Identifying Barriers to Being Offered and Accepting a Telehealth Visit for Cancer Care: Unpacking the Multi‐Levels of Documented Racial Disparities in Telehealth Use

ABSTRACT

Objective

To evaluate patient‐ and area‐level factors in relation to telehealth visit use in cancer care.

Study Setting and Design

We surveyed a cohort of adults with an upcoming healthcare visit related to their cancer treatment at two academic medical centers (one in central North Carolina and one in southeast Michigan) and their community affiliates. Black adults and those with a scheduled telehealth visit were purposively oversampled during recruitment. We linked respondent residential addresses to area‐level measures, including broadband access. The two patient‐reported outcomes of interest were (1) whether a choice in visit type (virtual or in‐person) was offered and (2) scheduled visit type.

Data Sources and Analytic Sample

We assembled a cohort of 773 adults (response rate = 15%). After excluding nonrecall for being offered a choice, the analytic sample was 725 adults.

Principal Findings

The sample was 46% aged < 65 years, 42% Black, and 67% women. Black respondents were less likely than non‐Black respondents to be offered a choice, 15% versus 23%, prevalence difference (PD) and 95% CI = (−8.7%, CI: −14.4, −3.0) and if offered a choice, less likely to accept a telehealth visit (20% vs. 67%; PD = −47.0%, CI: −62.0, −32.0). Compared to men, women had a lower frequency of visit choice (16% vs. 27%; PD = −10.9%. CI: −17.4, −4.4) and accepted telehealth visits (42% vs. 63%; PD = −20.8%, CI: −36.8, −4.7). Respondents who expressed technology‐related worries were less likely to accept a telehealth visit. Lower area‐level technology access (e.g., broadband ownership) and higher poverty were nonsignificantly associated with less offering and less scheduling of telehealth visits.

Conclusions

Interventions to improve access to telehealth in cancer care and mitigate structural inequities (namely racism and sexism) should consider patient‐ and area‐level barriers to being offered a choice in visit type and the ability to accept a telehealth visit.

Summary.

• What is known on this topic

  • ○The use of telehealth visits in cancer care can improve accessibility, timeliness, and convenience for patients; ensuring equitable access to telehealth should be a health policy priority.
  • ○However, disparities in its use persist, and few studies have evaluated both patient‐ and area‐level factors in relation to telehealth visit use in cancer care.
  • ○Further, no prior study has separately evaluated barriers to both steps to receiving a telehealth visit: being offered a choice of visit format and the decision to accept a telehealth visit.

• What this study adds

  • ○Black adults and women were less likely to be offered a choice in visit format, and if offered a choice, were less likely to have a telehealth visit scheduled.
  • ○Patients with concerns about their insurance covering telehealth visits were less likely to be offered a choice, whereas patients with technology concerns were less likely to accept a telehealth visit.
  • ○The causes of inequities in telehealth utilization are multifaceted, and appropriate policy solutions will need to target a complex suite of patient‐, health system‐, and environmental‐level barriers to care.

1. Introduction

Telehealth in cancer care is affordable, convenient, timely, and safe [1, 2, 3, 4]. The use of telehealth visits in oncology settings surged with the COVID‐19 pandemic [5, 6]. Studies have consistently found lower rates of telehealth utilization among Black and Hispanic adults with cancer [7, 8, 9, 10, 11, 12, 13] as well as among older and rural adults [9, 14, 15, 16]. There are also specific barriers that correlate with lower telehealth utilization, such as being uninsured, poor self‐reported health, negative perceptions of telehealth, and lack of technology/device access [9, 17, 18, 19]. Given the potential benefits of telehealth in oncology, ensuring equitable access should be a health policy priority.

Many prior studies of telehealth utilization in oncology have been based on electronic health record (EHR) or insurance claims data [8, 10, 11, 12, 13, 14, 18]. While informative, these studies also have several important limitations, including a lack of information on potential confounders such as visit purpose or symptom severity as well as a focus on one single outcome: the occurrence of a virtual visit. In practice, however, receiving a telehealth visit is the result of multiple steps, including the decision to offer a choice of visit format and the decision to accept or decline such an offer.

