An exploratory study to understand how rurality status and demographic characteristics are associated with enrollment, engagement, and retention in a digital health intervention targeting the Appalachian region

Abstract

Purpose

Digital health studies exploring group disparities across research phases are limited. As a secondary aim of a larger digital health trial, this study explored how rurality and other sociodemographics were associated with enrollment, retention, and engagement in a randomized controlled sugar‐sweetened beverage (SSB) reduction trial.

Methods

Participants from a primarily Appalachian sample were randomized into iSIPsmarter (experimental) or static Patient Education (control) websites. Enrollment, retention (6 months), and iSIPsmarter engagement (completion of metered program Core content and SSB and weight diaries) were collected from July 2021 to August 2023. Regression models assessed subgroup associations using Rural Urban Continuum Codes (RUCC), sex, race, age, income, education, and other sociodemographic predictors.

Findings

Of the 509 eligible participants, 249 (49%) enrolled, and 218 (88%) were retained. Participants were predominantly White (89%), college‐educated (59%) females (83%) with household incomes <$55,000/year (52%). Rurality varied: RUCC 1‐2 (medium‐large metro) = 15%, RUCC 3 (small metro) = 45%, and RUCC 4‐9 (nonmetro) = 41%. On average, iSIPsmarter participants (n = 127) completed 4.89/6 (SD = 1.69) Cores and 76% (SD = 29%) and 57% (SD = 31%) of SSB and weight diaries. Rurality was a nonsignificant predictor, but higher education and health literacy increased enrollment likelihood by 37% (95% CI = 1.12‐1.67) and 23% (95% CI = 1.03‐1.47), respectively. Greater education (OR = 1.51, 95% CI = 1.00‐2.29), age (OR = 1.04, 95% CI = 1.01‐1.07), and income (OR = 1.13, 95% CI = 1.00‐1.28) significantly predicted retention. Older age significantly (P<.05) predicted the completion of Cores and diaries.

Conclusions

Results suggested rurality was not significantly associated with enrollment, retention, or engagement, though this conclusion warrants caution. Future digital health studies targeting similar populations should consider additional sociodemographic differences.

INTRODUCTION

Digital behavioral health interventions offer the potential for high‐reach, low‐cost, scalable care to improve healthy lifestyles, prevent or reduce behavioral risk factors, and support the treatment and monitoring of chronic diseases. These interventions may be delivered via mobile or web‐based platforms and can incorporate automated behavioral self‐monitoring strategies and health‐monitoring devices. Evidence shows promising outcomes for promoting healthy behaviors (eg, healthy eating and physical activity, ¹ , ² smoking cessation ³ ), improving chronic disease care (eg, diabetes, ⁴ cardiovascular disease ⁵ ), and treating obesity and related health conditions. ² , ⁶ , ⁷

Digital health has the potential to enhance access to care among underserved populations and contribute to the reduction of health disparities. However, studies have documented persistent subgroup differences in engagement and outcomes across a variety of social determinants of health indicators, such as lower socioeconomic status, older age, lower educational attainment, racial and ethnic minorities, higher rates of comorbid conditions, and those living in rural communities. ⁸ , ⁹ , ¹⁰ Rural populations may be particularly vulnerable to disparities in access to and engagement with digital health interventions due to the overlapping of many of these social determinants of health. ⁹ , ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵

Key economic, environmental, and social factors that contribute to digital health disparities in rural regions include limited access to digital devices and reliable broadband; inequitable distribution of digital health technologies; lower literacy and health literacy; lack of technology skills and support; misaligned digital design; and cultural resistance to technology use. ⁹ , ¹⁰ , ¹⁶ , ¹⁷ , ¹⁸ Compared to urban residents, rural populations face lag in ownership of smartphones (88% vs 99% for suburban and 97% for urban), tablets (44% vs 54% suburban and 55% urban) and computers (72% vs 78% suburban and 80% urban), ¹² , ¹⁹ and are less likely to have reliable home broadband (73% vs 86% suburban and 77% urban) or own multiple connected devices (30% vs 43% suburban and 44% urban). Concerns that digital health may inadvertently widen health inequities for rural and other vulnerable poplations ¹⁷ have resulted in strategies to promote inclusivity such as codesigning interventions and partnering with community organizations to promote and implement studies. ²⁰ However, systematic inclusion of sociodemographic analyses across enrollment, engagement, retention, and outcomes of digital health trials are necessary to determine if disparities persist regardless of strategies. ²¹ , ²² , ²³

