Self-management intervention to reduce pulmonary exacerbations by supporting treatment adherence in adults with cystic fibrosis: a randomised controlled trial

Introduction Recurrent pulmonary exacerbations lead to progressive lung damage in cystic fibrosis (CF). Inhaled medications (mucoactive agents and antibiotics) help prevent exacerbations, but objectively measured adherence is low. We investigated whether a multi-component (complex) self-management intervention to support adherence would reduce exacerbation rates over 12 months. Methods Between October 2017 and May 2018, adults with CF (aged ≥16 years; 19 UK centres) were randomised to the intervention (data-logging nebulisers, a digital platform and behavioural change sessions with trained clinical interventionists) or usual care (data-logging nebulisers). Outcomes included pulmonary exacerbations (primary outcome), objectively measured adherence, body mass index (BMI), lung function (FEV1) and Cystic Fibrosis Questionnaire-Revised (CFQ-R). Analyses were by intent to treat over 12 months. Results Among intervention (n=304) and usual care (n=303) participants (51% female, median age 31 years), 88% completed 12-month follow-up. Mean exacerbation rate was 1.63/year with intervention and 1.77/year with usual care (adjusted ratio 0.96; 95% CI 0.83 to 1.12; p=0.64). Adjusted mean differences (95% CI) were in favour of the intervention versus usual care for objectively measured adherence (9.5% (8.6% to 10.4%)) and BMI (0.3 (0.1 to 0.6) kg/m2), with no difference for %FEV1 (1.4 (−0.2 to 3.0)). Seven CFQ-R subscales showed no between-group difference, but treatment burden reduced for the intervention (3.9 (1.2 to 6.7) points). No intervention-related serious adverse events occurred. Conclusions While pulmonary exacerbations and FEV1 did not show statistically significant differences, the intervention achieved higher objectively measured adherence versus usual care. The adherence difference might be inadequate to influence exacerbations, though higher BMI and lower perceived CF treatment burden were observed.

for the CFHealthHub Study Team      *Recurrent event survival was also calculated: hazard ratio 0.95 (95% CI 0.80, 1.13; p=0.567). † The difference between 'all exacerbations' and 'main -unadjusted' is the number of IV antibiotic courses that were offered by clinicians but declined by participants. The IV-declined rate was 32/558 (5.7%) for the usual care arm and 22/504 (4.4%) for the intervention arm. These values are far lower than the IV-declined rate observed in the general CF population of around 20%, 1 which provides evidence that the recruited participants may not be representative of the general CF population.
Statistical significance was not observed in any of the subgroups by baseline objectivelymeasured effective adherence. "Pulmonary exacerbation" was defined as the administration of intravenous (IV) antibiotics for any of the 12 Fuchs' symptoms/signs. As discussed in the main manuscript, there is a discretionary element to the use of IV antibiotics as rescue therapy to treat exacerbations. Increasing a person's adherence to inhaled therapies may also improve their engagement with other treatments, including improving their acceptance of IV rescue antibiotics. It is possible that the exacerbation rate appeared to have somewhat increased in those with baseline adherence 26-50% because this is the subgroup with greatest improvement in adherence, potentially leading to the greatest impact from ascertainment bias (see appendix D [ figure D2] for the subgroup analysis of adherence). It should be noted that interactions between exacerbations and adherence are complex, including for example the impact of engagement on IV acceptance, such that interpretation should be made with caution. BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)

Aim
The CFHealthHub intervention aims to support adults with cystic fibrosis (CF) to increase and maintain their adherence to prescribed nebulised medication in order to reduce exacerbations and improve or prevent decline in lung function.

