Improved Medication Adherence and Frequency of Blood Glucose Self-Testing Using an m-Health Platform Versus Usual Care in a Multisite Randomized Clinical Trial Among People with Type 2 Diabetes in India

Introduction

India is home to over 69.2 million people living with diabetes, about 1/6th of people with diabetes worldwide. ¹ The national prevalence in 2015 was estimated at 8.7% and is expected to rise to 10.5% by 2035 due to a combination of rapid urbanization, increasing life expectancy, and a diet containing significant amounts of sugar and fat. ¹

Uncontrolled diabetes is associated with an array of negative macrovascular and microvascular outcomes, including myocardial infarction, stroke, blindness, renal failure, and amputation. Close management of diabetes through exercise, diet, medication, and blood glucose (BG) self-testing can significantly reduce these risks. ² Many people struggle with maintaining these self-care behaviors and managing their BG consistently between physician visits ^3,4 as >5,000 waking hours each year are spent self-managing. As a result, in India less than half of people living with diabetes achieved the target A1c level of < 7% (53 mmol/mol) in 2011. ^{5 –7}

Although India's growing prosperity has contributed to the lifestyle changes leading to diabetes, it has also created new ways of reaching people. There are over 70 mobile phones per 100 people in India. ⁸ Although only 14% of the population owned smartphones in 2015, the number has been increasing exponentially year to year. ⁹

Globally, there is growing evidence for the benefit of telemedicine interventions. Although initial reviews found inconsistent impact, they included studies of varying quality and heterogeneous interventions and/or diseases. ^10,11 Meta-analyses restricted to high-quality studies of relatively similar interventions for diabetes show strong evidence of clinical effect. ¹² This is particularly true for interventions that support treatment modification within the electronic system ¹³ or utilize mobile phones (m-Health interventions). ¹⁴ A review of reviews confirmed the positive telemedicine findings for diabetes management. ¹⁵ However, this research has been largely based in developed countries, with limited evidence emerging from the developing world.

The Gather m-Health (mobile-health) platform provides a system for people with diabetes that supports self-management, facilitates patient–provider data exchange and communication, and enables treatment changes between routine visits. We hypothesized that use of this system would lead to improved diabetes self-care, communication with the healthcare team, and clinical outcomes.

Materials and Methods

This was a two-armed, open-label, randomized clinical trial comparing an m-Health intervention with a usual care control group. It was approved by the Astha Institutional Ethics Committee in Ahmedabad, India and the Prof. M. Viswanathan Diabetes Research Center Institutional Ethics Committee in Chennai, India. It was registered with the Clinical Trials Registry—India under #2015/02/5560 and the U.S. ClinicalTrials.gov site under NCT02587286.

Inclusion criteria were:

• Willing to participate and be randomized

Exclusion criteria were:

• Currently using an insulin pump, continuous glucose monitor, or glucocorticoids

The study was conducted in three diabetes-focused clinics: DHL Research Centre in Ahmedabad, Diabetes Action Centre in Mumbai, and Prof. M. Viswanathan Diabetes Research Centre in Chennai. All had at least one senior clinician to lead treatment and a health coach to provide regular participant interaction. Before study launch, coaches received training on the m-Health software and study protocols.

Potentially eligible participants were recruited during regularly scheduled visits. Study staff explained the project purpose and main features. If an individual was willing, initial eligibility was confirmed, informed consent obtained, and a baseline visit scheduled. At the baseline visit, staff conducted a questionnaire, physical examination, and fasting blood collection. Blood work confirmed if the individual's baseline A1c was eligible for enrollment.

The randomization sequence was investigator generated, stratified by site, with a 1:1 allocation, in blocks of 6, using the Sealed Envelope Ltd. online system. ¹⁶ The allocation sequence was concealed from implementing staff through sequentially numbered, opaque, sealed, and stamped envelopes. After a participant's eligibility was confirmed, research staff opened the next available envelope and assigned the participant to a group. At the time of randomization, participant ID and randomization date were written on the envelope and allocation paper to prevent allocation sequence tampering. After full enrollment, investigators compared randomization order and dates to the master allocation list. The outcomes assessor was blinded to assignment. There were no method changes after trial commencement.

