Text Message Intervention for Teens with Type 1 Diabetes Preserves HbA1c: Results of a Randomized Controlled Trial

Abstract

Aims: Teens with type 1 diabetes (T1D) often struggle with diabetes self-management, which may lead to suboptimal self-care and worsening hemoglobin A1c (HbA1c). Innovative strategies are needed to improve self-care and protect against glycemic decline, especially during adolescence. We aimed to assess the impact on HbA1c of two interventions, problem-solving and text messaging, in teens with T1D.

Methods: In a two-site randomized controlled trial, teens (N = 301) 13–17 years of age with T1D were randomized to one of the four groups using a 2 × 2 factorial design: Teenwork (TW), Text Messaging (Text), TW+Text, or Usual Care. TW intervention included problem-solving aimed at improving T1D self-care for blood glucose (BG) monitoring and insulin bolus dosing. Text intervention involved text reminders to check BG. The primary outcome was change in HbA1c from baseline to 12 months.

Results: At baseline, teens (51% female, 78% white, 59% pump-treated) were (mean ± SD) 15.0 ± 1.3 years, had diabetes duration of 6.5 ± 3.7 years, and HbA1c 8.5% ± 1.1%. There was no significant difference in HbA1c over time by study group. Responsiveness to text reminders by teens in the TEXT and TW+TEXT predicted glycemic benefit; TW did not.

Conclusions: Despite no HbA1c difference by study group, greater response to text message reminders to check BG led to better glycemic control and no deterioration in HbA1c; the problem-solving intervention did not. Given the high penetration of mobile phones and the wide acceptance of text messaging among teens in general, it is encouraging that a text messaging intervention can preserve HbA1c, thus preventing the expected deterioration in glycemic control often seen in teens with T1D.

Introduction

Teens with type 1 diabetes (T1D) often struggle with self-management as they simultaneously face the normal developmental tasks of adolescence and parents become less involved in teens' diabetes management. This often leads to suboptimal adherence and worsening glycemic outcomes during adolescence. Data from the T1D Exchange in 2016–2018 demonstrated a mean hemoglobin A1c (HbA1c) of 9.2% ± 1.9% in U.S. teens 13–17 years old (n = 6675), and only 17% of these youth met American Diabetes Association (ADA) glycemic targets for age.¹

Thus, it is important to identify strategies aimed at improving adherence and glycemic control in teens with T1D by strengthening the teens' self-efficacy for daily diabetes self-management to help prepare them for the transition between pediatric and adult care. Increasing a teen's self-confidence for self-care could involve behavioral interventions utilizing motivational interviewing or problem-solving strategies and daily reminders to perform self-management behaviors.^2–6 Text messaging for daily reminders has been utilized in many studies with some limited benefits, including modest glycemic improvements,^7–9 reduced diabetic ketoacidosis-related hospital visits,¹⁰ and improved diabetes self-efficacy and adherence.¹¹

Several large studies have shown that higher blood glucose (BG) monitoring frequency is associated with better glycemic control.^12,13 In a large database registry study of >26,000 children and adolescents with T1D with data from 1995 to 2006, HbA1c was 0.2% lower per each additional BG check per day after adjustment for several confounding factors, and a full 0.5% lower per each additional BG check when limiting the sample to those monitoring 0–5 times daily.¹³ Thus, one important target for a behavioral intervention in teens with T1D is to increase BG monitoring frequency.

Another potential target of a behavioral intervention for teen self-management is attention to frequency of insulin bolus dosing. Missed insulin boluses for meals and snacks are common among adolescents with T1D,¹⁴ and have been associated with worse glycemic control.^14–17 One study of youth with T1D receiving insulin pump therapy demonstrated that HbA1c was directly correlated with the number of missed insulin meal boluses; four missed meal boluses per week was associated with 1% higher HbA1c level.¹⁵ In addition, Patton et al.¹⁷ found that one additional mealtime bolus per day was associated with HbA1c values that were 1.5% lower.

This study aimed at implementing and evaluating approaches to enhance self-management of T1D in teens. This report describes the evaluation of two distinct interventions focused on increasing the frequency of BG monitoring and insulin bolus dosing to optimize glycemic control in teens with T1D.

