Outcomes of the Dulce Digital-COVID Aware (DD-CA) discharge texting platform for US/Mexico border Hispanics with diabetes

Abstract

Background:

Hispanics have higher type 2 diabetes (T2D) prevalence, poorer outcomes, and are disproportionately affected by COVID-19. Culturally-tailored, diabetes educational text messaging has previously improved HbA1c in this population.

Methods:

During the pandemic, hospitalized Hispanic adults with T2D (N=172) were randomized to receive Dulce Digital-COVID Aware (“DD-CA”) texting platform upon discharge plus diabetes transition service (DTS) or DTS alone. DD-CA includes diabetes educational messaging with additional COVID-safe messaging (e.g., promoting masking; social distancing; vaccination).

Findings:

Among adults with poorly-controlled diabetes (mean HbA1c = 9.6±2.2%), DD-CA did not reduce 30- or 90-day readmissions compared to standard care (28% vs 15%, p=0.06; 37% vs 35%, p=0.9, respectively). However, the improvement in HbA1c was larger among those in the DD-CA compared to DTS at 3 months (n=56; −2.69% vs. −1.45%, p=0.0496) with reduced effect at 6 months (n=64; −2.03% vs −0.91%, p=0.07). Low follow-up completion rates and the addition of covariates (to control for baseline group differences that existed despite randomization) impacted statistical power.

Interpretation:

During the pandemic, DD-CA offered an alternative digital approach to diabetes and COVID education and support for a high-risk Hispanic population and achieved trends toward improvement in glycemic control despite relatively low engagement and not reducing hospital readmissions.

Funding:

NIH/NIDDK: 3R01DK112322–05S1

Introduction:

Background:

Diabetes affects nearly 38 million US adults, and disparities in diabetes prevalence exist, such that ethnic minorities, including Hispanic populations, ¹ and individuals of low socioeconomic status (SES) are disproportionally affected. Total diabetes prevalence in Hispanic adults have been estimated up to 22%, which is nearly twice the prevalence observed in non-Hispanic whites². Hispanic individuals have also been disproportionally adversely impacted by novel coronavirus disease (COVID-19) and this disparity is especially notable on the US/Mexico border.³ The California Department of Public Health reports that Hispanic individuals make up 39% of California’s population but 57% of confirmed COVID-19 cases. ^4,5

Early in the COVID-19 pandemic, diabetes emerged as a leading risk factor for severe illness leading to hospitalization⁶, an independent predictor of intensive care unit placement and ventilation, ⁷ and as a factor associated with disease severity and mortality⁸. Maintaining good glucose control improves prognosis in COVID-19 among people hospitalized with pre-existing T2D. In general, once discharged from the hospital, patients with diabetes have higher 30-day readmission rates than all other hospitalized patients, ^9–12 and research has highlighted post-hospital discharge dysglycemia as a significant predictor of readmission. ¹³ During the pandemic, social distancing, quarantine, and stay-at-home/lockdown guidelines may have made it even more challenging to achieve and maintain adequate glycemic control once discharged to home. A digital approach to support high-risk individuals’ diabetes management during transition from the hospital and in the unique context of the pandemic was warranted to prevent the adverse (and synergistic) effects of poorly controlled diabetes and COVID-19 in this population.

Previous studies have shown that diabetes self-management education and support (DSME/S) improves glycemic control, reducing glycosylated hemoglobin (HbA1c) values by a clinically significant average of 0.74%.¹⁴ Project Dulce was developed as an in-person DSME/S program for diverse ethnic communities with low income status at high risk for diabetes-related complications and was effective in improving clinical, behavioral, and financial outcomes.^15–18 To overcome practical barriers inherent to accessing in-person DSME/S, Dulce Digital, a previously-tested text-messaging based intervention, was developed for under-resourced Hispanic adults with poorly controlled T2D to provide a telemedicine intervention integrating educational text messaging with two-way feedback of blood glucose values. Dulce Digital was found acceptable in this population and effective in reducing HbA1c by 1.0% and improving self-management behaviors after the 6-month program¹⁹. Dulce Digital’s format offered a unique opportunity to support high-risk people with diabetes discharging from the hospital setting during the early stages of the COVID-19 pandemic – i.e., a time when other diabetes care visits and programs were suspended or reduced significantly.

Objective:

The Dulce Digital-COVID Aware (DD-CA) program added COVID-19 educational messaging to the previously effective texting platform (Dulce Digital) that includes diabetes educational and motivational messages, medication reminders, and blood glucose monitoring prompts. DD-CA was offered to hospitalized adults with T2D (COVID-19 positive or negative to promote prevention efforts in a high-risk population) at the time of hospital discharge to virtually enhance diabetes self-management support, readmission prevention, and glycemic control. This randomized trial compared the effectiveness of standard of care diabetes transition service (DTS) versus DTS plus DD-CA in reducing 30- and 90-day hospital readmissions and improving HbA1c and diabetes self-management behavior at 3- and 6- months. We hypothesized that the adaptation of Dulce Digital to include COVID-centric messaging (e.g., regarding masking, social distancing, testing, and vaccination) would: (1) facilitate improvements in diabetes self-management behavior and clinical control, and thus reduced readmissions due to diabetes and its complications as well as (2) reduced severity of, or risk for COVID infection requiring hospitalization.