Intervention was use of Gather Health, an m-Health diabetes management platform, comprising of a smartphone app for people with diabetes and a Web portal and smartphone app for providers. The evidence-based platform was informed by theories of behavior change, including the health belief model, ¹⁷ health action process approach, ¹⁸ theory of planned behavior, ¹⁹ and Bandura's theory of self-efficacy. ²⁰ It used reminders, data visualization, and ongoing support to increase self-care behaviors and facilitate collaborative care decisions.

The provider Web portal had three components: a directory of all participants; a unique record for each participant with medical history, laboratory results, clinic visits, medications, and BG readings; and an algorithmically generated “feed” of alerts highlighting participants who needed attention. The feed included those who had sent messages; submitted a pattern of clinical information requiring review, such as a hypoglycemic reading; or had not been contacted by a provider in a set time. The provider smartphone app included chat and a concise medical history.

All participants received free clinical consultations, free laboratory testing, and transportation stipends for all visits. They also received sufficient BG testing strips and lancets for the study duration. At each visit, they reviewed their individual medication and BG testing schedule in collaboration with a provider.

Control participants were instructed to manage their diabetes as usual during the study and offered the m-Health app free for 6 months after study completion. Intervention participants received access to the m-Health smartphone app for 6 months, and a stipend to offset the cost of their mobile phone plan. During the baseline visit, a study staff member showed an instructional video, assisted in app download, and walked participant through sign-up. Study staff entered each participant's personalized medication list and BG testing schedule into the system.

The app automatically reminded participants to complete tasks each day, BG tests submitted out of standard ranges had automated question follow-up to identify issues, and participants could message questions to providers. Each site's health coach regularly responded to patient questions and system-generated alerts. Provider contact with participants outside the system was discouraged, except in cases of high-risk glycemic data or technical troubleshooting. No changes were made to the software during the trial.

Study participant data were collected at baseline, 3-month, and 6-month visits through a tablet computer. Information on providers' demographics and satisfaction with the system were collected after conclusion of all 6-month visits.

Collected patient information included sociodemographics, physical measurements (height, weight, waist), and clinical measures (blood pressure, fasting BG, A1c, lipids). Patient-reported outcomes were collected through previously established and validated scales. Treatment satisfaction was assessed with the 18-item Patient Satisfaction Questionnaire (PSQ-18), ²¹ a short form of the 50-item Patient Satisfaction Questionnaire III (PSQ-III). ^22,23 Self-care behaviors were assessed with the Summary of Diabetes Self-Care Activities (SDSCA), a 12-item instrument. ²⁴ Diabetes distress was determined through the 5-question Problem Areas in Diabetes scale (PAID-5), ²⁵ a shortened form of the PAID scale, ²⁶ demonstrated to be psychometrically robust. Self-efficacy was assessed by the Stanford Patient Education Research Center (PERC) 6-item instrument, ²⁷ measuring a person's confidence in regularly doing tasks to manage diabetes or prevent side effects from interfering with their life. Knowledge of diabetes care was assessed through a seven-item investigator-adapted version of the RAND Improved Chronic Illness for Care Evaluation (ICICE) study. ²⁸

Patient satisfaction with the smartphone app and interaction with the care team were assessed by an investigator-generated 13-item measure on a Likert scale ranging from “very satisfied” to “very dissatisfied.” The 21-item provider satisfaction measure was adapted by investigators from the Centers for Medicare and Medicaid (CMS) IDEATel Demonstration Project. ²⁹ Questions focused on the intervention's acceptability, impact on the practice, and impact on the patients, and were either on a 5-point Likert scale or open-ended.