Following a recent interview study from our team in which Black adults reported preferences for in‐person care [20], we hypothesized there might be differences in which patients schedule telehealth visits, or perhaps even who is offered telehealth visits. As indicated by well‐established frameworks for studying healthcare use such as Anderson's model of healthcare utilization, characteristics of patients, health systems, and the external environment all drive healthcare use patterns [21]. Although there are known patient‐level barriers to telehealth (e.g., difficulties accessing the internet), there is comparatively little study of the impact of structural barriers that may require community‐level interventions (e.g., broadband coverage) [18, 22].

We aimed to address these research gaps by surveying patients before a scheduled oncology appointment to ascertain information on whether they were offered a choice regarding visit format, their perceptions regarding barriers to care, care preferences, sociodemographic characteristics, and their community context. The objectives of this study were (1) to describe the frequency of being offered a choice of visit type (in‐person or virtual) and further, whether a virtual visit was scheduled among those offered a choice and (2) to assess the impact of patient‐ and area‐level barriers on each outcome.

2. Methods

2.1. Study Population

We recruited a cohort of adults who had undergone treatment for cancer within two integrated health systems in the United States (U.S.), one in North Carolina and one in Michigan. Eligible respondents were (1) aged 21 years or older with a cancer diagnosis; (2) scheduled for an oncology or urology appointment (telehealth or in‐person) within 3 weeks; and (3) treated for cancer no more than 3 years—but no less than 6 months—prior to patient identification. Included cancer types were head and neck cancer, prostate cancer, genitourinary cancers, breast cancer, gastrointestinal cancers, lung cancer, or hematological cancers. All adults with an upcoming virtual visit were eligible for the study. In addition, a random sample of patients with an upcoming in‐person visit was also drawn. We oversampled Black patients for both visit types with the goal of enrolling a study population that was 50% Black and 50% other racial groups. All eligible telehealth patients were contacted, whereas a subset of in‐person patients was randomly sampled. Individuals with evidence of hospice/comfort care in the prior 6 months, incarceration, cognitive impairment, or who were unable to read/comprehend English were excluded.

2.2. Respondent Recruitment and Data Sources

At the study onset, we aimed to enroll 1000 adults (N = 500 Black race and N = 500 other races). The sample size was determined to achieve 80% power to detect a 10% difference in a binary outcome by race (conservatively assuming a 15% prevalence of telehealth visits in the overall study population). Recruitment and data collection occurred between April 2022 and October 2023; this duration was determined by available funding. Eligible respondents were identified from EHR data. Those who gave informed consent to participate were asked to complete a telephone survey prior to and following their scheduled appointment. The survey was administered by a trained survey research team member. The present study leverages data from the pre‐visit survey, which was completed by 773 respondents. In the pre‐visit survey, information was collected for sociodemographic characteristics, barriers to care, pain and emotional worries, and care preferences. We geo‐coded respondents' residential addresses to the census tract level and joined them with area‐level information from the 2020 American Community Survey (ACS) 5‐year estimates [23].

2.3. Outcome Assessment

This study evaluated two self‐reported outcomes: (1) choice of visit format and (2) scheduled visit type. Visit choice was assessed by the question “When you scheduled this visit, were you offered a choice regarding the format of the visit—that is, did the person who helped you schedule the visit ask if you wanted your visit to be in‐person or via video or telephone?” Response choices were “yes,” “no,” or “unsure.” The unsure group (N = 48) was excluded from the analysis. The second outcome, scheduled visit type, was assessed by the question “Is this visit scheduled to be in‐person in the clinic or via video or telephone?” We assessed patient‐reported scheduled visit type—regardless of which type of visit ultimately happened—as this provided the best reflection of how patients evaluated their preferences and barriers when making a telehealth decision. Those who answered via video or telephone were defined as having a telehealth visit. We evaluated scheduled visit type among those who were given a choice, given our objective to identify barriers to accepting a telehealth visit when both virtual and in‐person are possible.