Unfortunately, digital health intervention studies examining sociodemographic differences across all of the research phases (ie, enrollment, follow‐up retention, and program engagement) are limited. ¹⁰ , ¹⁷ , ²³ , ²⁴ , ²⁵ To address the complexities of digital health disparities in rural populations, it is essential they are included in this research. Accordingly, this study addresses a secondary aim of a randomized control trial (RCT) comparing 2 sugar‐sweetened beverage (SSB) reduction digital health interventions (iSIPsmarter vs a static Patient Education [PE] website). Specifically, the purpose is to explore how rurality and other potentially overlapping sociodemographic characteristics of participants were associated with enrollment, retention, and outcomes in a primarily Appalachian sample.

METHODS

Study design

This study examines a secondary aim of a larger 2‐group RCT efficacy trial comparing the iSIPsmarter digital intervention to a static PE website. RCT details are found elsewhere. ²⁶ The study included rurality and other sociodemographic data from prescreening and baseline surveys as well as process data on enrollment, retention at 9‐week and 6‐month assessments, and engagement during the 9‐week iSIPsmarter intervention period. Enrollment and retention were measured and evaluated across the full sample, while engagement was only explored within the experimental condition. The study was approved by the University of Virginia Institutional Review Board, and participants provided informed consent. Participants received $50 gift cards for completing each of the assessments.

Study setting

The target population was adults from the Appalachian region of Southwest Virginia and surrounding counties. This area faces compromised social determinants of health: high poverty, low educational attainment, low health literacy, and limited health care. ²⁷ The median income in the region is approximately $41,540 compared to $90,974 statewide and about 51% of households are financially constrained, with 19% living at or below the poverty threshold and an additional 32% classified as Asset Limited, Income Constrained, Employed. ²⁸ Only 19% have a 4‐year degree compared to 33% at the state level ²⁹ and the racial/ethnic composition is predominantly White, Non‐Hispanic. Adults in this region consume approximately 38 ounces (475 calories) of SSB per day which is more than double the national average. ³⁰ , ³¹ , ³² Consequently, the region is disproportionately burdened by SSB‐related chronic diseases, such as obesity, diabetes, obesity‐related cancers, heart disease, hypertension, and dental decay. ²⁷ , ³³

Eligibility and recruitment

Eligibility criteria included: (1) English‐speaking adults 18 or older; (2) residing in Southwest Virginia or surrounding counties; (3) consuming >200 SSB calories daily; (4) not in another structured nutrition or diet program; and (5) willing to access the digital intervention on an internet‐enabled device and receive text messages. As part of the consenting process, participants were further screened for ownership of internet‐enabled devices and access to reliable internet service.

Recruitment of eligible participants was achieved through a partnership between the research team and key community stakeholders in Southwest Virginia and surrounding counties. Staff from 2 local Federally Qualified Health Centers, 1 local Health District, and a senior service agency served on the Leadership Council. In addition to recruiting from their own populations, the Leadership Council helped identify additional recruitment partners (ie, hospitals, Housing Development Authorities, public schools, local government agencies, and higher education institutions) and strategies. Recruitment strategies included both passive (print materials, emails to listservs, webpage or social media postings, press releases) and active (presentations and community events) methods. Recruitment messaging provided a brief study overview, highlighting SSB reduction content, the web‐based features of each program, and the RCT design. Potential participants were directed to a study website for further information and an electronic eligibility screener. Recruitment and enrollment progress were routinely assessed for representativeness. Adjustments in recruitment were made to garner more remote rural (eg, ads and press releases in targeted local newspapers), male (eg, social media targeted ads with increased images of men, recruitment at industrial worksites that have predominantly male employees), and lower socioeconomic (eg, recruitment through government assistance programs such as Head Start) participants.

Intervention description and engagement strategies

After the baseline assessment, participants were assigned into either the iSIPsmarter experimental condition or the PE website using simple randomization. ²⁶ Both digital interventions focused on reducing SSB, but differed in dosage, tracking, feedback, personalization, and engagement. A description of the 2 interventions can be seen in Table 1. Both interventions contain the same basic content. However, the PE website was presented as a simplified, online information resource that is primarily text‐based with minimal supporting graphics. In contrast, the iSIPsmarter intervention, which was cocreated using a human‐centered design approach, ³⁴ is grounded in the Theory of Planned Behavior and incorporates behavior change and health literacy techniques. ²⁶ iSIPsmarter has a built‐in tutorial Core, content in a variety of formats, deliberate metered delivery, guided action planning, technology‐supported behavioral monitoring, progress tracking, and personalized feedback on goal obtainment, as well as reminder prompts and stepped care protocols to keep participants engaged (Table 1).

