This telephone call was scripted, and conducted by a trained health coach (not a GHS coach). Participants set a goal for weight maintenance or further loss within 3 months, and then two goals for behavior change. For each behavioral goal, they were asked to identify: rewards for reaching their goal; expected benefits; preparatory behaviors for goal attainment; barriers and solutions; and a person who could support them to reach their goals. Participants selected their desired number of text messages (from 3-13 per fortnight), timing of texts (eg, 6 AM), and type of texts (from the four types described below). This information was recorded during the call and was used to tailor GHSH texts.

Four types of texts targeted different behavior change strategies, each with different permitted frequencies (an example of each text type has been previously published [27]):

Wording of the text messages (each ≤160 characters) was tailored to each individual’s name, gender, goals, identified barriers and strategies, preparatory behaviors to achieve their goals, expectations of behavioral change, and the first name of their identified support person. Texts were generated and sent by research staff, using a purpose-designed software package in which messages were preprogrammed in advance and scheduled to be sent at specific times. Replies to the goal check texts were stored and automatically triggered tailored responses whenever the participant replied with “yes,” “no,” or accepted variations of these. Whenever participants’ goal check reply contained additional words or an unrecognized variant of “yes” or “no” the program emailed research staff who manually decided which tailored reply to send. Unprompted reply text messages from participants did not receive a reply. At any stage participants, could change their text message preferences via text message or telephone call.

At 12 weeks, participants received a second telephone call from their coach to update their tailoring information. This call was made between weeks 12 and 14, and if contact was not made during this period, the existing tailoring information was used for the final 12 weeks.

During the CATI, participants reported their body weight in kilograms (while wearing light clothes and no shoes). They were encouraged to weigh themselves during the CATI if scales were present; otherwise, they were asked to report their most recent weighing. Participants were posted a measuring tape and instruction sheet on measuring waist circumference at baseline. The interviewer instructed participants to take the waist circumference measurement during the call. Body mass index (BMI) was calculated based on self-reported height at GHS baseline and self-reported weight at each assessment point. These self-report methods have been validated against objectively measured weight and waist circumference in a subsample of GHS users (n=38) [25]. This validation study showed strong correlations (Spearman rho >0.90) with the objective measures for both outcomes. There were 84% and 87% agreements in BMI and waist circumference classifications, respectively [25].

During the CATI, participants completed a validated, 3-item assessment tool (3Q-PA) which asked about the number of weekly sessions spent: walking for ≥30 minutes; doing other moderate-intensity physical activity for ≥30 minutes (termed “moderate”); and, doing vigorous-intensity physical activity for ≥20 minutes [30].

Participants were also posted a belt-mounted dual-axis accelerometer (Actigraph model GT1M) initialized to collect data in 10-second epochs, a wear-time log, and a reply-paid envelope to return their materials. Participants were asked to wear the monitor on the hip for 7 consecutive days during all waking hours, and to remove the monitor only for sleep (if desired) and during times the monitor could be damaged (eg, during water-based activities). The wear log included monitor-fitting instructions and asked about any monitor removals, sleep time, and whether the monitor was worn or removed during sleep. Accelerometer data were downloaded in Actilife (v 6.6.2). Both 10- and 60-second epoch files were processed in SAS version 9.3. Nonwear time was excluded (ie, ≥60 minutes of 0 counts per minute (cpm), allowing for up to 2 minutes of 1 to 49 cpm [31]) and only days with ≥10 hours of wear were deemed valid. Data were plotted (as heatmaps) and compared against wear logs. Any sleep not already excluded as nonwear was excluded, along with any days that had registered as valid but the monitor was in the post. All minutes with ≥1952 cpm (vertical axis [32]) were classed as MVPA, then summed for each day and averaged across valid days.

During the CATI, participants reported the following: number of daily servings of fruit and of vegetables [33]; average daily consumption of sweetened drinks (cordials, fruit juices, sports drinks, soft drinks not including diet soft drinks); and takeaway meals per week [34]. Participants also completed the FFBQ [35], which asks about consumption of high-fat and high-fiber foods over the previous month concerning cooking, eating and food choice behaviors, and two items regarding fruit and vegetable intake. The average of the relevant items (1-5, with higher values indicating healthier habits) form the three FFBQ indices (20-item Total Index, 13-item Fat Index, and 7-item Fiber Index). These have good reliability, acceptable validity and good responsiveness to change [35].

