Investigating Disparities in Physical Therapy Utilization: An Intersectionality Perspective

Abstract

Objective

To investigate the role of race, ethnicity, language, insurance payor, and socioeconomic status, both individually and through an intersectional framework, on outpatient physical therapy (PT) utilization. A secondary aim was to examine the differences in scheduling and attendance based on the aforementioned factors.

Design

A retrospective cohort study examining outpatient PT referrals, scheduled appointments, and attendance. Data included the status of visit completion, race, ethnicity, language, insurance payor, and zip code. Multiple logistic regression models, with and without interaction terms, analyzed the association of demographic factors with outcomes of interest.

Setting

This retrospective cohort study collected data between July 2021 and July 2022 from electronic medical records within a large academic medical system in New England.

Participants

N=61,125.

Interventions

Not applicable.

Main Outcome Measures

This study assessed 3 outcomes. Outcome 1 analyzed the rates of scheduling after referral. Outcome 2 analyzed the rates of attendance after scheduling. Outcome 3 analyzed the rates of attendance after referral.

Results

Race, language, and income were associated with differences in scheduling versus attending PT. Black or African American patients showed the highest appointment-making rates and lowest attendance rates after scheduling compared with White patients. Asian patients demonstrated the lowest appointment-making rates and highest attendance rates after scheduling compared with White patients. Non-English-speaking patients were less likely to schedule and attend PT compared with English-speaking patients. Higher socioeconomic status was associated with higher rates of scheduling and attendance. Further disparities were noted when examining the interaction of variables. Increasing income benefited most, but not all groups. Black or African American patients experienced a decrease in scheduling and attendance rates with rising income compared with White patients. Non-English-speaking patients experienced less of an increase in scheduling and attendance rates with rising income compared with English-speaking patients.

Conclusions

Findings highlighted disparities in PT utilization in scheduling and attendance with regard to race, language, and income. Disparities were amplified when examining interactions between race and income and language and income, underscoring the importance of an intersectional analysis.

Highlights

• •Disparities in PT scheduling and attendance by race, language, and income.
• •Black patients schedule more but attend less; Asian patients show opposite pattern.
• •Higher income boosts scheduling and attendance, but less for marginalized groups.
• •Language-concordant care improves scheduling and attendance rates.
• •Intersectional analysis reveals cumulative barriers for marginalized identities.

Patients who receive physical therapy (PT) improve in functional outcomes regardless of race.1, 2, 3, 4 Despite this, historically marginalized groups experience lower PT utilization.^1,2,5, 6, 7, 8 Literature has found a correlation of rehabilitation utilization factors with identity, social determinants of health, and structural determinants of health equity, such as gender, ethnicity, race, education, transportation access, employment, socioeconomic status (SES), insurance, and language-concordant care.9, 10, 11, 12, 13 When assessing factors that affect disparities for marginalized groups, it is important to frame systems of oppression as health risk factors rather than identity markers (i.e., racism is a risk factor, not a race). Data on race can be used as a proxy to measure the effect of structural racism and its associations with negative health outcomes.

While prior studies have examined individual factors that affect PT utilization, they do not assess the dynamic interactions of these factors with an intersectional lens. Intersectionality is an anti-oppression framework that examines how social identities converge to create unique experiences of oppression under systemic and institutional practices that unfairly advantage some and disadvantage others.14, 15, 16 While the use of intersectionality frameworks has been established in prior health care research,17, 18, 19, 20 this approach has not been used in PT. An intersectional application fosters a deeper understanding of the interplay between social determinants of health, structural determinants of health equity, and identity factors contributing to unique circumstances and health outcomes (fig 1).

Fig 1.

Framework of systemic inequities. SDHE, structural determinants of health equity; SDOH, social determinants of health.

Additionally, literature examining PT utilization has focused on both referral21, 22, 23 and attendance,^12,24 but there is limited information regarding the scheduling process. This unexplored step in the continuum of PT utilization can provide more information about how patients make decisions or attempt to access outpatient PT services. The primary aim of this study was to investigate the role of race, ethnicity, language, insurance payor, and SES, both individually and through an intersectional framework, on PT scheduling and attendance to identify potential disparities within a large health care system. The secondary aim of this study was to examine differences in scheduling and attendance based on the aforementioned factors.