TABLE 1.

Description of iSIPsmarter experimental and Patient Education website control interventions.

Components	iSIPsmarter	Patent Education website
Content	∼162 content pages (6 Cores with ∼27 content pages each)	13 content pages
Format	Text, audio, graphics, animation, video, interactive activities, and vignettes	Text and graphics
Delivery	Program content metered out via 6 Cores; each Core opens 7 days after completion of prior Core	Program content available all at once
Action planning	Guided goal setting with barrier and strategy identification	None
Sugary drink tracking	Daily sugary drink diary transmitted to a program dashboard via text message response	Paper diary
Weight tracking	Daily weight diary transmitted to a program dashboard via a cellular‐enabled scale	Paper diary to record weights from the cellular‐enabled scale
Progress tracking	Visualizations of sugary drink and weight trends	None
Personalization	Personalized feedback on progress	None
Automated reminders	Daily email reminders to complete SSB and weight diaries and notify when a new Core opens	None
Stepped‐care	Human‐supported texts and phone calls for participants who disengage	None

Enrollment and retention procedures

Participants were enrolled between August 2021 and December 2022 through a multiphase process: (1) online screener; (2) eligibility screening call and consent; (3) baseline electronic survey; (4) two 24‐hour dietary recalls via phone; and (5) weigh‐in using a cellular‐enabled scale. The process took approximately 2 weeks. Coordinators made up to 3 attempts to schedule prescreened eligible participants using emails, text messages, and phone calls. Text reminders were sent at 48 and 24 hours prior to each step of the enrollment process.

Participants completed 9‐week assessments between October 2021 and February 2023 and 6‐month assessments between April 2022 and August 2023. Like the baseline assessments, follow‐up assessments included electronic survey data collection, two 24‐hour dietary recalls completed via phone calls, and a weigh‐in on the cellular‐enabled scale. Participants in both conditions received automated reminders sent through the intervention platforms 2 weeks and 1 week prior to the assessment dates. Also, research coordinators made up to 5 email, text, and phone call attempts to schedule the 24‐hour dietary recalls. Coordinators tracked follow‐up progress and provided reminders when components of the assessment were not completed.

Measures

Table 2 outlines the outcome and predictor variables used in the regression models. Outcome variables included enrollment, retention, and 3 iSIPsmarter engagement measures: Completion of (1) Cores, (2) SSB diaries, and (3) weight diaries. Enrolled participants were those who completed consent, baseline assessment, and randomization into an intervention. Due to a lack of variability in retention at the 9‐week assessment time and to prioritize those most engaged in the study, retained participants were defined as those who completed both 9‐week and 6‐month assessments.

TABLE 2.

Outcome and predictor variables included in the enrollment, retention, and engagement models.

Variables	Variable Description	Time Collected	Regression Model
			Enrollment	Retention	Engagement
Outcome variables
Enrollment	Number of consented participants who completed baseline assessment	BL	X
Retention	Number of participants who completed assessments at both 9 weeks and 6 months	9 weeks and 6 months		X
Engagement, Core Completion a	Number of the 6 Cores completed within the 9‐week iSIPsmarter intervention	9 weeks			X
Engagement, SSB Diaries a	Percent of SSB diaries completed within the 9‐week iSIPsmarter intervention	9 weeks			X
Engagement, Weight Diaries a	Percent of weight diaries completed within the 9‐week iSIPsmarter intervention	9 weeks			X
Predictor variables
Rurality	County‐based Rural Urban Continuum Codes ranked as metro 1‐2, metro 3, and nonmetro 4‐9	PS	X	X	X
Race	Dummy coded for White	PS	X	X	X
Sex	Dummy coded for Male	PS	X	X	X
Age	Calculated as years between birthdate and date of enrollment	PS	X	X	X
Income	Annual household income ranked from 1 = <$5,000 to 12 = $55,0000 or more	BL		X	X
Education	Highest educational attainment ranked from 1 = grades 0‐8 to 6 = graduate school degree	PS	X	X	X
Health literacy	Sum score of 3 items (0‐12) assessing ability to process health information. Higher scores indicate higher health literacy.	PS	X	X	X
eHealth literacy	Sum score of 8 items (5‐40) to assess ability to digital health information. Higher scores indicate higher eHealth literacy.	BL		X	X
SSB intake	Average daily fluid ounce intake calculated through frequency and portion size of 5 SSB categories on the BEV‐Q 15	PS	X
BMI	Weight(kg)/height(m)2 calculated using self‐reported height and cellular scale weight transmissions	BL		X	X
Quality of life	Sum number of days in the past month (0‐30) that participants reported feeling physically or mentally unhealthy	BL		X	X

Abbreviations: BL, baseline assessment; PS, prescreen data.