Methods:

A schematic of the study design is presented in Figure 1. Briefly, the standard of care for hospital discharge of an individual with T2D at our hospital is to receive DTS, wherein a diabetes care and education specialist provides pre-discharge instructions and follow up calls are conducted by a peer health coach. The DD-CA intervention adds to DTS by providing diabetes self-management education and support text messaging along with COVID-related informational messaging over a 6-month period after discharge. The effect of DD-CA relative to DTS alone on 30- and 90-day hospital readmissions and 3- and 6-month HbA1c change was evaluated in the present trial.

Figure 1.

Summary of protocol for the Dulce Digital COVID-Aware trial for both DTS (Diabetes Transition Service) and DTS + Dulce Digital COVID Aware (DTS+DD-CA) arms

Trial Design, Randomization, and Blinding:

This study was a parallel arm, 1:1 individually randomized trial approved by the Scripps Health Institutional Review Board. Participant randomization was determined by a blinded study statistician using a binary random number generator and provided for each individual in sealed envelopes to prevent selection bias.

Participants:

Hispanic adults with T2D hospitalized at a large health system near the California, US/Mexico border were randomized to the DD-CA texting platform in their preferred language in addition to standard of care (DTS) or DTS alone. Inclusion criteria were Hispanic adults at least 18 years old; admitted to Scripps Mercy Hospital Chula Vista; diagnosed with T2D; English or Spanish as preferred language; owned a cell phone that could send text messages; and HbA1c ≥ 7.0% in the last 30 days. Exclusion criteria were pregnancy; current participation in other studies; anticipated move out of the San Diego region in the next 6-months; or any other condition deemed contraindicated for participation in the study.

Sample size:

An a-priori power analysis informed by our health system’s electronic health record data and prior studies in similar populations deemed a sample size of N=172 participants (n=86 per arm) sufficient to observe statistical significance in our primary outcome given estimated 30-day hospital readmissions base rate of 40% (in the standard care DTS group) and an absolute reduction of 20% among the DD-CA arm. These high rates of readmissions were anticipated amid the early stages of the pandemic when individuals with diabetes and Hispanic individuals were at higher risk of hospitalization due to COVID. Additional assumptions included: one-sided α = 0.05, and power = 0.80, with an attrition rate of 15% due to reasons including, but not limited to: loss to follow up and request to withdraw. Power was determined to be adequate to detect a small-to-medium effect size for additional primary and secondary outcomes, glycemic control at 3 and 6 months measured by HbA1c and self-reported indicators.

Procedures and Interventions:

Once a patient was determined eligible based on an electronic medical record (EMR) screening query, and confirmation of impending discharge (in next 24 hours), the patient was asked if they were willing to discuss the study with a bilingual research assistant, who obtained verbal consent to screen for remaining inclusion criteria. Eligible and interested patients were asked to provide written informed consent.

After consent and enrollment, research staff administered a baseline survey before unveiling group assignment. For all participants, DTS was offered as a standard service to diabetes patients discharged from Scripps Mercy Hospital Chula Vista. As part of DTS standard of care, patients received pre-discharge diabetes education from a certified diabetes care and education specialist. They were trained on blood glucose monitoring and were provided testing strips, lancets, and a cellular-connected blood glucose monitor (BioTel Care, Bethesda, MD). These transmitted blood glucose values were used for retrospective evaluation only and were not able to be viewed in real-time by the study team. After discharge, patients were contacted by a peer health coach using standardized protocols to coordinate care with outpatient health and other community resources and collaborate with patients in overcoming identified barriers. The coach called participants 2–3 times in the first month post-discharge to provide training on glucose pattern and excursion recognition. For DD-CA participants, this review/feedback was based on glucose data populated from prompted text-messages, and in standard care, it was based on self-report during the call. Standard DTS delivery was otherwise held constant across groups. The coach was trained on the latest COVID precautions/recommendations and answered participants’ COVID-related questions.

