Development of data processing algorithm to calculate adherence for adults with cystic fibrosis using inhaled therapy – a multi-center observational study within the CFHealthHub learning health system

ABSTRACT Objectives To develop a robust algorithm to accurately calculate ‘daily complete dose counts’ for inhaled medicines, used in percent adherence calculations, from electronically-captured nebulizer data within the CFHealthHub Learning Health System. Methods A multi-center, cross-sectional study involved participants and clinicians reviewing real-world inhaled medicine usage records and triangulating them with objective nebulizer data to establish a consensus on ‘daily complete dose counts.’ An algorithm, which used only objective nebulizer data, was then developed using a derivation dataset and evaluated using internal validation dataset. The agreement and accuracy between the algorithm-derived and consensus-derived ‘daily complete dose counts’ was examined, with the consensus-derived count as the reference standard. Results Twelve people with CF participated. The algorithm derived a ‘daily complete dose count’ by screening out ‘invalid’ doses (those <60s in duration or run in cleaning mode), combining all doses starting within 120s of each other, and then screening out all doses with duration < 480s which were interrupted by power supply failure. The kappa co-efficient was 0.85 (0.71–0.91) in the derivation and 0.86 (0.77–0.94) in the validation dataset. Conclusions The algorithm demonstrated strong agreement with the participant-clinician consensus, enhancing confidence in CFHealthHub data. Publishingdata processing methods can encourage trust in digital endpoints and serve as an exemplar for other projects.


