A Smartphone Is Not Enough: Telehealth Attendance and the Digital Divide

ABSTRACT

Background

Telehealth has emerged as a powerful tool for managing chronic diseases and mental health conditions, offering increased access to care and improved patient outcomes. However, inequities in digital connectivity and technological resources have created significant disparities in access to these potentially life‐changing services, disproportionately impacting marginalized and minoritized communities across the globe.

Methods

Data on 473,716 telehealth encounters occurring between January 1, 2022, and June 30, 2023 were retrieved from the electronic health records (EHR) system used by University Hospitals. These encounters were classified into three groups: attended, canceled, and no‐show. Relative risk was calculated based on age, sex, and race, and a multivariate linear regression was performed with age, sex, and race as inputs, to determine their effect on the encounter outcome.

Results

Our analysis identified significant differences in relative risk between demographic groups. Patients 20–39 years of age had a high relative risk of cancellation and no‐show, and Black patients demonstrated the highest relative risk for cancellation and no‐show. The regression analysis illustrated a statistically significant link between no‐shows and patients with a cellular plan with no other internet subscription (p < 0.001), smartphone ownership (p < 0.001), and not having a computer (p < 0.05).

Conclusions

This study highlights the clinical repercussions of the digital divide, as patients relying on a mobile phone and data plan to attend telehealth visits were more likely to no‐show. Current disparities in digital connectivity for historically marginalized populations heightens the risk of creating a digital underclass. There is evidence this study may be applicable in multiple countries across the world. Further research on the causes of the observed no‐shows is necessary to ensure equitable delivery of digital healthcare services.

Disparities in the quality of digital connectivity and internet connected devices have created access problems with telehealth visits. Our study identified significant differences in relative risk of missed visits between demographic groups, including age and race. Data also revealed a statistically significant link between missed visits and patients with a smartphone cellular plan with no other internet subscription (p < 0.01).

1. Introduction

The COVID‐19 pandemic brought unprecedented challenges to the healthcare industry, including the rapid adoption of telehealth services to ensure continuity of care while minimizing the risk of disease transmission [1]. This impact was felt across the world, as telehealth expanded in across the globe, from Europe to China, as society moved to a new normal with treatment of both acute and chronic conditions via remote care [2, 3]. The efficacy of telehealth for managing various health conditions, such as chronic disease and mental health, has been well‐documented [1, 4]. This has been seen across many different cultures and societies [1, 2, 3]. However, in the United States in particular, inequitable availability of digital connectivity and technological tools has created disparities in access to telehealth services, with marginalized and minoritized communities being disproportionately impacted [4]. This gap between those who have access to technology and those who do not has become known as the “digital divide” [5].

In Cleveland, Ohio, United States, the digital divide is a significant and longstanding issue, with upwards of 30% of residents lacking basic internet connectivity or reliable service, the latter of which has been driven by various factors, including historical (and more recent) “digital redlining” practices [6]. “Digital redlining” is the practice of increasing internet speeds in areas where telecommunications companies have more paying customers [6]. According to a 2020 study by Case Western Reserve University, more than 70% of Cleveland residents living in poverty lacked broadband internet connection in their home [7]. Moreover, African Americans in Cleveland are six times more likely than white residents to lack a broadband connection [8]. This lack of connectivity is a direct barrier to accessing telehealth services, which require reliable broadband internet connectivity, and in part reflect the impact of digital redlining.

The practice of digital redlining creates a cyclical feedback loop wherein higher paying customers get better internet speeds, while many impoverished and minoritized neighborhoods are progressively left behind. In turn, many low‐income households without access to high‐speed internet instead subsist on a mobile phone data plan [9]. Variations in both broadband access and the breadth of mobile data plans for low‐income, black, and other historically marginalized populations has the potential to limit access to telehealth opportunities [10].