Research Design and Methods

Study design

In this two-site (Joslin Diabetes Center and Texas Children's Hospital) randomized controlled trial (NCT01892280), we implemented and assessed a behavioral intervention (Teenwork) and a text messaging intervention in a 2 × 2 factorial design. Participants were randomized to one of the four groups: Teenwork (TW), Text Messaging (Text), TW+Text, or Usual Care. Teens were randomized in two strata: age (<15 years vs. ≥15 years) and HbA1c (<8.5% vs. ≥8.5%), with separate computer-generated randomization lists for each site. The primary outcome was change in HbA1c from baseline to 12 months, adjusting for baseline HbA1c. Safety outcomes included collection of adverse events related to diabetes-related hospitalizations or emergency department visits and severe hypoglycemia requiring assistance.

Participants

Eligibility criteria included age 13–17 years, diabetes duration of ≥6 months, daily insulin dose ≥0.5 U/kg, HbA1c 6.5%–11.0% (48–97 mmol/mol), and possession of a cell phone with text messaging ability. Institutional review boards at both sites approved the study protocol. Eligible teens/parents provided written informed assent/consent before initiation of any study procedures. Each site performed study procedures according to a unified protocol with unified staff training across sites. Review of the electronic health records and provider referrals served to identify teens/families eligible for recruitment. Research assistants contacted potentially eligible participants 2–4 weeks before their regularly scheduled clinical visit by letter or phone call to describe the study; occasionally, teens/families were approached on the day of their clinic visits. IRB-approved recruitment materials (e.g., study flyers) were also used.

Procedures

Study visits occurred concurrently with clinic visits every 3 months for data collection and intervention delivery by trained research assistants. Demographic and clinical data were obtained by parent–youth interview and chart review. BG monitoring frequency was calculated from meter/pump downloads reflecting the 2-week period preceding each study visit; 2 weeks of continuous glucose monitoring (CGM) glucose data have been shown to provide a good estimate of glycemic control for a full 3-month period.¹⁸ The study (conducted from late 2013 to early 2016) predated FDA labeling for nonadjunctive CGM use (in December 2016), and participants were instructed to only use BG results for insulin dosing. All study participants received compensation for time and effort with a maximum of $130 for attending five study visits over the 12-month period.

Study groups

Teenwork (TW) intervention

In the Teenwork (TW) intervention, teens met individually with a research assistant during each quarterly study visit to review strategies for improving diabetes self-care, focusing on BG monitoring and insulin bolus dosing. Teens participated in a maximum of four unique problem-solving sessions sequentially regarding these self-care tasks, one at each study visit between baseline and 9 months. A fifth module related to self-advocacy was delivered at the 12-month primary outcome timepoint (Supplementary Table S1). The modules were developed and designed by a multidisciplinary team (L.M.L., B.J.A., L.K.V., D.A.B., and W.L.L.) and delivered by trained research assistants who asked interactive questions throughout the intervention using motivational interviewing to engage the participant. At the end of each session a goal was set using problem-solving strategies. The research assistants delivering the interactive intervention modules used a tablet computer and the Explain Everything™ software application.

Integrity and fidelity of TW intervention delivery were assessed. All sessions were audio recorded. Supervisory staff (D.A.B. and W.L.L.) not delivering the intervention reviewed the audio recordings for the first 5–10 sessions delivered by each research assistant. Subsequently, ∼10% random sample of all TW sessions was reviewed to confirm consistency of intervention delivery across research assistants and fidelity over time. Supervisors provided weekly feedback to research assistants to maintain optimal and consistent intervention delivery.

Text messaging intervention

The text messaging (Text) intervention utilized a two-way SMS texting system (CareSpeak^® Communications) for text message reminders to check BG levels. At the baseline visit, participants received instructions in the use of the system from a research assistant and participants selected times at which they wanted to receive the text message reminders; times could be readjusted as needed. At each selected time, teens received a text message to check a BG and reply with their BG level. If teens did not respond within 10 min, they would receive a second automated reminder. After teens sent their BG value, they would receive a congratulatory text for providing the BG response, independent of the BG value (Supplementary Table S2 for examples of text reminders and congratulatory texts). Participants initially received one text per day on weekend days and the frequency of text message reminders increased every 1–2 months until they reached a maximum of four texts daily (Supplementary Table S3 for schema of text messaging reminder escalation). If a participant did not reply to any text reminders for 2 weeks, the number of text reminders was reset to one per weekend day with a gradual increase again of up to four texts daily. For resets that occurred after the initial 6 months of the study, there was an accelerated advancement in daily text message frequency to increase from one to four text reminders per day during the remaining 6 months of the randomized controlled trial (Supplementary Table S4). Participants received $5 per month toward the costs of an unlimited mobile text messaging subscription.

