Participants
Figure 1 shows patient flow through the study. Three participants consented but did not complete a T0 questionnaire (not randomised). Recruitment was faster than expected and 163 participants were randomised, 85 allocated to RESTORE and 78 the leaflet. (See Table 1 for sample characteristics.) Four participants experienced a recurrence during the study and were excluded from analysis.
Table 1 Baseline characteristics of RESTORE and comparator groups
Nineteen of the 81 invited participants completed a process evaluation interview, 8 from RESTORE and 11 from the comparator group. Most (n = 15) were female and ≤60 years (n = 14, age range 39–78 years). A range of cancer types were represented; the majority (n = 12) had breast cancer. Six in the RESTORE group had accessed ≥3 RESTORE sessions.
Feasibility
One hundred and sixty-three participants were recruited, an average of five participants per week. Forty-one percent of eligible participants consented to the study (16 % of those screened). Nine percent of patients screened were ineligible due to a lack of access to a computer/Internet, and a further 1 % declined due to the web-based nature of the study.
Randomisation worked well; baseline characteristics were generally well-balanced between groups (Table 1). However, the intervention group had a higher proportion ‘not working’ (48.8 vs 36.0 %), primarily due to more retired people in this group, and a greater number of days since last cytotoxic treatment (577.71 vs 484.61).
No process evaluation participants reported having to learn new skills to use RESTORE. However, concerns were raised that older people might struggle if they did not use computers regularly. Some participants encountered problems navigating RESTORE, experienced difficulties logging on, had password refused and reported screens freezing or closing down unexpectedly.
Acceptability
Half of participants in the process evaluation felt the timing of participation was ‘about right’. The remainder would have preferred RESTORE sooner. They were ≥1 year from diagnosis and suggested benefits of participating before treatment completion.
Most participants identified benefits of taking part in the trial, including feeling supported and reassured that someone was interested in their condition. A number made positive changes to their lifestyle as a result of using RESTORE or the leaflet. For some, RESTORE had improved their understanding of CRF and how to manage it; for others, involvement in the trial offered a period of reflection and allowed them to re-evaluate what they could do. Most reported an increase in confidence to self-manage the effects of CRF and considered their CRF to be less bothersome.
Of those who accessed both RESTORE and the leaflet, half preferred RESTORE, finding it more flexible and interactive. Others felt the leaflet was more convenient as it was immediately available and could be consulted any time.
Suggested improvements to the intervention included providing more cancer-specific information and more personalised feedback.
Total attrition rate (consent to T2) was 36 %. Nine people (5.5 %) actively withdrew and 36 (22.1 %) did not complete follow-up questionnaires. Seven participants from the RESTORE group, and 2 from the comparator group missed T1 but completed T2 questionnaires. Four people were excluded due to metastatic disease. No adverse events were reported.
Seventy-one percent of participants were deemed to have adhered with the intervention (logged on to sessions 1 and 2 and a third session). Sixty percent logged on to four sessions, 43 % to five sessions. The Work and home life session was the most visited (51 % of participants) of the three ‘optional’ sessions, and Talking to others was the least visited (27 % of participants). Within sessions 3–5, participants engaged most with the goal-setting section and least with the goal review section.
Potential to increase self-efficacy to manage CRF
Missing data were handled using multiple imputation where appropriate. Ninety-seven (61.0 %) complete cases were observed and 130 (81.8 %) people had ≤6 missing values. Imputation models were constructed for all outcomes at each time point. As a typical example, the imputation model for fatigue self-efficacy at T1 included fatigue self-efficacy at baseline and T2, each of the other outcomes at T1, centre, group, age, gender and time since last treatment. Examination of partial correlations between all outcomes suggested the inclusion of baseline values of fatigue self-efficacy and BFI in each imputation model. Time since last treatment also required multiple imputation (missing n = 15 [9.4 %]); the imputation model included all baseline values of outcomes, centre, group, age and gender.
Multiple imputation was performed using Stata 12.1. Predictive mean matching was used to ensure feasible values of the outcomes, with imputations drawn from the nearest three neighbours. Inspection of histograms suggested observed and imputed values had similar distributions.
The intended analysis (between-group, repeated-measures model with random intercept for centre and random coefficients for time) could not be implemented due to failure of the model to converge; we believe the highly variable nature of how fatigue evolves over time in an individual is the reason for this. Instead, simpler mixed-effects models were used to assess the effect of the intervention on outcomes at T1 and T2 separately (controlling for baseline scores and with a random intercept for centre).
Table 2 shows mean scores for all outcome measures over the three time points. There is evidence of improved fatigue self-efficacy at T1 (0.514, 95 % CI [−0.084, 1.112], P = 0.09), in the RESTORE group though the impact is lost by T2. There is no evidence of difference between groups for any other outcomes.
Table 2 Means, standard deviation and estimated group effects for fatigue self-efficacy and all secondary outcome measures
The pattern of fatigue self-efficacy for adherent and non-adherent individuals was similar to the patterns observed for the RESTORE and comparator groups: T0 5.466 (1.944) vs 5.028 (1.827) and T1 6.520 (1.789) vs 5.250 (1.250).
Based on data observed, a sample of at least 317 per group would be required to detect a standardised effect size of 0.5 (significance level 5 % and power of 90 %).
Health resource use
There was a high proportion of missing health economic data, making an economic evaluation challenging and formal statistical comparisons impossible. Examination of available descriptive data suggests comparable use of resources between groups: mean number of visits to a GP practice at T2 2.29 (1.27) in the RESTORE and 1.90 (1.04) in the comparator groups, and 1.41 (0.80) and 1.29 (1.27) visits to the oncologist respectively.