In line with the aforementioned considerations, University Hospitals caregivers have anecdotally noted that patients without a broadband internet subscription and/or limited cell phone data plans are less likely to successfully participate in telehealth visits, oftentimes resulting in nonattendance and/or “no‐show” events [11]. To better understand and determine if there is a relationship between the type of internet connection patients have access to, and the patients' ability to attend telehealth visits, University Hospitals subsequently compiled and analyzed organizational telehealth visit data. This data included missed appointments, canceled appointments, and attended appointments across the enterprise. Said data were then compared according to various demographic factors and the type of internet connection.

The primary aim of this paper is to explore the impact of digital inequity and access on patients' abilities to attend telehealth visits, with a particular focus on Cuyahoga County, including the metropolitan area of Cleveland, Ohio. In particular, reviewing the type of internet connection, including cellular versus other types, and how this impacts the ability to connect to telehealth visits. For the purposes of this paper, “telehealth” is defined per billing criteria at the center for Medicaid and Medicare Service as a real time, two‐way audio‐video communication between a remote provider and a patient [12]. “Digital inequity” also known as the digital divide, per the National Digital Inclusion Alliance refers to the disparities in access to and use of digital technologies and the internet among different groups, leading to unequal opportunities [13]. The applicability of the findings around digital inequity within the region of Cleveland, Ohio will also be further discussed in relation to other areas of the world.

2. Methods

The data for this study were extracted from the electronic health records (EHR) system employed by University Hospitals and comprised a total of 473,716 telehealth encounters that occurred between January 1, 2022, and June 30, 2023. These telehealth encounters were systematically classified into three distinct categories based on their final disposition: attended, canceled, and no‐show.

The attended category (n = 379,955) included appointments where the patient participated in the telehealth session at the scheduled time, and the healthcare provider successfully conducted the consultation. In contrast, the canceled category (n = 53,519) included telehealth appointments where the patient proactively canceled or rescheduled the session at least 24 h before the scheduled time. Finally, the no‐show category (n = 40,242) encompassed appointments where the patient neither attended nor provided adequate notice of cancellation, defined as failing to inform the healthcare provider or canceling within 24 h of the appointment time. Encounters falling into the no‐show category were identified using designated codes and flags within the EHR system, which were applied to distinguish between genuine cancellations and unanticipated no‐shows.

Relative risk was calculated based on age, sex, and race, with a 95% confidence interval (CI) established a priori. There were, however, a total of 78 people whose sex was not recorded in the EHR. These individuals were subsequently excluded from the relative risk calculation, as their sex could not be determined, resulting in a final sample of 473,638 individuals.

Following the relative risk calculation, a multivariate linear regression was performed with all the included variables (age, sex, and race) as inputs, to further isolate the effects of each variable on the outcome.

To ascertain the level of internet access available in the neighborhoods the patients resided in, the coordinates of each patient residence were geocoded using ESRI ArcGIS Pro 10.1 (ArcGIS Pro GIS software/Esri, Redlands, California, USA) to the street level or above. The census tract Federal Information Processing Series (FIPS) identifier was added to each patient record. Patients were initially organized by FIPS identifier, in conjunction with encounter type, sex, age category, and race. These variables were then grouped by their encounter type. Finally, the associated rate for each encounter type was calculated by taking the count of encounters for the encounter type and dividing it by the total number of encounters with the same FIPS identifier, and the resulting ratio was multiplied by 100.

This information was then joined to data collected by the American Community Survey, according to computer type and internet subscriptions. The resulting level of granularity allowed for a targeted analysis without compromising patient privacy. Variables included from the American Community Survey can be found in publication number S2801, and included computer types, subscription types, and subscription types by income level [14]. This data had a number of overlapping categories (e.g., cellular data plan and cellular data plan with no other type of internet subscription); a subset of variables was selected to avoid this overlap. A percentage value was used for each of the remaining variables, and Python 3.10.9 was used to join this data [15].

For each encounter type, the data were split into independent variables and the dependent variable, the latter of which represented the rate of encounters for that encounter type. The Python package scikit‐learn 1.2.1 was used to normalize the independent variables [16], and the package statsmodels 0.13.5 was used to run an ordinary least squares regression after adding a constant to the data [17].