For the DD-CA arm, in addition to the DTS resources described above, participants were enrolled in a digital texting platform (RipRoad, New York, NY) with educational, motivational and medication adherence messaging with additional COVID support messages that provided information addressing identified barriers in Hispanic under-resourced communities (e.g. obtaining testing supplies and medications, accessing routine medical care, and completing other important diabetes self-management behaviors such as healthful eating, exercise, social distancing, quarantine, and stay-at-home/lockdown guidelines). Participants received 13 core content messages per week at intervention-start, which tapered to 6/week across the 6 months of the study. Core content included educational and motivational messaging from the Dulce Digital program that has been previously published¹⁹ and includes messages relating to medication adherence, clinical indicators, physical activity, healthy eating, and well-being, as well as study-specific messaging related to COVID-19 support such as: “Wearing a mask can prevent you from catching the COVID-19 virus from others, but the mask needs to cover both your nose and mouth” and “Scientists are looking for the best medicines to treat COVID-19. When people volunteer for research studies, it helps scientists learn more quickly”. Additionally, participants received 1 – 3 text-message prompts per week to check their blood glucose and reply to the following prompt: “It’s time to check blood sugar! Respond with A – E based on your blood sugar level (mg/DL): A = below 70; B = 71–130; C = 131–180; D = 181–250; E = over 250”. While all values taken using the cellular-enabled glucose meter were automatically transmitted to and stored in a separate database for analysis, participants’ responses to the above prompt triggered automated feedback tailored to their reported blood glucose levels (e.g., “Nearly all of your blood sugars were in the ideal range of 80–180 this week. What you’re doing is working; keep it up!” or “Less than half of your sugars were in range this week. Make sure you are taking your medications every day as prescribed.”). Upon enrollment, research assistants educated the patients on the typical questions and requested responses to prepare participants for replying to study messages.

Outcomes:

The primary outcome for the study was 30-day all-cause hospital readmissions. Secondary outcomes included 90-day readmissions and changes in HbA1c at 3- and 6-months. EMR data was used to capture subsequent hospital admissions and HbA1c lab values were extracted by blinded data analysts. Surveys to measure changes in patient reported outcomes conducted at baseline and 3- and 6-month follow-up included: The Summary of Diabetes Self-Care Activities (SDSCA) ²⁰, Diabetes Distress Scale (DDS) ²¹, quality of life via The Patient-Reported Outcomes Measurement Information System (PROMIS) ²², and the COVID-19 Patient Survey (PhenxToolkit) ²³. The SDSCA is a 7-item survey that provides measures of overall and specific diabetes self-management behaviors, and higher scores indicate better diabetes self-care for each scale. The DDS is a 17-item questionnaire to measure overall, emotional, physical, regimen, and interpersonal distress that is common in individuals with diabetes. The PROMIS survey is a 10-item scale used to capture physical and mental health by measuring self-reported physical functioning, general mental health, emotional distress, satisfaction with social activities and relationships, ability to carry out usual social activities and roles, pain, fatigue and overall quality of life. Participants were also asked whether they had or suspected a new COVID-19 infection at each follow-up timepoint. All participants received a follow-up call at their 3-month and 6-month timepoint to complete the questionnaires via telephone and be reminded of study labs. After the surveys, participants were asked to go to any Scripps lab location in the next 7–10 days to get their blood drawn to measure HbA1c levels. All screening and survey data were uploaded into a Research Electronic Data Capture (REDCap) database²⁴. Participants received test results and compensation for completing surveys and labs.

Statistical methods:

The software R v.4.0.3 was used for all analyses. Descriptive statistics were generated for all outcomes (readmission rates, HbA1c, self-reported variables). Patient-reported indicators and baseline clinical variables were compared between groups to assess randomization among study arms by Fisher’s Exact tests to compare categorical variables and student’s t-tests to compare continuous variables. The EMR was used to identify readmissions during each patient’s unique follow up period. To conduct the planned outcomes analyses for which the study was powered, unadjusted between-group differences were first analyzed by comparing proportion of patients with any hospital readmissions within the 30-day period by a Fisher’s exact test. Additional analyses were conducted using the same methods but only among those who were alive at the follow-up timepoint. Further analyses were conducted using a logistic regression model for readmissions at 30-days as the dependent variable with group as fixed effect and controlling for variables with baseline differences that existed despite randomization between study. Differences in 90-day hospital readmissions between groups were also compared using the same methods. Similarly, unadjusted group mean differences in HbA1c were assessed with a student’s t-test at both the 3- and 6-month timepoints, and follow-up regression models were assessed with change in HbA1c as the dependent variable, group as fixed effect, and including covariates to control for variables that differed at baseline despite randomization. The 3- and 6-month HbA1c values were those closest to an absolute 90- and 180-days post enrollment, with a maximum difference of 76 – 120 and 166 – 210 days, respectively. HbA1c trends over time were also assessed using mixed effects models with study arm and time as fixed effects and subject as a random effect to account for all HbA1c measurements taken during the study period, rather than only assessing at discrete timepoints. Changes in self-report assessments were compared between groups by t-tests at each time point, and then assessed by linear regression models with a fixed effect of group controlling for variables that differed between groups at baseline. Rates of self-reported new COVID-19 infections over each follow-up period were identified as those reported in a follow up period that had not previously reported an infection and were compared by Fisher’s Exact tests and in follow up analyses by logistic regression models controlling for variables with baseline differences between groups. Rates of transmitted blood glucose values were captured and tested for association with change in HbA1c at both time points via linear regression models. Since a priori planned outcomes analyses anticipated randomization would successfully mitigate baseline differences, the resulting p-values from unadjusted analyses are presented in addition to p-values resulting from multiple regression models including covariates that differed between groups at baseline despite randomization.