^a Only assessed and relevant to iSIPsmarter intervention participants.

Predictor variables included participant rurality and other sociodemographic characteristics. Table 1 indicates their inclusion in the enrollment, retention, and engagement regression models. Rurality was determined using Rural Urban Continuum Codes (RUCC), with RUCC codes (1‐9) categorized into 3 strata: large/medium metro (1‐2); small metro (3), and nonmetro (4‐9) counties. ³⁵ Other variables were collected through the prescreener and the baseline survey, including race (dummy coded for White), sex (dummy coded for male), age, income, education, and health literacy. ³⁶ Two health literacy measures were used: a 3‐item assessing participants’ ability to process written health information ³⁷ , ³⁸ and an 8‐item eHealth Literacy Scale (eHEALS) assessing participants’ ability to process online health information. ³⁹ As part of their eligibility determination, participants completed an abbreviated Beverage Intake Questionnaire (BEVQ‐15). ⁴⁰ They also provided BMI data via self‐reported height and transmission of weight on the cellular‐enabled scale. Quality of life was assessed using the Centers for Disease Control and Prevention's Healthy Days core questions that sums the number of self‐reported physically and mentally unhealthy days in the past months. ⁴¹

Analysis

Three multivariable logistic regression models were used to examine the associations between the predictor variables and enrollment and 9‐week and 6‐month retention outcomes. A system of equations was estimated using a seemingly unrelated regression (SUR) model, where 3 equations—Core completion, SSB diaries, and weight diaries—were simultaneously analyzed to investigate how the predictor variables were associated with these 3 engagement outcomes. An SUR model is a statistical method used to analyze multiple, separate but related outcomes at the same time. Each equation has its own dependent variable and can be valid on its own, yet the error terms are assumed to correlate across the equations. The SUR model improves efficiency by accounting for those potential shared associations between equations. Robust standard error adjustments were applied to all models. Predictor variables in each model were based on the timing of their collection (eg, the enrollment model only included variables collected during prescreening).

RESULTS

Eligible, enrolled, and retained participants

Of the 778 potential participants who completed the online screener, 502 (65%) were eligible, and 249 (50%) were enrolled. Of those enrolled, 127 (51%) were randomized into iSIPsmarter and 122 (49%) into the PE website. Retention at 9 weeks was 93% (iSIPsmarter = 91%; PE website = 94%), 89% at 6 months (iSIPsmarter = 86%; PE website = 92%), and 88% at both time points (iSIPsmarter = 84%; PE website = 91%).

Enrolled participants were recruited from 48 different counties in and around Southwest Virginia, with 80% (n = 200) from designated medically underserved areas and 88% (n = 219) from Appalachian designated counties. ⁴² , ⁴³ Eligibility screening conducted during the consent process indicated that all participants owned an internet‐enabled device and 94% had access to reliable internet service. As shown in Table 3, enrolled participants rurality varied (RUCC 1‐2 medium‐large metro = 15%, RUCC 3 small metro = 45%, RUCC 4‐9 nonmetro = 41%). The majority were White (89%), female (83%), and with an average age of 42 (SD = 12.60) years. Most had college degrees (59%), household incomes <$55,000/year (52%), and relatively strong health literacy (M = 11.5 [SD = 0.99] out of 12 points) and digital health literacy (M = 31.0 [SD = 6.28] out of 40 points). SSB intake at prescreening for enrolled participants averaged 52.8 (SD = 36.63) fluid ounces, and 59% were with obesity. Participants reported an average of 14 (SD = 10.50) days that they felt physically or mentally unhealthy in the past month. Enrolled participants were largely similar to eligible participants, with slightly more college‐educated individuals choosing to enroll in the study. The retained sample also closely resembled the enrolled group.

TABLE 3.

Participant sociodemographic characteristics across study eligibility, enrollment, and retention samples.