3. Results

3.1. Demographics According to Virtual Visit Outcome

Demographic information pertaining to those patients accessing telehealth services during the observational study period are outlined in Table 1. Similar descriptive terms relative to attending, canceling, and no‐showing a telehealth visit are presented in Tables 2, 3, 4, respectively.

Table 1.

Demographic Information of patients accessing telehealth services (n, %).

Age (years)	Attended	Canceled	No‐show	Total number (percentage)
0–19	46,986 (10)	7311 (2)	4990 (1)	59,287 (13)
20–39	94,322 (20)	17,402 (4)	11,917 (3)	123,641 (26)
40–59	109,166 (23)	15,068 (3)	11,482 (2)	135,716 (29)
60+	129,481 (27)	13,738 (3)	11,853 (3)	155,072 (33)
Race
Asian	2109 (< 1)	364 (< 1)	246 (< 1)	2719 (1)
Black	56,157 (12)	10,492 (2)	7809 (2)	74,458 (16)
Other	51,843 (11)	8488 (2)	6288 (1)	66,619 (14)
White	269,846 (57)	34,175 (7)	25,899 (5)	329,920 (70)
Sex
Female	243,834 (51)	35,402 (7)	26,263 (6)	305,499 (64)
Male	136,083 (29)	18,108 (4)	13,948 (3)	168,139 (35)
Other/Unknown	38 (< 1)	9 (< 1)	31 (< 1)	78 (< 1)

Table 2.

Demographics of patients who attended a virtual visit.

Age (years)	Race (number of patients)
Male	Black	White	Asian	Other	Total
0–19	3234	15,340	156	4923	23,653
20–39	3236	19,737	206	4710	27,889
40–59	4155	24,984	205	4338	33,682
60+	4636	40,748	183	5156	50,723
Total	15,261	100,809	750	19,127	135,947
Female
0–19	3105	15,309	152	4752	23,318
20–39	10,700	44,133	584	10,993	66,410
40–59	15,524	50,169	406	9250	75,349
60+	11,565	59,411	302	7479	78,757
Total	40,894	169,022	1444	32,474	243,834
Unknown
0–19	0	5	0	6	11
20–39	2	7	0	13	22
40–59	0	3	0	1	4
60+	0	0	0	1	1
Total	2	15	0	21	38

Table 3.

Demographics of patients who canceled a virtual visit.

Age (years)	Race (number of patients)
Male	Black	White	Asian	Other	Total
0–19	823	2010	22	742	3597
20–39	751	3107	50	948	4856
40–59	658	2857	39	641	4195
60+	614	4154	19	673	5460
Total	2846	12,128	130	3004	18,108
Female
0–19	754	2131	31	797	3713
20–39	2847	7320	121	2252	12,540
40–59	2630	6680	57	1505	10,872
60+	1415	5911	36	915	8277
Total	7646	22,042	245	5469	35,402
Unknown
0–19	0	1	0	0	1
20–39	0	3	0	3	6
40–59	0	1	0	0	1
60+	0	0	0	1	1
Total	0	5	0	4	9

Table 4.

Demographics of patients who scheduled a virtual visit and did not attend or cancel.

Age (years)	Race (number of patients)
Male	Black	White	Asian	Other	Total
0–19	439	1461	18	501	2419
20–39	524	2152	23	635	3334
40–59	596	2404	23	446	3469
60+	587	3552	27	560	4726
Total	2146	9569	91	2142	13,948
Female
0–19	0	1503	0	85	1588
20–39	1867	5055	69	1590	8581
40–59	2018	4770	40	1184	8012
60+	1325	5002	29	771	7127
Total	5210	16,330	138	3630	25,308
Unknown
0–19	0	0	0	28	28
20–39	0	0	0	2	2
40–59	0	0	0	1	1
60+	0	0	0	0	0
Total	0	0	0	31	31

3.2. Relative Risk

As illustrated in Figures 1, 2, 3, our analysis revealed significant differences in relative risk of telehealth attendance, cancellations, and no‐shows across different demographic groups. The measurement of inequity is the difference in relative risk of no‐showing a telehealth visit based on different demographic factors, including age, race, and gender.