Other study groups

Participants in the TW+Text group received both the Teenwork and text messaging interventions. Participants in the Usual Care group received routine clinical care.

HbA1c collection

At the 0-, 6-, and 12-month study visits, blood samples for HbA1c were analyzed centrally at the Joslin Diabetes Center's clinical laboratory (Roche Cobas Integra™; Roche Diagnostics, Indianapolis, IN, Ref. range: 4%–6% [20–42 mmol/mol]). At 3 and 9 months, HbA1c levels for Joslin participants were analyzed at the Joslin clinical laboratory, whereas HbA1c levels for participants at Texas Children's Hospital were analyzed locally using a point-of-care DCA Vantage^® analyzer (Siemens Healthcare Diagnostics). For these participants at baseline, there was simultaneous collection of the point-of-care HbA1c and a capillary sample for central laboratory HbA1c measurement. We then calculated a regression equation [Adjusted_POC_A1c = 1.590074823 + (0.850707192 × POC_A1c)] to adjust the point-of-care HbA1c value to the central HbA1c assay for participants from Texas Children's Hospital at the 3- and 9-month visits. For participants with missing HbA1c at 12 months (n = 12), we carried forward the most recent HbA1c value when there was an HbA1c after baseline (n = 6); participants who had no HbA1c data beyond baseline (n = 6) were excluded from analysis of the primary outcome.

Statistical analysis

Statistical analyses were performed with SAS 9.4 (SAS Institute, Cary, NC). Descriptive statistics included mean ± standard deviation for continuous variables and frequencies or proportions for categorical variables. The primary analysis followed the intent-to-treat principle. All randomized participants were analyzed by group assignment. Analyses included chi-square tests, unpaired and paired t-tests, analysis of variance, general linear models, and multivariable longitudinal mixed modeling.

We constructed repeated-measures multivariable mixed linear models to examine the impact of treatment group and other potentially significant covariates on HbA1c over time. Variables for inclusion in the models were specified a priori based on clinical relevance and known predictors of glycemic control; variables included treatment group, sex, diabetes duration, BG monitoring frequency, CGM use, and pump use. HbA1c at each unique timepoint was accounted for by using time as a repeated-measures variable. Age was not included, as the age range of participants in the study was narrow. Site was included in the model as a random-effect variable to account for any heterogeneity between the two participating centers.

We performed secondary analyses assessing the separate impacts of the received dose of the TW intervention and the text message response rate on glycemic control. We evaluated the number of TW sessions received by an individual in the TW and TW+Text groups; there were four TW sessions relating to BG monitoring and insulin bolus dosing self-care. As the primary outcome was assessed at 12 months, simultaneously with receipt of the fifth module (self-advocacy), TW dose was based upon receipt of the first four modules. In secondary analyses that included the entire sample, those in the Text and Usual Care groups were assigned 0 for the number of TW sessions received.

In the Text and TW+Text groups, for the primary intention-to-treat analysis, we calculated the text message response rate by dividing the total number of BG responses sent by a participant by the total maximum number of text message reminders intended to be sent during the teen's study participation (based on the text messaging schedule, assuming no resets, and excluding any second reminders sent after 10 min of original reminder). In secondary analyses that included the entire sample, those in the TW and Usual Care groups were assigned 0 for the text message response rate.

A 5% (P < 0.05) level of significance was used in the primary analyses without adjustment for the 2 × 2 factor design. A value of P < 0.01 was considered significant in the secondary analyses.

Results

Participants

A total of 569 eligible teens were identified and approached; of those, 310 (54%) consented to participate, although 3 withdrew before randomization. Those who declined were 0.3 years older (P < 0.01) and had diabetes for 1.5 years longer (P < 0.0001) than those who enrolled. There were no significant differences in sex distribution, treatment modality, or HbA1c between those who enrolled and those who declined participation. The 307 randomized participants included 5 pairs of siblings; the 2 siblings in each pair were randomized to the same group based upon the sibling with longer diabetes duration. Only the sibling with the longer duration was included in analyses to avoid need for adjustment. During the study, one other randomized participant was identified as having maturity-onset diabetes of youth and was therefore excluded from analyses. This yielded a final sample size of 301 evaluable participants at baseline. Of these, six participants were either lost to follow-up, transferred care, or withdrew from the study before the 3-month visit, resulting in a final sample of 295 participants with follow-up data (Fig. 1).

