Low-dose oral theophylline combined with inhaled corticosteroids for people with chronic obstructive pulmonary disease and high risk of exacerbations: a RCT

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 11/58/15. The contractual start date was in July 2013. The draft report began editorial review in April 2018 and was accepted for publication in November 2018. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Graham Burns reports personal fees from Boehringer Ingelheim GmbH (Ingelheim am Rhein, Germany), Teva Pharmaceutical Industries Ltd (Petah Tikva, Israel), Chiesi Farmaceutici SpA (Parma, Italy), Pfizer Inc. (New York City, NY, USA) and AstraZeneca plc (Cambridge, UK), and non-financial support from Chiesi and Boehringer Ingelheim, outside the submitted work. Rekha Chaudhuri reports personal fees from AstraZeneca, GlaxoSmithKline (GSK) plc (London, UK), Teva and Novartis International AG (Basel, Switzerland) for advisory board meetings, outside the submitted work. Anthony De Soyza reports grants and non-financial support from AstraZeneca and Chiesi, non-financial support from Boehringer Ingelheim, grants and personal fees from GSK, Bayer AG (Leverkusen, Germany) and Pfizer, and grants from Forest Laboratories (New York City, NY, USA)/Teva, outside the submitted work. He is also a member of the National Institute for Health Research (NIHR) Health Technology Assessment (HTA) Elective and Emergency Specialist Care (EESC) Panel. Simon Gompertz reports personal fees from Pfizer and GSK, outside the submitted work. John Norrie reports grants from the NIHR HTA programme during the conduct of the study, was a member of the NIHR HTA Commissioning Board (2010–16) and is currently Deputy Chairperson of the NIHR HTA General Board (2016–present) and a NIHR Journals Library Editor (2014–present). Andrew Wilson reports grants from F. Hoffmann-La Roche (Basel, Switzerland), outside the submitted work. David Price reports grants and personal fees from Aerocrine AB (Stockholm, Sweden), AstraZeneca, Boehringer Ingelheim, Chiesi, Mylan NV (Canonsburg, PA, USA), Mundipharma International Ltd (Cambridge, UK), Napp Pharmaceuticals Ltd (Cambridge, UK), Novartis, Pfizer, Teva, Theravance Biopharma (San Francisco, CA, USA) and Zentiva Group a.s. (Prague, Czech Republic); personal fees from Almirall SA (Barcelona, Spain), Amgen Inc. (Newbury Park, CA, USA), Cipla Ltd (Mumbai, India), GSK, Kyorin Pharmaceutical Co. Ltd (Tokyo, Japan), Merck Sharp & Dohme (Kenilworth, NJ, USA), Skyepharma Production SAS (Saint-Quentin-Fallavier, France); grants from AKL Research and Development Ltd (Stevenage, UK), the British Lung Foundation, the Respiratory Effectiveness Group and the UK NHS; and non-financial support from the NIHR Efficacy and Mechanism Evaluation and HTA programmes, outside the submitted work. He also reports stock/stock options from AKL Research and Development Ltd, which produces phytopharmaceuticals and owns 74% of the social enterprise Optimum Patient Care Ltd (Cambridge, UK, and Australia and Singapore) and 74% of Observational and Pragmatic Research Institute Pte Ltd (Singapore).

Copyright statement

© Queen’s Printer and Controller of HMSO 2019. This work was produced by Devereux et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2019 Queen’s Printer and Controller of HMSO

The burden of chronic obstructive pulmonary disease on individuals and the NHS

Chronic obstructive pulmonary disease is a major personal and public health burden. 23,24 Data from 591 UK general practices comprising The Health Improvement Network (THIN) indicate that the prevalence of diagnosed COPD in the UK increased from ≈991,000 in 2004 to 1.2 million in 2012. 2 COPD is the fifth leading cause of death in the UK, accounting for ≈5% of all deaths (≈30,000 deaths in 2014). More than 80% of COPD patients, irrespective of disease severity, report a reduced QoL. 24–26 Comorbidities are an important feature of COPD, contributing to ill health and treatment burden. It has been estimated that, in the UK, 33% of people with COPD have hypertension, 19% have ischaemic heart disease, 18% have depression, 11% have diabetes mellitus and 6% have heart failure. 23 Over 50% of people currently diagnosed with COPD in the UK are < 65 years of age, and 24 million working days are lost each year as a result of COPD, with £3.8B per year being lost through reduced productivity. 23

Chronic obstructive pulmonary disease costs the NHS > £1B per year. In 2001, average annual NHS direct costs were £819 (> £1300 in severe COPD) for each COPD patient; 60% of this was accounted for by exacerbations and 19% was due to drug costs. 27 The Hospital Episode Statistics database shows that emergency hospital admissions for exacerbations of COPD in the UK have steadily increased as a percentage of all admissions, from 0.5% in 1991 to 1% in 2000 and to 1.5% in 2008/9. 28 In 2008/9, COPD exacerbations resulted in 164,000 hospital admissions in the UK, with an average length of stay of 7.8 days, accounting for 1.3 million bed-days. 28 COPD is the second leading cause of emergency admission to hospital in the UK and is one of the most costly inpatient conditions treated by the NHS. 23,24 At least 10% of emergency admissions to hospital are as a consequence of COPD, and this proportion is even greater during winter. Approximately 25% of patients who have been diagnosed as having COPD are admitted to hospital at some point, and ≈15% of COPD patients are admitted each year. 23,24 Over 30% of patients admitted to hospital with an exacerbation of COPD are re-admitted within 30 days, and an average of 12% of COPD patients die in the year following admission to hospital. 19

Despite advances in management that have led to the current National Institute for Health and Care Excellence (NICE) COPD guidelines,24 there is still an unmet need for improved pharmacological treatment of COPD, particularly the prevention of exacerbations.

Standard chronic obstructive pulmonary disease therapy

Standard COPD therapy remains suboptimal. At the time when the Theophylline With Inhaled CorticoSteroid (TWICS) trial was conceived, most international COPD management guidelines recommended the use of inhaled corticosteroids (ICSs) – usually in combination with inhaled long-acting β₂-agonists (LABAs), known as ICS/LABA – to reduce COPD exacerbation rates and to improve lung function and QoL. 1,24 Although more recent guidelines advocate the use of LABAs in combination with long-acting muscarinic antagonists (LAMAs), ICS/LABA and ICS/LABA/LAMA combinations remain major therapeutic options and continue to be used very widely in the treatment of COPD. 29,30 However, when compared with the marked responses observed in asthma, ICSs in COPD fail to fully suppress airway inflammation and patients continue to have exacerbations despite high ICS doses. Furthermore, little or no positive impact of ICS on mortality or disease progression is evident31,32 and concerns have been raised about long-term sequelae of high-dose ICS use in COPD. 33,34 A relative insensitivity of COPD airway inflammation to the anti-inflammatory effects of high-dose ICS has been demonstrated in induced sputum and airway biopsies of people with COPD. 35–37

In recent years, molecular mechanisms contributing to the reduced corticosteroid sensitivity of COPD have been elucidated. The chronic airway inflammation of COPD is driven by expression of multiple inflammatory genes regulated by acetylation of core histones, which open up the chromatin structure, allowing transcription factors and ribonucleic acid (RNA) polymerase II to bind to deoxyribonucleic acid (DNA), enabling gene transcription and increased synthesis of inflammatory proteins. 38 In COPD, there is increased acetylation of core histones associated with the promoter regions of inflammatory genes, with the degree of acetylation being positively associated with disease severity. 39 Histone acetylation is reversed by histone deacetylase (HDAC) enzymes. Corticosteroids appear to work by reversing histone acetylation through the recruitment of a specific HDAC called HDAC2,38,40,41 thereby switching off activated inflammatory genes. In people with COPD, increased histone acetylation appears to be a consequence of markedly reduced HDAC2 activity/expression in airways, lung tissue and alveolar macrophages. 39 It has been shown that the oxidative stress of COPD activates the enzyme phosphoinositide 3-kinase (PI3K)-δ, which then phosphorylates downstream kinases, resulting in the phosphorylation and inactivation of HDAC2. 41,42 The critical role played by reduced HDAC2 in the corticosteroid resistance of COPD is demonstrated by the finding that the corticosteroid resistance of COPD bronchoalveolar macrophages is completely reversed by overexpressing HDAC2 (using a plasmid vector) to levels seen in control patients without COPD. 40