2.4. Telehealth Barriers and Concerns

Technology barriers were assessed in the respondent survey with four questions (response options yes/no/unsure) asking respondents if they had worries regarding reliable internet access, having a necessary device, lacking someone to help connect, and privacy for a virtual visit [24]. Financial barriers were assessed with three questions. Two assessed respondents' beliefs about whether their health insurance covered virtual visits and whether their copayment would be equal to an in‐person visit (yes/no responses). The third question was about self‐perceived income, given prior research that found self‐perceived income adequacy to be associated with telehealth access [25]. We also identified a priori four area‐level factors (measured at the census tract level) that may reflect structural barriers to cancer care: percentage of people living below the federal poverty level (dichotomized as greater/less than 20%), percentage of people with a broadband internet subscription (greater/less than 80%), percentage of people owning a computer (greater/less than 80%), and percentage of people owning a vehicle (greater/less than 80%). These factors were selected for their potential relation to patient decision‐making regarding telehealth visits and whether patients were offered a choice in format (namely, if providers and staff made generalizations about residential areas based on locally held beliefs, perceptions, misperceptions, and biases).

2.5. Symptoms and Care Goals

Respondents rated their health‐related quality of life (HRQOL) on a 5‐point scale [26], and we combined responses of “good/very good/excellent” and “fair/poor.” Physical and emotional symptoms were measured using the patient‐reported outcomes version of the Common Terminology Criteria for Adverse Events (PRO‐CTCAE) scale [27]. For this analysis, we selected four PRO‐CTCATE symptoms with potential relevance to a patient's decision to seek virtual care: pain, anxiety, sadness, and inability to be cheered up. Each symptom was measured for domains of frequency, severity, and/or intensity. Using published PRO‐CTCATE grading algorithms, we generated composite scores (ranging from 0 to 3) for each symptom [28], and classified scores of 0 or 1 as “low burden” and scores of 2 or 3 as “high burden”. We also ascertained care‐related preferences and goals for the upcoming visit [29]. On a 7‐point scale ranging from “very strongly agree” to “very strongly disagree,” respondents indicated their expectation for a full body examination, a prescription, advice, arranging labs/tests, and a discussion of feelings.

2.6. Sociodemographic Characteristics

Age was categorized as < 65 years or 65+ years, facilitating comparison with previous studies using these categories [16]. We interpreted race as a social construct which may reflect experiences of racism and discrimination [30]; respondents were asked to report their racial self‐identity (“which of the following is your race?”) as well as their socially assigned race (“How would other people usually classify you?”). Given that both self‐identified race and socially assigned race correlate to differential health outcomes, we used both measures to construct our definition of race [31]. Respondents who indicated “Black” in response to either race question were categorized as “Black,” while respondents who did not report “Black” for either question were classified as “Non‐Black.” Other self‐reported demographic variables were ethnicity (Hispanic or non‐Hispanic), gender, sexual orientation. education, and self‐reported usual travel time to cancer care (less than 60 min, 60–120 min, or more than 120 min).

2.7. Statistical Analysis

Statistical analysis was conducted using SAS version 9.4 (SAS Institute), and the preparation of tables and figures was performed in R version 4.3.1 (R Foundation for Statistical Computing). Associations of symptoms and care goals with each outcome (visit choice and scheduled visit type) were assessed by chi‐squared tests with alpha = 0.05. The association of demographic factors with each outcome was calculated using linear‐binomial regression models to compute prevalence differences (PDs) and 95% confidence intervals (CIs). PDs for demographic characteristics were not adjusted for additional covariates, as we were interested in describing observed disparities in telehealth use rather than estimating a counterfactual “effect” of nonmodifiable factors like race and gender [32]. We estimated the effect of potentially modifiable (and thus intervenable) barriers on each outcome using both unadjusted and adjusted models; for adjusted models of sparse data, we used the weighted copy method [33, 34] to enable the convergence of all models. Covariate adjustment sets included factors that may affect the necessity for in‐person care; these variables were patient‐reported pain (composite score dichotomized as high or low), emotional stress (high/low), care preferences that we hypothesized would reflect a desire for in‐person services (tests/labs, physical examination, or both), and overall self‐reported health. Interpretation of whether a PD reflects a potentially important association was guided by both statistical significance (whether the 95% CI crossed the null) and the magnitude and precision of the point estimate. Finally, to assess the potential effect of nonresponse bias, we calculated the inverse probability of sampling weights based on patient race and scheduled visit type [35]. These weights were used to conduct sensitivity analyses for each regression model.