Methods

This retrospective cohort study collected data between July 2021 and July 2022 from electronic medical records within a large academic medical system in New England. All patients ≥18 years of age who were referred to outpatient PT and scheduled for an initial visit were included. Referral sources included both inpatient and outpatient providers. The structure of the electronic medical record system required running 2 distinct reports simultaneously to collect all necessary data from each patient’s chart. Report 1 encompassed patients referred to PT, and report 2 encompassed patients who scheduled an initial evaluation. All covariates were assessed before follow-up, and all data were collected at one time. Data collected included status of visit completion, insurance payor, zip code, and self-reported race, ethnicity, and primary language (Table 1, Table 2). See supplemental materials (available online only at http://www.archives-pmr.org/) for the utilization of inclusive language regarding the naming of groups. Duplicate patients were excluded, with data collected for the first referral and initial visits scheduled. Data analysis included the 300 insurance payors gathered from patient data. Payors with over 1000 patients in the data were analyzed individually. Payors with fewer than 1000 patients were grouped into Other^p. The first 6 months of referral data were analyzed to capture scheduled and completed appointments for the duration of the study timeframe, resulting in N=61,125 patients. This ensured every single patient had up to 6 months to schedule and attend an appointment, which was ample time to account for clinic wait times. All data were deidentified, and the study was approved by the institution’s Institutional Review Board. Informed consent was deemed not applicable due to the retrospective nature of this study.

Table 1.

Descriptive statistics of outcomes.

Variable	Category	Outcome 1: Schedule After Referral n/N (%)	Outcome 2: Attend After Schedule n/N (%)	Outcome 3: Attend After Referral n/N (%)
Race	Asian	577/2655 (21.7)	903/1272 (71.0)	455/2655 (17.1)
	Black or African American	1401/4148 (33.8)	1717/2799 (61.3)	971/4148 (23.4)
	Multiracial	108/418 (25.8)	163/242 (67.4)	77/418 (18.4)
	White	14,138/47,038 (30.1)	22,618/32,425 (69.8)	10,897/47,038 (23.2)
	Other*	1749/6720 (26.0)	2400/3715 (64.6)	1245/6720 (18.5)
Ethnicity	Hispanic	1657/5898 (28.1)	2079/3342 (62.2)	1163/5898 (19.7)
	Non-Hispanic	14,808/50,530 (29.3)	22,944/33,204 (69.1)	11,332/50,530 (22.4)
	Declined/unavailable	1538/4697 (32.7)	2825/3975 (71.1)	1174/4697 (25.0)
Language	English	16,881/56,202 (30.0)	19,430/2963 (69.5)	12,883/56,202 (22.9)
	Spanish	713/2684 (26.6)	718/1163 (61.7)	500/2684 (18.6)
	Other+	306/1679 (18.2)	347/535 (64.9)	206/1679 (12.3)
	Declined/unavailable	103/560 (18.4)	133/175 (76.0)	80/650 (14.3)
Zip code median income quintile	1 (lowest)	2856/12,218 (23.4)	3701/5855 (63.2)	2011/12,218 (16.5)
	2	3052/11,314 (27.0)	3978/5689 (69.9)	2330/11,314 (20.6)
	3	3202/11,953 (26.8)	4090/5834 (70.1)	2433/11,953 (20.4)
	4	3804/11,503 (33.1)	4094/5765 (71.0)	2914/11,503 (25.3)
	5 (highest)	4316/11,420 (37.8)	3886/5392 (72.1)	3398/11,420 (29.8)

^⁎Other: patients who self-selected Other or for whom data was unavailable.

⁺Other: grouped 86 languages for data analysis to as individual languages in this group were too small.

Table 2.

Descriptive statistics of payors.

