Background
The rapid increase in rates of diabetes poses a significant public health problem globally. Diabetes is currently estimated to affect 285 million adults worldwide, with the prevalence predicted to rise to 438 million by the year 2030 [
1]. Its complications contribute significantly to ill health, disability, poor quality of life and premature death. The associated global economic burden is projected to reach at least US$376 billion in 2030 [
2]. Although guidelines and targets for optimal diabetes management are well documented [
3], it is estimated that 40% of individuals with diabetes have sub-optimal glycaemic control [
4,
5], significantly increasing their risk of costly and debilitating diabetes-related complications [
6,
7].
Diabetes self-management education facilitates the acquisition of knowledge and skills to improve disease management and has been found to improve glycaemic control [
8], with program duration being a critical predictor of this success [
9]. Providing ongoing and long-term diabetes management support, particularly to those people living in rural and remote areas, is a major challenge for all health systems around the world. This highlights the need to develop and evaluate more feasible, accessible ways of providing such support for large numbers of people with diabetes than is traditionally offered. Using information and communication technology (ICT) to provide diabetes management education and support directly to patients offers such potential, by overcoming many of the barriers associated with more traditional modes of program delivery. Use of ICT has been shown to yield improvements in self-care knowledge and behaviour of patients and clinical outcomes associated with the prevention and control of chronic health conditions, including diabetes [
10‐
12]. Some studies have evaluated the role of automated or semi-automated telephone-delivered diabetes management interventions on glycaemic control, however, the results have been inconsistent with varying levels of reliance upon health professionals [
13‐
15].
The Telephone-Linked Care (TLC) program is an automated and interactive telephone system designed to emulate telephone encounters between patients and health professionals [
16] and to complement standard medical care. TLC systems have been previously used to effectively screen people with specific health conditions [
17,
18], promote self-care behaviours [
19‐
22] and provide monitoring of and feedback to patients with a range of chronic diseases [
23‐
26].
A randomised controlled trial was conducted to evaluate a TLC program - the Australian TLC Diabetes program - designed to improve type 2 diabetes management. This paper presents the six-month results for the study’s primary outcomes, glycosylated haemoglobin and health-related quality of life (HRQL), and it also describes the sample baseline characteristics, compared with a large Australian population study.
Results
Of the 52 individuals who did not wish to participate at the initial eligibility screening stage, the primary reason for non-participation was lack of interest (n = 21), with an additional 11 reporting potential difficulties with travel for the baseline data collection. Other reasons included lack of time due to work and other commitments. There were no age differences between those who were willing and unwilling to participate, although there was a higher proportion of women who were unwilling to participate compared with those who chose to participate (61.5% compared with 43%).
As shown in Table
2, which summarises the baseline characteristics of the TLC and usual care arms, the Australian TLC Diabetes sample had a mean age of 57.4 years (± 8.3), with a higher proportion of men (62.5%) than women. The vast majority of participants were born in Australia (70.0%), were married or cohabiting with a partner (75.0%), with education above secondary school level (65.0%). Approximately half of the sample were employed (45.8%) and had complementary private medical insurance (55.8%). The mean number of hours per week spent exercising was reported to be 6.1 (± 6.4), with the majority of the sample (55.9%) participating in the nationally-recommended level of weekly physical activity (>150 minutes of exercise per week in at least 5 sessions per week [
33]). Only 1.7% of the sample were current smokers. Approximately three quarters of the sample rated their health as good or higher (74.2%). Nearly two-thirds of the sample had been previously diagnosed by a doctor with hypertension (65.8%) and hypercholesterolaemia (63.3%), and therefore were likely to be receiving treatment for these conditions as was reflected in their blood pressure and lipid profiles that predominantly fell within the normal range.
Comparison of baseline sample characteristics between study arms
The baseline sample characteristics were compared across the usual care and intervention arms to evaluate the randomisation process (Table
2). Comparison of the baseline characteristics across usual care and intervention arms revealed important differences in e-GFR, which showed a significantly greater impairment in renal function in the intervention compared with usual care arm, and creatinine. Other differences observed were in age, education, and self-care behaviours (adherence to blood glucose testing recommendations and daily insulin/diabetes medications, and foot inspections). Adjustments were made for these variables in sensitivity analyses.