3. Results

This study recruited 773 participants with an overall response rate of 15%. Demographic characteristics of the analytic sample (N = 725) are reported in Table S1. Forty‐six percent of patients were aged less than 65 years, 42% Black, and 67% women. Self‐identified and perceived race were largely concordant: 284 patients self‐identified and were perceived as Black, 9 patients identified as Black but were not perceived as such, and 11 patients were perceived but did not identify as Black. Compared to non‐Black patients, Black patients were more likely to desire a full examination (86% vs. 66%, p < 0.001) and labs/tests (90% vs. 78%, p < 0.001) at their upcoming appointment (Table S2). Almost three‐quarters of patients (71%) reported overall good to excellent health, but 42% reported high severity of pain, and more than a quarter reported high severity of at least one mental health symptom. Black patients were more likely to report fair/poor health (35% vs. 25%, p = 0.005) and inability to be cheered up (19% vs. 11%, p < 0.003) compared to non‐Black patients; race differences for other symptoms were not statistically significant.

Black patients were significantly less likely than non‐Black patients to be offered a choice (15% vs. 23%; PD = −8.7%, CI: −14.4, −3.0), and women were less likely than men (16% vs. 27%; PD = −10.9%, CI: −17.4, −4.4) (Table 1). Usual travel time to cancer care was also associated with being offered a choice: relative to those with < 60 min travel, patients with 60–120 min (PD = 9.3%; CI: 1.1–17.5) and 120+ min (PD = 21.1%; CI: 10.7, 31.6) were more likely to be offered a choice. Patients wanting a full examination (16% vs. 30%, p < 0.001), wanting advice (12% vs. 21%, p = 0.09), reporting high severity anxiety (15% vs. 22%, p = 0.03), and high severity sadness (14% vs. 21%, p = 0.07) were less likely to be offered a choice regarding visit format (Table 2).

TABLE 1.

Bivariate associations of demographic characteristics with visit choice and scheduled visit type.

	Visit choice		(Among those offered a choice)
	Not offered a choice	Offered choice	Scheduled visit type
			In‐person	Virtual
	N = 581	N = 144	N = 69	N = 75
Age
< 65	273 (81%)	64 (19%)	31 (48%)	33 (52%)
65+	308 (79%)	80 (21%)	38 (48%)	42 (53%)
PD, % (95% CI)	ref	1.6 (−4.2, 7.4)	ref	0.9 (−15.5, 17.4)
Race
Non‐Black	316 (77%)	97 (23%)	32 (33%)	65 (67%)
Black	260 (85%)	45 (15%)	36 (80%)	9 (20%)
PD, % (95% CI)	ref	−8.7 (−14.4, −3.0)	ref	−47.0 (−62.0, −32.0)
Missing	5	2	1	1
Gender
Men	175 (73%)	65 (27%)	24 (37%)	41 (63%)
Women	405 (84%)	78 (16%)	45 (58%)	33 (42%)
PD, % (95% CI)	ref	−10.9 (−17.4, −4.4)	ref	−20.8 (−36.8, −4.7)
Missing	1	1	0	1
Urbanicity
Urban	486 (81%)	112 (19%)	56 (50%)	56 (50%)
Rural	81 (75%)	27 (25%)	12 (44%)	15 (56%)
PD, % (95% CI)	ref	6.3 (−2.5, 15.0)	ref	5.6 (−15.4, 26.5)
Missing	14	5	1	4
Usual travel to cancer care
< 60 min	430 (84%)	79 (16%)	39 (49%)	40 (51%)
60–120 min	94 (75%)	31 (25%)	16 (52%)	15 (48%)
PD, % (95% CI)	ref	9.3 (1.1, 17.5)	ref	−2.2 (−23.0, 18.5)
> 120 min	57 (63%)	33 (37%)	14 (42%)	19 (58%)
PD, % (95% CI)	ref	21.1 (10.7, 31.6)	ref	6.9 (−13.2, 27.1)
Missing	0	1	0	1

Abbreviation: PD (95% CI) = prevalence difference and 95% confidence interval.