Outcome 1: Scheduling After Referral			Outcome 2: Attendance After Scheduling			Outcome 3: Attendance After Referral
Insurance Payor	Total Patients N	Scheduling Rate %	Insurance Payor	Total Patients N	Attendance Rate %	Insurance Payor	Total Patients N	Attendance Rate %
Blue Cross Blue Shield	21,224	31	Medicare	11,389	70	Blue Cross Blue Shield	21,224	24
Medicare	18,583	30	Blue Cross Blue Shield	10,889	72	Medicare	18,583	29
MassHealth	10,286	31	MassHealth	5191	59	MassHealth	10,286	20
United Healthcare	5638	29	Always Health Partners	2820	72	United Healthcare	5638	22
Always Health Partners	5217	33	United Healthcare	2591	69	Always Health Partners	5217	26
Harvard Pilgrim	4349	34	Harvard Pilgrim	2543	72	Harvard Pilgrim	4349	27
Tufts Health Plan	3906	35	Tufts Health Plan	2237	73	Tufts Health Plan	3906	28
Workers Compensation	2245	30	UniCare GIC	1179	71	Workers Compensation	2245	23
Cigna	2206	19	Workers Compensation	1066	69	Cigna	2206	15
Aetna	2170	24	Motor Vehicle	820	64	Aetna	2170	18
UniCare GIC	1931	37	Aetna	859	70	UniCare GIC	1931	28
Health Safety Net	1742	27	Health Safety Net	744	61	Health Safety Net	1742	18
Motor Vehicle	1489	33	Cigna	673	71	Motor Vehicle	1489	24
Generic Commercial	1421	13	Commonwealth Care Alliance	276	59	Generic Commercial	1421	9
Health New England	704	8	Generic Commercial	272	61	Health New England	704	6

Abbreviations: GIC, Stands for Group Insurance Commission.

Three outcomes of interest were analyzed to best understand patterns in scheduling and attendance:

• •Outcome 1 – scheduling after referral: a binary outcome indicating whether a patient, internally referred to PT, scheduled an appointment after their first referral. Only internally referred patients were included.
• •Outcome 2 – attendance after scheduling: a binary outcome indicating whether a patient who scheduled an evaluation attended their first scheduled appointment. Rescheduled appointments were ignored. Both internally and externally referred patients were included.
• •Outcome 3 – attendance after referral: a binary outcome indicating whether a patient, internally referred to PT, both scheduled an appointment and attended an appointment at any point in time during the data range, not limited to a patient’s first scheduled appointment. Only internally referred patients were included.

A total of 61,125 patients were internally referred between July and December 2021. Outcome 1 analyzed whether these patients scheduled an appointment between July 2021 and July 2022 after referral, and outcome 3 analyzed whether these patients attended their scheduled appointment after referral between July 2021 and July 2022. A total of 18,803 internally referred patients scheduled an appointment. Additionally, there were 21,718 externally referred patients who scheduled an appointment between July 2021 and July 2022, resulting in 40,521 patients analyzed in outcome 2 (fig 2).

Fig 2.

Data collection.

Multiple logistic regression models, with and without interaction terms, analyzed the association of demographic and clinical factors with the 3 outcomes of interest. Regressions were run separately for each outcome. Patients never appeared more than once in the analyzed data sets. The main models created for outcomes 1, 2, and 3 had McFadden pseudo R-squared statistics of 0.038, 0.016, and 0.035, respectively. These relatively low goodness-of-fit results indicated that, while the reported trends are true on average, there are still many patients who do not fit the trend (see appendix S1, available online only at http://www.archives-pmr.org/). Pseudo R-squared values for the rest of the models, as well as area under the curve (AUC) statistics for all models, can be found in the supplementary data appendix. As a robustness check, quasi-Poisson regression models were also used to replicate each logistic regression model (see supplementary data appendix).