Rationale
The CFHealthHub intervention is underpinned by the Capability Opportunity Motivation-Behaviour (COM-B) model. 1 It has been developed using the Behaviour Change Wheel approach alongside a person-based approach to intervention development. This process is described in detail elsewhere 2 but broadly consisted of the following stages: • Identification of barriers and facilitators for nebuliser adherence using the Theoretical

Domains Framework
• Identification of appropriate intervention functions and behaviour change techniques to address barriers identified • Iterative development of the CFHealthHub intervention with patients, using feedback from interviews and 'think aloud' to refine the intervention • Creation of an intervention manual and training programme for interventionists • Pilot and feasibility trial including a process evaluation which was used to further refine the intervention, manual and training process

Conceptual framework and theory
The conceptual framework that describes the intervention is provided in figure A1. Consistent with the COM-B model, the framework considers issues of capability, opportunity and motivation, all of which must be present in order for repetition of the behaviour (i.e. medication adherence) to occur. Initially we anticipate that repetition will require effortful self-regulation, but with repetition and strategies to promote habit formation we aim for the behaviour to become more automatic.  The intervention addresses a range of different barriers and is tailored to meet the specific needs of the person. The intervention draws on key theories in order to address different parts of the proposed process: Social Cognitive Theory, 3 Control Theory, 4 and Habit Theory, 5 as follows: • Social Cognitive Theory (SCT) 3 proposes that behaviour is influenced by two core constructs: i) perceived self-efficacy, i.e. an individual's beliefs in their capability to adhere to treatment; and ii) outcome expectancies, i.e. an individual's beliefs about the likely consequences of their actions. Self-efficacy can be enhanced through: i) mastery; ii) vicarious experiences, where a role model, similar to the individual successfully achieves behavioural change in a similar situation; or iii) verbal persuasion. Outcome expectancies include beliefs about the positive and negative and short-and long-term consequences of adherence, and in this context include perceived necessities and concerns. 6 According to SCT, outcome expectancies may result in intentions to change one's behaviour. Self-efficacy then influences the translation of that intention into action through the pursuit of goals.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) • Control Theory 4 explains the processes of self-regulation. When a behavioural standard or goal has been set, an individual directs their attention through monitoring behaviour to the discrepancy between their current behaviour and their goal. They then use this feedback to regulate their behaviour to meet their goal through action control.
This in the context of adherence, once an adherence goal is set, self-monitoring of treatment-taking provides the feedback to prompt action to enable self-regulation of behaviour.
• A habit is where a behaviour is prompted automatically by a situational cue. Habits are created due to the repetition of a behaviour in a specific context 7 which, over time results in a learned cue-behaviour association. 8 In the context of adherence, the repeated taking of treatment in a specific context or in the presence of a specific cue should over time result in the formation of a habit. Habits are particularly advantageous because theory predicts that, once formed, they do not rely on motivational processes and therefore should persist even if motivation wanes. 9 They may therefore play a particularly important role in the promotion of long-term maintenance of behaviour, 10 in this case adherence which is a key aim of the programme.

Materials
The CFHealthHub intervention includes a range of materials as follows:

Intervention providers
Intervention providers were recruited from each site. The majority of sites recruited individuals who were already members of the multi-disciplinary teams working in CF at that site. Other sites recruited from other parts of the hospital or recruited externally.
Thus, interventionists had a range of backgrounds including: • Physiotherapists working in CF or other respiratory conditions

Interventionist training, assessment and support
Interventionists received training in how to deliver the intervention in a variety of ways:

Training in use of equipment
Interventionists received training in how to use the eTrack nebuliser and 2net Hub, how to pair the devices, and how to register a new participant onto the CFHealthHub platform and PARI Track system, as part of their research procedures training. This was delivered face-to-face by the study manager and PARI, and supported with a research procedures manual and adhoc telephone support throughout the trial.

Training in delivery of CFHealthHub intervention
Interventionists received training in how to use the CFHealthHub web platform and how the deliver the CFHealthHub intervention. Training was delivered over a 2-day face-to-face training session, followed by a schedule of online training to be completed over the equivalent    Competencies to deliver a review visit and a phase review visit were assessed by listening to the first audio-recorded visit of that kind for each interventionist. Two members of the training team looked at the completed worksheet for the session and listened to the accompanying audio-recording. They then discussed and agreed marks. Agreed marks for each section were averaged and the pass mark was 90%. Interventionists received individual feedback on their performance and tutorial support where they had failed. The next audio-recorded visit of that kind was assessed where the assessment was failed.