TABLE 2.

Visit goals and symptoms according to visit choice and scheduled visit type (N = 725).

	Visit choice (N = 725)		p	Scheduled visit type among those offered a choice (N = 144)		p
	Not offered a choice	Offered choice		In‐person	Virtual
	N = 581	N = 144		N = 69	N = 75
Want full examination			< 0.001			< 0.001
No	127 (70.2%)	54 (29.8%)		9 (16.7%)	45 (83.3%)
Yes	452 (83.9%)	87 (16.1%)		60 (69.0%)	27 (31.0%)
Missing	2	3		0	3
Want prescription			0.16			0.06
No	361 (81.9%)	80 (18.1%)		33 (41.3%)	47 (58.8%)
Yes	211 (77.6%)	61 (22.4%)		35 (57.4%)	26 (42.6%)
Missing	9	3		1	2
Want advice			0.09			0.50
No	64 (87.7%)	9 (12.3%)		3 (33.3%)	6 (66.7%)
Yes	517 (79.3%)	135 (20.7%)		66 (48.9%)	69 (51.1%)
Want labs/tests			0.80			0.02
No	99 (81.1%)	23 (18.9%)		6 (26.1%)	17 (73.9%)
Yes	480 (80.1%)	119 (19.9%)		62 (52.1%)	57 (47.9%)
Missing	2	2		1	1
Want to discuss feelings			0.55			0.03
No	227 (81.4%)	52 (18.6%)		19 (36.5%)	33 (63.5%)
Yes	354 (79.6%)	91 (20.4%)		50 (54.9%)	41 (45.1%)
Missing	0	1		0	1
Self‐reported health			0.22			0.77
Good/excellent	404 (78.9%)	108 (21.1%)		51 (47.2%)	57 (52.8%)
Fair/poor	175 (82.9%)	36 (17.1%)		18 (50.0%)	18 (50.0%)
Missing	2	0		0	0
Pain			0.41			0.70
Low severity	340 (81.1%)	79 (18.9%)		39 (49.4%)	40 (50.6%)
High severity	240 (78.7%)	65 (21.3%)		30 (46.2%)	35 (53.8%)
Missing	1	0		0	0
Anxiety			0.03			0.65
Low severity	422 (78.3%)	117 (21.7%)		55 (47.0%)	62 (53.0%)
High severity	159 (85.5%)	27 (14.5%)		14 (51.9%)	13 (48.1%)
Cannot be cheered up			0.31			0.19
Low severity	491 (79.8%)	124 (20.2%)		56 (45.2%)	68 (54.8%)
High severity	85 (84.2%)	16 (15.8%)		10 (62.5%)	6 (37.5%)
Missing	5	4		3	1
Sadness			0.07			0.17
Low severity	455 (78.7%)	123 (21.3%)		56 (45.5%)	67 (54.5%)
High severity	124 (85.5%)	21 (14.5%)		13 (61.9%)	8 (38.1%)
Missing	2	0		0	0

Note: Patient‐reported outcomes (PROs) of pain, anxiety, ability to be cheered up, and sadness were assessed by frequency, severity, and intensity. For each PRO we calculated a composite severity score based on published methodology [28]. Patients with a composite score of 2 or 3 were classified as high severity, and those with scores of 0 or 1 were classified low severity.

Although our results did not reach significance in the adjusted models, at the patient level, insurance concerns showed the strongest associations (among barriers evaluated) with being offered a choice (Figure 1; Table S3). After adjustment, respondents who indicated their insurance did not cover telehealth visits were less likely to be offered a choice (10% vs. 24%; PD_adj = −9.4%, CI: −23.1, 4.4), as were respondents who believed their co‐payments would be unequal for telehealth and in‐person visits (19% vs. 25%; PD_adj = −7.4%, CI: −20.9, 6.1). On the other hand, connectivity/technology worries were not associated with choice (estimates were nonsignificant and close to null). At the census tract level, respondents living in high‐poverty tracts were less frequently (though nonsignificantly) offered a choice: 13% vs. 21%; PD_adj = −6.1%, CI: −13.3, 1.1.