FIG. 1.

CONSORT trial flow diagram. The CONSORT flow diagram shows participant flow through each stage of the randomized controlled trial (enrollment, intervention allocation, follow-up, and completion).

Baseline characteristics are given in Table 1. Teens (N = 301, 51% female) had mean age of 15.0 ± 1.3 years and diabetes duration of 6.5 ± 3.7 years. The majority (59%) of participants used insulin pump therapy; only 12% used CGM. Mean HbA1c was 8.5% ± 1.1% (69 ± 12 mmol/mol). There were no significant differences for any baseline characteristics across the four study groups (Table 1).

Table 1.

Baseline Characteristics

	All participants (N = 301)	Usual Care (n = 76)	TW only (n = 74)	Text only (n = 74)	TW+Text (n = 77)	P*
Age (years)	15.0 ± 1.3	15.1 ± 1.3	15.0 ± 1.4	14.9 ± 1.3	14.9 ± 1.2	0.69
Diabetes duration (years)	6.5 ± 3.7	5.8 ± 3.5	6.2 ± 3.5	7.3 ± 3.7	6.9 ± 3.9	0.06
Sex, female	51%	53%	43%	55%	51%	0.49
Race/ethnicity, white	78%	80%	74%	78%	78%	0.85
Family structure, 2-parent	84%	87%	85%	84%	82%	0.85
Household income, ≥$100K	52%	51%	53%	52%	53%	0.99
Parental education, college+	69%	70%	61%	72%	73%	0.38
BMI z-score (SDS)	0.82 ± 0.81	0.90 ± 0.78	0.82 ± 0.90	0.86 ± 0.80	0.69 ± 0.73	0.39
Daily insulin dose (U/kg)	0.96 ± 0.26	0.93 ± 0.26	1.00 ± 0.28	0.97 ± 0.26	0.95 ± 0.25	0.33
BG monitoring (times/day)	4.5 ± 1.9	4.5 ± 2.1	4.2 ± 1.5	4.6 ± 2.1	4.8 ± 1.9	0.35
A1c (%) [mmol/mol]	8.5 ± 1.1 [69 ± 12]	8.4 ± 1.2 [68 ± 13]	8.6 ± 1.0 [70 ± 11]	8.6 ± 1.1 [70 ± 12]	8.4 ± 1.1 [68 ± 12]	0.66
Pump use	59%	54%	61%	62%	61%	0.72
CGM use	12%	9%	15%	12%	13%	0.76

^*P-value for ANOVA or chi-square for difference between all four groups.

ANOVA, analysis of variance; BG, blood glucose; BMI, body mass index; CGM, continuous glucose monitoring; SDS, standard deviation score.

Study retention and intervention delivery

Of the 301 teens who were eligible and randomized, 280 (93%) completed the 1-year study; completion rate did not differ by group assignment (P = 0.49) (Fig. 1). Majority of participants (64%) completed four of the four possible visits before the 12-month visit; visit completion did not vary significantly between groups (P = 0.13).

Among the teens randomized to the Teenwork intervention (TW group and TW+Text group), most participants (75%) completed all four TW sessions related to BG monitoring and insulin bolus dosing self-care. Among the teens randomized to receive text messages (Text group and TW+Text group), 60% of teens advanced according to the predetermined schedule and never required a reset (i.e., never went 2 weeks without replying to any text reminders), whereas 13% were reset once, 12% were reset two or three times, and 15% were reset four or more times.

Glycemic control over time

Mean HbA1c increased for the entire study sample over the 12 months from 8.5% ± 1.1% (69 ± 12 mmol/mol) to 8.7% ± 1.2% (72 ± 13 mmol/mol) (P = 0.006). However, the primary outcome of change in HbA1c from 0 to 12 months, with adjustment for baseline HbA1c, did not differ among the four study groups. There were also no differences in HbA1c across the study groups at any of the interval follow-up visits (3, 6, or 9 months) (Fig. 2).

FIG. 2.