Ongoing support
Ongoing support for interventionists was delivered via a weekly teleconference, email and telephone support with the training team, technical support via telephone and email. The weekly teleconference provided a space where interventionists could discuss problems, successes and case studies (anonymised), to aid group learning. Individuals could also access members of the team individually and individual interventionists were targeted with support where they had failed their earlier competency assessment or where there were any problems identified.

Intervention schedule of delivery
The intervention schedule of delivery is described in figure A2. The content of each kind of intervention session is described below. Within this schedule there are a number of different paths that were determined during delivery.

Consent visit and set-up
All participants receive their eTrack nebuliser and 2net Hub at the consent visit. They also complete the COM-BMQ screening tool at this visit. An account is created on CFHealthHub into which is added the current prescription data for the participant and the data from the COM-BMQ screening tool. The consent visit takes place ≥4 weeks prior to the first intervention visit.
During this time adherence data is transmitted automatically from the eTrack nebuliser via the 2net Hub, which is plugged into their home, to the CFHealthHub platform. Figure A3 shows this process.

Intervention sessions received by all participants
All participants receive their first intervention visit ≥4 weeks following consent (so that the consultation is based on ≥4 weeks' worth of objectively-measured adherence data). This visit is always done face-to-face although can be in a variety of locations, including hospital (inpatient), clinic or home. All participants then receive an intermediate review phone call one week later. Subsequent visits depend on their objectively-measured effective adherence level.
Participants with an adherence level of ≥80% follow the 'Very high adherence' pathway while those with adherence level of <80% follow the normal pathway.

Normal pathway (adherence less than 80%)
Participants on this pathway have intervention sessions over a 12-week period. In addition to the first intervention session (at week 0) and an intermediate review (at week 1), they receive a review session at week 4, an intermediate review at week 6, a second review session at weeks 8 or 9, and a phase review at week 12. This pattern of delivery constitutes a phase.
They then receive a phase review session every 12 weeks, or every 6 weeks if their objectively-measured effective adherence level is <25%.

Very high adherence pathway (adherence 80% or more)
Participants on this pathway have intervention sessions over a 4-week period. In addition to the first intervention session (at week 0) and an intermediate review (at week 1) they receive a phase review at week 4. They then receive a phase review session every 12 weeks. Participant requested support. This can be a request for additional support at a phase review in which case an additional intervention delivery period is triggered without a break, or at any other time.
ii. Additional periods of delivery are offered to participants if one or both of the following triggers occurs following the first phase review. iii.
A drop of ≥20% in objectively-measured effective adherence since the phase review.
iv. An exacerbation requiring intravenous treatment.
In any of these situations, participants are contacted, and additional support is offered. If participants agree then the triggered pathway commences with a review session at week 0, an intermediate review 1 week later, a review visit 4 weeks later, and a phase review 6 weeks later. Participants then revert back to phase reviews every 12 weeks intervals (or every 6 weeks for those with an objectively-measured effective adherence level of <25%).

Access to CFHealthHub
Participants have an individual login providing access to the CFHealthHub platform throughout the intervention. It can be accessed on a laptop or via an app available for Apple or Android devices.
Participants are encouraged to access the site regularly and are provided with a participant guide with instructions on how to access and information about what to find where.

Intervention modules
The CFHealthHub contains a number of distinct modules each of which focuses on a different aspect using a range of specific behaviour change techniques (described using the definitions in the behaviour change taxonomy) 11 and modes of delivery.

Tailoring and personalisation
The CFHealthHub intervention is not one-size-fits-all and is designed to be tailored and personalised so that it can best meet the needs of a wide range of participants. While the entire content of the CFHealthHub website is available for participants to browse, tailored aspects are emphasised or added into a specific personal 'favourites' area called 'My Toolkit'. Table A2 describes the ways in which the intervention is tailored.

Types of intervention visit
Broadly, the intervention visits all have the same aim, which is to enable participants to look at their data, reflect on why adherence is important, set goals to increase their adherence and make plans as to how they will achieve these, and problem-solve any barriers that are likely to get in the way. However, the intervention visit types do differ somewhat in their set-up, focus and how in-depth they are, as follows. Detailed information about the structure of the delivery for each type of session is provided in the intervention manual and the relevant worksheets.