Post-intervention results at six months
Attrition
Of the total sample, 92.5% completed the six-month assessment (see Figure
1). Overall, nine participants (two women and seven men) withdrew from participation in the study, four in the intervention arm and five in the usual care arm. The reasons given for withdrawal from the usual care arm were all related to frustration at ‘missing out’ on the intervention. The participants receiving the Australian TLC Diabetes intervention withdrew for a range of reasons, including relocation, being unable to use the blood glucose meter, and disappointment with the intervention. The sociodemographic, behavioural or biological profiles were compared between those people who remained in the study and the nine people who withdrew. There were no significant differences at baseline across any of the domains of risk factor profiles.
Use of Australian TLC Diabetes system
The mean number of completed calls for the Australian TLC Diabetes participants during the six-month intervention was 18 (± 6), ranging between 2 and 27 calls, with a mean call duration of 11 minutes (± 1). The mean percentage of completed calls out of the expected weekly calls for all individuals in the intervention condition was 76% (± 22). More detailed analyses of the usage of the Australian TLC Diabetes system are beyond the scope of this paper and are to be presented in a future manuscript.
A small number of people in the intervention arm (n = 5) discontinued participation in the intervention but still completed the six-month assessment (Figure
1). Out of these, two participants made less than five calls and one made only seven calls.
Study outcomes
These analyses were based on intention-to-treat. There was a statistically significant difference in HbA
1c at six months between the usual care and TLC Diabetes arms. The geometric mean (arithmetic means provided in parentheses) of HbA
1c decreased from 8.7% (8.8%) to 7.9% (8.0%) in the TLC Diabetes arm, compared with 8.9% (9.0%) to 8.7% (8.9%) in the usual care arm, with the adjusted ratio of six-month geometric means of 0.91 (95% CI 0.86-0.93, p = 0.002) (Table
3). The ratio of 0.91 means that the geometric mean HbA
1c at six months in the TLC arm is 0.91 of the value in the usual care arm after adjustment for baseline covariates. There was slight evidence that the difference in HbA
1c at six months between study arms increased with baseline HbA
1c (
p = 0.09 for the interaction term in regression model). This suggested that the difference in six-month HbA
1c between TLC and usual care patients was greater in patients with high baseline HbA
1c values than in patients with low values. Of participants in the intervention arm, 20 % achieved HbA
1c levels of 7.0% or lower (95% CI 9.6-29.7), compared with 15% (95% CI 4.4-24.7) in the usual care arm (
p = 0.32).
Table 3
Baseline and post-intervention primary outcome values between usual care and Australian TLC Diabetes arms
HbA
1c
(%)
| | | Ratio |
Usual care | 8.9 (8.6-9.2) | 8.7 (8.7-9.0) | 0.91 (0.86-0.93, p = 0.002) |
TLC Diabetes | 8.7 (8.4-9.0) | 7.9 (7.6-8.3) | |
Health-related quality of life - mental
| | | |
Usual care | 49.5 (47.1-50.3) | 48.7 (47.1-50.3) | 3.0 (0.8-5.2 p = 0.007) |
TLC Diabetes | 49.8 (47.5-52.0) | 51.7 (50.2-53.3) | |
Health-related quality of life - physical
| | | |
Usual care | 45.4 (43.0-47.9) | 45.2 (43.8-46.6) | 0.4 (−1.7-2.4, p = 0.7) |
TLC Diabetes | 45.5 (43.0-47.9) | 45.6 (44.1-47.0) | |
In terms of HRQL, the mental component summary score was found to be significantly different between the two arms at six months (difference = 3.0,
p = 0.007), after controlling for baseline mental HRQL, plus other covariates (Table
3). Mental HRQL improved in the TLC Diabetes group, compared with those in the usual care group where mental HRQL decreased marginally. There was no interaction between study arm allocation and baseline levels for mental HRQL (
p = 0.4). No differences were observed in physical HRQL between the usual care and intervention arms (
p = 0.7).
Comparison of sample characteristics between Australian TLC and AusDiab samples
To determine the representativeness of the TLC sample at baseline, we used a comparable subsample of individuals from the AusDiab study, obtained from applying the Australian TLC Diabetes criteria for age range and HbA
1c levels (≥ 7.5%) to the subsample (n = 643) of those classified in AusDiab as having diabetes. 156 AusDiab participants were identified for comparison with the Australian TLC Diabetes sample. Overall, the AusDiab and TLC samples were similar (Table
2). There were no significant differences between the TLC sample and the AusDiab subsample across demographic variables, HRQL, and self-reported health variables. Behaviourally, there were no differences in nutrition self-reports between the study populations, however the TLC sample reported markedly lower smoking rates and were more likely to perform the recommended levels of exercise. In terms of their clinical profiles, the TLC sample appeared healthier, with lower systolic blood pressure, and generally better glucose and lipid profiles. These results, however, are likely to reflect the increased levels of doctor-diagnosed hypertension and hypercholesterolaemia, and therefore probably high levels of treatment in the TLC sample. Interestingly, despite their reported healthier behavioural profiles, the TLC sample were significantly more likely to be obese using both BMI and waist circumference classifications.