The primary outcome was change in A1c from baseline to 6 months. Secondary outcomes were change in A1c from baseline to 3 months as well as change in body mass index, waist circumference, blood pressure, fasting BG, and lipids from baseline to 3 and 6 months. Self reported measures of medication adherence, BG testing, communication with doctors, treatment satisfaction, diabetes self-care activities, diabetes distress, self-efficacy, and diabetes knowledge were also compared at 3- and 6-month visits versus baseline.

Sample size was calculated for a two-sample t-test with equal allocation for the primary outcome of change in A1c from baseline to 6 months. Assuming a minimum significant difference of 1% point in A1c with a standard deviation (SD) of 1.5% points and 20% attrition, 90 people were enrolled to result in 80% power at a significance level of 0.05 (two-sided).

Descriptive statistics were calculated for baseline values using means and SDs (continuous) or frequencies and percentages (categorical). As previously reported, a two-sample t-test of difference in mean change from baseline to 6 months was used as the primary test of statistical significance in an intent-to-treat analysis with both p-value and 95% confidence intervals (CIs) with subgroup analyses for gender, age, and length of time with diabetes.

Secondary analyses also used t-tests to determine the intervention effect on all other endpoints 3 and 6 months after randomization. All tests were two sided with 5% type I error and reported with both p-values and CIs. Associations between categorical variables were assessed using chi-squared or Fisher's exact tests. Descriptive statistics were conducted on patient and provider intervention satisfaction and patient usage data. All statistical analyses were performed using STATA MP Version 11.1 (Stata Corp., College Station, TX).

Results

The 90 eligible participants enrolled at baseline (Fig. 1) had a mean age of 48.4 years (SD ± 9.2) and 30% were female (Table 1). Most had owned a smartphone for longer than a year (median: 14 months, interquartile range [IQR]: 7–36) and used it for phone calls (97.8%), text messages (90.0%), and social media (64.4) (multiple responses allowed). Median diabetes duration was 10 years (IQR: 4–15) and over two-thirds visited a doctor less than the recommended quarterly schedule (68.9%). Only 32.2% tested their blood sugar more than once a week and one-third acknowledged missing a medication dose in the last week. Participants were recruited between March 20, 2015 and May 27, 2015 and followed up until January 1, 2016 at latest.

OPEN IN VIEWER

Fig. 1.

Study flowchart.

A1c results have been previously reported. ³⁰ Briefly, overall baseline mean A1c was 9.3% (SD ±1.2) and at 6 months the 80 returning participants had a mean A1c decrease of 1.5% in the intervention group and 0.8% in the usual care group, a statistically significant difference (p = 0.02; 95% CI mean difference: 0.10–1.37). There were no statistically significant differences when stratified by gender, age, or length of time with diabetes.

Among the 80 returning participants (39 control; 41 intervention), there were significantly more in the intervention group than the control group who improved self-reported medication adherence from baseline (39.0% vs. 12.8%; p = 0.03) (Table 2). Participants receiving the intervention were more likely to increase their BG testing from baseline than the control group (39.0% vs. 10.3%; p = 0.01). No other outcomes showed statistically significant differences between the treatment arms. Results from the 3-month visit were similar to those from the 6-month visit, with the following exception: differences in medication adherence were not statistically significant at 3 months. All analyses were conducted using originally assigned groups.

Participants were defined as active if they opened the app, recorded a medication, recorded a BG test, or sent a message. At week 12, 79.5% (35/44) and at week 24, 75% (33/44) of participants were active. Most (88.6%, 39/44) were active for at least half of the study weeks. Participants entered 29,668 medications and 2,575 BG readings, sent 497 messages to providers, and received 890 messages from providers.

Intervention participants had high satisfaction for all aspects of the app (80% “satisfied” or “very satisfied”), with highest satisfaction for “ability to view own data” at 95% (Table 3). All providers found the m-Health software “very acceptable” and the time they spent on it “very” or “slightly” acceptable. All found it “very” or “slightly” helpful to their practice and felt their patients also found the addition of telemedicine services to usual care “very” or “slightly” helpful. The software was perceived to have the greatest impact on the level of patient satisfaction with healthcare and the clinic's diabetes management of patients (both 57.1% “very positive impact”). No unintended effects were reported.

