A Text Messaging Intervention With Financial Incentive for Adolescents With Type 1 Diabetes

Abstract

Background:

Adolescents with type 1 diabetes (T1D) have higher hemoglobin A1C (HbA1c) levels than others. In general, adolescents engage with text messaging (TM) and financial incentives, both associated with improved diabetes outcomes. This study aimed to assess the impact of a TM intervention with financial incentives on self-care behaviors and HbA1c.

Methods:

A six-month randomized controlled trial compared MyDiaText™, a TM education and support application, with standard care. The sample included 166 teens with T1D, 12-18 years old, attending a diabetes clinic. The intervention group received one daily TM and were instructed to respond. Participants who responded to TMs for the most consecutive days were eligible for a financial reward biweekly via lottery. All participants received prompts to complete the self-care inventory (SCI) at baseline, 90, and 180 days. HbA1c was collected at clinic visits. Changes in SCI and HbA1c were analyzed using a multilevel mixed-effects linear regression model. Intention-to-treat and per-protocol analyses were performed.

Results:

The median TM response rate was 59% (interquartile range 40.1%-85.2%) and decreased over time. After adjustment for baseline characteristics, in per-protocol analysis, there was a statistically significant difference in SCI score increase in those receiving one TM per day vs control (P = .035). HbA1c decreased overall, without significant difference between groups (P = .786).

Conclusions:

A TM intervention with financial incentives for adolescents with T1D in suboptimal control was associated with increasing self-care report; however, glycemic control did not differ from controls. Further research is needed to develop digital health interventions that will impact glycemic control.

Introduction

The majority of adolescents with type 1 diabetes (T1D) fail to achieve the 2018 International Society for Pediatric and Adolescent Diabetes and 2020 American Diabetes Association hemoglobin A1c (HbA1c) targets of <7.0%,^1,2 likely due to multiple factors related to pubertal growth, ongoing neurocognitive development, and competing socioemotional demands. ³

To help engage adolescents with T1D in their self-care, it is appealing to consider mobile technologies, given that 88% of teens own a cell phone and >50% text with friends daily. ⁴ Surveys of teens with T1D demonstrated that they prefer text messaging (TM) over other communication technology to help manage diabetes.^5,6 Furthermore, studies suggest T1D education and support via TM may improve outcomes in adolescents via reminders ⁷ or education and support.^8,9 For example, “Sweet Talk,” a TM intervention, was associated with improved HbA1c when used with an intensive insulin regimen. ¹⁰ “Superego,” another TM intervention, was associated with stable as opposed to rising HbA1c levels in the control group. ¹¹ In another pilot TM intervention, adolescents receiving TMs reported higher adherence to physical activity and nutritional goals.^7,8,12 Systematic reviews of TM interventions in youth with T1D concluded that additional larger studies, grounded in behavioral theory and addressing TM timing/frequency, are needed.^13,14

Given the recognition that TM may have limited engagement for adolescents with T1D, we sought to enhance its usability by incorporating external motivation. Financial incentives have been used to encourage self-care behaviors (SCBs) related to blood glucose (BG) monitoring in adolescents with T1D.^15,16 Additional bonus incentives were awarded based on durability of the BG monitoring frequency on consecutive days. ¹⁵ Further, a lottery system may add further external motivation.^17-19 Given the inordinate challenge of improving glycemic control in adolescents with T1D, the current study combined TMs with financial incentives as an intervention compared with routine care aimed at improving diabetes SCBs. The six-month randomized controlled trial utilized the web-based TM application MyDiaText™, which had demonstrated feasibility in adolescents with T1D. ²⁰

Methods

Study Design and Participants

The trial was conducted from August 1, 2018 to October 1, 2019 following review and approval by the local institutional review board (#16-01332). All participants provided signed informed consent (≥18 years) or assent (12-17 years) along with parent/guardian consent as required. Inclusion criteria included youth with T1D ages 12-18 years, duration of T1D diagnosis of at least 1 year, and a clinic visit to the Children’s Hospital of Philadelphia (CHOP) Diabetes Center for Children within the last 6 months, HbA1c ≥8.0%, insurance approval for the point-of-care (POC) HbA1c testing at CHOP, English comprehension, and youth ownership of mobile phone with unlimited TM. Adolescents with a major cognitive or organ disability or who had taken steroid medication in the past three months were ineligible. The study was registered on ClinicalTrials.gov (#NCT02927639).