Results:

Participant flow and recruitment:

Total participants on the automated recruitment lists generated based on EMR criteria, screened for additional inclusion/exclusion criteria, deemed eligible, consented, and enrolled in the study are presented in Figure 2. Participants discharged prior to screening were not able to be enrolled. As shown in the CONSORT Flow Diagram, readmissions were compared for all living participants, but for follow-up measures, completion rates were higher (>50% completed) for HbA1c than the low rates of follow-up patient-reported outcome surveys (< 50% completed) at each time point (3- and 6-months post enrollment).

Figure 2.

CONSORT Flow Diagram

Baseline data:

During the COVID-19 pandemic from February through December 2021, Hispanic patients with T2D (N=172) hospitalized near the US/Mexico border in California were randomized to DD-CA+DTS or DTS alone. Participants were a mean age of 58 ± 13.3 years, 52% male, 9.3% COVID+, with a mean BMI of 32.7 ± 8.2kg/m² and baseline HbA1c of 9.6 ± 2.2%. Despite randomization, patients in the DD-CA arm were younger (mean age: 54.2 ± 14.2 versus 61.3 ± 11.2 years, p = .0004), had lower rates of healthcare coverage (77% versus 91%, p = .02), and had longer median hospitalization LOS (5 versus 4 days, p = .02) compared to the DTS arm, respectively. BMI, baseline HbA1c, sex, COVID positive status during index hospitalization, and preferred language did not significantly differ between groups. For self-report distress, diabetes management behaviors, and quality of life measured at baseline, the only between-group difference was that the DD-CA arm reported lower blood glucose checking behaviors on the SDSCA assessment (p = .01). Detailed baseline characteristics are presented in Table 1.

Table 1.

Participant and baseline characteristics

	Overall (N=172)	DTS (N=86)	DD-CA+DTS (N=86)
	Mean (SD), n (%)	Mean (SD), n (%)	Mean (SD), n (%)	p-value
Sociodemographic Characteristics:
Age	57.78 (13.26)	61.34 (11.24)	54.23 (14.2)	0.0004
Sex				0.0671
Female	83 (48.26%)	48 (55.81%)	35 (40.7%)
Male	89 (51.74%)	38 (44.19%)	51 (59.3%)
Preferred Language				0.6147
English	50 (29.07%)	27 (31.4%)	23 (26.74%)
Spanish	122 (70.93%)	59 (68.6%)	63 (73.26%)
Race
White	69 (40.12%)	32 (37.21%)	37 (43.02%)	0.5339
Black/African American	0 (0%)	0 (0%)	0 (0%)	-
American Indian/Alaskan Native	8 (4.65%)	5 (5.81%)	3 (3.49%)	0.7200
Asian/Native Hawaiian/Pacific Islander	0 (0%)	0 (0%)	0 (0%)	-
Other	98 (56.98%)	52 (60.47%)	46 (53.49%)	0.4414
Employment				0.2341
Retired, Disabled, or Not Employed	124 (72.09%)	66 (76.74%)	58 (67.44%)
Employed Part- or Full-Time	48 (27.91%)	20 (23.26%)	28 (32.56%)
Household Income				0.1537
≤$20,000	93 (54.07%)	52 (60.47%)	41 (47.67%)
>$20,000	65 (37.79%)	28 (32.56%)	37 (43.02%)
Decline to answer	14 (8.14%)	6 (6.98%)	8 (9.3%)
Married or Living with a Partner	95 (55.23%)	46 (53.49%)	49 (56.98%)	0.7592
High school education or greater	89 (51.74%)	45 (52.33%)	44 (51.16%)	0.9353
Health/Healthcare-related Characteristics:
Healthcare Coverage	144 (83.72%)	78 (90.7%)	66 (76.74%)	0.0219
BMI	32.7 (8.23)	32.83 (9.13)	32.57 (7.27)	0.8370
COVID+ at Index Admission	16 (9.3%)	7 (8.14%)	9 (10.47%)	0.7938
Comorbidities/Diagnosis History
Congestive Heart Failure	41 (23.84%)	26 (30.23%)	15 (17.44%)	0.0728
Peripheral Vascular Disease	28 (16.28%)	16 (18.6%)	12 (13.95%)	0.5361
Chronic Pulmonary Disease	23 (13.37%)	17 (19.77%)	6 (6.98%)	0.0243
Diabetes Chronic Complications	91 (52.91%)	48 (55.81%)	43 (50%)	0.5413
Renal Disease	56 (32.56%)	30 (34.88%)	26 (30.23%)	0.6256
Baseline A1c (%)	9.55 (2.15)	9.38 (1.99)	9.73 (2.29)	0.2814
Index Admission LOS (Med, IQR)	5 (3–7)	4 (3–6)	5 (4–8)	0.0241
Baseline Self-Report Surveys:
Diabetes Self-Management Behaviors
Diet	4.18 (1.78)	4.2 (1.62)	4.15 (1.92)	0.8470
Physical Activity	2.55 (2.92)	2.38 (2.96)	2.72 (2.9)	0.4570
Blood Sugar Check	3.51 (3.21)	4.13 (3.12)	2.88 (3.21)	0.0111
Insulin (#NA: 83; 33; 50)	5.74 (2.44)	5.62 (2.54)	5.92 (2.32)	0.5739
Diabetes Pills (#NA: 52; 27; 25)	5.72 (2.44)	5.78 (2.41)	5.66 (2.5)	0.7824
Diabetes Distress
Overall (n= 153; 75; 78)	1.89 (0.86)	1.93 (0.86)	1.85 (0.85)	0.5976
Emotional Burden	1.67 (0.82)	1.69 (0.8)	1.66 (0.84)	0.7952
Physical Distress	2.16 (1.04)	2.19 (1.08)	2.14 (1.01)	0.7513
Regimen Distress	2.2 (1.18)	2.32 (1.19)	2.08 (1.16)	0.1949
Interpersonal Distress	1.47 (0.79)	1.51 (0.77)	1.42 (0.81)	0.5095
Quality of Life (n=167; 84; 83)
Physical	12.19 (3.24)	11.98 (3.23)	12.41 (3.25)	0.3811
Mental	13.4 (3.18)	13.31 (3.27)	13.5 (3.09)	0.6987