Logistic regression models included the following categorical independent variables: race, ethnicity, language, and insurance payor. Primary languages were English, Spanish, declined/unavailable, or Other⁺. Other⁺ grouped 86 languages for data analysis. Similar to insurance, there were not enough patients in each individual language to be treated individually. This resulted in treating all languages other than English or Spanish as “Other*.” Racial groups were American Indian or Alaska Native, Asian, Black or African American, Hispanic or Latino, multiracial, Native Hawaiian or Other Pacific Islander, White, or Other*. Other* denotes patients who self-selected Other or for whom data were unavailable. Ethnic groups were Hispanic, non-Hispanic, or declined/unavailable. Zip codes were paired with the estimated zip code median income from US census data²⁵ as a proxy to measure SES.26, 27, 28 These median incomes were broken into quintiles (Qs) and were the only numeric independent variable. Table 1 shows descriptive results of these variables. In some instances, certain covariates were unavailable for all patients, resulting in models running in 2 ways: excluding the variables with incomplete data, allowing all or most participants to remain in the model; and retaining the variables of interest and excluding the patients for whom data on these variables were unavailable. As a result, one model was run without zip code median income data but included externally referred patients, and another model was run with zip code median income data and excluded externally referred patients. For example, in Outcome 2, the data set lacked information on zip code median income for patients externally referred to the hospital system.

Results

A total of 61,125 patients were referred to PT, and 40,251 patients scheduled an appointment (fig 2).

Individual variables

Outcome 1 analyzed scheduling after referral. When looking at race, Black or African American patients were most likely to schedule an appointment (33.8%), followed by White (30.1%), Other* (26.0%), multiracial (25.8%), and Asian patients (21.7%). Logistic regression confirmed these differences when controlling for all covariates. When considering population-level results, Black or African American patients had 1.395 times the odds of scheduling an appointment after referral (95% CI, 1.30-1.50, P<.001) compared with White patients, while Asian patients had 0.69 times the odds (95% CI, 0.63-0.76, P<.001). Results for Other* and multiracial patients were less certain at the population level. When looking at ethnicity, there was no significant difference between Hispanic (28.1%), non-Hispanic (29.3%), or declined/unavailable (32.7%) patients. Logistic regression estimated that Hispanic (OR=1.21, 95% CI, 1.10-1.32, P<.001) and declined/unavailable (OR=1.33, 95% CI, 1.24-1.42, P<.001) patients were more likely than non-Hispanic patients to schedule appointments. English-speaking patients had the highest scheduling rates (30.0%), followed by Spanish-speaking patients (26.6%), declined/unavailable (18.4%), and Other⁺ (18.2%). Logistic regression showed that patients in the Other⁺ and declined/unavailable groups had significantly lower odds of scheduling compared with English-speaking patients (OR=0.61, 95% CI, 0.54-0.70, P<.001 and OR=0.53, 95% CI, 0.42-0.67, P<.001, respectively). No statistically significant difference was found between Spanish- and English-speaking patients. Patients in the highest-earning Q (Q5) scheduled an appointment 37.8% of the time, while those in the lowest-earning Q (Q1) scheduled an appointment 23.2% of the time. Logistic regression revealed that patients in Q5 had approximately double the odds of scheduling compared with patients in Q1 (OR=2.00, 95% CI, 1.89-2.12, P<.001). Patients with Medicare were most likely to schedule, with 37% of patients scheduling after referral. Patients with Health New England were least likely to schedule, with 8% of patients scheduling after referral. The population-level results supported these findings. Logistic regression results showed that patients with Medicare (OR=1.73, 95% CI, 1.66-1.80, P<.001) are substantially more likely than Other^p patients to schedule an appointment, while patients with Health New England (OR=0.32, 95% CI, 0.24-0.42, P<.001) are substantially less likely.