Introduction
Digital measures utilizing real-world patient data can be used in routine care or as endpoints in clinical trials.Compared to traditional endpoints captured in clinical settings, digital endpoints may offer greater insights into real-life patient experiences which are not reliably captured in the clinical setting [1].
Adherence can be described as the extent to which a person follows healthcare provider recommendations, such as taking medicines [2].Low adherence to prescribed medicines in longterm conditions is a key contributor to suboptimal clinical benefits and worse health outcomes [2,3].Electronic adherence monitoring devices enable adherence to function as a digital measure which offers greater objectivity than alternatives, such as self-report [4][5][6].
Cystic fibrosis (CF) is an archetypal long-term condition where life expectancy is approximately 50 years, driven by respiratory failure, resulting from chronic lung inflammation and recurrent infection [7,8].Most adults with CF are prescribed medicine regimens including multiple daily doses of inhaled therapy, usually delivered by an electronic nebulizer device.Higher adherence to inhaled therapy is associated with better outcomes, but realworld adherence is low at <40% [5,[9][10][11].
CFHealthHub is a UK-based multi-center Learning Health System.A Learning Health System is described as 'a health system in which outcomes and experience are continually improved by applying science, informatics, incentives, and culture to generate and use knowledge in the delivery of care' [12] CFHealthHub centers have access to the cloud-based CFHealthHub digital platform, which continuously captures and displays objective adherence data from nebulizer devices.This platform can be accessed by clinicians and adults with CF at all times.Certain clinicians within each CFHealthHub center are also trained to deliver a behavioral intervention which supports adults with CF to improve adherence through habit formation [13,14].CFHealthHub increased percent adherence to inhaled therapy and reduced perceived treatment burden in a 607-participant randomized controlled trial (RCT) [15].CFHealthHub is now active in over 50% of adult CF centers in England, as an evidence-based digital platform and behavioral intervention, which empowers adults with CF to self-manage their condition.A recent report from The Health Foundation recognized CFHealthHub as the only conditionbased, full Learning Health System with national reach in the UK, and is an exemplar for other long-term conditions [16].
The CFHealthHub digital platform requires the ability to accurately measure objective adherence data from nebulizer devices with electronic data capture (EDC) capability.The CFHealthHub digital platform is device agnostic, and compatible with both of the EDC-capable nebulizer devices used in the UK: the I-neb Adaptive Aerosol Delivery (AAD) System (Philips Respironics, Chichester, UK) and eFlow Technology nebulizers with an eTrack data-logging Controller (PARI Pharma GmbH, Starnberg, Germany), subsequently referred to as 'eTrack nebulizers.' AAD devices, such as the I-neb, can accurately determine whether a dose is completely administered, as aerosolized medicine is only released on breath activation of the user.Non-AAD devices, such as the eTrack nebulizer, do not have this functionality and therefore require data processing algorithms to determine completeness of the dose delivery with accuracy.Therefore, this work focuses on data from eTrack nebulizers only.The component parts of the eTrack nebulizer referred to throughout this article, are shown in Figure 1.
To ensure accurate calculation of percent adherence, the nebulizer data must be processed to count the number of doses of medicine which have been delivered 'completely' each day.This figure, produced for each day, is referred to as the '"daily complete dose count."'Each time a dose of medicine is initiated via an eTrack nebulizer, a log is created with the timestamp, duration of the dose, and a numeric code (known as an interruption code) recording whether the dose was considered 'complete' or not (Table 1).An interruption code of '4' denotes a 'complete' dose.Alternative interruption codes suggest the dose may have been 'incomplete.'For example, an interruption code of '1' suggests the dose was interrupted due to loss of power supply to the eTrack nebulizer controller.Most medicines delivered via an eTrack nebulizer are expected to take between 2-8 minutes to complete; therefore, all doses with a very short duration (<60s) are likely 'incomplete' (Personal Communication, Dr C Fuchs, PARI GmbH, Email, Jan 2021).
The method of processing doses with varying duration and interruption codes can result in different 'daily complete dose counts.'The example presented in Appendix A demonstrates that the data processing algorithm must be carefully considered, to accurately reflect the true 'daily complete dose count.'It is possible for a singular dose to be misclassified, but the algorithm still yields an accurate 'daily complete dose count' as explained in Appendix 1.During the CFHealthHub RCT, nebulizer data were processed using an algorithm based on expert advice from PARI GmbH [15].This involved considering all doses with duration ≥60s, and an interruption code of 2 (indicating disconnection of the aerosol head from the eTrack controller) or 4 (indicating that the dose completed as expected) as 'complete.'The algorithm excluded the following doses from the 'daily complete dose count:' • all duplicate doses, based on start time, duration, and interruption code, • all doses of duration less than 60 seconds, • all doses with interruption code 3 (suggesting there was no medicine in the device at the initiation of the dose), • all doses conducted in the EasyCare cleaning mode (identified by an interruption code >100), • all doses with a date of '01JAN2015' (suggesting device corruption).
After exclusions, the following doses are combined in calculating the 'daily complete dose count.' • all doses which were interrupted due to power failure or pre-set timeout (based on interruption codes 1,5,6,7,8) and duration ≥60s would be classified as partial dose, contributing 0.5 to the 'daily complete dose count.'If the subsequent dose is started within 1500s and had an interruption code of 2 (cable disconnection) or 4 (dose complete as expected), then these two doses would be combined to give 1 complete dose.
There is ongoing, real-world learning in the CFHealthHub Learning Health System, where many more adults are using eTrack nebulizers.In the 12 months prior to this work, approximately 6% of all doses recorded on the CFHealthHub digital platform had a duration of <60s and 24.5% were potentially 'incomplete,' as per the interruption code.The most accurate method of processing data from these doses is uncertain and requires stronger evidence than advice from the manufacturer.The objectives of this sub-analysis within CFHealthHub were: first, to understand how doses could be identified as 'complete' based on their duration and interruption code.Second, by triangulating eTrack nebulizer data with participants' records of taking each dose, to develop and validate a data processing algorithm to optimize the accuracy of the 'daily complete dose count' used in percent adherence calculations.These objectives align with the key aim of developing the CFHealthHub digital platform as one which maximizes the salience of adherence data and may also serve as an exemplar for other platforms capturing digital adherence data remotely.