First Intervention visit
This session always happens face-to-face, although this can be in a hospital/clinic setting or at home. It lasts between 40 and 60 minutes. It is the first time that the participant accesses the CFHealthHub platform and sees their data. Interventionists must prepare for this session by entering the data from the COM-BMQ screening tool and checking that data are coming through to CFHealthHub from the nebuliser.
The key things that happen in this session are: • Participant receives their log-in details and accesses CFHealthHub

Intermediate review
The intermediate review is a short session that is designed to trouble-shoot 'quick' and easy to solve problems (e.g. an action plan that isn't working). It is normally delivered by telephone and lasts 5 to 15 minutes. The review is less structured than other visits.

Ad-hoc review
This follows the same structure as the intermediate review but is delivered where there is unplanned face-to-face contact with a participant (e.g. in clinic).

Review visit
This session normally last 30 to 45 minutes and can be delivered face-to-face or by telephone.
The session focuses on the data and what has happened in terms of adherence since the last visit. The precise focus will vary depending on the individual participant, e.g. a session with a participant who has met their goal would have a different focus to one with a participant who has not met their goal (or did not set one).
Broadly thought, the session covers the following modules: • My treatment

Phase review
The focus of this appointment is to facilitate reflection on progress since the intervention (or the current phase of delivery) began and to consider whether continued support is required or whether the participant wishes to manage their adherence independently. Ideally this should be delivered face-to-face but can be delivered by telephone. It normally lasts 20 to 30 minutes.
It covers the following modules: • My treatment

Fidelity of intervention delivery
Fidelity of delivery was assessed throughout the delivery of the intervention to ensure that interventionists continued to deliver the intervention as specified in the manual and training (assessment of drift). Two reviewers independently assessed a purposive sample of audiorecordings and worksheets associated with the delivery of intervention sessions with participants (first intervention session, review and phase review) using a scoring sheet that was developed and piloted during the feasibility trial.
Sessions were selected to represent a range of different sites, types of sessions with particular focus on interventionists who: • Had initially failed any of their certification assessments • Had high withdrawal rates (more than two participants withdrawn from the interventionist contact) • Had submitted <80% audio-recorded sessions from those participants who provided consent for them to be recorded • Had completed a lower than expected number of intervention visits and/or had fewer than average action and coping plans recorded in CFHealthHub