FIGURE 1.

Impact of multilevel barriers on receipt of choice in visit format. Prevalence differences (PDs) and 95% confidence intervals (CIs) were estimated using linear‐risk regression. Models are presented unadjusted (blue) and adjusted (red) for patient‐reported pain (composite score 2/3 vs. 0/1), emotional stress (composite score 2/3 vs. 0/1), desire for in‐person services (including tests/labs or physical examination), and overall health status (average/poor vs. good/excellent). Sample sizes for the regression models are reported in the “N” column.

Among those who received a choice, race, and gender were strongly and significantly associated with scheduled visit type (Table 1). Black patients (20% vs. 67%; PD = −47.0, CI: −62.0, −32.0) and women (42% vs. 63%; PD = −20.8, CI: −36.8, −4.7) were less likely to accept a virtual visit than non‐Blacks and men. Age and usual travel to cancer care were not significantly associated with scheduled visit type. Multiple self‐reported care goals were associated with decreased frequency of accepting a virtual visit, including wanting a full examination (31% vs. 83%, p < 0.001), a prescription (43% vs. 59%, p = 0.06), labs or tests (48% vs. 74%, p = 0.02), and discussing feelings (45% vs. 63%. p = 0.03) (Table 2).

When assessing patient worries, we found that each connectivity worry was significantly associated with decreased frequency of scheduled telehealth visits in the unadjusted models (Figure 2; Table S4). Results were not statistically significant in the adjusted models, but we still observed potentially meaningful associations between a scheduled telehealth visit and internet worries (33% vs. 57%; PD_adj = −11.9%, CI: −32.9, 9.2), device worries (27% vs. 55%; PD_adj = −12.5%, CI: −37.3, 12.4), help/assistance worries (29% vs. 56%; PD_adj = −14.4%, CI: −35.9, 7.0), and privacy worries (29% vs. 56%; PD_adj = −13.6%, CI: −40.9, 13.7). Given that so few respondents with insurance worries were offered a choice, the only financial barrier assessed in relation to scheduled visit type was self‐reported income adequacy, which showed no association. Respondents living in census tracts with higher poverty and lower broadband/computer ownership were somewhat less likely to accept a telehealth visit; however, after restricting to those offered a choice, there were few patients living in disadvantaged communities, and so these estimates were not precise. For example, 5/12 (42%) patients from high‐poverty census tracts accepted a telehealth visit compared to 66/127 (52%) from all other tracts: unadjusted PD (95% CI) = −10.1% (−39.3, 19.1); PD_adj (95% CI) = −3.7% (−31.9, 24.5). Similarly, patients living in census tracts with low broadband ownership had fewer accepted telehealth visits: 18/41 (44%) versus 53/98 (54%); unadjusted PD (95% CI) = −10.0% (−28.1, 8.1) and PD_adj (95% CI) = −1.2% (−18.9, 16.4).

FIGURE 2.

Impact of multilevel barriers on scheduled visit type, given a choice was offered. Prevalence differences (PDs) and 95% confidence intervals (CIs) were estimated using linear‐risk regression. Models are unadjusted (blue) and adjusted (red) for patient‐reported pain (composite score 2/3 vs. 0/1), emotional stress (composite score 2/3 vs. 0/1), desire for in‐person services (including tests/labs or physical examination), and overall health status (average/poor vs. good/excellent). Sample sizes for the regression models are reported in the “N” column.

For each model, we performed sensitivity analyses using inverse probability of sampling weights. These were determined based on strata‐specific response rates: 11% for Black adults with a telehealth visit, 16% for non‐Black adults with a telehealth visit, 14% for Black adults with an in‐person visit, and 16% for non‐Black adults with an in‐person visit. Associations for both outcomes were very similar in the weighted versus unweighted analyses, and statistical significance was unchanged (Tables S5 and S6).