Participants and methods
In this sub-analysis, data collected from eTrack nebulizers were triangulated against real-world records of what happened

Article highlights
• Supporting adherence to medicine regimens in long-term conditions requires accurate measurement of adherence.• The CFHealthHub Learning Health System offers a digital platform which can collect inhaled medicine usage data from nebulizer devices capable of electronic data capture.• Clinicians and people with cystic fibrosis collaborated to develop a data processing algorithm for these usage data to calculate the number of complete doses taken each day ('daily complete dose count').• The resultant data processing algorithm was considered highly accurate for calculating the "daily complete dose count."• Accurate nebulizer usage data processing allows for calculation of accurate adherence measurement, which can be used as both a digital study endpoint in but also as part of optimizing routine care.
during each dose, created by adults with CF.Participants were all eTrack nebulizer users who had consented to the CFHealthHub Learning Health System.Regulatory approval was provided by the London-Brent Research Ethics Committee (Reference number: 17/LO/0032).This analysis included adults with CF who had ≥20 nebulizer doses that were either <60s in duration or potentially 'incomplete,' as per their interruption code.There were no previous data to inform a target sample size.Since approximately 30% of the doses were expected to be of short duration or potentially 'incomplete,' 300 doses was chosen as a pragmatic target to provide 100 doses of interest, which should have encompassed an adequate range of different interruption codes.Due to constraints in clinical resources, we enriched the sample with doses of interest (short duration and/or potentially 'incomplete') so that an adequate range of different interruption codes could be captured over a shorttime duration.Therefore, purposive sampling was used to identify participants with a particularly high number of doses of interest (≥20 doses of interest per week), such that 10 participant-weeks of data each for derivation and validation datasets was determined as sufficient.
Participants were included from three centers, which are part of the CFHealthHub Learning Health System.These centers were selected due to the relatively high prevalence of eligible participants and the availability of clinicians to complete this work.Approximately 8,000 doses from the CFHealthHub digital platform were screened between 15 October 2021 and 31 October 2021 (two weeks prior to the sub-analysis start date).Data collection was between 1 November 2021 and 15 December 2021.
Local clinicians approached eligible adults with CF from these three centers using a standardized script to facilitate the initial discussion (Appendix B).In the first telephone call, participants were informed about this work and invited to provide verbal consent to participation.If they agreed, a longer call would be arranged at a future time to discuss their nebulizer data.

Phase 1 (data calibration)
Once relevant participants were identified for inclusion in this analysis, local clinicians were provided with a log of each participant's nebulizer data for the preceding week, extracted from the CFHealthHub digital platform.These data included the timestamp, duration, and interruption code for each recorded dose.To mitigate recall bias, clinicians used these data to help prompt participant recall of: 1) the time the dose was started; 2) the medicine used for each dose; 3) if they considered the dose 'complete' or not; and 4) if relevant, a reason why the dose was considered '(in) complete.'Discussions around nebulizer usage are part of routine clinical care in CF, and the data used to inform these discussions are available to all clinicians providing care to adults with CF enrolled in the CFHealthHub Learning Health System on request.
The clinician and participant reached a consensus as to whether each nebulizer dose was likely to have been 'complete.'For example, the participant recognizing an appropriate residual volume of the medicine in the medication reservoir suggests the dose was 'complete' even though the eTrack nebulizer had not recognized the dose as complete.Clinicians then asked participants to keep a record of their nebulizer usage for prospective data collection in Phase 2.
Participants were asked to record the name of the medicine being nebulized, and the date and time the dose was started.They were also asked to note anything remarkable about that dose, for example, if they experienced a power failure or disruption, and if they considered the dose to be 'complete.'A follow-up call was then arranged with each participant to review their prospective record.
The purpose of Phase 1 was to familiarize participants with the process of discussing their nebulizer usage and to consider ways of determining whether a dose was 'complete,' in preparation for the prospective data collection.Data from Phase 1 were not used in the analysis.