Abbreviations: LOS – Length of stay; IQR – interquartile range (Q1 – Q3)

Engagement:

As shown in Table 2, engagement with responding to automated text-messaging was low, and participants responded to a median of 11% of blood glucose check prompts in the DD-CA arm. Rates of blood glucose checks on the digitally connected meter did not differ between study arms (Med: 1, Min: 0, Max: 252) and predicted change in HbA1c from baseline at both 3 and 6 months (β = −0.013, p = .012, β = −0.014, p = .002, respectively).

Table 2.

Study engagement and outcomes by arm

	DTS (N=86)	DD-CA+DTS (N=86)
	n (%), Mean (SD)	n (%), Mean (SD)	p-value	adj. p-value*
Engagement (Totals over 6-months)
Blood glucose text responses (as % of those received) (Med, IQR)	-#	11.02 (1.69–41.10)	-	-
Transmitted blood glucose checks on connected meter (Med, IQR)	1.00 (0.00–16.00)	1.50 (0.00–49.25)	0.4752	-
Readmissions
30-day: Among all enrolled	13 (15.12%)	24 (27.91%)	0.0626	0.0765
30-day: Among alive at 30-days	13 (15.29%)	23 (27.38%)	0.0622	0.0967
30-day: All emergent/urgent admissions	10 (11.63%)	21 (24.42%)	0.0461	0.1130
30-day: All elective/non-urgent admissions	3 (3.49%)	6 (6.98%)	0.4960	0.1718
90-day: Among all enrolled	30 (34.88%)	32 (37.21%)	0.8739	0.8130
90-day: Among alive at 90-days	29 (34.94%)	30 (36.14%)	>0.9999	0.9909
90-day: All emergent/urgent admissions	26 (30.23%)	26 (30.23%)	>0.9999	0.8870
90-day: All elective/non-urgent admissions	13 (15.12%)	10 (11.63%)	0.6549	0.6939
Mortality
30-day Mortality	1 (1.16%)	2 (2.33%)	>0.9999	0.3256
90-day Mortality	3 (3.49%)	3 (3.49%)	>0.9999	0.6974
New COVID-19 Infections
Month 3 (no prior infection N=113, 57, 56)	2 (3.51%)	5 (8.93%)	0.2712	0.4230
Month 6 (no prior infection N=98, 49, 49)	3 (6.12%)	2 (4.08%)	>0.9999	0.4931
HbA1c (%)
Absolute M3 A1c (n=57; 29, 28)	7.53 (1.13)	7.32 (1.36)	0.5234	0.458
Absolute M6 A1c (n=65; 34, 31)	7.85 (1.49)	7.71 (1.46)	0.7088	0.511
M3: Change from Baseline (n=56; 29, 27)	−1.45 (2.04)	−2.69 (2.53)	0.0496	0.155
M6: Change from Baseline (n=64; 34, 30)	−0.91 (2.02)	−2.03 (2.66)	0.0651	0.161
Self-Report Surveys: 3-month change from Baseline
DDS: Overall	−0.31 (0.69)	−0.42 (0.77)	0.4702	0.7128
DDS: Emotional Burden	−0.27 (0.83)	−0.42 (0.81)	0.3411	0.4721
DDS: Physical Distress	−0.24 (1.18)	−0.57 (0.97)	0.1093	0.3632
DDS: Regimen Distress	−0.38 (1.03)	−0.46 (1)	0.7004	0.9980
DDS: Interpersonal Distress	−0.12 (0.84)	−0.15 (1.06)	0.8657	0.8930
SDSCA: Diet	1.06 (1.89)	0.99 (1.99)	0.8635	0.7750
SDSCA: Physical Activity	0.47 (3.45)	1.77 (3.41)	0.0471	0.0939
SDSCA: Blood Sugar Check	1.64 (3.11)	3.08 (3.37)	0.0218	0.1661
SDSCA: Insulin	0.54 (1.6)	0.7 (2.44)	0.7884	0.7150
SDSCA: Diabetes Pills	0.33 (2.15)	1.15 (2.56)	0.1647	0.3339
PROMIS Physical	0.55 (3.21)	0.87 (2.73)	0.5779	0.5650
PROMIS Mental	−0.16 (2.92)	0.91 (2.74)	0.0493	0.0985
Self-Report Surveys: 6-month change from baseline
DDS: Overall	−0.35 (0.58)	−0.56 (0.68)	0.1258	0.2230
DDS: Emotional Burden	−0.21 (0.62)	−0.43 (0.78)	0.1284	0.2030