Outcome 2 analyzed attendance after scheduling. When looking at race, Asian patients had the highest rate of attendance after scheduling (71.0%), followed by White (69.8%), multiracial (67.4%), Other* (64.6%), and Black or African American (61.3%) patients. Population-level logistic regression results did not show statistically significant differences between White patients and Asian, multiracial, or Other* patients, while Black or African American patients had lower odds of appointment attendance compared with White patients (OR=0.79, 95% CI, 0.71-0.87, P<.001). When looking at ethnicity, declined/unavailable patients had the highest attendance rates after scheduling (71.1%), followed by non-Hispanic (69.1%) and Hispanic (62.2%) patients. However, when controlling for all covariates, there was no evidence of differences between these groups at the population level. The following results are only for internally referred patients because not all covariate data were available for externally referred patients. Patients whose language was declined/unavailable had the highest rates of attendance (76.0%), followed by English (69.5%), Other* (64.9%), and Spanish (61.7%). Logistic regression revealed that there was no statistical difference between language groups and the odds of attendance at the population level. Patients in Q1 were least likely to attend after scheduling (63.2%) compared with patients in Q2-Q5 (69.9%-72.1%). Logistic regression showed that patients in Q2-Q5 had significantly greater odds of attending appointments compared with patients in Q1. Patients with Tufts and Cigna had the highest rates of attendance, with 73% of patients attending an appointment after scheduling. Patients with MassHealth had the lowest rates of attendance, with 59% of patients attending after scheduling. Logistic regression showed that patients with Tufts (OR=1.17, 95% CI, 1.03-1.32, P=.02) were substantially more likely to attend, while patients with MassHealth (OR=0.69, 95% CI, 0.63-0.77, P<.001) were substantially less likely than Other^p patients to attend an appointment.

For outcome 2, more limited models were run on both internally and externally referred patients with the available independent variables. When comparing internally and externally referred patient attendance after scheduling, 68.2% of internally referred patients attended appointments, while 69.1% of externally referred patients attended appointments. Logistic regression showed that internally referred patients had slightly higher odds of attendance on average than externally referred patients (OR=1.06, 95% CI, 1.01-1.11, P=.017). This is important because the referral source was not a significant variable in outcome success compared with other covariates.

Outcome 3 analyzed attendance after referral. The groups with the highest attendance rates were Black or African American (23.4%) and White (23.2%) patients. Patients identifying as Other* (18.5%), multiracial (18.4%), and Asian (17.1%) had lower rates of attendance after referral. At the population level, Black or African American patients had increased odds of attendance after referral compared with White patients (OR=1.25, 95% CI, 1.15-1.36, P<.001). Other* and Asian patients had decreased odds compared with White patients (OR=0.89, 95% CI, 0.81-0.97, P=.006 and OR=0.74, 95% CI, 0.67-0.83, respectively). There was no statistical difference in the odds of attendance among multiracial patients compared with White patients. When looking at ethnicity, declined/unavailable patients had the highest attendance rates (25.0%), followed by non-Hispanic (22.4%) and Hispanic (19.7%) patients. However, when adjusting for all covariates, logistic regression estimated that declined/unavailable or Hispanic patients had higher odds of attendance after referral than non-Hispanic patients (OR=1.29, 95% CI, 1.19-1.39, P<.001 and OR=1.15, 95% CI, 1.04-1.27, P=.04, respectively). Attendance rates were highest for English- (22.9%) and Spanish-speaking patients (18.6%), followed by declined/unavailable (14.3%) and Other⁺ (12.3%) patients. Logistic regression showed that patients in the Other⁺ and declined/unavailable language groups had significantly lower odds of appointment attendance after referral compared with English-speaking patients (OR=0.60, 95% CI, 0.51-0.69, P<.001 and OR=0.58, 95% CI, 0.45-0.75, P<.001, respectively). No statistically significant difference was found between Spanish- and English-speaking patients. Patients in Q1 attended an appointment after referral 16.5% of the time compared with patients in Q5, who attended an appointment after referral 29.8% of the time. Per logistic regression results, patients in Q2-Q5 had significantly greater odds of attendance after referral compared with patients in Q1. Patients with Medicare had the highest rates of attendance, with 29% of patients attending an appointment after referral. Patients with Health New England had the lowest rates of attendance, with 6% of patients attending an appointment after referral. Logistic regression showed that patients with Medicare (OR=1.68, 95% CI, 1.61-1.76, P<.001) are substantially more likely than Other^p patients to attend an appointment, while patients with Health New England (OR=0.35, 95% CI, 0.26-0.49, P<.001) are less likely to attend than Medicare patients.