Sociodemographic characteristics	Eligible (n = 502)	Eligible and enrolled a (n = 249)	9‐week retention (n = 231)	6‐month retention (n = 221)
RUCC
1‐2 n (%)	84 (16.7%)	36 (14.5%)	35 (15.2%)	30 (13.6%)
3 n (%)	206 (41.0%)	111 (44.6%)	103 (44.6%)	99 (44.8%)
4‐9 n (%)	212 (42.1%)	102 (41.0%)	93 (40.3%)	92 (41.6%)
Race
Black n (%)	22 (4.4%)	13 (5.2%)	11 (4.8%)	10 (13.6%)
White n (%)	444 (89.0%)	221 (88.8%)	205 (88.7%)	196 (88.7%)
Multi n (%)	20 (4.0%)	8 (3.2%)	8 (3.5%)	8 (3.6%)
Other n (%)	13 (2.6%)	7 (2.8%)	7 (3.0%)	7 (3.2%)
Sex
Male n (%)	71 (14.3%)	41 (16.5%)	40 (17.3%)	38 (17.2%)
Female n (%)	427 (84.7%)	207 (83.1%)	190 (82.3%)	182 (82.4%)
Other n (%)	1 (< 1%)	1 (<1%)	1 (<1%)	1 (<1%)
Age M (sd)	40.3 (12.73)	41.6 (12.60)	42.0 (12.63)	42.3 (12.57)
Household income
≤ $14,999 n (%)	—	17 (6.9%)	15 (6.6%)	12 (5.5%)
$15,000‐$34,999 n (%)	—	49 (19.8%)	44 (19.2%)	40 (18.3%)
$35,000‐$54,999 n (%)	—	62 (25.1%)	60 (26.2%)	57 (26%)
≥ $55,000 n (%)	—	119 (48.2%)	110 (48.0%)	110 (50.2%)
Education
< College degree n (%)	248 (50.4%)	103 (41.4%)	93 (40.3%)	84 (38.0%)
College degree n (%)	250 (49.6%)	146 (58.6%)	138 (59.7%)	137 (62%)
Health literacy b M (sd)	11.3 (1.20)	11.5 (0.99)	11.5 (1.02)	11.5 (1.00)
eHealth literacy c M (sd)	—	31.0 (6.28)	30.8 (6.23)	30.8 (6.00)
SSB intake fl oz M (sd)	58.2 (39.40)	55.2 (36.63)	54.9 (36.41)	53.2 (34.19)
BMI M (sd)		33.2 (8.68)	33.1 (8.72)	33.0 (8.74)
Underweight (%)	—	1 (<1%)	1 (<1%)	1 (<1%)
Normal (%)	—	39 (16.3%)	38 (17.0%)	38 (17.8%
Overweight (%)	—	58 (24.3%)	54 (24.2%)	49 (22.9%)
Obese (%)	—	141 (59.0%)	130 (58.3%)	126 (58.9%)
Quality of life d M (sd)	—	14.5 (10.50)	14.6 (10.63)	14.4 (10.62)

^a Eligible participants were randomized into iSIPsmarter (n = 127) or the Patient Education website (n = 122).

^b Sum score of a 3‐item scale (0‐12); higher scores indicate higher health literacy.

^c Sum score of an 8‐item scale (5‐40); higher scores indicate higher eHealth literacy.

^d Sum number of days in the past month (0‐30) that participants reported feeling physically or mentally unhealthy.

Predictors of enrollment and retention

As shown in Table 4, rural status (RUCC 4‐9) was not significantly associated with enrollment when compared to either medium‐large metro (RUCC 1‐2) (OR = 0.79, 95% CI = 0.46‐1.37, P = .41) and small metro (RUCC 3) (OR = 1.28, 95% CI = 0.86‐1.91, P = .23) counties. However, higher education was a significant predictor with more educated participants being 1.37 times (95% CI = 1.12‐1.67, P = .002) more likely to enroll in the study than their less educated counterparts. Perceived health literacy also significantly predicted enrollment. Participants with higher health literacy scores were 1.23 times (95% CI = 1.03‐1.47, P = .02) more likely to enroll in the study than were those with lower health literacy scores. It should be noted that there appears to be a trend of older age‐predicting enrollment (OR = 1.01, 95% CI = 1.00‐1.03, P = .09).

TABLE 4.