Metrics for fidelity of intervention delivery
There were 32 interventionists and a total of 213 fidelity of delivery assessments conducted during the randomised controlled trial.    In the trial, to reflect effective medication use, adherence was calculated as normative (effective) adherence using objective data from Weeks 3-52 as the outcome and Weeks 1&2 as the "baseline". Objectively-measured effective adherence was adopted as an outcome measure because it better reflects the effectiveness of medication use in comparison to simply calculating percent adherence according to an agreed regimen between adults with cystic fibrosis (CF) and their clinical team, as we have detailed elsewhere. [1][2][3] The calculation of objectively-measured effective adherence involves numerator adjustment (capping daily maximum nebuliser use at 100%) and denominator adjustment (to define the minimum effective treatment regimen) according to a person's Pseudomonas aeruginosa status, as described in section 9.2.1 of the statistical analysis plan (available as supplementary material).
For example, a person with chronic Pseudomonas aeruginosa infection should be on at least a nebulised muco-active agent and an antibiotic (i.e. three daily doses). If a person with chronic Pseudomonas aeruginosa infection only agreed to use nebulised dornase alfa once daily, adherence levels in outcome calculation will use the denominator of three daily doses.
If that person was on aztreonam thrice daily and hypertonic saline twice daily, no denominator adjustment will be carried out because denominator adjustment only applies for less than ideal regimen. In particular, the denominator adjustment is important because there is a wide variation in the prescription of inhaled therapies between different centres. 4 By standardising the denominator given the person's clinical characteristics in calculating objectively-measured effective adherence, it is ensured that an increase in percentage adherence is due to an increase in nebuliser use (i.e. increase in the numerator) rather than simply due to a reduction in agreed prescriptions (i.e. decrease in the denominator).
It is important to distinguish the concept of standardisation for effectiveness used as an outcome measure from individualised feedback to participants. Objectively-measured effective adherence allows standardisation based on randomised controlled trial evidence of what treatment is likely to work. Individualised target setting between clinical teams and people with CF continued to be informed by both considerations of effective treatments and considerations of what the person feels they wish to aim for. On occasions within the trial, clinicians and participant may have agreed on regimens that exceed the minimum number of doses that would be considered effective given a participant's characteristics. Since effective adherence denominator adjustments are intended simply to ensure minimum level of effectiveness, no adjustments were necessary in the case of these participants. That is to say the denominator adjustment was a strategic instrument to ensure minimal level of effectiveness is being reflected in the calculation of percent adherence. CFHealthHub interventionists were trained to continue to record prescriptions that fulfil the minimum effective dose requirement. In the example of figure B1, the participant is a person with chronic Pseudomonas aeruginosa infection who aimed for three daily doses of aztreonam but muco-active agent was not part of their agreed treatment target. Data feedback within CFHealthHub captured individualised targets by displaying a target reflecting the treatment(s) that the participant chose to aim for. If they used all three daily doses of aztreonam, their effective adherence on the day would be 3/(3 + 1) = 75%. This personalised decision would appear as the target set on CFHealthHub in the form of the green line on the graph (at 75% adherence). When fewer than three daily doses were used, the daily adherence bar will be displayed in orange (for example on 23 November). When three daily doses were used, which met their individualised target though it did not achieve an effective adherence of 100%, the daily adherence bar will be displayed in green (for example on 08 November). Therefore, the data display of CFHealthHub feeds back the individualised target. In this example, the agreed prescribed regimen was still recorded as three daily doses of aztreonam and a daily dose of dornase alfa, though the participant was only aiming for three daily doses of aztreonam. By lowering the treatment target rather than reducing prescribed doses, any deviation from effective targets will still be visible on CFHealthHub. It is important to emphasise that this data The green bar indicates that the participant's individualised target has been achieved on that particular day. The individualised target is NOT necessarily 100% The orange bar indicates that the participant's individualised target has not been achieved on that particular day The primary analysis of adherence for randomised clinical trial reporting was standardised using the concept of adherence to a regimen considered to be effective. Any deviation from the guidance to enter effective prescription into CFHealthHub or errors that were made based on a lack of awareness, for example of Pseudomonas aeruginosa status, were corrected for in the analysis which ensured that adherence at all sites and for all participants were being compared on an equal basis, i.e. effective adherence. That is to say the analysis of participant data for someone with chronic Pseudomonas aeruginosa infection only using inhaled antibiotic recognises it to be a regimen not considered to be maximally effective by international consensus. [5][6][7] Thus denominator adjustment in this case would ensure that the adherence level analysed against the primary outcome of exacerbation would be not be 100%, but would be capped to a maximum of 75%, as in the example of figure B1. Without such standardisation, a person with more effective nebuliser use would not be identified in the calculation of percent adherence. Rigour around effectiveness is an important element in understanding the relationship between adherence and health outcomes. For example, without denominator adjustment, a person with chronic Pseudomonas aeruginosa infection using an average daily dose of one inhaled antibiotic and one dornase alfa but prescribed a total of three daily doses would have adherence of 67% yet a similar person using just a daily dose of dornase alfa but prescribed a total of one daily dose would have adherence of 100%.
By using objectively-measured effective adherence as the method of calculating adherence, we can be confident that an increase in percent adherence reflects more effective medication use. It is important, given the use of effective adherence, to highlight that participants with chronic Pseudomonas aeruginosa infection were equally distributed between intervention and usual care. Yet usual care had slightly higher prescribed daily doses (mean 3.1 vs 2.9, see table 2 of main manuscript for breakdown of prescribed doses), meaning that the denominator adjustment would have reduced effective adherence among intervention participants to a greater extent compared to usual care. That is to say the use of objectively-measured effective adherence if anything, would bias against the intervention group.   It is reassuring that as the intervention was delivered, a clear between-group divergence in the mean inhaled doses emerged (figure B2). That is to say the intervention group used more doses of nebulisers, which mirrors the divergence in objectively-measured effective adherence ( figure B3). Therefore, the difference in calculated percent effective adherence was driven by the number of doses taken (numerator) among intervention participants rather than prescription (denominator) adjustments. The fact that the absolute number of doses between intervention and control diverges indicates that the use of objectively-measured effective adherence is capturing a difference in absolute treatment use between intervention and control.
The percent objectively-measured adherence without any adjustments also mirrors the difference observed with objectively-measured effective adherence, as shown in table B1.