Participants were screened for eligibility by review of the electronic health record (EHR). At enrollment, participants confirmed presence of mobile phone for TM and provided sociodemographic and contact information. They then chose, without input from guardians or providers, an AADE 7 SCB ²¹ (Figure 1) for which they sought improvement. All participants received the first Self-Care Inventory (SCI) ²² link via TM at enrollment.

Figure 1.

AADE7 SCBs selected by 165 study participants at enrollment. Percentage of total participants selecting SCB are indicated above each bar.

SCB, self-care behavior.

Randomization

Upon study enrollment and establishment of a web-based account, participants were randomly assigned to TM intervention or control arms. Participants, study staff, technology support personnel, and statistician were unblinded due to study design and need for application technical support.

Procedures

In addition to assessing eligibility, the EHR was used to extract clinical information used in the analysis. HbA1c levels were recorded from clinic visits during the study period (generally every three months). POC HbA1c results were measured on the DCA Vantage analyzer (reference range 3.8%-5.9%; maximum reportable numeric value “>14%”); levels >14% were recorded as 14.1% in the study database. Clinical depression included depression in the patient’s problem list or mention of “depression” or “coping” in the narrative assessment. Mental health counseling was assessed by encounters with team psychologist or provider report of patient receiving outside counseling. Medical problems other than T1D were recorded from the patient’s EHR problem list.

The previously validated SCI was used, with permission from the author, to assess SCBs. ²² The SCI is a 14-item self-report measure developed to assess the patients’ perceptions of the degree to which they adhered to treatment recommendations in the past month, a higher score indicating more self-care. The questions were transcribed on to the website, and radiobuttons allowed participants to select a response from 1 to 5, or N/A, adding up to a total possible 70 points. Scores could be reported either as a mean of 1-5 out of 5, or a total score out of 70 as in previous studies of adolescents.^23,24 Questions answered “N/A” were not included in the mean score calculation. Advantages of the SCI include assessing self-care in patients managed with varying insulin regimens, internal consistencies over 0.80 in various pediatric studies, and previous studies demonstrating that adolescents’ SCI scores predicted glycemic control independent of anxiety and depression. ²² All participants were prompted to complete three SCIs with compensation of $5, $10, and $15 for completed surveys at baseline, 90 days, and 180 days, respectively. TM reminders were sent to fill out SCI2 and 3. If they were not completed after initial prompt, weekly reminders were sent out for two weeks. Along with SCI3, intervention participants received an “Exit Survey” at study conclusion.

TM Intervention

Participants in the intervention arm received one daily TM, defining the per-protocol (PP) intervention. The Twilio REST API services were used to send TMs. The TM defaulted to a 4 PM send-time based on results of focus groups of teenagers with T1D, ²⁰ but participants had the opportunity to change this time as they preferred. Participants who texted “STOP” did not receive further text messages and had exit survey sent to them via email. Messages had been created by a multidisciplinary team of T1D experts, including healthcare providers, certified diabetes educators, and nutritionists. Intervention participants received both declarative messages related to their preselected SCB and quiz-type questions about general T1D knowledge with automated replies indicating the correct answer. TMs were shown to be useful and enjoyable by study participants in a feasibility study of the same web application. ²⁰ Figure 2 shows example TMs in the way they may have been seen by study participants. TMs were selected randomly. Participants could receive repeated messages once the message pool was exhausted during the six months. Participants were asked to respond in any way to the TMs to indicate that it had been read. Every 2 weeks, one participant received a $10 reward via lottery drawn from the participants with the longest period of consecutive days with TM response (maximum of 14 responses every 2 weeks). Payments were administered via reloadable debit card.

Figure 2.

Example declarative (a) and quiz-type (b) text messages as they may have been viewed by study participants.

Outcomes

The primary study outcome was change in SCI score from day 0 to 180. The secondary outcome was change in HbA1c over this time period. Additionally, response rate, calculated as TM responses divided by days on study, determined engagement with the TM intervention. In a post hoc analysis, characteristics associated with higher engagement with TM were assessed.