HbA1c over time. Mean HbA1c level is shown at each study time point for each study group. There were no differences in HbA1c among the study groups at any of the study time points (baseline, 3, 6, 9, or 12 months). HbA1c, hemoglobin A1c.

In a linear-mixed model assessing HbA1c over time with adjustment for sex, diabetes duration, BG monitoring frequency, CGM use, pump use, and time since study entry, group assignment did not predict change in HbA1c, over time. In this model, the only variables that were independently related to lower HbA1c over time were higher BG monitoring frequency (P < 0.0001), CGM use (P = 0.003), and pump use (P = 0.0002).

BG monitoring

Mean BG monitoring frequency decreased for the entire study sample over the 12 months from 4.5 ± 1.9 to 4.1 ± 1.9 times per day (P < 0.0001). This decrease was statistically significant in all study groups except the Usual Care group (Fig. 3). At the 3-month time point, BG monitoring frequency was significantly higher in the TW+Text group compared with the Usual Care group (P = 0.01) and the TW group (P = 0.003) but did not differ from the Text-only group (P = 0.10). There were no differences in BG monitoring frequency among the study groups at any subsequent follow-up visits (6, 9, or 12 months). There were no differences in the change in BG monitoring frequency over 1 year, controlling for baseline BG monitoring frequency, among the four treatment groups.

FIG. 3.

BG monitoring frequency over time. Mean BG monitoring frequency is shown at each study time point for each study group. At the 3-month time point, BG monitoring frequency was significantly higher in the TW+Text group compared with the Usual Care group (P = 0.01) and the TW group (P = 0.003) but did not differ from the Text-only group (P = 0.10). There were no differences in BG monitoring frequency among the groups at any other time points (0, 6, 9, or 12 months). *P < 0.05 for difference from TW+Text group, **P < 0.01 for difference from TW+Text group. BG, blood glucose.

In a linear mixed model assessing BG monitoring frequency over time, with adjustment for sex, diabetes duration, CGM use, pump use, and time since study entry, treatment group assignment did not predict change in BG monitoring frequency, whereas pump use predicted higher BG monitoring frequency (P < 0.0001). The model confirmed the significant decline in BG monitoring frequency over time (P < 0.0001). In a separate mixed model comparing those who received text reminders (Text and TW+Text groups) with those who did not (TW and Usual Care groups), receiving text message reminders did not predict BG monitoring frequency over time.

“Dose–response” of the Teenwork intervention

In a general linear model assessing change in HbA1c over 1 year adjusting for baseline HbA1c, the number of TW sessions attended over the 12-month study was not associated with change in HbA1c. In a mixed model assessing impact of dose of the TW intervention, the number of TW sessions attended over the 12-month study did not predict change in HbA1c over time, with adjustment for sex, diabetes duration, BG monitoring frequency, CGM use, and pump use. Similarly, TW session attendance did not impact change in BG monitoring frequency. In a general linear model assessing change in BG monitoring frequency over 1 year adjusting for baseline BG monitoring frequency, the number of TW sessions attended was not associated with change in BG monitoring frequency. In a mixed model assessing impact of dose of the TW intervention, the number of TW sessions attended did not predict BG monitoring frequency over time, with adjustment for sex, diabetes duration, CGM use, and pump use. Results were similar in models that included the entire study sample (n = 301) where those in the Text and Usual Care groups received no TW sessions, and in models that included only those participants receiving the TW intervention (n = 151, TW and TW+Text groups).

“Dose–response” of the Text Messaging intervention

There was no significant difference in text messaging response rate (BG text responses divided by intended text reminders) between those in the Text-only group compared with the TW+Text group (34% ± 22% vs. 29% ± 23%, P = 0.19). Text message response rates were not different based on race/ethnicity, family structure (two-parent families vs. other), family income, or parental education. For this reason, these variables were not included in the models. In a general linear model assessing change in HbA1c over 1 year adjusting for baseline HbA1c, the text message response rate was significantly associated with change in HbA1c (β = −0.96, P = 0.008) in those assigned to the Text and TW+Text groups (n = 151). In a mixed model assessing impact of text message response rate on HbA1c in those assigned to the Text and TW+Text groups with adjustment for sex, diabetes duration, BG monitoring frequency, CGM use, and pump use, a higher text message response rate was associated with lower HbA1c over time (−1.3, P < 0.0001). In that model, higher BG monitoring frequency (−0.1, P < 0.0001) and pump use (−0.3, P = 0.006) were also related to lower HbA1c over time. For example, if a participant responded to every scheduled text message, HbA1c would be 1.3% lower at any given timepoint compared with those responding to none. In addition, in the model, pump users were predicted to have a mean HbA1c that was 0.3% lower than nonpump users at any given timepoint.