Figure 1.

Relative risk of attending a telehealth visit by age, race, and sex.

Figure 2.

Relative risk of canceling a telehealth visit by age, race, and sex.

Figure 3.

Relative risk of no showing a telehealth visit by age, race, and sex.

Patients 60 years of age and older were more likely to attend their scheduled telehealth visit (relative risk: 1.06, 95% CI:1.02–1.11). Conversely, patients aged 20–39 years exhibited a markedly higher likelihood of both canceling and failing to show up for their appointments. Their relative risk of cancellation was elevated at 1.36 (95% CI: 1.26–1.47), and their risk of no‐show was also significantly higher, with a relative risk of 1.19 (95% CI: 1.09–1.30).

Racial disparities were also evident in telehealth engagement patterns. Black patients were found to have the highest relative risk for cancellations (relative risk: 1.44, 95% CI:1.30–1.59) and no‐shows (relative risk: 1.37, 95% CI: 1.25–1.50), while their White peers were most likely to attend (relative risk: 1.03, 95% CI: 1.00–1.06).

Gender differences in telehealth engagement were more modest but still noteworthy. Males demonstrated slightly higher attendance (relative risk: 1.01, 95% CI: 0.99–1.03) as compared to females, and an overall lower risk of cancellation (relative risk: 0.94, 95% CI: 0.89–0.99) and no‐show (relative risk: 0.93, 95% CI: 0.89–0.97).

3.3. Regression Analysis

The regression analysis was conducted to examine the factors influencing attendance, cancellation, and no‐show rates. Key findings of the regression analysis are outlined below.

(1) Attendance: The regression model illustrated a statistically significant relationship with attendance, with the R‐squared value of 0.382, indicating that approximately 38.2% of the variance in attendance was explained by the independent variables. The analysis revealed several significant predictors of attendance, including age groups with individuals in the younger age brackets (e.g., 18–24, 25–34) were more likely to attend their telehealth appointments compared to older age groups. (2) Cancellation: The regression model demonstrated a moderate explanatory power, with the R‐squared value of 0.253, indicating that approximately 25.3% of the variance in cancellations was explained by the independent variables, such as age, gender, race, and socioeconomic factors. While this R‐squared value suggests that the model captures some of the key drivers behind cancellation behavior, it also highlights that a significant portion of the variance remains unexplained by the included predictors. (3) No‐show: The regression model showed a moderate level of explanatory power, with the R‐squared value of 0.246, indicating that approximately 24.6% of the variance in no‐show rates was explained by the independent variables mentioned previously. While the model captures a portion of the factors influencing no‐show behavior, a substantial amount of variance remains unexplained, suggesting that the variables included in the model may not fully capture the primary drivers of this behavior.

Overall, the regression analysis illustrates a statistically significant link between no‐shows and those patients with a cellular plan with no other type of internet subscription (p < 0.001), those who retain a smartphone (p < 0.001), and those who lack a computer (p < 0.05). Beyond these three areas, there were no additional significant associations between the array of predictors for attendance, cancellation, and no‐show rates, within the specific context studied. Further research and consideration of additional variables may be necessary to gain a deeper understanding of these dynamics, and to develop more accurate predictive models.

These results indicate that other factors may be responsible for the differentiation in telehealth appointment outcomes. Internet access may be a better predictor for attendance of telehealth visits. Moreover, relying on a smartphone, and not having access to a computer, was associated with telehealth visit no‐shows.

Finally, all of the included patients were geocoded and aggregated by census tract. The patterns of internet use available from the American Community Survey was joined to these census tracts to elicit any correlations that may explain the attendance patterns, as was the Centers for Disease Control and Prevention's Social Vulnerability Index. There was no significant relationship between the Social Vulnerability Index score of a patient's census tract of residence, and the outcome of their telehealth visit.