Intersectional analysis: interactions between variables

Logistic regression models were run for individual variables and interaction terms for each outcome. When examining the statistical interactions of the independent variables, similar relationships were identified for outcomes 1 and 3.

Outcomes 1-3 interactions

When examining interactions between language and race, White and English-speaking patients were the reference groups. In isolation, English-speaking patients had higher rates of completion across all 3 outcomes compared with non-English-speaking patients. When examining interactions between language and racial groups, there were no significant interaction effects on the odds of scheduling and/or attending an appointment. This indicates that while English-speaking patients are more likely to schedule and attend appointments, this positive effect was experienced similarly across all racial groups.

When examining interactions between language and zip code median income Q, English-speaking patients were the reference group. For scheduling after referral (outcome 1) and attendance after referral (outcome 3), the odds of scheduling an appointment increased with a 1 Q increase in zip code median income. This effect was amplified or dampened based on language. English-speaking patients’ odds of scheduling increased by 1.195 times for each 1 Q increase in zip code median income Q for outcome 1 and by 1.198 times for outcome 3. However, this positive effect was dampened for non-English-speaking patients (Spanish and Other*) (outcome 1: Spanish OR=0.78, 95% CI, 0.72-0.84, P<.001; Other* OR=0.80, 95% CI, 0.73-0.89, P<.001; outcome 3: Spanish OR=0.800, 95% CI, 0.733-0.87-0.944, P<.001; Other* OR=0.83, 95% CI, 0.74-0.92, P=.001). There was no difference for declined patients in outcomes 1 and 3. There were no significant language interactions for attendance after scheduling (outcome 2). This indicates that while increasing income Q increases the odds of scheduling, this effect was experienced similarly across language groups in outcome 2.

When examining interactions between race and zip code median income Q with White patients as the reference group (fig 3), the odds of scheduling after referral (outcome 1) increased by 1.23 times for each 1 Q increase in zip code median income Q for White patients (OR=1.23, 95% CI, 1.21-1.25, P<.001), and the odds of attendance after referral (outcome 3) increased by 1.22 times (OR=1.22, 95% CI, 1.20-1.25, P<.001). However, interaction data indicate that this effect was reversed for Black or African American patients for outcomes 1 and 3 (outcome 1: OR=0.78, 95% CI, 0.75-0.83, P<.001; outcome 3: OR=0.79, 95% CI, 0.75-0.84, P<.001). The results for Asian and multiracial patients did not deviate significantly from those of White patients. This effect was dampened for Other* patients compared with White patients (outcome 1: OR=0.83, 95% CI, 0.79-0.87, P<.001; outcome 3: OR=0.85, 95% CI, 0.80-0.89, P<.001). The results did not demonstrate any significant interaction effect for race and zip code median income Q on attendance after scheduling (outcome 2), suggesting that the relationship between odds of appointment attendance and zip code median income Q might not vary across racial groups.

Fig 3.

Interaction between race and zip code median income quintile.

Discussion

While directly examining how systemic inequities, such as structural racism, affect PT utilization was beyond the scope of this study, we aimed to identify whether disparities in PT utilization exist in this health care system, and the findings should be understood within the context of systemic inequities (fig 1). Although it is well-established that inequities contribute to disparities in health care utilization,29, 30, 31, 32, 33 objectively demonstrating the existence of disparities is essential for designing effective solutions.

To better define the issue, PT utilization was analyzed at multiple points after physician referral, focusing on 3 key outcomes related to scheduling and attendance rates. When examining individual factors, patients who hold aspects of a marginalized identity, such as having a minoritized racial identity, having a lower SES, or receiving language-discordant care, had lower rates of scheduling and/or attendance, consistent with existing literature.^1,2,5, 6, 7, 8 Additionally, the findings revealed that a single group’s utilization patterns could vary between scheduling and attendance. Specifically, Black or African American patients had the highest scheduling rates but the lowest attendance rates after scheduling compared with White patients. In contrast, Asian patients exhibited the opposite pattern, with the lowest scheduling rates but the highest attendance rates compared with White patients. These differences underscore the need to examine all 3 outcomes, as focusing solely on attendance would have overlooked important patterns in scheduling, and patients may have experienced barriers at 1 or more stages. Identifying stage-specific barriers and understanding how they differ across groups can inform targeted interventions that address a greater scope of access issues. For example, with regard to improving scheduling and attendance rates for patients based on language-concordant care, a language-matched scheduler can ensure clear communication and may improve scheduling rates, while a provider who speaks the patient’s language fosters trust and comfort⁹ and may improve attendance rates.