Multivariable logistic regression exploring associations between demographic characteristics and study enrollment of eligible participants (n = 496) and retention of enrolled participants at both 9 weeks and 6 months (n = 234). ^a

Sociodemographic characteristics	Enrollment of eligible participants OR (95% CI) (P value)	Retention of enrolled participants 9 weeks and 6 months OR (95% CI) (P value)
Rurality Nonmetro (RUCC 4‐9) as base
Metro (RUCC 1‐2)	0.79 (0.46‐1.37) (.41)	0.44 (0.12‐1.61) (.22)
Metro (RUCC 3)	1.28 (0.86‐1.91) (.23)	0.57 (0.21‐1.54) (.27)
Other sociodemographics
Race (1 = White)	0.97 (0.54‐1.73) (.92)	0.54 (0.13‐2.27) (.40)
Sex (1 = Male)	1.46 (0.85‐2.51) (.17)	2.77 (0.68‐11.30) (.16)
Age	1.01 (1.00‐1.03) (.09)	1.04 (1.01‐1.07) (.01)
Income	N/A	1.13 (1.00‐1.28) (.05)
Education	1.37 (1.12‐1.68) (.002)	1.51 (1.00‐2.29) (.05)
Health literacy	1.23 (1.03‐1.47) (.02)	0.92 (0.64‐1.33) (.67)
eHealth literacy	N/A	0.95 (0.87‐1.03) (.23)
SSB intake fl oz	1.00 (0.99‐1.00) (.38)	N/A
BMI	N/A	0.97 (0.87‐1.02) (.20)
Quality of life	N/A	1.02 (0.93‐1.06) (.45)

Abbreviations: CI, confidence interval; OR, odds ratio.

^a The reductions in eligible (n = 496) and enrolled (n = 234) samples are due to missing predictor variable data that removed participants from the regression analysis.

Similar to enrollment, rural status (RUCC 4‐9) was also not significantly associated with retention when compared to large metro (RUCC 1‐2) (OR = 0.44, CI 95% = 0.12‐1.61, P = .22) or small metro (RUCC 3) (OR = 0.57, 95% CI = 0.21‐1.54, P = .27) counties. However, older participants were 1.05 (95% CI = 1.01‐1.07, P = .02) more likely to be retained than younger participants. Similarly, those with greater household incomes (OR = 1.13, 95% CI = 1.00‐1.28, P = .05) and educational attainment (OR = 1.51, 95% CI = 1.00‐2.29, P = .05) were also more likely to be retained than their lower income and less educated counterparts.

Predictors of iSIPsmarter engagement

Over half (58%) of iSIPsmarter participants completed all 6 Cores during the 9‐week intervention with an additional 16% completing 5 Cores. On average, participants completed 5.2 Cores (SD = 1.6) and tracked 76% of their daily SSB diaries and 57% of their daily weight diaries. Only 3 participants did not engage at all, and 7 completed the orientation Core (Core 1) only.

Table 5 summarizes engagement predictors. Rural status (RUCC 4‐9) was not significantly associated with either Core, SSB diary, or weight diary completion when compared to medium‐large metro (RUCC 1‐2) (Core ß = −0.42, SE = −1.36‐0.51, P = .38; SSB ß = −6.43, SE = −23.32‐10.26, P = .45; Weight ß = −9.02, SE = −25.97‐7.93, P = .30) or small metro (RUCC 3) (Core ß = −0.23, SE = −0.81‐0.34, P = .43; SSB ß = −7.65, SE = −17.18‐1.88, P = .12; Weight ß = −0.61, SE = −10.86‐9.64, P = .91) counties. However, older age significantly predicted all 3 engagement indicators. Specifically, for every 10‐year increase in age, Core completion increased by 0.3 Cores (SE = .004‐.06, P = .02), SSB diary tracking increased by 6.2 days (SE = 0.17‐1.07, P = .007), and weight diary tracking increased by 10.5 days (SE = 0.60‐1.49, P<.001). While not statistically significant, higher quality of life (ß = 0.02, SE = −0.002‐0.05, P = .07) and income (ß = 1.55, SE = −0.62‐3.15, P = .06) approached significance for predicting Core and SSB diary completion, respectively.

TABLE 5.

Seemingly unrelated regression of demographic characteristics influencing iSIPsmarter engagement in core and diary completion during the 9‐week intervention time (n = 119).