Phase 2 (prospective data collection)
Clinicians contacted participants at the agreed time to review 1-2 weeks of nebulizer data extracted from the CFHealthHub digital platform, as described in Phase 1.These data were discussed with the participants and triangulated with their records of the corresponding doses, which the clinicians then cross-checked against the nebulizer data.Clinicians completed a data collection form using Microsoft Excel (version 16.62).As in Phase 1, the clinician and participant reached consensus as to whether each recorded dose was considered 'complete' or not, along with a brief description, e.g.'participant reported their device timed-out after 20 minutes.'An example of a completed data collection form is shown in Figure 2.
Following collection, the prospective data were divided into derivation and validation datasets, prior to any analysis being undertaken.Therefore, clinicians were not aware of the resultant algorithm at the time of data collection.For participants providing two separate weeks of data, one week of data was allocated to derivation and the other week's data to validation.This was done to ensure both datasets contained an adequate range of interruption codes, given the small number of participants (n = 12) and doses (approximately 300 in each dataset).
Researchers reviewed the derivation dataset, consisting of nebulizer data (date & time, duration, and interruption code for each dose), and whether the dose was considered 'complete' by the clinician-participant consensus, with associated free-text comments where available.First, all doses with duration of <60s were reviewed.Next, all doses with duration ≥60s were stratified by the interruption code listed in Table 1, and each resultant group was reviewed separately.With this information, an algorithm to calculate a 'daily complete dose count' from the nebulizer data was developed, which used dose start time, duration, and interruption code only, to determine if a dose was likely to be 'complete.'Appendix C contains a full description of the number of doses in each combination of duration and interruption code, with a justification for how the algorithm would process these combinations, based on the triangulated nebulizer data and consensus 'daily complete dose count.'If a dose was likely to be 'complete,' then it would be included and counted as a 'complete' dose, however if it was likely to be 'incomplete,' it would be excluded or combined with another dose to create a single 'complete' dose.
The agreement between algorithm-derived 'daily complete dose count' and consensus-derived 'daily complete dose count' in the derivation dataset was determined using both percent accuracy and kappa values.In view of the clustered nature of the dataset, bootstrapping was used to calculate kappa and agreement values [17].This involved bootstrapping of 1000 samples from the original dataset, calculation of kappa and agreement values for each sample (i.e.1000 values were calculated for each participant) and then ascending re-order of those values to provide a median, 2.5 th and 97.5 th centile as measures of central tendency and dispersion.In addition, the extent to which the algorithm under-or over-estimated the consensus-derived 'daily complete dose counts' were quantified with absolute differences in both 'daily complete dose counts' and percent adherence between the two measures.
An a-priori target was to proceed to validation if the algorithm-derived 'daily complete dose count' was ≥80% accurate in comparison to the consensus-derived 'daily complete dose count,' which was considered as the 'reference standard.'If the accuracy was <80%, then the derivation dataset would be rereviewed to refine the algorithm.

Results
Twenty-two adults with CF receiving care in Center 1 (n = 8), Center 2 (n = 8) and Center 3 (n = 6) were identified as potentially eligible for inclusion.
Eight of these 22 adults were excluded after approach, and two excluded after review of their nebulization data prior to Phase 2. Twelve participants were included in the analysis.Their baseline characteristics are shown in Table 2.The flow of recruitment, reasons for exclusion and allocation are shown in Figure 3.
One week of data from 10 participants comprised the derivation dataset, with one week of data from 10 participants comprising the validation dataset.Eight of the 12 participants contributed data to both derivation and validation datasets, as they each provided two weeks of data, compared to the four other participants, contributing one week of data each who were assigned to either the  derivation or validation datasets in a 1:1 ratio.A total of 74 patient days (with 295 doses) from 10 patients were used in the derivation dataset and 69 patient days (with 309 doses) from 10 patients in the validation dataset.Dose durations and interruption codes for the derivation dataset were reviewed and results are reported in Table 3.