Discussion

This randomized trial demonstrated that the users of the Gather m-Health diabetes management platform improved their medication adherence and BG testing more than individuals in usual care over 6 months. Among intervention participants, there were high levels of app usage at 12 and 24 weeks. Both participants and providers were highly satisfied with interactions with the software. As previously reported, mean A1c reduction was 1.5% in intervention and 0.8% in control, a statistically significant difference. ³⁰ There were no significant changes in other clinical or patient-reported outcomes.

The observed changes in self-reported medication adherence and BG testing may have been the intermediate steps that generated a beneficial change on A1c. Previous research in India has shown that text message programs for people with prediabetes can help improve dietary habits ³¹ and prevent development of diabetes. ³² However, a study among people with diagnosed diabetes did not find that receiving text messages significantly improved A1c over usual care. ³³ It is possible that the richer interaction and engagement in a smartphone app, in contrast to a text message, may have been more supportive of the necessary behavior changes. Indeed, the nature of engagement with a text message, a smartphone app, or a Web portal is qualitatively different. Grouping interventions using these different mediums under the umbrella term of “telemedicine” may obscure important differences and create ambiguity over intervention effectiveness.

Unfortunately, neither the studies of text messages in India nor Web portals ^34,35 or smartphone apps ^36,37 internationally specifically report on medication outcomes or BG testing. These results must be interpreted cautiously as there is no “gold-standard” measure for treatment adherence and self-reported values are subject to recall and social desirability biases and thus are often overestimates. However, such measures are widely used to evaluate self-care, and typically correlate with more stringent measures such as blood tests and electronic pill caps.

Study strengths include a classic two-arm randomization design to control for bias and balance baseline characteristics across groups. Broad age and A1c eligibility criteria make results generalizable to the larger Indian population with diabetes. Similarly, inclusion of clinics from multiple Indian states makes results more applicable across settings.

However, results need to be evaluated within the context of the study's limitations. Only people with type 2 diabetes were included, so the impact of the m-Health app on other diabetes types is unclear. While smartphone ownership is rapidly increasing in India, <20% of the population own one. As younger and/or wealthier individuals generally use smartphones, these results may not be applicable to groups less familiar with this technology. All clinics had a diabetes care specialization and a staff member to run the system. Similar results may not be found in primary care settings or at clinics with fewer staff. Lastly, as the system is being evaluated holistically, it cannot be determined which aspect of the m-Health program led to improved outcomes.

In addition, this trial utilized a classic fixed design, where the software was frozen at study initiation. This is in sharp contrast with the typical experience of a mobile phone app, where updates and changes may be sent to users monthly or even weekly. Indeed, there have been calls to study m-Health interventions in more real-word scenarios, allowing for revision and adaptation during a trial, along a core set of behavioral principles. ³⁸ While three-quarters of users remained active at 6 months, one of the greatest side effects of m-Health interventions is user boredom. An ability to keep content and interactions fresh through updates may mitigate that concern.

The results of this study help build the evidence base for telemedicine interventions in general and m-Health interventions for diabetes in India in particular. To our knowledge, there are no studies of any smartphone-based m-Health interventions on diabetes in India of any kind, let alone one of this rigor, including a randomized control group.

There is a rising burden of diabetes in the developing world, with over three-quarters of people with diabetes living in low- or middle-income countries. ¹ Healthcare systems around the world are best suited to handling acute, emergent disease and struggle to provide the type of ongoing care most effective in managing chronic diseases. Many low and middle-income countries have healthcare systems that are further strained by limited numbers of healthcare providers and facilities. Solutions are urgently needed to extend the impact of each healthcare provider and make high-quality chronic disease care accessible. This intervention could help in this effort and provide considerable benefit to many people living with diabetes.