4. Discussion

The results of this study support a clear link between the patient's type of internet connection, and their subsequent ability to attend a telehealth visit. Compared to prior studies that mostly focus on broadband showing a positive correlation to attendance of visits, our study shows an inverse correlation for patients with only a cellular data plan [18, 19]. Those patients that relied solely on a cellular phone or smartphone to engage in a telehealth visit, and did not have access to a traditional laptop, desktop computer, and/or internet subscription, were more likely to no‐show for a scheduled telehealth appointment. We believe this is an important finding, as cell phone ownership and data plans are near ubiquitous, while access to laptops and other internet‐connected devices along with a broadband internet subscription are not. This disparity risks creating a worsening digital divide with access segregation along socioeconomic and racial groups [20, 21, 22, 23, 24, 25]. Moreover, our identification of higher relative risks for no‐show or cancellation of telehealth visits in black populations raises questions as to if this is due to the known higher rates of cell phone data dependency is this historically marginalized community [20].

This access characteristic may be a proxy for broadband availability and be further compounded by the constraints of the respective patient's cellular phone plan (data usage costs and/or limits, cellular speed, gaps in service, etc.). Additionally, it aligns with previous findings from Broffman and colleagues relative to high mobile phone use in areas of limited broadband access [20]. Broffman et al. also identified that patients from marginalized communities were twice as likely to access a telehealth visit from their mobile phone [20]. As demonstrated here, Black patients across all age groups were at the highest risk of no‐show with respect to race, while those 20–39 years of age across all races demonstrated the greatest relative risk for no‐show (Figure 3).

Where these two characteristics converge presents a potentially synergistic disparity in healthcare utilization patterns and socioeconomic resources as younger adults, those with lower incomes, and those with a high school education or less have the highest smartphone dependency; irrespective of age, 17% of Black Americans are considered smartphone‐dependent, as compared to 12% of their White peers [21]. Overall cellphone ownership is also highest among Black Americans [21], and it is unclear if this ownership is directly related to living in areas with physical or financial barriers to broadband internet.

Napoli and Obar in their research have documented concerns related to the development of a marginalized group of internet users [22]. They point out that access to the internet solely by way of a smartphone represents a new “underclass” of internet users, as smartphone‐limited internet is characterized by less content availability, restricted platform and network connectivity, and has less speed, memory, and functionality when compared to accessing the same content from a desktop or laptop computer [22].

In their critique of mobile‐only access, Pearce and Rice also discuss the multitude of functions beyond healthcare visits that are not as easily performed from a smartphone compared to a computer [26]. Additionally, Nationally, Black Americans are disproportionately impacted in terms of broadband access in both rural [24] and urban areas [16], a reality that may in part reinforce the usage of a cellular phone or smartphone.

The risk of this developing digital divide is that our society's access to health care may widen, and subsequently worsen current disparities in health outcomes. As it is, health inequity is seen in worse health outcomes for historically marginalized minority populations [27, 28, 29]. This is exemplified in consistently higher rates of obesity, cancer, and premature heart disease for African American populations [30, 31, 32]. And, while there are a multitude of historical, social, political, and community‐related reasons, access to health care is one that has been highlighted as a contributor to health inequities [33]. Given the recent explosion of digital connectivity, we must be vigilant in the years to come in how well patients and providers connect via digital means. Simply having a smartphone and a data plan is not enough. The results of this study highlight the need for further in‐depth evaluation and subsequent interventions to bridge the digital divide. This has already started in many ways. There has been recent investment by the federal government in funding for historically marginalized populations to procure an internet connected device or subsidized broadband internet [34]. As well, there have been efforts to promote utilization of alternative forms of digital health that may reduce the impact of digital inequity such as asynchronous technologies and portals [35]. Better screening has been a recent focus of research as there is a need for a validated way to identify those most at risk for disrupted digital connectivity and illiteracy [36, 37, 38].