An additional factor explored in this study that could have affected scheduling and/or attendance patterns was insurance payor. While positive correlations between private insurance and PT utilization exist in the literature,¹² variable copays and reimbursements may be a barrier. For example, if a patient’s insurance does not cover PT, they may not schedule an appointment. On the other hand, if a patient schedules an appointment but then learns about a significant copay, they may not attend. In this study, patients with specific types of insurance tended to have higher rates of scheduling and attendance. However, this trend varied based on payor and whether the payor was public or private. Future studies should examine interactions between insurance payors and demographic factors to better understand how these relationships influence PT utilization, specifically, both scheduling and attendance.

Lastly, this study uniquely contributes to the body of literature regarding PT utilization due to its intersectional analysis that explored how the interaction of multiple factors associated with marginalized identity status influenced scheduling and/or attendance. Findings demonstrated that individuals with multiple marginalized identities had fewer scheduled and attended visits compared with those with fewer marginalized identities or from dominant groups, indicating that these individuals likely faced cumulative barriers to access. For instance, while increased SES was associated with higher utilization rates, this trend did not apply consistently across all racial or language groups. Black or African American patients experienced a decrease in both scheduling and attendance with rising income compared with White patients. While Other* patients saw higher scheduling and attendance rates with increasing income, the effect was less pronounced compared with White patients. Additionally, Spanish-speaking patients experienced a smaller increase in scheduling and attendance rates with rising income compared with English-speaking patients. By running logistic regression models with interaction terms, this study was able to capture this trend.

These findings can inform meaningful interventions to improve equitable access to PT with an emphasis on targeted approaches based on the intersection of unique patient factors. For example, patients who schedule but do not attend their first visit may require different interventions than those who do not schedule at all. While it is not known whether similar patterns occur in other health care systems, the overall attendance rates in this system align with those reported in the existing literature,^3,34,35 suggesting that these findings may be generalizable to other systems in the United States. Given the evidence that patients who attend PT experience positive outcomes regardless of identity,1, 2, 3, 4 implementing appropriate interventions to optimize access along the continuum of care is critical. A qualitative follow-up would provide a deeper understanding of specific reasons why patients may face barriers with scheduling and/or attendance to outpatient PT.

Study limitations

While authors sought to control as many variables as possible, this study included some notable limitations. Data collection could not capture patients who attended PT out-of-network. However, the rates of PT utilization in this study were similar to other reported studies,^3,34,35 ensuring consistency with previous research. Underrepresented language groups were categorized using “Other⁺” due to their small sample sizes. This study used zip code median income as a proxy for SES, as supported in previous literature,^26,36 but this may not fully capture the complexity of an individual’s unique financial circumstances. The utilization data may have been affected by the COVID-19 pandemic; however, data were not collected until after the mandated federal and local lockdowns were lifted. Lastly, the study relied on data collected across a large health care system, and potential biases in data collection, such as selection and referral biases, could have occurred due to factors such as incomplete records, health literacy, and systemic differences in access to care.

Conclusions

Study findings highlighted disparities in PT scheduling and attendance with regard to race, language, and income. These disparities were amplified when examining variable interactions, demonstrating how individuals from multiple marginalized identities face cumulative barriers and further underscoring the importance of an intersectional analysis. Despite the documented benefits of PT across demographic lines, reduced PT participation by minoritized groups warrants further investigation to elucidate factors influencing PT access and utilization. Intersectional research is crucial for understanding and addressing the multifaceted nature of these disparities to identify and address systemic inequities resulting in unique access challenges for marginalized groups.

Disclosure

None.