APPENDIX C Participant recruitment
As is discussed in appendix D, "baseline adherence" of the participants may have been exaggerated by novelty effect and white coat adherence. Nonetheless, we also acknowledge that the study may have recruited a convenience sample that was more focused on adherence compared to the general cystic fibrosis (CF) population. The baseline median adherence of the participants was 52% whereas real world median adherence among adults with CF has been reported as closer to 35%. 1,2 Another observation supporting the contention that a more engaged sample was recruited in this trial is the intravenous (IV) antibiotic rejection rate for exacerbations among this sample of around 5% (see table S3 footnote), which is four-fold lower than in real-world dataset where the IV rejection rate is around 20%. 3 In the CONSORT diagram (figure 2 of main manuscript), we report that 3510 adults with CF were screened and 608 were recruited. The discrepancy between screening and recruitment was driven by a decision to prioritise rapid recruitment because more than two-thirds of large publicly funded trials in the United Kingdom (UK) failed to recruit to time and target. 4 As such, all adults with CF in participating centres were screened using data from the UK CF registry and investigators may have also first approached those they thought would be most amenable Although a biased sample that was more focused on adherence may have been recruited as the result of the recruitment strategy, it is important to consider the direction of any resultant bias. As is discussed in appendix D, there is a ceiling effect associated with high baseline adherence. 5,6 It may follow that scope for improvement in adherence in our trial was curtailed in the intervention arm by ceiling effects associated with high baseline adherence and nearly 30% of the participants having baseline adherence >75%. Therefore, any bias associated with the recruitment strategy would be towards null effect and the overall adherence difference of adjusted mean difference of 9.5 percentage points (95% confidence interval 8.6, 10.4) may have been an under-estimate.  Table 2 of the main manuscript suggests there may be some imbalance between usual care and intervention groups at baseline. The intervention group was around 1 year older (mean age 31.1±10.6 years versus 30.3±10.8 years) yet percent predicted forced expiratory volume in one second (FEV1) was higher by around 2 (60.7±23.5 versus 58.3±22.6) and annual intravenous (IV) antibiotics use was lower by around 3 days (24.2±27.9 days versus 27.7±33.0 days). This may suggest that the intervention participants had slightly better lung health at baseline, which may be due to higher adherence prior to recruitment. Indeed, there is also imbalance of "baseline adherence", that is the objectively-measured effective adherence level measured in the first two weeks post randomisation, which was around 9% in favour of the intervention group (54.1±33.0% versus 45.5±34.1%).
In this section, we deal with the following five issues: 1) Explain how the randomisation process could result in baseline imbalance despite 608 participants being randomised 2) Explore the likely impact of age on the baseline adherence 3) Provide analyses which explore the adherence trajectory for intervention versus usual care after minimising the imbalance of baseline adherence 4) Explore the impact of baseline imbalance in terms of the direction of bias on the observed effect size 5) Explore how these limitations can be minimised to make future trials more efficient

1) The randomisation process
The imbalance in baseline parameters is likely due to a randomisation process which involved two levels of stratification (centre and past-year IV days, as described in Section 9.1 of the protocol [available as supplementary material]) which limits the block size. Each centre recruited around 35 participants and the aim was to achieve approximately similar numbers of usual care and intervention participants in each centre, so that the centre interventionists were not overwhelmed by excess number of intervention participants. Thus the play of chance is not acting on 608 participants but is acting on a maximum block size of 35 with two levels of stratification to randomise participants into usual care and intervention; i.e. the play of chance is constrained by limited block size.