Statistical Analysis

A priori sample size calculations were based on SCI score. A clinically significant change in SCI score was considered half of a standard deviation (SD) based on a study of 164 youth ages 11-18 who had a mean SCI score of 36.9 (SD 6.6) out of 70; therefore, 3.3 out of 70 was considered a clinically significant difference in SCI between groups. ²⁴ With 80% power, two-sided type 1 error of P = .05, and 20% dropout rate, we estimated a target sample size of 160 participants.

A multilevel mixed-effects linear regression model assessed for differences in change in SCI score as well as HbA1c from enrollment to six months, in both an intention-to-treat (ITT) and PP analysis. These models allowed for missing and misaligned data due to nonadherence to timely SCI completion and/or clinical visits with POC HbA1c that were not reliable at baseline, three months, and six months by modeling actual time of assessment as a continuous measure. The models were adjusted for characteristics found to be different between groups at baseline. HbA1c levels obtained more than 90 days prior to enrollment were not included in the outcome analysis. SCI and HbA1c values were accepted up to 194 days following enrollment, since SCI completion prompts were sent up to that day. Analyses were performed using STATA v15.1 (StataCorp, College Station, TX, USA). A P value of < .05 was considered statistically significant.

Results

Study Sample

A total of 166 participants were enrolled (Figure 3); 86% of those approached agreed to participate. Five participants subsequently enrolled in another interventional study; their baseline and TM response data were included in our analysis, but outcomes data were not included due to potential confounding. One teen was inadvertently enrolled in our study following previous entry into another; all data for that participant were eliminated, leaving a sample of 165 (47% female, 41% non-white) at baseline (Table 1). There were nine participants known to have texted “STOP,” causing them to not receive further TM. As none of these participants actively elected to drop out of the study, their data were included.

Figure 3.

Flowchart of participant recruitment and progression through study.

HbA1c, hemoglobin A1c; SCI, Self-Care Inventory; TM, text messaging.

^*No outcomes data analyzed; ^**Baseline characteristics, HbA1c, and TM response data included in analysis; ^***SCI and HbA1c prior to co-enrollment only; all TM response data included.

Table 1.

Baseline Characteristics of Control and Intervention Groups.

Parameter	Intervention (n = 83)	Control (n = 82)	P
Age in years (SD)	15.8 (1.9)	15.5 (2.3)	.40
Gender (%)			.41
- Male	47 (56.6%)	38 (46.3%)
- Female	35 (42.2%)	43 (52.4%)
- Other	1 (1.2%)	1 (1.2%)
Race (%)			.35
- White	43 (51.8%)	54 (65.8%)
- Black	26 (31.3%)	20 (24.4%)
- Asian	1 (1.2%)	1 (1.2%)
- Mixed race	8 (9.6%)	3 (3.7%)
- Other	5 (6.0%)	4 (4.9%)
Ethnicity (%)			.18
- Hispanic	14 (16.9%)	8 (9.8%)
- Non-Hispanic	69 (83.1)	74 (90.2)
Diabetes duration, years (SD)	7.2 (4.2)	7.0 (3.8)	.65
Baseline HbA1c, % (SD)	10.1 (1.9)	9.8 (1.7)	.22
Baseline mean SCI score (SD)	2.7 (1.5)	3.1 (1.4)	.11
BMI percentile (SD)	73.9 (23.8)	74.1 (24.3)	.90
Technology use (%)
- Pump	46 (55.4%)	48 (58.5%)	.16
- CGM	54 (65.1%)	48 (58.5%)	.74
- Closed loop	5 (6.0%)	4 (4.9%)	.11
Mental health services (%)
- Depression and/or coping	23 (27.7%)	11 (13.4%)	.02
- Receiving treatment	17 (20.5%)	7 (8.5%)	.03

P values were generated using t-test and chi-square, where appropriate. Significant differences between groups indicated by P value less than .05 are indicated in bold.

BMI, body mass index; CGM, continuous glucose monitoring; HbA1c, hemoglobin A1c; SCI, Self-Care Inventory; SD, standard deviation.