Sociodemographic characteristics	Core completion ß (SE) (P value)	SSB diary ß (SE) (P value)	Weight diary ß (SE) (P value)
Rurality Nonmetro (RUCC 4‐9) as base
Metro (RUCC 1‐2)	−0.42 (−1.36‐0.51) (.38)	−6.43 (−23.32‐10.26) (.45)	−9.02 (−25.97‐7.93) (.30)
Metro (RUCC 3)	−0.23 (−0.81‐0.34) (.43)	−7.65 (−17.18‐1.88) (.12)	−0.61 (−10.86‐9.64) (.91)
Other sociodemographics
Race (1 = White)	−0.69 (−1.50‐0.12) (.10)	−8.46 (−23.52‐6.61) (.27)	6.77 (−10.53‐24.07) (.44)
Sex (1 = Male)	0.05 (−0.70‐0.81) (.89)	1.55 (−11.81‐14.92) (.82)	4.47 (−7.80‐16.75) (.48)
Age	0.03 (0.004‐0.06) (.02)	0.62 (0.17‐1.07) (.007)	1.05 (0.60‐1.49) (<.001)
Income	0.08 (−0.02‐0.18) (.10)	1.55 (−0.62‐3.15) (.06)	0.18 (−1.50‐1.87) (.83)
Education	0.13 (−0.14‐0.41) (.33)	0.13 (−4.52‐4.78) (.96)	0.40 (−5.27‐6.06) (.89)
Health literacy	0.003 (−0.31‐0.31) (.99)	1.31 (−3.89‐6.51) (.62)	4.04 (−0.87‐8.94) (.11)
eHealth literacy	−0.03 (−0.08‐0.03) (.32)	−0.47 (−1.26‐0.32) (.24)	−0.56 (−1.32‐0.21) (.15)
BMI	0.001 (−0.03‐0.03) (.93)	0.16 (−0.30‐0.62) (.50)	0.01 (−0.52‐0.55) (.96)
Quality of life	0.02 (−0.002‐0.05) (.07)	0.37 (−0.14‐0.88) (.16)	0.12 (−0.34‐0.58) (.62)

Abbreviations: ß, beta coefficient; SE, standard error.

DISCUSSION

Digital health interventions hold promise for improving access to evidence‐based resources and health care in underserved rural areas. ⁸ , ¹⁷ However, concerns persist that limited digital access and skills could potentially increase health inequities. ⁴⁴ Overall, rurality status was not associated with our study's enrollment or retention, nor was it associated with engagement in the iSIPsmarter intervention. This is in contrast to rurality disparities identified in larger scoping reviews, ⁹ , ¹⁰ , ¹¹ yet similar to pilot digital health trials that integrated strategies to overcome rural population barriers (eg, access to virtual health care providers, codesign of intervention, community‐based participatory research approaches). ²⁰ , ⁴⁵ Consistent with the latter pilot studies, the iSIPsmarter study used a human‐centered design process to create highly personalized content, ²⁶ , ³⁴ community‐based recruitment partners, reminders and incentives, and human‐supported stepped‐care strategies to reach, engage, and retain an Appalachian sample. ²⁶ These efforts likely contributed to the limited influence of rurality status in the iSIPsmarter study. However, other social determinants of health were influential.

Findings from our study indicated that lower educational attainment, health literacy, and income, as well as younger age, were significantly predictive of lower enrollment, engagement, and retention. The link between these social determinants of health and disparities in digital health uptake, engagement, and outcomes is persistent across the literature. ⁹ , ¹⁰ , ¹⁷ Thus, interpreting these findings requires attention not only to rurality status but also to the demographic composition of our sample, which may help explain why certain subgroup differences were more salient than others.

Given these findings, it is important to consider the representativeness of the enrolled sample when interpreting our results. Although the racial and ethnic composition generally reflects the regional population, the sample was disproportionately female (83%) and college‐educated (59%). The overrepresentation of females aligns with patterns observed in our previous SSB reduction trials ⁴⁶ , ⁴⁷ and is consistent with broader trends in lifestyle behavioral intervention trials. ⁴⁸ , ⁴⁹ This trend may also reflect the generally higher acceptability of SSB consumption among men. ⁵⁰ , ⁵¹ , ⁵² , ⁵³ The higher proportion of college‐educated participants likely is a result of inclusion criteria related to technology access and the use of higher education recruitment partners. Furthermore, higher educated participants, especially those affiliated with academic institutions, may have more exposure to clinical trials and greater trust in health research and digital tools than their less‐educated or more remote rural counterparts. ⁵⁴ , ⁵⁵ This somewhat restricted sample variability may have attenuated or exaggerated subgroup associations due to selection effects. For example, technology requirements may have masked rural‐urban differences by enrolling highly connected rural participants demographically similar to urban peers. Likewise, while older adults are often underrepresented in digital health studies due to limited technology access and skills, ⁹ , ¹⁷ we observed a positive association between age and both enrollment and engagement—likely reflecting selective participation by older adults with sufficient technological capacity. However, despite our study sample's less vulnerable profile, significant associations with education, health literacy, income, and age suggest these social determinants continued to shape outcomes.