DDS: Physical Distress	−0.46 (0.86)	−0.61 (0.92)	0.4154	0.8412
DDS: Regimen Distress	−0.55 (0.85)	−0.65 (1)	0.6127	0.7060
DDS: Interpersonal Distress	−0.3 (0.89)	−0.32 (0.81)	0.9375	0.9680
SDSCA: Diet	0.82 (1.69)	1.08 (2.08)	0.5090	0.7190
SDSCA: Physical Activity	1.33 (3.13)	1.89 (3.41)	0.4034	0.7653
SDSCA: Blood Sugar Check	0.76 (3.41)	2.38 (3.19)	0.0175	0.0455
SDSCA: Insulin	0.13 (1.98)	0.4 (2.16)	0.6610	0.4318
SDSCA: Diabetes Pills	0.48 (2.36)	0.73 (2.8)	0.7090	0.8070
PROMIS Physical	0.38 (3.37)	1.59 (3.33)	0.0773	0.0310
PROMIS Mental	−0.21 (2.82)	0.81 (3)	0.0921	0.1115

^#Only applicable to DD-CA arm

^*p-value reported after controlling for variables age, insurance coverage, and index hospitalization LOS that differed at baseline between groups Bolded values indicate significant associations (p < .05)

Outcomes:

Results for all outcome measures are summarized in Table 2. DD-CA did not significantly reduce 30- or 90-day readmissions compared to DTS alone (28% versus 15%, p = .06; and 37% versus 35%, p = .9, respectively). In follow-up analyses, there were no improvements in readmissions rates at either timepoint when controlling for baseline differences between study arms, excluding patient deaths (n=3), or in examining emergent/urgent admission only.

However, the change in HbA1c was significantly larger among those in the DD-CA compared to DTS at 3 months (n=56; 2.69% versus −1.45%, p = .05) with reduced effect at 6 months (n=64; −2.03% versus −0.91%, p = .07) and when controlling for variables with baseline differences between groups. To maximize available data and include additional HbA1c measurements, a mixed-effects model including all HbA1c measurements taken for participants across the study time period (including both routine care and those ordered as part of the study), there was a significant study group by time interaction toward improvement in HbA1c over time in DD-CA versus DTS (β = −0.01, p = .005, see Figure 3).

Figure 3.

A group by time interaction reveals greater reduction in HbA1c over time among DD-CA arm

For changes in participant reported outcomes from baseline to 3-months, there was a greater improvement in physical activity (p = .05), and blood sugar checking behaviors (p = .02) via the SDSCA subscales, and greater improvements in mental health quality of life (p = .05), measured by the PROMIS survey, in the DD-CA versus DTS arm, though these effects diminished when controlling for baseline differences between groups. At 6-months, the only significant change from baseline was a persistent improvement in reported blood sugar checking behaviors (via the SDSCA) in DD-CA versus DTS arm (p = .02), which persisted when controlling for baseline differences. When controlling for these baseline differences, a significant improvement in self-reported physical health measured by the PROMIS survey was also observed (p = .031). There were no significant differences in new self-reported COVID-19 infections post-discharge between study arms at either time point.