Baseline characteristics including age, diabetes duration, baseline HbA1c, gender, ethnicity, race, BMI, use of diabetes technology, and other medical problems are detailed in Table 1; there were no significant differences in these characteristics between control and intervention groups. However, the intervention group had a significantly higher proportion of youth with clinical concern for depression, 27.7% of the TM group vs 13.3% of the control group (P = .02), and mental health counseling, 20.5% of the TM group vs 8.5% of the control group (P = .03). Over half the sample used technologies including continuous glucose monitors (61.8%), insulin pumps (57.0%), and/or hybrid closed-loop systems (5.5%). The most commonly selected SCBs were “Monitoring” and “Healthy Eating,” selected by 38% and 24% of participants, respectively (Figure 1). Completion of SCI and HbA1c testing decreased over the course of the study (Figure 3).

TM Intervention Delivery

A system malfunction began in June 2019 resulting in delivery of two immediately sequential TMs instead of one, differing from the planned protocol intent of one TM daily. This issue affected 29 (24.9%) of intervention participants toward the end of the study (median days on study at time of error 132 days, interquartile range [IQR] 127-153 days). To harmonize evaluation of the intervention as planned, we performed a PP analysis. Data for each affected participant were truncated when the two daily TMs began for them. This resulted in a loss of 4 SCI scores and 12 HbA1c levels attributable to 15 participants total.

SCI

A multilevel mixed-effects linear regression, adjusted for depression and counseling treatment, showed a significant difference between the TM and control groups in the change in SCI score from enrollment to 180 days in the PP, truncated data analysis (P = .035), but not in the ITT analysis (P = .164) (Figure 4). Notably, participants in the intervention group demonstrated a predicted rate of change over twice that of the control group in the PP analysis: score change of 0.47 (confidence interval [CI] 0.28, 0.66) in 180 days, vs score change of 0.21 (CI 0.05, 0.36) in the control group. The ITT analysis resulted in a slightly different score change in the intervention group of 0.37 (CI 0.20, 0.55).

Figure 4.

Multilevel mixed-effects linear regression model-predicted values of SCI and HbA1c. (a) and (c) Show predicted mean SCI score over study period, intention-to-treat and per-protocol, respectively. (b) and (d) Show predicted mean HbA1c score over study period, intention-to-treat and per-protocol, respectively. Only predicted SCI score change in the per-protocol group showed a significant difference between groups and is shown in bold (P = .035).

HbA1c, hemoglobin A1c; SCI, Self-Care Inventory; TM, text messaging.

Hemoglobin A1c

The predicted HbA1c levels of both groups decreased over the study period. The mixed-model analysis showed no significant difference in HbA1c decrements between groups in either ITT (P = .876) or PP analysis (P = .786) (Figure 4).

TM Response and Engagement

Response rates were calculated for the intervention group. The mean response rate from ITT data was 57.7% (SD 29.9%) and median 61.0% (IQR 34.5%-85.0%). The mean response rate using PP data was 58.6% (SD 28.9%) and median 62.7% (IQR 40.1%-85.2%) (Figure 5). Response rates were not significantly associated with age, diabetes duration, gender, race, ethnicity, baseline HbA1c, use of diabetes technology, use of mental health services, or number of other medical problems.

Figure 5.

Response rate over time using all data as used in intention-to-treat analysis (orange) and only one TM per day as used in per-protocol analysis (blue). Response rates were calculated as TM responses received divided by days on study.

TM, text message.

Participant Perspectives

Survey results from a limited number of participants (n = 20) in the intervention group appear promising. The majority (80%) reported that the application helped their diabetes self-care; 85% reported that they enjoyed receiving TM’s “somewhat” or “a lot”; and 75% responded that the frequency of TMs was “just right.” However, 80% reported the TMs arrived too late in the day. Participant quotations included, “It was kind of like a reminder to care of [sic] myself in the midst of my chaotic life,” and “I liked that these messages are inspirational and very insightful.”

Discussion

In our study of a diverse sample of adolescents with T1D and suboptimal glycemic control, an intervention delivering one TM per day along with financial incentives showed promise for improving self-reported SCBs compared with standard care. To our knowledge, this is one of the first studies to combine TM motivation and education with a financial incentive reward paradigm. While there were issues with attrition and technology, our mixed model showed a greater predicted rate of change in SCI scores favoring the intervention group in the PP analysis, and participant feedback about the intervention was generally positive.