Low-dose theophylline may have synergistic anti-inflammatory effects with corticosteroids

Oral theophylline has been used in the treatment of COPD for > 70 years, but usually at doses required to achieve relatively high blood concentrations (10–20 mg/l). It has been observed that the reduced HDAC2 activity of COPD can be reversed in a dose-dependent manner by low doses of theophylline; moreover, low-dose theophylline reduces corticosteroid insensitivity in COPD such that there is a marked synergistic interaction between theophylline and corticosteroids in suppressing the release of inflammatory mediators from alveolar macrophages from COPD patients. This in vitro work has shown that at (low) concentrations of 1–5 mg/l theophylline increases HDAC2 activity (sixfold) but at (high) concentrations of over ≈10 mg/l theophylline inhibits rather than stimulates HDAC2 activity. 43,44 These studies show that at concentrations of 1–5 mg/l, there is a marked synergistic effect between theophylline and corticosteroids, with theophylline inducing a 100- to 10,000-fold increase in the suppressive effect of corticosteroids on the release of pro-inflammatory mediators. Such an increase in corticosteroid potency is worthy of clinical interest, particularly if associated with reduced exacerbation rate. An explanation for the ability of low-dose (i.e. 1–5 mg/l) theophylline to increase HDAC activity has been described: it specifically inhibits the enzyme PI3K-δ with consequent restoration of HDAC2 activity to normal in COPD macrophages, rendering them steroid responsive. In mice exposed to cigarette smoke,42 steroid-resistant lung inflammation has also been found to be reduced by low-dose theophylline when given together with steroids. Similarly, rats exposed to cigarette smoke were found to have markedly decreased lung HDAC2 expression, and that reduced HDAC2 expression was correlated with increased lung destruction index. 45 The increased lung destruction index was restored to normal with ICS treatment in combination with low- (but not high-) dose theophylline. It was concluded that low-dose theophylline might provide protection from cigarette smoke damage and improve the anti-inflammatory effects of steroids by increasing HDAC2 activity.

In human peripheral blood mononuclear cells, corticosteroid insensitivity and reduced HDAC2 activity after oxidative stress have been shown to be reversed with low concentrations of theophylline. 46 In a study of human alveolar macrophages extracted from resected lung samples, the addition of hydrogen peroxide reduced HDAC expression and was associated with an increase in interleukin 8 (IL-8) and matrix metalloproteinase-9 (MMP-9) release. 47 The addition of low-dose theophylline restored HDAC expression to levels above that observed with LABA, ICS and ICS/LABA.

These basic research studies suggest that low-dose (i.e. 1–5 mg/l) theophylline could increase HDAC activity and hence reduce corticosteroid resistance in COPD patients, thereby enabling ICS to switch off inflammation and potentially reduce exacerbation rates more effectively. This is supported by findings from two small randomised controlled trials (RCTs) and a population-based health administration database study. The first RCT in 35 patients with acute COPD exacerbations found that low-dose theophylline increased responsiveness to corticosteroids as measured by increased HDAC activity and further reduced concentrations of pro-inflammatory mediators in induced sputum compared with ICSs alone. 48 In the second small (n = 30) pilot RCT of COPD patients, the combination of low-dose theophylline with high-dose ICSs was associated with increased HDAC activity, improved lung function and reduced sputum inflammatory cells and mediators, whereas either drug alone was ineffective. 49 A Canadian health administration database study of 36,492 COPD patients reported that treatment with theophylline alone or in combination with ICS was more protective against exacerbations than treatment with LABA or ICS/LABA [relative risk (RR) 0.89, 95% confidence interval (CI) 0.87 to 0.92]. 50

More recent studies, however, have not replicated the results of earlier studies. Fexer et al. 51 used data from a German ambulatory COPD management programme and closely matched 1496 COPD patients commenced on theophylline with 1496 COPD patients not commenced on theophylline. The use of theophylline was associated with an increased likelihood of exacerbation [hazard ratio (HR) 1.41, 95% CI 1.24 to 1.60] and hospital admission (HR 1.61, 95% CI 1.29 to 2.01). Although it was concluded that theophylline is associated with an increased incidence of exacerbations and hospitalisations, it should be noted that this study did not identify those patients on low-dose theophylline. 51 The Spanish Low-dose Theophylline as Anti-inflammatory Enhancer in Severe Chronic Obstructive Pulmonary Disease (ASSET) trial recruited patients with COPD while hospitalised for a COPD exacerbation and randomised to low-dose theophylline (100 mg twice a day) or matched placebo in addition to usual ICS/LABA treatment. 52 In total, 70 patients were randomised (theophylline, n = 36; placebo, n = 34) and 46 completed the year of treatment (theophylline, n = 23; placebo, n = 23). The addition of theophylline had no effect on the COPD exacerbation rate or plasma/sputum concentrations of HDAC and inflammatory mediators. It should be noted that the study was small and designed to detect a 50% reduction in exacerbations.

Conventionally, oral theophylline has been used as a bronchodilator in COPD; however, to achieve modest clinical effects, relatively high blood concentrations (of 10–20 mg/l) are required. The bronchodilator effect of high-dose theophylline is the consequence of inhibition of phosphodiesterase (PDE) and the consequent relaxation of airway smooth muscle. However, non-specific inhibition of PDE by theophylline is also associated with a wide range of well-recognised side effects that may occur within the conventional therapeutic range of plasma theophylline, namely nausea, gastrointestinal upset, headaches, insomnia, seizures, cardiac arrhythmias and malaise. Theophylline toxicity is dose related, and this is an issue with conventional theophylline use because the therapeutic ratio of theophylline is small and most of the beneficial bronchodilator effect occurs when near-toxic doses are given. 53 Theophylline is metabolised by cytochrome P450 mixed function oxidase; as a consequence, theophylline use is further complicated by significant drug interactions with drugs commonly prescribed to people with COPD, for example clarithromycin or ciprofloxacin. 54 The narrow therapeutic index, modest clinical effect and side effect profile of theophylline, together with drug interactions, the need for blood concentration monitoring and the availability of more effective inhaled therapies, have resulted in current COPD guidelines relegating high-dose theophylline to third-line therapy. 1

The TWICS trial was a pragmatic, double-blind, randomised, placebo-controlled clinical trial that was built on emerging evidence that low-dose (i.e. 1–5 mg/l) theophylline may produce a beneficial synergistic effect in COPD by increasing the corticosteroid sensitivity of the airway inflammation underlying COPD and, as a consequence, reduce the rate of COPD exacerbation when used in conjunction with ICSs.

Hypothesis

The hypothesis being tested was that the addition of low-dose theophylline to ICS therapy in COPD reduces the risk of COPD exacerbation requiring treatment with antibiotics and/or oral corticosteroids (OCSs) during the year of treatment, delivers QoL improvements and is cost-effective.

Objectives

The primary objective of the trial was to determine the clinical effectiveness and cost-effectiveness of adding low-dose theophylline to ICS therapy in patients with COPD and a history of two or more exacerbations treated with antibiotics and/or OCSs in the previous year in relation to the number of exacerbations in the 1-year treatment period requiring therapy with antibiotics and/or OCSs.

The secondary objectives were to compare the following outcomes between participants treated with low-dose theophylline and those treated with placebo:

• hospital admissions with a primary diagnosis of exacerbation of COPD

• total number of episodes of pneumonia

• total number of emergency hospital admissions

• lung function

• all-cause and respiratory mortality

• drug reactions and serious adverse events (SAEs)

• health-related QoL

• disease-specific health status

• total ICS dose/usage

• health-care utilisation

• incremental cost per exacerbation avoided

• lifetime cost-effectiveness based on extrapolation modelling

• modelled lifetime incremental cost per quality-adjusted life-year (QALY).