Proposed screening algorithm
We proposed the following process for identifying 'complete' doses from the nebulizer data. 1) Initially screen out: • All doses with duration <60s.

2) Combine
• Any 2 or more doses starting within 120s of each other.

3) Finally screen out:
• Doses with duration <480s due to loss of supply voltage or battery power to the eTrack nebulizer (interruption code = 1 or 6).

Accuracy of the proposed screening algorithm
In the derivation dataset, there was a high level of agreement between the algorithm-derived 'daily complete dose count' and the consensus-derived 'daily complete dose count.'The kappa coefficient was 0.85 with 95% confidence interval of 0.71-0.91,and accuracy was 87.5% (77.0-95.7).Similar agreement and accuracy were seen in the validation dataset (kappa co-efficient 0.86 [0.77-0.94],accuracy 89.9% [84.3-95.5]).These results along with the total numbers of doses considered 'complete' by both the algorithm and consensus are reported in Table 4.The absolute differences in 'daily complete dose count' between these two measures were 10 (out of 266 'complete' doses by consensus) in the derivation dataset and 7 (out of 267 'complete' doses by consensus) in the validation dataset.The absolute differences in mean percent adherence calculated using the 'daily complete dose count' from these two measures were 3.2% and 2.8%, respectively, as reported in Table 5.