We ran a separate mixed model assessing impact of text message response rate on BG monitoring frequency in those assigned to the Text and TW+Text groups with adjustment for sex, diabetes duration, CGM use, and pump use; a higher text message response rate was associated with higher BG monitoring frequency over time (+2.87, P < 0.0001), and pump use was also related to higher BG monitoring frequency over time (+0.62, P = 0.001).

For the analyses above, we ran separate models that included the entire study population (n = 301) where those in the TW and Usual Care groups received no text messages (and response rate was 0) to assess the impact of text message response rate on HbA1c and BG monitoring frequency over 1 year. In these models including the entire study population, results were similar as those above that included only those participants receiving the text messaging intervention (n = 151, Text and TW+Text groups). All models adjusted for baseline covariates as above in the multivariable mixed linear models.

Safety outcomes

Over the 12-month study, there were 33 diabetes-related emergency department visits (26 participants) and 11 diabetes-related hospitalizations (11 participants). There were 92 episodes (51 participants) of severe hypoglycemia requiring assistance by another person for oral treatment, and 16 episodes (14 participants) of severe hypoglycemia resulting in seizure or loss of consciousness or requiring IV dextrose or glucagon. For each of these events, there was no significant difference across the four study groups in the percent of participants who experienced one or more event during the study.

Discussion

The aim of this study was to improve glycemic control in teens with T1D by increasing attention to diabetes self-care behaviors, specifically BG monitoring frequency and insulin bolus dosing. The interventions utilized motivational interviewing techniques and problem-solving strategies targeting self-care behaviors (TW) and text reminders to encourage teens to check BG levels (through Text). Neither the behavioral intervention, the text messaging intervention, nor the combination of the two, showed efficacy with respect to the primary outcome of change in HbA1c over 1 year. However, engagement with the text intervention (evidenced by a higher text response rate), after adjustment for baseline BG monitoring frequency and HbA1c, was associated with improvement in HbA1c.

Studies of behavioral interventions that improve glycemic control in teens with T1D have yielded variable results. In a multicenter randomized controlled trial of motivational interviewing in teens with T1D, Channon et al.¹⁹ demonstrated significant reduction in HbA1c over 12 months that was sustained at 24 months in the intervention group. In a family-based intervention, Wysocki et al.²⁰ showed improvement in glycemic control among adolescents with high baseline HbA1c. Conversely, a recent 18-month randomized controlled trial of an adaptive behavioral intervention (Flexible Lifestyles Empowering Change; FLEX) for adolescents with T1D did not demonstrate efficacy with respect to glycemic control.²¹

Studies of text messaging interventions in youth with T1D have frequently observed substantial drop-off in responsiveness and efficacy over time, particularly when there has been automated text messaging devoid of human interactions.^9,22,23 Our text reminder system did include nonprescriptive, automated feedback regarding the act of BG monitoring, without attention to actual BG value to assist in self-management decision-making. Therefore, future design of mobile technology systems should likely require incorporation of actionable feedback along with some form of human interaction and ongoing monitoring and support.⁹ Thus, there is a need for tailored interventions that meet the needs and preferences of this high-risk population. Central to this is the utilization of mixed methods research and qualitative work to inquire exactly what the adolescent population needs and desires to maintain teen self-care engagement.

Our findings of improved glycemic control with text message responsiveness are promising. This effect may be mediated by the increase in BG monitoring frequency observed among participants with higher text message responsiveness. This study focused on BG monitoring frequency as increased BG monitoring frequency has repeatedly been associated with lower HbA1c levels.^12,13 Other recent publications have focused on increasing BG monitoring frequency through financial incentives to improve HbA1c.^24–26 These studies, and this study, were performed in an era that pre-dated nonadjunctive use of CGM. Since completion of this project, CGM use has become increasingly prevalent; recent data from the T1D Exchange notes an increase in CGM use from 3% in 2010–2012 to 24% in 2016–2018 for the adolescent population of similar age to our current study sample.¹ Although more frequent use of CGM has been associated with improved glycemic control,²⁷ many CGM users may not be using the glucose data for optimal insulin bolus dosing and other self-care behaviors. In fact, text message reminders may have utility in the CGM era to remind users to look at their glucose readings and to use them for insulin bolus dosing.