Discussion
This randomised controlled trial evaluated the efficacy of an automated, interactive telephone intervention for improving the management of diabetes. As far as we are aware, this is one of the first studies in the world to formally evaluate an automated telephone system for diabetes management that involves tailoring to individual needs and the findings offer promising results for the longer term use of this kind of program for people with diabetes. We have demonstrated that the Australian TLC Diabetes program significantly improved glycaemic control and mental HRQL after six months for those who participated in the program compared with the routine care condition.
Participation in the Australian TLC Diabetes intervention led to a significant improvement of HbA
1c, compared with the routine care available to people with diabetes in Brisbane, Australia. The mean reduction in HbA
1c of 0.8 % in the intervention arm is of substantial clinical significance if maintained long-term. Results from the UKPDS study highlight the substantial reductions in all diabetes endpoints associated with 1% reduction in HbA
1c[
7], such as 21% of deaths related to diabetes, 14% of myocardial infarction and 37% microvascular complications [
30]. A meta-analysis reported comparable levels of HbA
1c improvement from the pooled effects of 31 previous interventions providing education on self-management of diabetes [
9]. The majority of studies cited in the review, however, directly involved healthcare professionals/health workers for the provision of diabetes management education. Another meta-analysis evaluating the use of mobile phone interventions to improve glycaemic control showed a pooled change of 0.5% over six months, however, again with heavy involvement of healthcare personnel for intervention delivery [
11]. One previous study of another fully-automated telephone intervention aimed at improving glycaemic control failed to show significant post-intervention differences between intervention and control groups in levels of HbA
1c[
13]; however, that system did not provide tailored feedback to individuals. Therefore, a major advantage of the Australian TLC Diabetes program is its successful impact on glycaemic control and the potential for reduced costs and increased accessibility associated with an automated telephone-linked system for the provision of tailored diabetes management.
In addition to the observed improvements in glycaemic control, mental HRQL was significantly enhanced in people who received the intervention compared with those who did not, despite this not being a specific focus of the TLC program for the trial. The burden of daily management of diabetes and the development of complications lead to compromised HRQL in populations with diabetes [
34,
35], and therefore enhancing well-being, in addition to diabetes management per se, is an additionally important outcome. Despite this improvement reflecting only a small effect size (0.20) [
36], the literature in this field indicates that even small effect sizes of HRQL improvement may be of clinical significance in the longer term [
37‐
39]. Interestingly, the physical component of HRQL did not improve during the six-month intervention period. A brief computer-assisted diabetes self-management intervention on quality of life outcomes showed no change in HRQL, however, their two-month follow-up might not have been long enough to detect changes [
40]. In contrast, the pooled results from 20 publications showed that people with diabetes experience improved HRQL after receiving interventions designed to develop their diabetes self-management behaviours [
37], although this meta-analysis did not differentiate between the mental and physical components of HRQL.
Another important aspect of this study is the focus on people with poor glycaemic control (HbA1c ≥ 7.5%), indicating difficulty in their self-management of diabetes with the available routine care. These people are likely to be most at risk of the development of complications associated with diabetes, and therefore, given the results achieved, Australian TLC Diabetes has the potential to improve the health of the highest risk groups. Consequently, this program also provides the opportunity to significantly reduce the financial burden of type 2 diabetes on the healthcare system. Subsequent analyses will examine the cost-effectiveness of the program, which will have important implications for the widespread implementation of the program.
Our comparison of the TLC sample with a ‘matched’ subgroup from the AusDiab study sample suggests that the TLC participants did not differ significantly in terms of demographic characteristics from the best available data from a general population-based diabetes sample in Australia. The baseline AusDiab study, conducted in 1999–2000, offers benchmark national data on the prevalence of diabetes, obesity, hypertension, and kidney disease in Australia. This indicates the representativeness and external validity of our results and their applicability to other diabetes populations.