It is important to note that these findings around digital inequity are not limited to the USA. In particular, the rural‐urban divide has shown to exacerbate health outcomes in part due to lack of access to digital tools in Europe, Asia and Africa [39, 40, 41]. In 2022, the WHO performed a scoping review of European countries adoption of digital health technology and the use for improving the health and wellness of their populations [39]. The study explored the range of inequities in digital health tools across a comprehensive range of demographic characteristics. They found digital health technologies were much more widely in urban areas, and less by rural populations along with people identifying as an “ethnic minority” and those facing language barriers [39]. This study also found more frequent use of digital health tools by people with higher education levels and socioeconomic status [39]. Younger people were also found to use tools more than older adults.

As a result of these findings, the WHO developed a document that describes steps to intervene on these worsening disparities, titled Regional digital health action plan for the WHO European Region 2023–2030 (RC72) [42]. The plan calls for developing patient centered approaches to improving the distribution of digital technologies within populations. This includes discovery of a common framework to monitor engagement and comprehension of technology, identifying inequities in digital infrastructure, and interventions that improve access for those with disabilities or language barriers.

At University Hospitals we have identified practical ways to improve the digital divide in developing pilots for screening for patients with limited connectivity (no broadband) or lacking an internet connected device beyond a smartphone. We have started intervening by collaborating with like minded organizations that can provide free laptops along with broadband connections and even digital literacy training. The hope is we can start developing models that show the benefits of sustained improvement in digital connectivity within marginalized communities.

There are several limitations to our data and this study. One potential limitation of this analysis is the lack of information relative to insurance type. Health insurance coverage is a major force relative to patient attendance for any healthcare visit, and telehealth is no different. Additionally, our data analysis indicated potential issues related to multicollinearity and singularity, which should be further investigated. Addressing these concerns is crucial to ensure the accuracy and reliability of the estimated coefficients. Future research could explore the role of other factors, such as patient demographics, geographic location, and healthcare provider characteristics, in telehealth adoption

The study revealed significant deviations from normality and kurtosis in the distribution of residuals, as evidenced by the Omnibus and Jarque‐Bera tests. These results imply potential limitations in the model's assumptions and warrant caution when interpreting the findings. Still, our findings warrant further investigation to determine the effect that the type of internet connectivity has on access to telehealth visits.

5. Conclusion

The type and quality of a patient's digital connection to their healthcare provider is a critical factor when engaging in a telehealth appointment. Current disparities in digital connectivity for historically marginalized populations such as African‐Americans heighten the risk of creating a digital underclass. This study provides evidence of clinical repercussions of this digital divide, as patients that relied solely on a mobile phone and data plan to attend telehealth visits were more likely to no‐show for the scheduled telehealth appointment. We believe this study can be extrapolated to other areas of the world where digital inequities have been shown to exist in similar dimensions. In the end, we suggest that owning a smartphone is not enough to ensure successful participation in telehealth services. Further research on the causes of these no‐shows and interventions to address them will be needed to stem the continued evolution of access disparities in the digital delivery of healthcare services.

Author Contributions

James Labadorf: data curation (equal), formal analysis (lead), investigation (equal), methodology (equal), software (equal), writing–review and editing (equal). Matthew Nichols: conceptualization (equal), data curation (equal), formal analysis (supporting), writing –original draft (equal), writing–review and editing (equal). Tayana Williams: conceptualization (equal), project administration (equal), validation (equal), writing–original draft (equal), writing–review and editing (equal). Celina Cunanan: supervision (equal), validation (equal), writing–review and editing (supporting). Brian D'Anza: conceptualization (lead), data curation (equal), project administration (equal), supervision (lead), writing–original draft (equal), writing–review and editing (equal).

Ethics Statement

This study was evaluated for by the Institutional Review Board (IRB) at University Hospitals for any ethical conflicts and none were found. It was deemed exempt from formal IRB review as it did not include personally identifiable health data.

Consent

The Institutional Review Board (IRB) determined this study was exempt from the need for consent as personally identifiable patient level data was not used.

Conflicts of Interest

Matthew Nichols is a staff member of Four Springs Health LLC. All authors declare no conflicts of interest.

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 or ethical restrictions. Data is available upon request from the authors.