Discussion
Through examination of nebulizer data and triangulation of these data with participant records, we have developed an algorithm to generate a 'daily complete dose count.'This algorithm involved excluding all doses of <60s, combining doses which start within 120s of each other and then using a combination of the interruption code and dose duration to determine which other doses are likely to be 'complete.'The resultant 'daily complete dose count' was 87.5% accurate in the derivation dataset and 89.9% accurate in an internal validation dataset.By outlining the process for designing and validating a data processing algorithm in collaboration with adults with CF, we aim to inspire trust in adherence data from the CFHealthHub digital platform as a digital measure.At a patient-level, adherence data from the CFHealthHub digital platform is central to the development of personalized care plans, an essential part of caring for people with long term conditions [18].A tangible benefit of the greater objectivity is that actual pattern of nebulizer use can be understood by clinicians, who can then provide personalized advice on how to fit nebulizer use within the other routines of the person with CF.
An erroneously high '"daily complete dose count"' risks overestimating adherence, which risks then falsely reassuring both adults with CF and clinicians that adherence is higher than it is.The consequence of this is that some people may be under-served by the health care system by not being offered adherence support when they could benefit from it.Furthermore, overestimating adherence may result in unnecessary treatment escalation in the event of clinical deterioration.Conversely, underestimating adherence could create conflict between adults with CF and their clinicians and lead to both parties losing faith in the adherence data available on the CFHealthHub digital platform.
We recognize that the algorithm produced a marginally higher 'daily complete dose count' than the participant-clinician consensus, which was considered the 'reference standard' in this project.However, the difference in percent adherence derived from the 'daily complete dose count' (around 3% against an average adherence exceeding 90%) was clinically negligible.It is worth noting Among the 74 days of data, there were 66 days (89%) with identical "daily complete dose counts" by both algorithm & consensus, 4 days (5%) with a higher count by consensus and 4 days (5%) with a higher count by algorithm.β Among the 69 days of data, there were 62 days (90%) with identical "daily complete dose counts" by both algorithm & consensus, 1 day (1%) with a higher count by consensus and 6 days (9%) with a higher count by algorithm.that a participant-clinician consensus for whether each dose of treatment is 'complete' is not feasible outside of a dedicated research project.It would be unreasonably burdensome for all participants on CFHealthHub to keep a detailed daily diary of all their nebulizer doses.Therefore, we are reassured by the small differences noted in this study.Within a Learning Health System where data used to generate knowledge which drives and measures improvement work, optimizing data quality is critical [12].Previous quality improvement work, underpinned by large datasets, has focussed on measures of completeness, conformance and plausibility, through the production of automated functions with statistical software [19].In this work, we have developed an algorithm to improve calculation of 'daily complete dose counts.'This was strengthened by working alongside adults with CF to gain a qualitative understanding of circumstances of doses, from which quantitative data were produced.
A key strength is that this is the first report triangulating nebulizer data with the real-world experiences of adults with CF using eTrack nebulizers within the CFHealthHub Learning Health System, using a parsimonious study design to minimize the burden of adults with CF.Putting people at the center of research into their condition is a key priority for improving care in long-term conditions [18].Continuous patient engagement is recommended during the evaluation phase of digital measures such as this [20,21].
There are, however, some limitations.To minimize the burden of adults with CF, there is a need to use a parsimonious study design enriching the cohort with participants with relatively high numbers of short or 'potentially incomplete' nebulizer doses.By applying a purposive sampling strategy within three of the 15 CFHealthHub centers, the sample of participants could be criticized as being less generalizable.For example, the mean adherence of the sample exceeded 90% when real-world median adherence is only around 30% [9].However, this study design allowed us to capture an adequate range of short and/or 'potentially incomplete' doses to enhance the applicability of the resultant algorithm in a larger population.
Due to the limited number of participants imposed by scarce resources, data from different weeks by the same participant were included in both the derivation and validation datasets.This ensured an adequate range of interruption codes in both datasets.Whilst no individual dose appeared in both datasets, the inclusion of the same participant in both datasets meant that the validation dataset is not external to the derivation dataset.Further validation of this algorithm in other CFHealthHub centers would be useful.The fact that CF is a rare disease, with approximately 7,000 adults with CF in the UK and the relative infrequency of potentially incomplete doses (<25% of all doses on the CFHealthHub digital platform) contributed to the small sample size of 12 participants and 604 doses [22].
Another limitation was reliance on self-report as to which medicine was being administered for each dose, and circumstances around doses which were considered potentially 'incomplete.'Currently, eTrack nebulizers lack the technology to identify the specific medicine being administered.We also cannot identify, from data alone, whether prolonged nebulization duration is due to equipment malfunction or person factors.We mitigated potential recall bias by prospectively asking participants to keep contemporaneous records for data collection during the study period, rather than relying on retrospective recall.We also crossreferenced their records against the nebulizer data.An alternate approach of direct observation of nebulizer usage in a controlled environment would have allowed the gold standard data collection around whether a dose was 'complete' or not.This was considered unfeasible given the time and resource burden for clinicians and participants, which is a known barrier to participation in research within CF [23].Our chosen methods were parsimonious and better captured the real-world experience of adults with CF using eTrack nebulizers where factors, such as consumable wear and dose interruptions come into play.
Finally, this study was limited to adults with CF who were using eTrack nebulizer devices, which represents 88% of the approximately 1400 adults with CF who were enrolled in CFHealthHub.At the time of this study, only two data-logging nebulizer devices are used in the UK: the eTrack nebulizer and the I-neb.As an adaptive aerosol delivery device, the I-neb already provides dose completeness information in the following scale: 'Full;' '>12.5%;<100%;' '<12.5%' and 'none.'Therefore, such an algorithm is not required for I-neb users.
This data processing algorithm will now be embedded within the CFHealthHub digital platform, where further validation in larger and more diverse cohort is recommended.These data are used to support adherence in the real-world setting [24].CFHealthHub Learning Health System also has a research function and is currently undertaking a large observational study, exploring the role of co-adherence to inhaled therapy for adults with CF who are taking novel oral treatments [25].Digital endpoints may present unique challenges in the value assessment of pharmaceuticals or cost evaluation of consumed medications.Recognizing this, CFHealthHub adherence data are also used to optimize medicines supply by aligning supply with actual usage, with the potential to realize significant cost savings [26,27].For both of these workstreams to be effective, data accuracy, which is strengthened by this work, is critical.
Inspired by information uncovered during this work, we have since completed a formal study of how these data can identify adults with CF who are having frequently prolonged nebulizer durations.Troubleshooting and replacement of consumable parts led to mean 37% reduction in the time adults with CF spent on nebulizer treatment each day [28].This is a further demonstration of how paying attention to data from digital measures can have real-world benefits for people with long-term conditions.