An additional secondary objective was the time to the first exacerbation of COPD.

Outcomes

Trial design

The trial protocol has been published in an open-access journal. 55

The TWICS trial was a pragmatic, double-blind, randomised, placebo-controlled, parallel-arm, UK multicentre clinical trial that compared the addition of low-dose theophylline or placebo for 52 weeks with current COPD therapy that included ICSs in patients with COPD who had experienced two or more exacerbations of COPD in the previous year treated with OCSs and/or antibiotics. The aim was to recruit 1424 participants, with at least 50% recruited in primary care. The trial was approved by Scotland A Research Ethics Committee (REC) (reference number 13/SS/0081) and the Medicines and Healthcare products Regulatory Agency (MHRA) (EudraCT 2013-001490-25, Clinical Trial Authorisation 21583/0218/001). All participants provided written informed consent, which included consent to inform a participant’s GP of involvement in the trial and consent to pass on a participant’s name and address to a third-party distributer that delivered the trial drug to participants’ homes. Figure 1 provides a schematic representation of the trial design and schedule. Face-to-face trial assessments were carried out at recruitment/baseline and at 6 and 12 months, as shown in Figure 2.

FIGURE 1.

Trial design.

FIGURE 2.

Flow diagram of trial schedule.

The trial was registered on 19 September 2013 as ISRCTN27066620.

Participants

Identification

Potential participants were recruited from both primary and secondary care sites across the UK. To ensure generalisability, the intention was that the majority of participants (> 50%) would be recruited from primary care. Recruitment strategies differed between centres depending on local geographic and NHS organisational factors.

Recruitment/baseline visit

At the recruitment visit, a participant’s eligibility was confirmed by a medically qualified doctor and fully informed consent was recorded in writing. Baseline data (see Data collection) were also collected.

Randomisation/treatment allocation

Participants were randomised, usually by a research nurse, using a computerised randomisation system available as both an interactive voice response telephone system and an internet-based application; the internet application was used for all randomisations within the trial. The randomisation service was created and administered by the Centre for Healthcare Randomised Trials (CHaRT) in the University of Aberdeen. Consenting participants were stratified by trial centre (for participants recruited in secondary care) or area (for participants recruited in primary care) and by where the participant had been identified (primary or secondary care) and then randomised with equal probability to the intervention (low-dose theophylline) and control (placebo) arms.

The random allocation sequence for the TWICS trial was generated using permuted blocks. This provided randomly generated blocks of entries of varying sizes permuted for each combination of trial centre/area and where the participant had been identified (primary or secondary care). Each entry was assigned a treatment according to a randomly generated sequence utilising block sizes of two or four. Each treatment option was assigned an equal number of times within each block, ensuring that the total number of entries assigned to each treatment remained balanced. The sequence of blocks was also random, so it was not possible for anyone to determine the next treatment to be allocated based on previous allocations made during the randomisation process.

It was possible to randomise a participant only if the relevant eligibility criteria had been met. In addition to trial centre/area and where the participant had been identified (primary or secondary care), sex, height, weight, smoking status (and, for smokers, number of cigarettes per day) and date of birth were captured during the randomisation process to calculate the correct dosage of trial medication for that participant and assign an appropriate drug pack.

With this information captured, the randomisation process assigned a trial number (i.e. a participant identification), allocated a treatment and assigned a drug pack. The user/caller was notified of the trial number and drug pack either on screen or during the randomisation telephone call. The allocated treatment remained blinded throughout, with neither the user/caller nor the participant (or anyone involved in the participant’s care or the assessment of outcomes) made aware of the allocation. All of the data captured or assigned were saved to a secure database.

The random permuted blocks that defined how treatments were allocated to participants were created by the CHaRT programming team during the system development process. The system built to utilise these permuted blocks was tested by a run of simulated randomisations that allowed the outcomes to be cross-checked and validated. Before the randomisation system went ‘live’, enough blocks were created to ensure that entries existed for the maximum expected number of participants across the maximum expected number of trial centres/areas. However, the randomisation system was flexible enough to allow the option to add further permuted blocks to the list if more were required during the lifetime of the trial. In such circumstances, randomly generated sequences in blocks of two and four continued to be utilised.

Intervention

The active intervention was 200-mg tablets of Uniphyllin modified release (MR) taken once or twice a day for 52 weeks. The placebo was manufactured to be visually identical, and was also taken once or twice a day for 52 weeks. The packaging and labelling of active and placebo interventions were identical. The intervention was for 52 weeks of therapy. The 200-mg tablets of Uniphyllin MR and placebo were supplied by Napp Pharmaceuticals Ltd (Cambridge, UK). Napp Pharmaceuticals Ltd is the holder of the marketing authorisation for 200-mg tablets of Uniphyllin MR (marketing authorisation number: PL 16950/0066–0068). Uniphyllin Continus 200 mg, 300 mg and 400 mg is licensed for the treatment and prophylaxis of bronchospasm associated with COPD, asthma and chronic bronchitis; consequently, theophylline was administered within licensed indication. 54 Placebo tablets were manufactured by Mundipharma Research Ltd (Cambridge, UK).

Supply of trial medication

Each participant received their first bottle of 4 weeks of trial medication (or placebo) from a participating clinical trials pharmacy. For secondary care sites, this was usually the clinical trials pharmacy based at that secondary care site. For participants recruited in primary care study sites, the first bottle of medication was dispensed from the clinical trials pharmacy in NHS Grampian and couriered to a participant’s address.

Each participant also received two further supplies of six bottles (each bottle being a 4-week supply). These supplies were dispatched to participants by a third party (AndersonBrecon, Hereford, UK) and delivered to participants’ addresses via a courier. These shipments were made around weeks 3 and 27 to enable continuity of supply. Receipt of trial medication to a participant’s home address was confirmed by signature on receipt.

Data collection

Baseline, outcome and safety data were collected by face-to-face assessments conducted at recruitment/baseline (week 0), 6 months (week 26) and 12 months (week 52). Participants were telephoned 2 weeks after starting the trial medication to ensure that they were tolerating the medication. The schedule for data collection in the trial is outlined in Table 2. If a participant was unable to attend a scheduled follow-up assessment visit because of an acute illness, for example exacerbation of COPD, or other reasons, the visit was postponed and the participant was assessed within 4 weeks of the scheduled assessment visit. Participants unable to attend a face-to-face assessment at 6 and 12 months were followed up by telephone or home visit, or were sent the questionnaires to complete at home.

The following data were collected.

Participant withdrawal

Participants who withdrew from treatment (e.g. because of unacceptable side effects, or because they were prescribed a contraindicated medicine for > 3 weeks) but who agreed to remain in the trial for follow-up were followed up at 6 and 12 months. Those who did not want to attend for clinical follow-up at 6 and 12 months could be followed up by telephone or home visit, or could opt to receive questionnaires at home. Participants who wanted to withdraw from trial follow-up could continue to contribute follow-up data by agreeing to have data extracted from their primary and secondary care medical records.

Sample size

The sample size of 1424 was estimated on the basis of the ECLIPSE study reporting the frequency of COPD exacerbations in 2138 patients. 21 For patients identical to the target population (who in a 1-year period have had at least two self-reported COPD exacerbations requiring antibiotics or OCSs), the mean number of COPD exacerbations within 1 year was 2.22 (SD 1.86). 21 Given a similar rate in the placebo arm, 669 participants were needed in each arm of the trial to detect a clinically important reduction in COPD exacerbations of 15% (i.e. from a mean of 2.22 to 1.89) with 90% power at the two-sided 5% significance level. Allowing for 6% loss to follow-up,83 this was inflated to 712 participants in each trial arm, giving 1424 in total.

The sample size of 1424 included a 6% loss to follow-up based on a Cochrane Review of oral theophylline in COPD. 83 During the present trial, a higher proportion of participants than expected ceased their trial medication (although most were not lost to follow-up). With the appropriate REC and regulatory approvals, recruitment continued beyond 1424 in the time available, with the total number recruited being 1578 to counteract this loss of person-years on medication. Recruitment ended in August 2016.