4. Discussion

In a racially diverse cohort of adults undergoing treatment for cancer, we examined factors that may affect decisions regarding who gets offered a choice of visit format and a patient's decision to opt for a telehealth visit in oncology care. Black adults and women were less likely to be offered a choice, and if offered a choice, they were less likely to have a telehealth visit scheduled. We evaluated multilevel potential barriers to telehealth visits and found different barriers salient for each outcome. Patients with concerns about their insurance covering telehealth visits were less likely to be offered a choice, whereas patients with device and connectivity worries or privacy concerns were less likely to accept a telehealth visit when offered the choice. However, these PDs were only statistically significant in the unadjusted models. When examining area‐level factors, we also found that patients from higher poverty and lower computer ownership census tracts were somewhat less likely to be offered a choice and to accept a telehealth visit, but these associations were imprecise and none reached statistical significance in adjusted models. Low broadband census tracts were (nonsignificantly) associated with decreased scheduling of telehealth in the unadjusted model only.

A key finding of this study was that different barriers were associated with being offered a choice of visit format and accepting a telehealth visit when offered the choice. Shorter distances to care and insurance concerns were associated with decreased frequency of being offered a choice. Home address and patient insurance are readily knowable by health care organizations; hence, perhaps, why these factors inform whether patients were identified as candidates for telehealth visits and thus offered a visit format choice. However, health care organizations routinely lack information regarding other factors that determine their patients' access to virtual care, including social determinants of health (particularly broadband connectivity) or care preferences, such as the desire for a full examination or to discuss one's feelings [36, 37, 38]. Given our finding that technology concerns (e.g., not having reliable internet) were inversely associated with accepting a telehealth visit, more robust offerings of assistance to patients who might be lacking resources to make use of telehealth visits should be undertaken.

Importantly, we evaluated both patient‐ and area‐level factors. Prior literature identified variation in telehealth use according to area‐level poverty and other indices of social deprivation, suggesting there are structural reasons for disparate access to telehealth [39, 40]. Our study extends this literature in the cancer care context, finding that higher poverty and lower rates of computer ownership were nonsignificantly associated with decreased probability of both being offered a choice and accepting a telehealth visit, while area‐level broadband was (nonsignificantly, and only in the unadjusted model) associated with accepting a telehealth visit. Interpretation of these associations is limited by poor precision, as relatively few patients in our study lived in disadvantaged census tracts. Still, some of these PDs were borderline with respect to statistical significance and/or the magnitude was large, which provides some evidence of a meaningful association. Furthermore, such associations are consistent with a well‐documented ‘digital divide’ in the United States, characterized by community‐level differences in broadband accessibility and infrastructure [41]. A related concept is ‘digital redlining’ due to unequal investments in technology infrastructure that disadvantages Black Americans and other racial/ethnic minority groups [42, 43]. Such practices have particular implications for the ability to use video visits (where high‐speed internet is particularly important), and are likely one reason, along with cost and digital literacy, why some prior studies found that Black and Hispanic adults are more likely to use voice‐only visits relative to White adults [44, 45]. Thus, while patient‐level interventions (e.g., digital literacy) are important, macro‐level investments are needed to ensure access to telehealth visits across diverse geographic areas, as some analyses have found broadband access to be problematic in urban as well as rural areas [46, 47].

Differences in telehealth utilization across populations (e.g., by race or gender) are not inherently unjust because preferences for visit type could reflect differences that are not disparities. For example, variation in cultural norms and preferences has been suggested as one (albeit not exclusive) explanation for racial differences in telehealth use [18, 48]. In a separate analysis of this same population, our study team found that perceptions of telehealth were less favorable among Black adults [49]. However, while some patients may desire in‐person services, we caution that self‐reported attitudes toward telehealth can also reflect inequities in access. For example, a qualitative study found that Black adults expressed privacy and trust‐related concerns with telehealth [50]. If such concerns (and therefore preferences for in‐person visits) are attributable to mistrust in the healthcare system or crowded housing, for example, then such differences would represent a disparity. Notably, we found Black adults and women were less likely to have a scheduled telehealth visit and less likely to be offered a choice. As such, ensuring these patients are given a choice regarding visit format as well as supported in their ability to comfortably participate in a telehealth visit should be a health equity priority.