Conclusion
We have developed a data processing algorithm by triangulating nebulizer usage data with participants' real-world records, which was then tested in a multi-center dataset.The algorithm has high levels of accuracy.Co-designing and validating this algorithm helps optimize the accuracy of, and trust in, adherence data from the CFHealthHub digital platform.These data can be used to optimize clinical interactions at a patient-level, underpin quality improvement work at an organization-level, and facilitate national benchmarking at a system-level.The methods we

Figure 2 .
Figure 2. Example of a completed data collection form for a single participant.

Table 2 .
Baseline characteristics of participants and datasets.

Table 4 .
Kappa and accuracy scores for the algorithm-derived 'daily complete dose count' against the gold standard consensus-derived 'daily complete dose count' in the derivation and validation datasets.

Table 5 .
Differences in total 'complete' doses and mean adherence calculations when using the algorithm-derived and consensus-derived 'daily complete dose counts.

Table 3 .
Distributions of combinations of dose duration and interruption codes (IC) in the derivation dataset, with decisions of how to use these in the data processing algorithm.
22. Cystic Fibrosis Trust.UK Cystic Fibrosis Registry Annual Data Report 2022.2023.23. Lee M, Hu XY, Desai S, et al.Factors influencing clinical trial participation for adult and pediatric patients with cystic fibrosis.J Cyst Fibros.2021;20(1):57-60.doi: 10.1016/j.jcf.2020.08.019 24.Sandler RD, Antrobus S, Cameron S, et al.P220 the CFHealthHub learning health system -supporting a community of practice to deliver a normal life expectancy in cystic fibrosis.J Cystic Fibrosis.2023;22:S132.doi: 10.1016/S1569-1993(23)00593-3 25.Daniels T. 284: explaining the efficacy-effectiveness gap for ivacaftor: the potential impact of adherence to maintenance inhaled therapy on outcomes.J Cystic Fibrosis.2021;20:20.doi: 10.1016/ S1569-1993(21)01709-4 26.Bevan A, Hoo ZH, Totton N, et al.Using a learning health system to understand the mismatch between medicines supply and actual medicines use among adults with cystic fibrosis.J Cyst Fibros.2022;21(2):323-331.doi: 10.1016/j.jcf.2021.09.007 27.Bevan A, Hoo ZH, Totton N, et al.Corrigendum to "Using a learning health system to understand the mismatch between medicines supply and actual medicines use among adults with cystic fibrosis".J Cyst Fibros.2022;21(2): 323-331.doi: 10.1016/j.jcf.2021.09.007 28.Sandler RD, Lai L, Anderson A, et al.P207 CFHealthHub allows clinicians to identify people with long nebuliser durations and intervene to reduce duration.J Cystic Fibrosis.2023;22:S128.doi: 10.1016/S1569-1993(23)00581-7 [If yes, arrange time to call again in a week.Ask participant if they might be happy to keep a log of the times and days they do a treatment for the next week (e.g. on their phone, or a piece of paper).Ask them to note the: 1) date; 2) time; 3) name of treatment; 4) and anything to note with each treatment or the eTrack in general e.g.Did they see two ticks -one when the treatment had finished and one when the data had transferred?Did the device lose power?Did they do an 'easycare' clean?Did the grey cable disconnect?Did they pause their treatment?Did they turn the device off or did it turn off itself?etc.]