Statistical analysis

All analyses were prespecified in the statistical analysis plan, which was approved by both the Trial Steering Committee (TSC) and the Data Monitoring Committee (DMC) in advance of analysis. The statistical analysis plan can be found on the project web page at www.journalslibrary.nihr.ac.uk/programmes/hta/115815/#/ (accessed 23 April 2019). Unless prespecified, a 5% two-sided significance level was used to denote statistical significance throughout and estimates are presented alongside their 95% CIs. No adjustments were made for multiple testing. All analyses were conducted in accordance with the intention-to-treat (ITT) principle, with a per-protocol analysis performed as a sensitivity analysis. The per-protocol analysis excluded participants who were not compliant, with compliance being defined as taking ≥ 70% of their expected doses of trial medication. All analyses were undertaken in Stata^® version 14 (StataCorp LP, College Station, TX, USA).

Categorical variables are described with number and percentage in each category. Continuous variables are described with mean and SD if normally distributed and median and interquartile range (IQR) if skewed. The number of missing data is reported for each variable.

Health economics

Patient and public involvement

A person with COPD was an independent voting member of the TSC. Initially, this was a patient from the Aberdeen Chest Clinic who was nominated by Chest Heart & Stroke Scotland (CHSS) as part of its Voices Scotland initiative. In 2015, this person had to resign from the TSC because of ill health and was replaced by another patient, who is living with COPD, from the Aberdeen Chest Clinic.

Early versions of the trial protocol and PILs were reviewed by a representative from the British Lung Foundation–North Region and by a person who lives with COPD and attends the chest clinic at the Freeman Hospital, Newcastle. They both attended the trial initiation meeting, purposely held in Newcastle in February 2013, and contributed suggestions and changes to the final trial design that were reflected in the protocol and PIL.

The TWICS trial was publicised in 2014 by a press release that included supportive quotations from the British Lung Foundation and CHSS. This publicity resulted in members of the public with COPD volunteering to participate; with their permission, their details were passed on to their local TWICS trial site.

We anticipate that the patient and public involvement (PPI) member of the TSC will comment on the results letter to be sent to trial participants. It is also anticipated that the publication of the trial results will be co-ordinated with press releases from the participating academic/NHS institutions, the British Lung Foundation and CHSS. Members of the trial team will be participating in local public engagement with research activities.

Protocol amendments

There were seven protocol amendments, which are summarised in Table 4.

Trial oversight

A TSC, with independent members, including PPI members, oversaw the conduct and progress of the trial. An independent DMC oversaw the safety of participants in the trial.

Breaches

Breaches of trial protocol or good clinical practice were recorded and reported to the sponsor. A summary of breaches is included in Appendix 3. Participants who were the subject of a breach remained in the ITT population, the safety population and the per-protocol population (if compliance criteria were met).

Recruitment

Participants were recruited to the trial between February 2014 and August 2016. During this 31-month period, 141 UK sites were opened to recruitment. Once opened, some sites (n = 20) failed to recruit any participants to the trial. Reasons for this included staff changeover, lack of eligible patients, competing priorities, practice closure and eligible patients who did not agree to take part.

In total, 1578 participants were recruited from 121 sites (Table 5). A detailed summary of recruitment, by site, is given in Appendix 4. In summary, across 33 secondary care sites, 1101 participants were recruited, and, across 88 primary care sites, 477 participants were recruited. Of those recruited in secondary care, 464 were identified in primary care. Overall, 59.6% of participants were identified in primary care.

The initial funding included a 24-month recruitment period. Delays in manufacturing and packaging the trial medication led to the projected recruitment being reprofiled across 21 months. After ≈6 months of recruitment, it became clear that we were unlikely to meet the recruitment target within 21 months. To address this, several measures were successfully implemented: ‘second’- and ‘third’-wave sites were opened earlier than planned; additional primary and secondary care sites were identified and a rolling programme of opening these sites was established; a 6-month extension to recruitment was granted by the funder; and additional recruitment time was accommodated within the existing funding. In the 31-month recruitment period we were granted approval to over-recruit beyond the original target of 1424 participants. The justification for this was the higher than anticipated numbers of participants who ceased taking the trial medication (see Chapter 4, Treatment adherence/compliance).

Figure 3 shows the original recruitment targets, the reprofiled recruitment targets (to accommodate the delay in manufacturing/packaging), the revised recruitment targets (after the extension to recruitment was granted) and the actual recruitment.

FIGURE 3.

Recruitment.

Post-randomisation exclusions

Eleven participants were recruited in error and were then excluded. None of these participants took any dose of trial medication and all are excluded from all trial analyses. Reasons for these post-randomisation exclusions are given in Table 6.

Sixteen participants who were recruited into the TWICS trial were subsequently noted to be ineligible for the trial at the point of recruitment. All 16 participants had taken at least one dose of trial medication and were retained in the follow-up and included in the trial analyses. Seven of these were taking a form of diltiazem at recruitment, and were identified during a review of all baseline medication recorded for participants. Diltiazem can cause a slight increase in serum theophylline concentration; however, any effect is usually clinically insignificant. 94 One of these participants took the trial medication for ≈10 days but stopped because they experienced symptoms considered likely to be related to theophylline. A further participant experienced some symptoms that may have been side effects related to the trial medication and stopped after ≈4 months. One further participant experienced some symptoms that may have been side effects related to the trial medication but did not cease taking the trial medication.

In this same review, a further five participants were noted to have been taking a contraindicated medication at baseline. Three participants were noted to be taking a form of oestrogen. Serum theophylline concentration is slightly increased by concomitant oestrogen but no toxicity has been reported. 94 Two participants who were co-prescribed oestrogen continued to take the trial medication through their 12-month follow-up with no ARs. The third participant experienced symptoms (thought to be related to theophylline) and stopped taking the trial medication after 14 days. One participant was noted to have been taking febuxostat at recruitment. They had taken the trial medication through their 12-month follow-up and experienced symptoms that may have been side effects related to the trial medication. High-dose febuxostat has been reported to possibly increase serum theophylline. 94 One participant was noted to have been taking roflumilast at recruitment. Although roflumilast has no reported effect on serum theophylline concentrations, the two drugs act through phosphodiesterase enzymes,94 albeit theophylline is a phosphodiesterase inhibitor at serum concentrations of 10–20 mg/l when used at ‘high-dose’ levels. This participant had taken trial medication throughout their 12-month follow-up without any ARs.

Three participants were taking a form of theophylline at recruitment. In two cases, this was noted only after the participant had completed their 12-month follow-up (and the participants had taken the trial medication throughout their 12-month follow-up). In the third case, this was noted after the participant had taken the trial medication for 8 days; the trial medication was then stopped. In all three cases, no ARs relating to the trial medication were noted.

One participant was recruited into the TWICS trial when they were already participating in another CTIMP for an unrelated condition. The participant did not disclose this at the time of recruitment and it was not clearly documented in their hospital notes. No interaction between the TWICS trial medication and the medication used in the other trial is likely. The participant continued to take the trial medication for ≈11 months, at which point the participant ceased taking the trial medication as a result of a non-related throat problem.

Baseline characteristics

Baseline characteristics are presented for the 1567 included participants (after exclusion of the 11 post-randomisation exclusions). The theophylline and placebo groups were well balanced in terms of demographic and disease characteristics at baseline.

The mean age of participants was 68.4 years (SD 8.4 years) (Table 7). Just over half of the participants (53.8%) were male. Approximately one-third (31.7%) were current smokers; the remainder were ex-smokers. The median number of pack-years smoked was 42 (IQR 27.7–56.0 pack-years). The mean BMI was 27.2 kg/m² (SD 6.1 kg/m²).