Discussion:

Interpretation:

Amid the COVID-19 pandemic’s reduction of in-person clinical care and emphasis on staying at home, the DD-CA intervention provided a modality to disseminate DSME/S to recently-hospitalized people with T2D and simultaneously provide culturally tailored COVID-19-centric messaging to a high-risk Hispanic population. Participants were randomized to standard of care DTS or the DD-CA intervention, which was delivered via two-way text messaging in addition to standard of care DTS. Despite randomization, baseline differences were present in age, insurance coverage, and index hospitalization LOS. Less than 10% of enrolled participants were COVID-19+ during their index hospitalization, likely due to our lack of enrollment in the intensive care units, where many COVID-19+ patients with T2D were receiving care, and the possibility that more acutely ill participants with COVID-19 may have been less likely to engage with study staff.

Despite hypotheses that the DD-CA intervention would reduce hospital readmissions, we saw a trend toward increased admissions at 30-days with little evidence of difference between study arms at 90-days. While unexpected, this finding is similar to that of a related RCT: the Mi Puente Care Transitions Program, which was conducted in the same hospital with Hispanic adults with cardiometabolic disease. In this study, participants randomized to the Mi Puente arm received post-discharge telephone follow-up to address social, medical, and behavioral health needs and experienced higher 30- and 180-day readmission rates than those receiving usual care at discharge. ²⁵ Similarly, another RCT of a post-discharge program for urban elderly adults saw a trend toward increased ED visits. ²⁶ It is possible that the education provided on conditions and symptoms may increase proactive self-identification and awareness of “red flags” (health concerns), prompting higher utilization of emergency care evidenced by the increase in urgent/emergent admissions at 30-days, though this difference did not persist through 90-days. Despite randomization, patients in the DD-CA arm did have significantly longer LOS at their index admission, which may have indicated more severe conditions that were more likely to need follow-up care or pose higher risk for subsequent admission. While not captured in the present study, a retrospective evaluation of causes for and severity of disease in the initial and subsequent hospitalization(s) and/or complications arising during the index hospitalization may provide additional insight toward the ideal patients who may benefit from a post-discharge texting program. Similarly, medications received during the hospitalization, upon discharge, or in the follow-up period have not been investigated and could also contribute to readmission risk mitigation and warrant further study.

Further, since the participants in this study were largely from low-income communities and may not have had reliable outpatient care, messages prompting proactive care seeking (e.g., telling the patient to “call their doctor” depending on the results of their blood glucose checks) may have contributed to participants seeking emergent/urgent hospital care more frequently in the DD-CA arm. It is also noteworthy that, despite randomization, individuals in the DD-CA arm had lower reported rates of insurance coverage, and that may have caused increased emergent/urgent visits in the absence of an established care provider. Similar to the Mi Puente trial, the DD-CA intervention did not have a direct connection to outpatient care services or include telehealth or nurse hotline resources. While potentially useful in improving outcomes, these services were practically limited by constraints of resources amid the pandemic and due to each participants unique circumstances (e.g., insurance or health coverage, primary care provider network). Future work is necessary to find an optimal approach to reduce readmissions in this high-risk population, particularly at heightened vulnerability post-discharge. ²⁷ It is also critical to acknowledge that the present study was designed within the first 3 months of the pandemic in the United States, and due to the disparities of this population, the poor outcomes observed for both individuals with diabetes and in high disparity populations, we anticipated higher than observed rates of readmissions overall, which diminished our statistical power to detect effects in readmissions differences, had they existed.

While DD-CA did not significantly reduce readmissions relative to standard care, this text message-based DSME/S intervention showed trends toward improvements over time in glycemic control (HbA1c) despite low follow-up rates and thus power to detect significant changes. The observed trend is consistent with prior trials of Dulce Digital programs that documented the effectiveness of the program in improving HbA1c. ^19,28 However, the current study was the first application and modification of Dulce Digital for care transition after hospital discharge and the first to examine hospital re-admission outcomes. The discrepant findings for readmissions and HbA1c in the current study suggest that DD-CA may be beneficial for diabetes outcomes but not for reducing overall hospital utilization during the COVID-19 pandemic. Although sustained reductions in HbA1c among those with T2D is expected to reduce risk for diabetes complications and related hospitalizations over a longer period, a more comprehensive transition intervention may be needed to reduce short-term readmission rates in this population. A recent review of interventions to reduce readmissions in people with diabetes concluded that the evidence for those delivered via telemedicine is equivocal but limited, ²⁹ highlighting the need for more studies that investigate remote or largely automated interventions to reduce readmissions. Additional research on the most effective, relevant, and targeted educational content for this specific population and context is also needed to guide future intervention development efforts.