The trial was completed in accordance with the Medical Research Council’s guidelines for the effective design and evaluation of complex intervention trials [
41]. Principal components of any effective complex intervention include feasibility, participant-engagement, identification of mechanisms for intervention outcomes, and trial fidelity [
42]. The feasibility and relevance of the Australian TLC Diabetes program are demonstrable within the current context of type 2 diabetes. The accessibility of the telephone-delivered intervention over the long-term is particularly important for a widespread chronic condition, such as diabetes, which requires ongoing management and affects a large proportion of the population. The very high usage of the Australian TLC Diabetes system and results to date indicate that the participants in the intervention arm engaged with the program, with over three quarters of weekly calls being completed. Full details of system usage were recorded as part of the data collection and will be reported elsewhere for full process evaluation of the system’s usability and participant satisfaction, as well as whether the cost of the intervention provides acceptable value for money. Furthermore, the intervention was able to affect pathways that led to improvements in glycosylated haemoglobin and therefore diabetes management, as well as improvement in mental health-related quality of life for the participants. The fidelity of the trial implementation in accordance with the original design and protocol [
27] was strong. Difficulties were encountered during recruitment and this led to increased recruitment opportunities via enhanced presence at Diabetes Australia – Queensland shops and seminars and hospital diabetes clinics. The sample size was smaller than originally planned, however, as discussed, the sample obtained is powered to detect group differences that will be both statistically and clinically significant at 12-month follow-up. No changes were applied regarding the randomisation process or implementation of the intervention.
Although only glomerular filtration rate significantly varied across the study arms at baseline, other baseline characteristics (Table
2) showed some differences. Separate analyses tested the impact of the inclusion of these variables individually on the main results and the main outcome results did not change. As with most research, it is possible that a selection bias operated in this study, with people willing to participate being more likely to prioritise their health and/or have the social, educational, and economic resources to accommodate participation. The study requirement of access to a telephone meant that there may have been a socioeconomic selection bias; however in the geographic area from which we recruited, over 96% of households have a fixed phone connection, so we are confident that this criterion did not appreciably influence participation. It is also possible that the reduced sample size and some of the challenges associated with trial recruitment may limit generalisability. More research is required to investigate generalisability and to explore uptake by others with diabetes. Although there was a suggestion of an increasing effect of intervention with increasing baseline HbA
1c values (from the interaction test), this did not reach conventional levels of statistical significance and should be reassessed in future studies.
A substantial body of research conducted over the last 30 years has drawn attention to the importance of ongoing support and follow-up to sustain improvements in diabetes management and management of other chronic conditions, with strong links to health and self-care behaviours [
43‐
45]. Therefore a diabetes management support program such as this, designed to provide easy access to long-term (potentially cost-effective) support, is of paramount importance, and hence, this kind of program also requires detailed evaluation in the longer term as well. A subsequent paper will elucidate the changes in behaviour that may have facilitated the improvements observed.
Acknowledgements
The study is funded by a National Health Medical Research Council project grant (ID 443214), by the HCF Health and Medical Research Foundation and by Queensland Health. We wish to thank all study participants, Diabetes Australia for provision of educational material and Diabetes Australia - Queensland for its assistance with recruitment. We acknowledge the dedication of the investigator team: Prof Mary Courtney, Prof Richard Wootton, and Prof Kerrie Mengersen. We also acknowledge the commitment of the project staff: Mandy Cassimatis, Lyndall Kopp, Megan Rollo, Wei-I Wu, Adrienne O'Neil and Vivien Harris. We would like to acknowledge the Baker IDI Heart and Diabetes Institute for access to the AusDiab data. Furthermore, we would like to thank Dr Stephan Gaedhe, one of the principal developers of the TLC Diabetes system, and the Australian Diabetes Educators Association for their contributions. We are grateful for the input from staff at the Medical Information Systems Unit, Boston University. Finally, we also wish to thank Roche Diagnostics ACCUCHEK for their supply of the glucose meters and Alive Technologies for all of their technical advice. EDW was funded by a Diabetes UK Moffat Fellowship.
Competing interests
Dr. Friedman has stock ownership and a consulting agreement with Infomedics, the company that owns commercial rights to the TLC technology used in the computerized intervention. He is also a member of its Board of Directors. The other authors declare that they have no competing interests.
Authors’ contributions
EDW analysed the data and wrote the manuscript. DB collected the data, contributed to study development, discussion and manuscript writing. AF, AR, SA, PS, RF and BO contributed to study development, and discussion, reviewing/editing of the manuscript. All authors read and approved the final manuscript.