The median number of participant-reported exacerbations in the 12 months prior to recruitment was 3 (IQR 2–4) and the mean number of exacerbations was 3.6 (SD 2.2) (Table 8). The majority of participants (79.9%) were prescribed the ‘triple therapy’ combination of ICS, LABA and LAMA at baseline. Almost one-fifth (16.7%) were prescribed ICS and LABA. The remainder were prescribed ICS only (2.0%) or ICS and LAMA (1.5%).

Comorbidities, as reported by participants, were relatively common. Almost one-fifth of participants (18.3%) had a concurrent diagnosis of asthma. Four per cent of participants reported a diagnosis of bronchiectasis. Just over one-third of participants (38.2%) reported a diagnosis of hypertension. Thirteen per cent reported ischaemic heart disease and 6.7% reported a previous cerebrovascular event. Almost one-third (28.0%) reported anxiety or depression in the previous 5 years. Eleven per cent had a diagnosis of diabetes mellitus and 12.8% had a diagnosis of osteoporosis.

Measurement of lung function at baseline revealed that the mean FEV₁ was 51.7% (SD 20.0%) predicted. Using the GOLD classification,1 13.6% were classified as having very severe COPD, 37.7% as having severe COPD, 39.6% as having moderate COPD and 9.2% as having mild COPD.

The mean score on the CAT was 22.6 (SD 7.7), indicating that, overall, COPD was having a high impact on the lives of participants (Table 9). Considering the cut-off points used to interpret the scores derived from the CAT, COPD was having a low impact on the lives of 5.3% of participants, a medium impact on the lives of 29.9% of participants, a high impact on the lives of 44.4% of participants and a very high impact on the lives of 20.4% of participants.

The mean EQ-5D-3L utility score was 0.63 (SD 0.28). The mMRC dyspnoea score revealed that 7.1% of participants were too breathless to leave the house, 27.6% had to stop for breath after walking ≈100 m, 31.5% walked more slowly than contemporaries on level ground because of breathlessness, 28.3% became short of breath when hurrying or walking up a slight hill and only 5.5% of participants were not troubled by breathlessness except on strenuous exercise.

A comparison of the participants recruited in primary and secondary care indicated that those identified in secondary care were slightly younger, were more likely to be ex-smokers and had experienced a greater number of exacerbations in the previous 12 months. Furthermore, a higher proportion of those identified in secondary care had more severe COPD, were on triple (ICS/LAMA/LABA) therapy and long-term antibiotic use, and had a significantly greater prevalence of comorbidities, such as bronchiectasis, ischaemic heart disease and osteoporosis (see Appendix 5, Table 35). Participants recruited in secondary care had a higher CAT score and a slightly lower QoL.

Clinical effectiveness of low-dose theophylline compared with placebo

In this chapter, we report the results of people with COPD being treated for 1 year with low-dose theophylline compared with placebo. A total of 1578 participants were randomised to theophylline or placebo; 11 post-randomisation exclusions resulted in 1567 participants being eligible to initiate trial medication and for whom baseline characteristics have been reported (see Chapter 3). Follow-up data were unavailable for 31 (2%) participants (theophylline, n = 16; placebo, n = 15); therefore, the results presented for the ITT analysis are based on 1536 participants (theophylline, n = 772; placebo, n = 764). Figure 4 shows the Consolidated Standards of Reporting Trials (CONSORT) flow diagram for the ITT analysis. In total there were 1489 person-years of follow-up data: 747 person-years in the theophylline group and 742 person-years in the placebo group (Table 10).

FIGURE 4.

The CONSORT flow diagram for the ITT analysis. a, Reasons for ineligibility were as follows: 16 did not meet the inclusion criteria for established COPD diagnosis or had a predominant respiratory disease other than COPD, 10 had not had two exacerbations in the previous year, seven did not meet the smoking history criteria, seven were taking contraindicated medication, three were not currently using ICSs, one was not clinically stable, two were participating in another clinical trial, one was currently taking theophylline, one had known or suspected hypersensitivity to theophylline, one was pregnant, two had severe heart disease and 11 did not meet two or more of the inclusion criteria.

Intention-to-treat analysis

Secondary outcomes

Safety outcomes (safety population)

The safety population comprised all participants who were randomised and included in the trial (n = 1567) and initiated their trial medication. Five out of 788 (0.6%) participants allocated theophylline did not initiate medication; 9 out of 779 (1.2%) participants in the placebo group did not initiate placebo. The safety population consisted of 1553 (99.1%) participants (theophylline, n = 783; placebo, n = 770).

Subgroup analysis (intention to treat)

Appendix 5, Table 40 details the results of the subgroup analysis for the prespecified subgroups. Given the exploratory nature of the analyses, we present 99% CIs. Figure 5 displays this information, alongside the p-values for the interaction in the adjusted model. There was no evidence at the 1% level of statistical significance that any effect of low-dose theophylline differed between subgroups of age, sex, smoking status, BMI, COPD treatments, exacerbation history, COPD severity, baseline ICS dose or use of maintenance OCSs.

FIGURE 5.

Forest plot of estimates from the subgroups. Vertical dashed line represents no effect (IRR = 1), vertical solid line indicates the overall treatment effect for exacerbation (IRR = 0.992). a, Upper limit of CI truncated to 1.5, actual value is 4.09; b, upper limit of CI truncated to 1.5, actual limit is 2.20.

Treatment adherence/compliance

Adherence/compliance was defined as having taken ≥ 70% of expected doses of trial tablets. In the ITT population (n = 1536), 1180 (76.8%) participants fulfilled the definition of adherent/compliant (and made up the per-protocol population). In the theophylline-allocated group, 181 out of 772 (23.4%) participants were classed as non-adherent/non-compliant; three of these never initiated treatment, 171 were non-persistent with (i.e. ceased) trial medication and seven persisted with trial medication but from returned medication it was evident that they were non-adherent/non-compliant (Table 19). In addition, 32 out of 591 low-dose theophylline participants fulfilled the adherent/compliant definition despite not persisting with the trial medication, usually very late in the treatment period (Table 20). In the placebo group, 175 out of 764 (22.9%) participants were classed as non-adherent/non-compliant; six never initiated medication, 159 were non-persistent with trial medication and 10 persisted with trial medication but medication returns demonstrated poor implementation (see Table 19). A further 34 participants were non-persistent with medication but fulfilled the definition of adherent/compliant because they ceased trial medication late into the treatment period (see Table 20). In summary, the per-protocol population comprises 1180 participants (591 from the theophylline arm and 589 from the placebo arm) and there were 1146 person-years of follow-up data (see Table 19). A comparison of the proportion who were non-adherent/non-compliant (23.4% in the theophylline arm vs. 22.9% in the placebo arm) was not significant (p = 0.802). In total, 203 out of 772 participants in the theophylline arm were non-persistent with medication, compared with 193 out of 764 in the placebo arm (unadjusted IRR 1.05, 95% CI 0.84 to 1.32).

Per-protocol analysis

The per-protocol population comprised the 1180 participants of the ITT population who met the trial definition of adherence with the trial medication. The per-protocol analysis comprised 591 participants allocated to low-dose theophylline and 589 allocated to placebo (Figure 6).

FIGURE 6.

The CONSORT flow diagram for the per-protocol analysis.

Primary outcome: total number of exacerbations of chronic obstructive pulmonary disease necessitating a change in management

In the per-protocol population, 591 theophylline-allocated participants had a mean of 2.20 (SD 1.96) exacerbations, compared with a mean of 2.14 (SD 1.92) exacerbations among the 589 placebo participants. There were 1298 exacerbations in the theophylline group and 1258 in placebo group. The adjusted IRR for COPD exacerbation was 1.00 (95% CI 0.91 to 1.10), indicating no difference in the exacerbation rate during the 12-month follow-up period for those on low-dose theophylline compared with those receiving placebo who were adherent/compliant with the trial medication (Table 21).

Secondary outcomes

Sensitivity analysis

We undertook a sensitivity analysis for the primary outcome and a number of secondary outcomes that excluded the 33 participants who died during the 12-month follow-up period. This left 1503 participants in the ITT population (753 theophylline and 750 placebo). Appendix 5, Tables 41 and 42, give the details of these analyses.