Our findings linking certain social determinants of health with digital health intervention participation and engagement highlight opportunities to develop strategies to enroll, retain, and engage a more representative sample. While stakeholders were integral in implementing recruitment strategies, their expertise in their communities could have been leveraged to create tailored study messaging for reaching and retaining a wider audience within the targeted population. ⁵⁶ Furthermore, prescreener participant data could be used to trigger enhanced onboarding processes for more vulnerable participants (e.g., lower‐literacy participants could recieve greater detail on program design accomodations). ⁵⁷ The program itself could employ nonintrusive pop‐ups to instruct individuals throughout the Cores and provide reminders of various features. Personalized touch points throughout the study as well as time‐sensitive incentives could encourage engagement of younger participants in post‐assessments and enhance study retention of all vulnerable subgroups for follow‐up assessments. ⁵⁶ With intentional refinements to recruitment approaches, engagement strategies, and support mechanisms, it may be possible to enhance enrollment and retention in digital health trials, particularly for underserved populations.

Limitations and strengths

Several limitations should be considered when interpreting our findings. First, as previously discussed, the representativeness of our sample may limit generalizability. Second, because this study was an exploratory aim of a larger RCT, it was not powered to detect differential effects by rurality status. While sample size is one consideration, variation in latent factors such as willingness to enroll, engage, and remain in the study is also critical for detecting differences in subgroup associations. Third, our approach to rural classification using 3 RUCC groupings (medium‐large metro, small metro, and nonmetro) aimed to reflect the nuanced mix of urban and rural communities within small metro counties may have masked important intracounty variation. ⁵⁸

These limitations are countered by our study strengths which include: (1) trusted local agency recruiters; (2) multiple reminders for research and program activities; (3) a human‐centered design intervention with tailored content and personalized feedback; (4) digital technology for tracking SSB and weight; and (5) a human‐supported stepped‐care program to re‐engage participants. In addition, while randomization does not eliminate limitations related to statistical power or heterogeneity in rural populations, it may have helped to balance unmeasured confounding variables that could otherwise influence study participation and engagement.

Future research

Our study identifies several key areas for future research. First, fully powered studies are needed to more definitively assess the associations of rurality and other sociodemographic factors with enrollment, retention, and engagement in digital health interventions. Second, future research should use inclusive study design and recruitment strategies to attract and enroll representative samples. This includes codesigning interventions with end‐users and using community‐based recruitment and implementation practices. ²⁰ Prioritizing baseline systematic measurement of key social determinants of health—geographic location, age, sex, race, income, educational attainment, and health literacy—and examining for disparities across the phases of research is essential for evaluating the potential of digital health to reduce health disparities. ²⁵ Third, collecting qualitative data on participants’ motivations to enroll, retain, and engage in digital health trials could provide additional insight into specific barriers faced by subgroups, especially younger participants whose disengagement may stem more from perceived burden or lack of benefits than from educational or health literacy challenges. Finally, future RCTs should focus on the development and testing of targeted strategies for enrollment, retention, and engagement, including addressing disparities in device ownership and reliable access to high‐speed broadband. These tailored approaches could not only improve study outcomes but also enhance the broader applicability of digital health interventions in addressing health disparities among underserved populations.

CONCLUSIONS

The findings from this exploratory study offer valuable insights into the role of digital health interventions in the rural Appalachian region. These results suggest that demographic factors—such as educational attainment, health literacy, income, and age—were key determinants for participation in a digital health intervention, rather than rurality status. From a public health perspective, these results underscore the potential of digital interventions to bridge evidence‐based behavioral intervention access gaps in rural regions facing significant health disparities. However, achieving equitable reach and engagement in rural communities will require further efforts to address social determinants of health barriers, including tailored strategies for those with lower educational attainment, health literacy, and income, as well as younger age.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this manuscript.

Brock D‐JP, Ritterband LM, You W, et al. An exploratory study to understand how rurality status and demographic characteristics are associated with enrollment, engagement, and retention in a digital health intervention targeting the Appalachian region. J Rural Health. 2025;41:e70042. 10.1111/jrh.70042