The text message software collected data on delivery (ie, number and type of text messages sent) and intervention engagement (ie, number of responses to goal check texts and achievement of weekly behavioral goals). The duration of the tailoring interviews and participant alterations to text preferences were tracked by research staff.

At 6 months, intervention participants rated satisfaction and usefulness in five categories (“not at all” to “extremely” satisfied/useful), of the GHSH intervention overall and specifically regarding support for achieving behavioral and weight loss goals. Intervention participants were invited to complete a telephone interview with research staff (approximately 10 minutes) involving open-ended questions regarding intervention usage, satisfaction, and potential program improvements. These interviews were audio-recorded, transcribed verbatim, and coded independently by two authors. Discrepancies between coding were discussed until consensus was reached and recruitment stopped once thematic saturation was reached.

Personnel time was tracked for delivering all intervention-related tasks. Coach time was costed at AU$37.56 per hour and research staff time was costed at AU$31.56 per hour. Average duration per participant (minutes) was multiplied by the relevant personnel cost. Direct costs of sending the texts (at AU$0.15 each) were tracked, totaled and then divided by the total number of intervention participants to provide a per participant cost.

Intervention effects were re-examined using multiple imputation by chained equations in STATA (m=20 imputations), adjusting for the same covariates as the completers models as well as predictors of missing data ( P<.2). Alternative models assessed whether conclusions were affected by model choice, as the main models sometimes failed to meet requisite assumptions. For count outcomes, intervention effects were retested using negative binomial models or zero-inflated negative binomial models when these fit the data better (Vuong test) [36]. The zero-inflated models were best suited to modelling the behaviors (physical activity, soft drink, and takeaway meal consumption) in which many participants did not engage.

It is complex to interpret results from extended contact interventions where “no change” in outcomes can be interpreted as a positive finding. Within-group changes (“worsened” or “improved”) and between-group differences (intervention “better” or “worse” than control) were only claimed when these were statistically significant. As nonsignificant findings can indicate either no change/difference or an insufficient sample size to yield a conclusive finding, we only described outcomes as “maintained” or as groups being “similar” if in addition to the finding being nonsignificant, the likely true effect size for the change/difference as indicated by the 95% confidence interval (CI) was also very small (<MDI).

Intervention participants showed significant reductions in both body weight (−0.89 kg, 95% CI: −1.53, −0.25) and waist circumference (−1.34 cm, 95% CI: −2.31, −0.36; Table 2). Intervention group improvements were significantly greater than controls for weight loss (−1.35 kg, 95% CI: −2.24, −0.46, P=.003), but not for waist circumference ( P=.115; Table 3).

The intervention group maintained their accelerometer-assessed MVPA and self-reported vigorous activity, and changed significantly only in self-reported walking (−0.55, 95% CI: −0.99, −0.11 sessions/week; Table 2). The control group changed significantly in moderate activity and MVPA, which both declined (−0.68, 95% CI: −1.11, -0.26 sessions/week, and −16.10, 95% CI −28.60, −3.61 min/week; Table 2). The intervention group did significantly better than the control group in moderate activity (0.56, 95% CI: 0.15, 0.96 sessions/week) and in accelerometer-assessed MVPA (24.16, 95% CI: 5.07, 43.25 min/week; Table 3). Intervention effects for walking and vigorous activity were not significant (Table 3).

In the intervention group, small, but statistically significant improvements in dietary outcomes were observed for the FFBQ Fiber and Total Indices, while remaining dietary outcomes, except takeaway meals per week, were all maintained (Table 2). By contrast, in the control group, there were no statistically significant improvements, only statistically significant declines in vegetable intake and fruit intake (Table 2). Other dietary outcomes were maintained. No significant or meaningful intervention effect for dietary outcomes was observed.