Summary

There was no evidence that, overall, low-dose theophylline significantly reduced the number of COPD exacerbations requiring treatment compared with placebo. There was some evidence that low-dose theophylline reduced exacerbations that necessitated hospital admission, with most benefit being evident in a small [1% (13/1556)] subgroup of patients frequently hospitalised with COPD. The total number of emergency hospital admissions (non-COPD related) did not significantly differ between groups, and nor did total episodes of pneumonia or mortality. Lung function was similar across the 12-month follow-up in the two groups. The impact of the disease on patients was measured by the CAT, mMRC dyspnoea scale and HARQ; no significant differences were found. The safety profile of low-dose theophylline was similar to placebo. There was no evidence that the treatment effect differed in any of the prespecified subgroups.

Baseline resource use and costs

Baseline resource use and costs are presented in Table 27.

There is no significant difference between the arms for any of these baseline resources.

Resource use

Table 28 reports the mean resource use per participant for complete cases, during the 12-month follow-up period.

As discussed in Chapter 4, the treatment of exacerbations at hospital was significantly different between groups: more exacerbations were treated in hospital in the placebo group than in the theophylline group (p = 0.02).

Missing data

The disaggregated level of missing data affecting resource use is reported below; these are broken down into exacerbations, COPD maintenance treatment and non-COPD emergency hospital admissions.

Costs

Table 30 reports complete-case costs (unadjusted). Differences between arms are calculated using a GLM with identity link, gamma family and a cluster for centre number. Regular medication was not included in these costs because there was no significant difference between arms in regular medication count.

There is a significant difference of £452 (95% CI £133 to £771) in the mean total costs between arms, with placebo being more costly than theophylline. This difference is driven by the difference in exacerbation mean costs between arms: £447 (95% CI £186 to £709) higher in the placebo arm. The difference in exacerbation costs is driven by the location of the treatment of an exacerbation. The mean ‘location of exacerbation treatment’ cost is £422 (95% CI £171 to £673) higher in the placebo arm than in the theophylline arm. As presented in Chapter 4, this is driven by a higher number of exacerbations treated in hospital in the placebo arm than in the theophylline arm. This is reflected in the health economics analysis for which the location of treatment costs are further broken down into ‘treatment at home’, ‘care by services to prevent hospitalisation’ and ‘admitted to hospital’. The ‘treatment at home’ and ‘care by services to prevent hospitalisation’ resource use costs show no significant differences between arms; however, the ‘admitted to hospital’ cost is £416 (95% CI £177 to £655) higher in the placebo arm than in the theophylline arm, which is a statistically significant result.

At a per-exacerbation level, this difference can be explored further. The mean cost per exacerbation treated in hospital is £3613 [standard error (SE) £342] in the placebo arm and £2671 (SE £220) in the theophylline arm, a significant difference of £941 (SE £386) (95% CI £140 to £1743). The 10 most costly observations (> £10,000) were all in the placebo arm, and were the result of hospital stays of > 40 days. Because of the lack of treatment effect, we believe this difference to be a chance finding and not a real result of the trial. The distribution for length of hospital stay is similar for both arms apart from a small excess of participants in the placebo arm with longer stays. It is important to note that the proxy for hospital length of stay is length of exacerbation and that this is likely to overestimate length of stay in hospital. In total, 319 exacerbations were treated in hospital: 185 in the placebo arm and 134 in the theophylline arm.

The cost of treatment of exacerbations differed significantly between arms, the mean cost per participant being £25 lower in the theophylline arm. At a per-exacerbation level, this is driven by treatment with oxygen. The mean cost of oxygen use per exacerbation treated with oxygen was £141 (SE £52) (95% CI £40 to £243) lower in the theophylline arm than in the placebo arm. The difference in oxygen treatment is driven by the fact that seven participants, six of whom were in the placebo group, were treated with oxygen for > 51 days, at a cost per exacerbation of > £1000.

The wide SDs for hospitalised exacerbations, treatment of exacerbations, non-COPD emergency hospital admissions and other health services use indicate a wide range of individual participants’ costs in these resource groups.

No other resource use costs are significantly different between arms, which is reflected in the fact that there was no difference between arms for the non-intervention, non-exacerbation costs presented in Table 30.

Economic outcome

Complete-case EQ-5D-3L data and QALYs are reported in Table 31.

Utilities from the EQ-5D-3L data at baseline and the 6- and 12-month follow-ups and QALYs are higher in the placebo arm than in the theophylline arm; however, these differences are not significant.

Multiple imputation

Multiple imputation results are presented in Table 32 for costs and QALYs.

Multiple imputation results mirror the complete-case results, with costs significantly higher in the placebo arm (a difference of £439). Total QALYs are higher in the placebo arm; however, this is not a statistically significant result (a difference of 0.004).

Bootstrapping

To explore the robustness of these results, 1000 non-parametric bootstrapped samples were taken from the observed data. The results were plotted using a cost-effectiveness plane to illustrate the mean differences between the arms in incremental costs and QALYs.

Non-adjusted bootstrapped results are presented in Figure 7. This cost-effectiveness plane clearly illustrates that the majority of total mean costs are lower in the theophylline arm than the placebo arm, with the majority of incremental samples falling in the south-east and south-west quadrants of the cost-effectiveness plane (below the horizontal axis of £0). The majority of total mean QALYs are lower in the theophylline arm than in the placebo arm, represented by the majority of bootstrapping samples falling in the south-west quadrant where the placebo arm has higher mean QALYs than the theophylline arm. The cost-effectiveness plane includes an ellipse to illustrate the 95% confidence level.

FIGURE 7.

Cost-effectiveness plane (unadjusted).

This uncertainty is explored further using cost-effectiveness acceptability curves.

The unadjusted bootstrapped results are presented in Figure 8. At a willingness-to-pay threshold of £20,000, there is a 75% chance of theophylline being cost-effective. At £30,000, there is a 64% chance of theophylline being cost-effective. However, these results should be viewed with caution as there is no significant difference in QALYs or clinical effect, and the difference in costs is driven by a very small number of participants with prolonged hospital admissions and the likelihood that the finding of a difference between arms for exacerbations treated in hospital is a chance finding. Moreover, as discussed in Adjusted analysis, the cost benefits of theophylline are not evident in multivariate models.

FIGURE 8.

Cost-effectiveness acceptability curve (unadjusted).

Adjusted analysis

Multiple imputation total mean costs were adjusted for baseline variables that were significant predictors of cost. These were medication count at baseline, EQ-5D-3L data at baseline, offset time (time spent in the trial), age, number of hospitalisations for exacerbations in the 12 months prior to randomisation and number of exacerbations in the 12 months prior to randomisation. A cluster command was used for centre number.

Multiple imputation total mean QALYs were adjusted for baseline variables that were significant predictors of QALYs. These were baseline EQ-5D-3L data, medication count at baseline, offset time, age, sex, hospitalisation for exacerbations in the 12 months prior to randomisation and exacerbations in the 12 months prior to randomisation. A cluster command was used for centre number. These results are presented in Table 33.

When multiple imputation total costs are adjusted, there is a trend towards higher costs in the placebo arm; however, this difference is not significant.

Adjusting QALYs for baseline characteristics results in theophylline having higher QALYs than placebo; however, this difference is not significant.

Figure 9 illustrates that, when the results are adjusted for baseline characteristics, the results are more uncertain: the majority of total mean costs in the theophylline arm are still lower than in the placebo arm, although this is now not a significant result. In addition, the QALYs are marginally higher in the theophylline arm, again not a significant result. The ellipse represents the 95% confidence levels.

FIGURE 9.

Cost-effectiveness plane (adjusted).

The adjusted bootstrapped results are presented in a cost-effectiveness acceptability curve in Figure 10. At a willingness-to-pay threshold of £20,000, there is a 90% chance of theophylline being cost-effective. At £30,000, there is an 85% chance of theophylline being cost-effective. Again, these results should be viewed with caution as there was no significant difference between arms for QALYs or treatment effect.

FIGURE 10.

Cost-effectiveness acceptability curve (adjusted).