2) Impact of age on the baseline adherence
The intervention group was around 1 year older. The adherence imbalance at the initial part of the trial may in part be influenced by differences in the proportion of participants according to age categories. Multiple studies have demonstrated a strong association between the age categories (16-18 years, 19-25 years, 26-34 years, ≥35 years) and adherence levels. 1,2 The usual care arm has an excess of younger participants with lower adherence and the intervention arm has an excess of older participants with higher adherence ( figure D1). There were 27 usual care and 17 intervention participants aged 16-18 years, where the mean baseline adherence for 44 participants was 31%. There were 75 usual care and 91 intervention participants aged ≥35 years, where the mean baseline adherence for 166 participants was 62%. By plotting adherence according to age categories, the effect of age imbalance at the start of the trial is clearer. There is less adherence imbalance at the start of the trial when participants were grouped by age ( figure D1) except for the few participants aged 16-18 years (n=44, 7%). Some of the baseline adherence imbalance following age stratification may be due to the transient effect of enhanced white coat adherence in the intervention group who were aware from Day 1 that a planned 3-week meeting with interventionists to review their data would occur. This is consistent with the behaviour change technique of feedback used as part of the intervention and contrasts with the usual care group who were aware that adherence measurement would simply be used for research and neither fed back nor reviewed by interventionists.

3) Explaining the adherence trajectory for intervention versus usual care
Trial participants in both arms had never previously been monitored with data-logging nebulisers and would likely be susceptible to novelty effect and whitecoat adherence at the initial part of the trial, [3][4][5][6] with the consequence that adherence in the first two weeks was unrepresentative of their steady-state adherence. Of note, the baseline adherence imbalance was discussed in the previous section and was unrelated to novelty effect or white coat adherence. It is known that novelty effect and whitecoat adherence are relatively short-lived [3][4][5][6] and this is reflected in the initial sharp adherence decline for both arms seen in the study (figure 3 of main manuscript; figure D1). Among control participants who did not receive any intervention, this decline continued over the next 12 weeks to around 35%, which is the realworld objective adherence level for inhaled therapies among adults with CF, 7,8 and stayed at this level until the end of the trial. In the intervention group, the initial rate of decline was similar to the controls until the behavioural-change intervention started from Week 3 and adherence subsequently improved. It is also important to note that the separation in adherence curves between intervention and usual care participants occurred regardless of baseline adherence when curves were plotted by adherence categories ( figure D2).

4) The impact of baseline imbalance on the direction of bias
There is a ceiling effect associated with high adherence. 9,10 Indeed, subgroup analysis according to baseline adherence (figure D2) indicates minimal end of study between-group difference in objectively-measured effective adherence among those with baseline adherence >75%. It is therefore likely that a preponderance of high adherers among the intervention group would bias the overall adjusted adherence results towards null effect, i.e. the overall adjusted mean difference in objectively-measured effective adherence of 9.5 percentage points (95% confidence interval 8.6, 10.4) may have been larger had those with baseline adherence >75% been excluded.