Results from analyses using multiple imputation mostly supported the interpretations from the main analyses (Multimedia Appendix 4), except that due to slightly narrower CIs in the multiple imputation analysis, the intervention group maintained takeaway meals/week (rather than being ‘inconclusive’) and the intervention effect for vegetable serves/day was ‘similar’ (rather than ‘inconclusive’). Conclusions using the alternative models mostly supported the main findings (Multimedia Appendix 5) and revealed significant differences in walking ( P=.018). The intervention group showed a greater tendency to walk than controls (odds ratio = 2.17, 95% CI: 0.45, 10.43, P=.334), but also to report fewer walking sessions when they did walk than controls (relative rate = 0.77, 95% CI: 0.64, 0.92, P=.005).

Overall, the GHSH intervention was feasible to deliver and acceptable to participants. It led to significantly better outcomes compared with the control group in weight loss and some forms of physical activity, but not in dietary behaviors. For most dietary outcomes, meaningful intervention effects were also unlikely, based on the CIs. The study findings were inconclusive due to insufficient sample size for vegetable intake, walking, vigorous activity and waist circumference.

The GHSH weight loss of 0.89 kg (95% CI: −1.53, −0.25) during extended care was better than or consistent with comparable extended contact interventions. Spark and colleagues [23], in a single-group group trial of breast cancer survivors (n=29), found when receiving text message extended contact, women on average regained 1.3 kg (95% CI: −0.5, 3.1 kg) over 6 months. Donaldson and colleagues [22], in a small (n=34), nonparallel controlled trial, found intervention participants lost significantly more weight than controls (−1.6 vs 0.7 kg, 95% CIs not reported) over 12 weeks of extended contact that involved both text messages and face-to-face group sessions. The magnitude of change observed in this trial is also congruent with the intervention effects summarized in two recent meta-analyses of extended contact interventions [12,13] delivered via other modalities (ie, not text messaging).

The GHSH intervention was designed to follow on from a telephone coaching program, in which participants developed a rapport with a health coach. It is positive to note that a semiautomated text message program (with two brief telephone contacts) maintained the perception of accountability and that most participants felt personal engagement with their coach. It is encouraging that participants updated their text message preferences throughout the trial, given that continued engagement is a known facilitator of maintenance [12]. Particularly encouraging was that this highly tailored contact was achieved at a relatively low cost, which supports the scalability of GHSH.

Strengths of this trial include: the range of outcomes evaluated to inform service delivery; the high retention rate (216/228, 95%); its conduct in partnership with the service delivery agents and funders; and, the rigorous research design embedded within real-world delivery [38]. In line with public health principles and how the GHS is delivered, all participants were recruited and analyzed in this trial regardless of whether they complied with national guidelines at baseline, strengthening the findings further. These strengths enabled trial findings to inform the uptake of GHSH as part of standard delivery of the GHS program. Future evaluation of the GHSH impact once implemented into practice will advance our understanding of its uptake and the outcomes that can be achieved when scaled up for population-wide delivery.

While self-report of anthropometric outcomes was a limitation, the measures showed acceptable validity against objective measures in a GHS subsample [25]. The study was adequately powered for some but not all outcomes. The differences between trial participants and others completing GHS during the recruitment period may indicate the types of groups willing to receive a text message-delivered intervention, but may also reflect biases in research participation and eligibility criteria. This comparison of the trial sample to graduating GHS users is appropriate given this is the intended audience of the extended contact program; however, it is also useful to consider whether GHS users are broadly representative of Australian adults. Previous evaluations [39] have shown that GHS users were representative in relation to education, employment status, Aboriginal status, and fruit and vegetable consumption, but there was a disproportionately high representation of women, which may be expected given evidence of their higher likelihood to access health advice and attempt to lose weight compared with men. In the current trial, there was minimal missing data from the baseline and 6-month data collection points. Our sensitivity analysis using multiple imputation showed that missing data had very minimal impact on the interpretation of our findings.

The GHSH extended contact intervention was feasible to deliver, acceptable to GHS clients and led to significantly better outcomes than standard practice in weight loss and physical activity. Supporting individuals to maintain positive lifestyle changes through cost-effective programs is paramount to the success of the GHS and more broadly to public health in Australia.