Exacerbation costs were also adjusted separately to explore the adjustment on the significant difference in exacerbation costs between arms. Strong predictors of exacerbation costs were offset time, hospitalisation for exacerbations in the 12 months prior to randomisation and exacerbations in the 12 months prior to randomisation. A cluster command was used for centre number. These results are presented in Table 34 and show that, for adjusted exacerbation costs, although there is a trend for higher costs in the placebo arm, this difference is not significant. The mean cost difference has decreased from £447 to £67.

Cost-effectiveness

For cost per QALY, the unadjusted results suggest that, although theophylline is cheaper than placebo (significant result), the QALY gain is higher in the placebo arm than in the theophylline arm (non-significant result). The adjusted results suggest that theophylline dominates: it is cheaper than placebo and results in a higher QALY gain. However, this result should be interpreted with caution because the difference in QALYs is not significant. This is mirrored by the trial primary outcome: theophylline is not clinically effective in terms of reducing exacerbations.

Main results

The results of this trial show that, for people with COPD at high risk of exacerbation, the addition of low-dose oral theophylline to a drug regimen that includes an ICS confers no overall clinical or health economic benefit. This result was evident from both the ITT and the per-protocol analyses. The primary outcome measure for this trial was the total number of exacerbations of COPD requiring changes in management (minimum management change: use of OCSs and/or antibiotics) during the 1-year treatment period, as reported by the participant. For the 11 prespecified secondary outcome measures, the addition of low-dose theophylline had no clinical or health economic benefit in 10 of these measures. The addition of low-dose theophylline did reduce the number of COPD exacerbations necessitating hospital admission (often classified as ‘severe’)68 (adjusted IRR 0.72, 95% CI 0.55 to 0.94); however, further inspection of the data indicated that this difference was the consequence of a small excess of participants allocated to placebo (n = 10) having three or more hospital-treated exacerbations, which accounted for 39 of the extra 51 hospital-treated exacerbations in the placebo arm. This effect on hospital admissions was also evident in the per-protocol analysis. Given that adjustments for multiple comparisons were not performed, it is possible that this finding could be due to type I error. However, in the light of a 2018 report96 showing that another PDE inhibitor (roflumilast) is most beneficial in people with prior COPD hospitalisation for exacerbation and greater exacerbation frequency, this finding warrants further investigation. The safety data demonstrated that the addition of low-dose theophylline was not associated with an increase in SAEs or ARs.

Relevance to existing literature

Oral theophylline has been used in the treatment of COPD and asthma for > 70 years. Conventionally, oral theophylline has been used as a bronchodilator in COPD, this effect being mediated by inhibition of PDE. However, in order to achieve modest clinical effects, relatively high blood concentrations (of 10–20 mg/l) are required, but, at these concentrations, non-specific inhibition of PDE is also associated with a wide range of well-recognised side effects, for example nausea, palpitations and headaches. A Cochrane review published in 2002 and updated in 201083 identified 20 randomised placebo-controlled trials of theophylline in COPD, all of crossover design, using dosing schedules to obtain conventional plasma theophylline levels in the therapeutic range (10–20 mg/l), that is conventional high-dose theophylline. The number of participants in these trials ranged from 8 to 60; the total number of participants in the 20 trials was 488. The duration of the studies was 9–90 days, the mean age of participants ranged from 58 to 69 years and four of the studies were graded as being of high quality. The systematic review demonstrated that use of high-dose conventional theophylline resulted in a small but significant increase in FEV₁ of 100 ml (95% CI 40 to 160 ml). This was derived from 13 studies with 244 participants. Two studies with a total of 45 participants reported on the incidence of exacerbations, concluding that high-dose conventional theophylline had no effect on the incidence of exacerbations. Three studies with a total of 64 participants reported data on nausea, with the risk of experiencing nausea when on theophylline treatment being significantly increased (RR 7.67, 95% CI 1.47 to 39.94). When compared with previous trials of conventional high-dose theophylline in COPD, the current trial of low-dose theophylline that recruited 1578 participants is clearly somewhat larger and the treatment period is longer in duration. Moreover, in contrast to conventional high-dose theophylline trials with their focus on lung function, the primary outcome of the current study was exacerbations of COPD and the study population comprised participants at high risk of exacerbating. When compared with these trials of high-dose conventional theophylline, the current trial, as expected, showed no effect of low-dose theophylline on lung function (FEV₁) and, reassuringly, no increase in side effects. One of the findings from the Cochrane review83 was that very few participants withdrew from intervention trials of high-dose conventional theophylline for any reason. In the review,83 nine studies reported no dropouts, and in the remaining studies the dropout rate was generally very low; the only exception to this low dropout rate was the study of Guyatt et al. ,97 who reported eight withdrawals out of 27 recruited participants (30%). The sample size of the current trial included an estimate of 6% of participants ceasing taking their trial medication, based on the four high-quality studies reported in the Cochrane review,83 in which three out of 51 (6%) participants dropped out. The 26% of participants ceasing trial medication in the current study is greater than anticipated (although balanced across the arms) and more in keeping with the study of Guyatt et al. ,97 probably reflecting the pragmatic nature of the current trial, the fact that participants in the current trial were older than those in the Cochrane review and that the trial lasted longer. 83

The use of high-dose conventional theophylline has declined over the years because of its narrow therapeutic index, modest clinical effect, side effect profile, drug interactions, the need for blood concentration monitoring and the availability of more effective inhaled therapies. 98 High-dose conventional theophylline is now included in current COPD guidelines as a third-line therapy. 1

The concept of using low-dose theophylline to augment the anti-inflammatory effects of corticosteroids on the airway inflammatory processes in COPD originated from in vitro and animal studies investigating the molecular mechanisms contributing to the reduced corticosteroid sensitivity of COPD. 32,38–40,43,46 The key observation was that the reduced HDAC2 activity of COPD can be reversed by low concentrations (1–5 mg/l) of theophylline; moreover, theophylline reduces corticosteroid insensitivity in COPD, such that there is a marked synergistic interaction between theophylline and corticosteroids in suppressing the release of inflammatory mediators from alveolar macrophages obtained from COPD patients. 43,44 These basic research studies suggest that low-dose (i.e. 1–5 mg/l) theophylline could increase HDAC activity and, hence, reduce corticosteroid resistance in COPD patients, thereby enabling ICSs to switch off inflammation and potentially more effectively reduce exacerbation rates.

Prior to commencing the current study, the concept of using low-dose theophylline in conjunction with corticosteroids in COPD had been explored in two small RCTs. The first RCT was in 35 patients admitted to a Spanish hospital with an acute exacerbation of COPD who were treated with a regime that included systemic corticosteroids. 48 Participants were randomised to receive additional low-dose theophylline or nothing in a single-blind design; participants not on ICSs at admission were commenced on ICSs. After 3 months of treatment, low-dose theophylline increased sputum macrophage HDAC activity and reduced sputum concentrations of the pro-inflammatory mediators IL-8 and tumour necrosis factor alpha (TNF-α). There were no clinically significant effects in this small study,48 although fewer participants in the theophylline group than in the control group had a subsequent exacerbation (12.5% vs. 26%). This study48 differed from the current study in the following ways: small sample size, single blinded, no placebo control, 3-month follow-up, participants were recruited only during hospitalisation with exacerbations of COPD, all were male and only 14% had experienced two or more exacerbations in the previous year. Notably, 26% of participants were not followed up at 3 months.

The second small (n = 30) RCT of COPD patients was a double-dummy (low-dose theophylline vs. placebo, standardised dose of ICS vs. placebo), randomised, double-blind, parallel study, based in the UK. 49 After 4 weeks of low-dose theophylline, there was no effect on the primary outcome of absolute number of sputum neutrophils. The combination of low-dose theophylline/ICS significantly reduced a number of secondary end points (e.g. sputum percentage neutrophils, sputum total eosinophil count). In an open-label extension of the trial, the combination of low-dose theophylline/ICS increased peripheral blood mononuclear cell HDAC activity ninefold. The study concluded that the combination of ICS and low-dose theophylline may attenuate airway inflammation in patients with COPD. One of the limitations of this study was that the significant findings were for low-dose theophylline/ICS versus theophylline, rather than low-dose theophylline/ICS versus ICS, suggesting perhaps that the observed effects were a consequence of the ICS and not the low-dose theophylline. This study49 differs from the current study in the following ways: 4-week duration, small numbers, 83% male and younger average age (61 years). Lung function (mean FEV₁ 54%) was similar.