5) How these limitations can be minimised to make future trials more efficient
As discussed in the main manuscript, the measurement of "baseline adherence" in the first two weeks post randomisation is a limitation of the trial. It would have been ideal to obtain an understanding of the study participants' actual baseline adherence by measuring adherence over longer periods prior to randomisation, which may allow white coat adherence among adults using data-logging nebulisers for the first time to wear off. The decay of usual care participants' adherence to baseline took approximately 12 weeks, suggesting the importance BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) of providing objective adherence monitoring technology to participants for at least 12 weeks before baseline adherence is captured. This would impact time scales for an adherence trial and the funding envelope requested. In our subsequent trials, we plan to nest the evaluation of adherence interventions within a digital learning health system (ISRCTN14464661) so that baseline adherence can be understood prior to randomisation. This has a number of benefits, including recruiting participants in whom adherence can be seen to improve from baseline (effectively removing the impact of whitecoat adherence) and greater efficiency by avoiding the recruitment of potential participants with maximal adherence at baseline.  In the main manuscript, the limitations of using exacerbation as the primary endpoint which may bias the result towards null effect and the difficulty of discerning the trajectory of intervention effect due to a lack of pre-randomisation steady-state adherence level were discussed. In this appendix, we discuss the other limitations of the trial.
One of the potential limitations is the delivery of both behavioural change and research processes by interventionists. The intervention was delivered via CFHealthHub, which was unavailable to usual care participants. Mixed-methods process evaluation of our two-centre pilot study, which specifically addressed contamination, demonstrated negligible contamination among usual care participants. 1 Outcome data were objective measures unlikely to be biased by interventionists' data collection. 2 Three other limitations of the trial might bias the observed results towards a null effect. First, a convenience sample was recruited with around 30% of the participants having baseline adherence >75%, a subgroup in whom an impact on outcome measures would be unlikely, rendering the trial less efficient. It is noteworthy that the intravenous (IV) antibiotic rejection rate in this trial was around 5% whereas the real-world IV rejection rate is typically four-fold higher at around 20%, 3 supporting the contention that a more engaged sample was recruited in this trial. The ceiling effect among high adherers means that the effect size would have been larger if high adherers were excluded (see appendices C and D). With this limitation, any observed difference in adherence in the trial could be considered particularly noteworthy.
Since trial participants may have better health outcomes than non-participants, 4 there may also be ceiling effect on health outcomes as well as ceiling effect on adherence. If we assume the intervention is able to impact people with lower levels of adherence, the outcomes seen in this opportunistic sample might have a larger effect size in the whole population where median adherence is ~30%. Interestingly, the FEV1 difference did not include unity in the subset of participants with adherence <25%. This further supports the assertion that focusing an adherence intervention study on participants with lower levels of adherence has the advantage of both trial efficiency and increased probability of impacting health outcomes such as FEV1.
Second, there was a period of server downtime which affected intervention delivery.
Adherence data were not lost but simply inaccessible during the downtime. Interventions were delivered over 80 weeks (9 months for recruitment) and the CFHealthHub server experienced a 43-day outage at one point, which delayed the receipt of data to the server such that the platform was inaccessible to all participants during this period. Intervention sessions would be rescheduled if adherence data were unavailable, meaning that no intervention took place during this period. The server hosting infrastructure was improved following the downtime, BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) reducing the likelihood of future issues. Fidelity assessments throughout the trial, which required the use of objective adherence data during sessions, showed reassuring scores of 93-97%. Given the importance of the platform for intervention delivery, unavailability would reduce the intervention effectiveness and bias the results towards null effect. However, in the spirit of intent-to-treat analysis, we did not make any adjustments to avoid over-estimating treatment effect. It is important to emphasise that periods where data transfer was delayed did not result in data loss as data were simply backed up and transferred once system transfer was restored.
Third, the trial was underpowered to detect the observed point estimate in forced expiratory volume in one second (FEV1). Sample size calculation in table S1 showed that the trial has under 80% power to detect a 6 point difference in between-group percent predicted FEV1. The observed between-group point estimate of 1.4 in percent predicted FEV1 at 12 months may simply be due to chance but is within the range observed for hypertonic saline at 48 weeks. 5 Overall, these four limitations (alongside the limitations of exacerbation as the primary outcome) reduced the trial's ability to demonstrate statistically significant improvements in lung health. The significant albeit small difference in body mass index (BMI) with the intervention versus usual care should be noted, and higher BMI has shown an association with higher FEV1. 6 It is possible that FEV1 improvement may emerge gradually over time with longer follow-up.
It is also possible that improvement in health outcomes may not be linearly associated with the increase in adherence; for example, there may be both a threshold effect and a ceiling effect. The relationship between improvement in treatment adherence and improvement in health outcomes among people with CF is relatively unexplored, in part because previous adherence trials did not demonstrate improved adherence. Further analyses would be performed using the ACtiF dataset to better understand the relationship between adherence to chronic therapies and health outcomes.
nutrition-related management of children and adults with cystic fibrosis and pancreatic insufficiency: results of a systematic review. J Am Diet Assoc 2008;108:832-9.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)