Of note, our analysis of dose response of text messaging was based on the intention-to-treat principle: text message response rate was the number of BG responses divided by the total number of scheduled reminders according to the text messaging schedule during the time of the teens' study participation. In reality, there were time periods when teens did not have access to their cell phone to respond to text message reminders (e.g., when they were at summer camp); teens generally informed study staff of these time periods and reminder messages were not sent; hence, this time was not counted as nonresponse time for schedule resets. On the contrary, teens who did not respond to any messages for 2 weeks (apart from known time away at camp, when traveling abroad, etc.) were reset to initial frequency of one reminder daily on weekend days. To consider that some participants received fewer text messages (and thus had fewer actual opportunities to respond), we also calculated a response rate that was the number of BG responses a participant sent during the 12-month study period divided by the total number of reminders that a participant actually received during the 12-month study period. For this calculation, the denominator was lower for participants who took a break from receiving texts or were reset because of lack of responsiveness, yielding a higher response rate. Nonetheless, our findings were no different when using this alternative response rate, that is, Text message responsiveness remained a significant predictor of HbA1c and BG monitoring over time.

Our unadjusted data were consistent with known patterns of rising HbA1c and declining frequency of daily BG monitoring in teens over time.^1,28,29 Of interest, when modifiable factors such as pump use and BG monitoring frequency were included in our mixed models, the trend of significant worsening of HbA1c over time was no longer present.

One study limitation is the lack of objectively collected data on number of insulin boluses per day or number of missed insulin doses per day, and relationship to meals or snacks for the study participants, especially given that insulin bolus frequency was one of the two primary behaviors targeted by the Teenwork intervention. Future studies should assess self-care related to this important management behavior. In addition, we recognize that our study population had a relatively high baseline frequency of BG monitoring, potentially reflecting high baseline engagement with diabetes self-care, and potentially impacting generalizability to other samples of teens with T1D. Another limitation stems from the predetermined frequency of text message reminders in our study procedures; participants may have preferred to self-select the number of daily reminders, although the text system did allow participants to choose the times for reminders. Finally, these interventions were of relative low intensity, and more clinically responsive interventions, such as recommending insulin dose adjustments based on BG text responses, may have impacted HbA1c more directly. Future studies can consider more intense or more clinically responsive interventions, potentially involving remote or virtual outreach to teens with T1D in efforts to improve their glycemic control.

Despite these limitations, our study included a large, diverse sample of teens receiving care at two geographically distinct centers in the United Sates. There was substantial retention of study participants during the 1-year study and we did demonstrate a benefit of the text messaging component on glycemic control, although neither the behavioral intervention, the text messaging intervention, nor the combination impacted change in HbA1c after 12 months. However, those who responded to the text reminders increased their frequency of BG monitoring, which in turn likely led to improved glycemic control. As the number of teens using nonadjunctive CGM continues to increase, text messaging may have a role in reminding teens to use their CGM data in self-management, especially for insulin bolus dosing. Furthermore, newer Bluetooth-enabled devices to collect both glucose and insulin delivery data may allow for remote assessment and subsequent clinical guidance.³⁰ Future studies can consider use of interactive two-way text messaging or other more intensive video modalities for teens to interact with members of their diabetes health care team at more frequent intervals apart from routine clinic visit using remote data capture and other innovative strategies. Such approaches should be the topics of future investigations.

Disclaimer

The content is solely the responsibility of the authors and does not necessarily represent the official views of these organizations. The study sponsors were not involved in designing the study; collecting, analyzing, or interpreting the data; writing the article; or deciding to submit the article for publication.

Author Disclosure Statement

No competing financial interests exist.

Funding Information

This research was supported by the National Institutes of Health under grants R01DK095273, K12DK094721, and P30DK036836, the JDRF under grant 2-SRA-2014-253-M-B, the Katherine Adler Astrove Youth Education Fund, the Maria Griffin Drury Pediatric Fund, and the Eleanor Chesterman Beatson Fund.

Supplementary Material

Supplementary Table S1

Supplementary Table S2

Supplementary Table S3

Supplementary Table S4