scores should be considered for inclusion in multivariable outcomes models, not that they are the dominant driver of outcomes.

This is one of the largest studies to date to prospectively assess the association between geocoded home address-based ADI and mid-term postoperative outcomes after orthopaedic surgery using a repeated measures design. The study focused on a mixed population undergoing outpatient extremity surgery, which is a population that has undergone unavoidable selection bias. It is, therefore, important to validate the scope of involvement of SES factors on PROs in this context, as findings from other contexts will not necessarily generalize here. The repeated measures design, where each patient is tested both perioperatively and postoperatively, is a unique strength of the paper and allows assessment of the sensitivity of PROs to SES across a spectrum of analytic frameworks.

Our findings demonstrate a clear association between socioeconomic deprivation and outcomes at 2 years postoperatively. More importantly, they demonstrate that improvement in function after orthopaedic surgery may be subject to an SES-based moderating effect if more globally sensitive PROs are used. Finally, the findings demonstrate that geography-based SES scores may offer additional dimensionality for outcomes models which is not otherwise accounted for using traditional self-reported metrics. Compensation structures or research studies focused on PROs are likely to systematically misrepresent outcomes based on SES unless they use a repeated measure design and incorporate SES measures into their statistical models, or use PROMs or a research design that is otherwise SES-insensitive.

Patients selected for outpatient extremity orthopaedic surgery have postoperative function that is associated with their SES; patients with low SES report worse function on many PROs. However, the magnitude of improvement appreciated after surgery appears to be largely unrelated to SES for joint-specific PROMs. Although an SES-based moderating effect may be present for more globally sensitive PROMs, it is unlikely to reach MCID threshold across the spread of patients typically indicated for outpatient surgery. Orthopaedic surgery can be used to effectively improve symptoms related to an orthopaedic problem regardless of SES, but does not relieve other social stressors affecting overall life satisfaction or global activity level that may color reporting of postoperative function.

CRediT authorship contribution statement

Evan L. Honig: Data acquisition, Analysis, Interpretation, Design. Samir Kaveeshwar: Data acquisition, Analysis, Interpretation, Design. Nathan N. O'Hara: Analysis, Interpretation. Dominic J. Ventimiglia: Analysis, Interpretation. Isaiah Harris: Data acquisition, Interpretation. Samuel Q. Li: Data acquisition. Craig Shul: Data acquisition. Natalie R. Danna: Interpretation. R. Frank Henn, 3rd: Interpretation, Design. Christopher G. Langhammer: Interpretation, Design.

Guardian/patient's consent statement

All patients provided informed consent to participate in the registry.

Ethics approval

This study was approved by the University of Maryland, Baltimore (UMB) Institutional Review Board (HCR-HP-00062261-9).

Funding

The authors disclose a grant from The James Lawrence Kernan Hospital Endowment Fund, Incorporated (BL1941007WS).