There was no significant difference in SCI change scores in the ITT analysis. This could be due to burnout in the group receiving two TMs per day due to the software malfunction. Alternatively, results may have been confounded by the timing of initiation of the erroneous delivery of two TMs per day, as it began close to the end of the school year. Nonetheless, there is promise in the observation that SCBs were positively improved in teens receiving a TM intervention delivering one TM per day along with financial incentives.

Given the high baseline levels of HbA1c, it is not surprising that youth in both groups experienced a decrease in HbA1c over the study period; as such, the intervention did not lead to a significant difference when compared with the control group. Previous studies of TM interventions and financial incentives used individually similarly did not reveal differences in HbA1c between groups.^11,12,16,25 Participants may have experienced a study effect and/or elected to participate because they were already motivated to improve their glucose control. Of note, the SCI has been correlated with glycemic control ²⁴ but not in the current investigation; perhaps if our sample was larger, experienced less attrition, or if follow-up was longer, we may have observed this as well.

Response rate to the intervention decreased over time. The overall mean response rates of 58.6% in the PP analysis and 57.7% in the ITT analysis were similar to the past reports. Previous TM studies in adolescents with T1D showed that engagement was 60% over six months ⁸ and 76% over a shorter period of eight weeks. ²⁵ Both these studies also showed a decline in response rate over time.^8,25 As our control group did not receive TMs, our results do not establish to what degree the financial incentive promoted engagement.

Our study encountered some limitations. Despite financial compensation for time and effort, few participants completed the 180-day SCI. The log-in step required to fill out the SCI could have been a potential barrier to survey completion. It is possible that the $5-15 compensation for survey completion was modest or that an additional guaranteed incentive in addition to the lottery would have promoted engagement. In-person or telephone follow-up by research staff to those who did not complete the SCI when reminded via TM might have increased completion rate. While the mixed model created a calculated estimate of mean SCI scores at a given time point, the accuracy of this estimate decreased with the increasing loss of data due to attrition and/or with removal of data points in the PP analysis. Despite low completion of the final SCI, over one-third of participants were still responding to TMs at study end, perhaps indicating more promising engagement with the TM intervention than with study surveys. Another potential limitation relates to the use of a third-party TM service, which maintained propriety error reporting and recordkeeping structure. This somewhat limited assessments of error messages relating to TM delivery and receipt, including that which caused double TMs to be sent. Newer services such as Way To Health that integrate TM into the application itself may limit these issues. ²⁶ Finally, the maximum numerical HbA1c value was 14%; those coded as 14.1% could represent a range of HbA1cs above 14%. Fortunately, this affected both groups similarly, with five in control and six in intervention.

Although assignment was random, we found that the intervention participants had over twice the prevalence of clinical concern for depression and mental health treatment as recorded in the EHR. The mixed model analyses controlled for this difference. It is possible that those with clinical depression and/or a history of counseling visits may have been more receptive to the intervention. Studies in digital health have shown lower engagement in interventions by those with mental health issues, but this group’s susceptibility to the effects of the intervention is less clear. ²⁷

Despite the above limitations, the relatively persistent TM response and changes in self-reported SCBs indicate that TMs with financial incentives remain an important area of research for adolescents with T1D and suboptimal control. For future studies, attrition could be decreased through one-step survey completion, personal follow-up, and increased compensation. Technological issues could be reduced by utilizing a healthcare-specific TM service. ²⁶ The influence of the financial incentive could be better understood by comparing a control group with both a TM-only and TM plus financial incentive condition.

Conclusion

As a group, adolescents with T1D have suboptimal glucose control and are highly engaged in communication technology, specifically with TM. In our study, we were able to recruit a large number of such adolescents into a randomized controlled trial of a TM intervention with financial incentives that aimed to increase diabetes self-care. Results showed persistent engagement with and potential for increase in self-care using this intervention. We have demonstrated that TM is a promising method by which to engage adolescents with T1D and suboptimal control in their self-care and deserves further investigation. There remains a need to develop a digital health intervention that significantly impacts glycemic control in this group.