In addition to improvements in glycemic control, there were trends toward self-reported improvements in physical activity and mental health quality of life at 3-months, and increased blood glucose monitoring at both 3- and 6- months, and significant improvement in self-reported perception of physical health for the DD-CA arm versus DTS alone. This may be due to the specific core content messaging received by the DD-CA arm focused on these components (e.g., increasing physical activity, managing stress, healthy eating). While there were not statistically significant differences between groups in changes from baseline in diabetes distress, diabetes self-management behaviors, and quality of life, nearly every measure showed trends toward more favorable change (i.e., reduced distressed, increased self-management behaviors, improved quality of life) in the DD-CA versus DTS arm, and it is likely that the low survey completion rates precluded power to detect significant differences. While the lack of in person lab follow-up may have been impacted directly by the pandemic, the rates of telephone-based survey completion were lower than anticipated, and additional analyses are warranted to examine characteristics of those who did versus did not follow-up to inform future research implementation efforts. Further analyses are underway to determine if there were differences in any outcomes depending on engagement with the daily texting intervention or adherence to follow-up assessments. Additionally, with advances in diabetes technology, future research integrating sensor-based technology like continuous glucose monitoring (CGM) may present an opportunity for more real-time following of individuals glycemic control without requiring adherence to prompted manual reporting.

Limitations:

The pandemic presented unique challenges with ever-changing guidelines and updates to best practices. However, the utility of this digital intervention in an unprecedented time for ambulatory diabetes management and glycemic control is promising and could be adapted into many settings and contexts. As mentioned, engagement with the text-based intervention and the use of the cellular-connected blood glucose meter were low. The present technology does not capture whether messaging was opened or read, so it is unknown whether prompts not responded to influenced study outcomes. Some participants disclosed that they shared phones with their family members and missed the text, they weren’t savvy with technology and found the blood glucose text complicated to respond and some felt that the texts received were too frequent. This highlights the challenges of digital intervention uptake in this population and warrants additional research into how to bridge the gaps in digital interventions in underserved or under-resourced populations.

Participants in the DD-CA group received the intended core content messages as anticipated; however, responses to the blood glucose text prompts were low. Many participants disclosed to study staff that they had difficulty replying to messages with blood glucose value responses, and the use of the connected blood glucose meter was very low, with half of the participants transmitting blood glucose values never or only once. It is possible participants checked their blood glucose using a personal meter or was unable to connect the study-provided meter to a network to transmit the readings.

The randomized study design was selected to minimize any baseline differences between groups, and therefore the planned outcomes analyses and a priori power calculations did not include adjustment or control for these differences. Despite randomization, age, insurance coverage, and LOS at the index hospitalization differed between groups, such that individuals randomized to the DD-CA intervention were older, had less insurance coverage, and longer hospital LOS, which collectively may have increased the groups likelihood of readmitting due to increased disparity or disease severity/acuity. It is possible that despite the importance of controlling for these unanticipated baseline differences in the outcomes evaluation, including additional covariates in the analyses may have further diminished our statistical power; thus we have included both adjusted and unadjusted findings to increase transparency in interpretation.

Follow-up lab completion rates were much lower than anticipated, which may be likely due to multiple surges in COVID-19 cases over the course of the study period impacting healthcare operations (including lab services) and heightening precautionary behaviors (e.g., emphasis on staying at home, distancing). Low completion rates limited statistical power, and though we saw an independent effect of the DD-CA intervention on 3-month change in HbA1c, we did not achieve a statistically significant effect at 6-months, and the effect at 3-months was diminished when accounting for baseline differences (age, health coverage, and LOS) likely due to insufficient power. However, by binning labs into discrete 3-month and 6-month time frames for analysis, additional labs drawn or labs in the intervening period between these timeframes would have been excluded from analyses. To maximize the available data to better understand the impact of the intervention, the mixed-effects model approach was used and ultimately showed greater improvement in HbA1c values over time among the DD-CA arm compared to DTS alone.

Additionally, EMR-derived outcomes present additional challenges including a lack of detail regarding the circumstances, cause, or acuity of readmissions. It is also possible that participants in the study may have sought care at other regional health centers, which may have impacted readmissions rates. We also did not have information available on the reasons for the index or subsequent admissions to include as covariates in our analyses. However, the challenges with capturing details and accurate rates of readmissions would likely affect both study arms equally. Similarly, COVID-19 infections were captured through self-report due to the inherent difficulties with accurately capturing COVID-19 testing across the region, since many individuals went to local, county, state, or privately operated testing centers, and that data was not captured in our EMR. Thus, self-report bias may have impacted the reported new infection rates; though, we would assume both arms would have been affected by this bias equally.

Conclusions:

During the pandemic, DD-CA offered an alternative digital approach to diabetes and COVID education and support for a Hispanic population at high risk of complications in a post-discharge setting, particularly during a highly vulnerable time, and achieved clinically meaningful improvements in glycemic control despite not reducing hospital readmissions. Further research is needed to identify effective interventions for acutely reducing hospital readmissions and providing long-term sustainable strategies to improve glycemic control and diabetes management.