Our study also had several limitations. First, whether someone was offered a choice could be prone to recall error; however, misclassification was likely mitigated by allowing respondents to respond as “unsure” (a relatively small proportion of the sample which we elected to exclude from consideration here). Similarly, knowledge of whether one's insurance covers telehealth visits (a barrier) may similarly be subject to misclassification. Prior studies have shown fairly high sensitivity (> 80%) in recall of insurance status and plan type but we would expect lower awareness of more specific coverage details [51, 52, 53]. Responses to this question, however, are likely reflective of one's typical experience with using their health insurance: someone with previous difficulties (denied reimbursement, high co‐pays, etc.) may be skeptical that a telehealth visit would be covered. The significant association of this variable with being offered a choice is also likely indicative of the accuracy of respondents' perceptions of their insurance, or at least that the patients' insurance beliefs are shared by the provider and/or other staff. Nevertheless, the omission of EHR‐determined insurance status is a limitation of this work and is an important consideration for future studies. Second, the scheduled visit type outcome was evaluated among only those who were offered a choice (about 20% of the total sample) limiting the precision of associated estimates. Third, despite prioritizing recruitment of Black adults with telehealth visits, there were only five Black respondents with scheduled telehealth visits in our study, reflective of both their reduced likelihood of being offered a choice regarding visit format as well as less trust and willingness to participate in health research [54]. Also, women (67%) and those with higher educational attainment (49% with a 4‐year college degree or higher) were overrepresented in our study, consistent with documented variations in survey participation [55, 56]. Fourth, this study achieved a low response rate overall (15%), an increasingly common challenge with telephone‐based health surveys and may limit generalizability [57]. It may also be important that in health services research, response rate has been shown to correlate with greater patient satisfaction [58]. Thus, although we found a substantial prevalence of barriers and concerns with telehealth visits, the true burden might be greater among nonrespondents. To quantify potential nonresponse bias, we conducted sensitivity analyses using inverse probability of sampling weights. Weighted models based on strata‐specific response rates (for race and scheduled visit type) were similar to the main models, suggesting that estimates were not biased by these two factors. However, available data on nonrespondents was limited, and nonresponse bias due to other factors cannot be ruled out. Finally, the appointment scheduling process likely varied across the study population. For example, appointments may have been scheduled over the phone, electronically, or while checking out of the prior appointment. Since we did not collect detailed information on how appointments were scheduled, we were unable to assess the potential impact on receipt of choice or scheduling decisions.

In conclusion, our results indicate that inequities in access to telehealth stem from both who is offered the choice and who has the resources to feel comfortable using telehealth visits—two decision junctures along the multi‐step process that underlies the use of telehealth visits. We found that individual‐level technology and insurance barriers, as well as residence in disadvantaged communities (albeit with less statistical certainty) were associated with less choice and fewer decisions to schedule a telehealth visit when offered the choice. It is not surprising that the causes of inequities in telehealth utilization are multifaceted or that appropriate policy solutions will need to target a complex suite of patient‐, health system‐, and environmental‐level barriers to care. But such complexity should not prevent oncology practices and others that are rapidly sustaining telehealth services either from proactively providing digital support and exploring other innovative solutions to ensure equity in the use of telehealth visits.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1. Characteristics of the study population (N = 725).

Table S2. Patient‐reported visit goals and symptoms for non‐Black and Black adults (N = 718).

Table S3. Patient‐ and area‐level barriers in relation to visit choice.

Table S4. Patient‐ and area‐level barriers in relation to scheduled visit type.

Table S5. Sensitivity analyses for receipt of choice models using inverse probability of selection weights.

Table S6. Sensitivity analyses for scheduled visit type models using inverse probability of selection weights.

Dunn M. R., Fridman I., Kinlaw A. C., Neslund‐Dudas C., Tam S., and Elston Lafata J., “Identifying Barriers to Being Offered and Accepting a Telehealth Visit for Cancer Care: Unpacking the Multi‐Levels of Documented Racial Disparities in Telehealth Use,” Health Services Research 61, no. 2 (2026): e14461, 10.1111/1475-6773.14461.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy restrictions.