While the current trial was being conducted, two trials investigating the therapeutic consequences of low-dose theophylline were published. 52,99 The first study,99 from India, was a hospital-based, single-blinded, prospective, randomised, placebo-controlled study that investigated the effects of adding low-dose theophylline to the combination of formoterol plus budesonide. A total of 58 patients with moderate/severe COPD were commenced on a standardised ICS/LABA therapy (budesonide and formoterol) and were randomised to receive in addition either low-dose theophylline or placebo for 60 days. Fifty participants completed the trial and their data were presented. 99 The addition of low-dose theophylline resulted in a greater improvement in total symptom scores, a greater increase in FEV₁ and a greater increase in 6-minute walking distance than did the addition of placebo. Of note, however, is that the method of randomisation was not described, the actual number of participants randomised to each treatment group was not presented, the nature of the ‘single blind’ was not explained and there was no ITT analysis. The randomisation appeared not to have eliminated potential sources of bias, as the participants allocated to low-dose theophylline were clearly more severely affected by COPD: their respiratory rate was greater (20.7 vs. 18.7 breathes per minute; p = 0.003), their FEV₁ values were lower (49% vs. 57% predicted; p = 0.05), their symptom scores were greater (10.17 vs. 8.37; p = 0.003), their 6-minute walking distance was shorter (373 m vs. 409 m; p = 0.07) and more were classified as having severe COPD (54% vs. 27%; p = 0.09). 99 Moreover, the placebo tablets were described as similar rather than identical. These differences could reflect a bias for the more severely affected participants to be preferentially allocated to the low-dose theophylline arm of the trial. This study99 differs from the current study in the following ways: the sample size was much smaller and hospital based, 92% of participants were male, participants were younger (≈55 years), BMI was lower (≈17 kg/m²), there was a 60-day treatment period and the trial was single blinded. In addition to the issues regarding blinding and randomisation, the results of the trial99 also raise the possibility that, although the intention was to investigate low-dose theophylline, in reality, conventional high-dose theophylline was being tested: an improvement in FEV₁ was described with theophylline treatment and the dosing regimen for this study was 400 mg of theophylline for a weight of > 50 kg, 300 mg for a weight of 40–50 kg and 200 mg for a weight of < 40 kg. However, a significant proportion of participants appeared to be underweight, with a mean BMI of ≈17 kg/m², and theophylline treatment resulted in higher incidences of typical high-dose theophylline toxicity symptoms such as nausea, vomiting, headache, palpitation and insomnia. In the current study, this was avoided by basing theophylline dose on IBW and smoking status.

The second study, the Spanish ASSET trial,52 was a multicentre, randomised, double-blind, placebo-controlled trial that recruited patients with COPD while they were hospitalised for a COPD exacerbation. Participants were randomised to low-dose theophylline (100 mg twice a day) or matched placebo in addition to ICS/LABA treatment; participants not routinely taking ICS/LABA were established on ICS/LABA. In total, 70 patients were randomised (theophylline, n = 36; placebo, n = 34) and 46 completed the year of treatment (theophylline, n = 23; placebo, n = 23). The co-primary outcomes were change in HDAC and exacerbation frequency during the 1-year treatment period. The addition of low-dose theophylline had no effect on plasma/sputum HDAC concentrations and no effect on COPD exacerbation rate [theophylline 0.97 (SD 0.94) vs. placebo 0.88 (SD 0.89)]. This trial52 has some similarities with the current trial: primary outcome of exacerbation, same definition of exacerbation, 1-year treatment period, similar participant age, CAT score and levels of cardiovascular comorbidity at baseline, and no significant difference in ARs between groups. However, there are some important differences between the ASSET trial52 and the current trial (TWICS). The current trial is much larger (n = 1578) than the ASSET trial (n = 70), being designed to detect a 15% reduction in exacerbations with 90% power, whereas the ASSET trial was designed to detect an arguably implausibly large 50% reduction in exacerbations with 80% power. The exacerbation rate in the ASSET trial was about half that observed for the TWICS trial (0.92 vs. 2.23 per year). Perhaps the most plausible explanation for this is that all participants in the ASSET trial were recruited while hospitalised with an exacerbation of COPD, irrespective of exacerbation history, whereas participants in the TWICS trial were clinically stable; 60% were identified from primary care and all had a history of two or more exacerbations in the previous year requiring treatment with antibiotics and/or corticosteroids. The proportion of participants ceasing trial medication was higher in the ASSET trial than in the TWICS trial (34% vs. 26%); however, it should be noted that 14% of participants in the ASSET trial ceased trial medication because their FEV₁ improved to > 50% predicted during the 1-year treatment period. This is most likely to be a consequence of the ASSET trial recruiting in the peri-exacerbation period and the TWICS trial recruiting when participants were clinically stable. When compared with the TWICS trial, participants in the ASSET trial were more likely to be male, had more severe COPD (lower FEV₁) and were more likely to be hospitalised during the treatment period, but were less likely to be diabetic [probably reflecting the higher mean BMI of the TWICS trial participants (27 vs. 22 kg/m²)].

The GOLD management strategy guideline1 highlights that the clinical relevance of low-dose theophylline has not been fully established and that clinical evidence for an effect of low-dose theophylline, particularly on exacerbations, is limited and contradictory. The TWICS trial is the first large, pragmatic, community-based trial to investigate the effect of adding low-dose theophylline to the treatment regimen of people with COPD who are at a high risk of exacerbating despite a treatment regime that includes maintenance ICSs in COPD.

Pre-clinical work convincingly demonstrates that the combination of low-dose theophylline and corticosteroids has a strong biological effect, increasing HDAC and inhibiting the release of pro-inflammatory mediators. 38,39,41–44 The trials conducted to date have been small (n = 30–70), hospital based and have tended to focus on biological outcomes with short treatment periods. 48,49,52,99 The largest trial52 to date in this field has reported that low-dose theophylline had no effect on HDAC or exacerbations; however, as the authors of the ASSET trial acknowledge, ‘we might have overestimated the potential clinical benefit when we calculated the sample size, which may have precluded us from identifying a clear-cut clinical effect’. 52 The TWICS trial avoids many of the limitations of previous studies and clearly demonstrates that, in a NHS setting, the addition of low-dose oral theophylline to a drug regimen that includes an ICS confers no overall clinical benefit for people with COPD. The participants in the TWICS trial were a group of people with COPD who were at high risk of experiencing an exacerbation based on their history of exacerbating in the previous year. This group was deliberately chosen because of their impact on the NHS and it enabled us to design a trial of realistic (but ambitious) sample size. Although the TWICS trial did not investigate whether or not people with COPD who are at low risk of exacerbation would benefit from low-dose theophylline, the combination of the findings of the TWICS trial and the absence of a biological effect (HDAC concentrations) in the ASSET trial, despite a sample size that was ‘more than enough to demonstrate a biological effect of the intervention’,52 make it highly unlikely that low-dose theophylline would be beneficial in COPD patients who are at low risk of exacerbation. A possible explanation for the disparity between the biological effects observed in previous studies, with short treatment periods, and the absence of beneficial effects in the TWICS trial, with a year-long treatment period, is that any biologically beneficial effect of low-dose theophylline is not sustained in the long term.

Cost-effectiveness

The health economics results indicate that, after adjustment for baseline characteristics, there was no significant difference between the total health economic costs associated with treatment with low-dose theophylline and the costs associated with placebo treatment: adjusted mean difference £222 (95% CI –£27 to £472). With unadjusted complete-case data, the total costs are higher in the placebo arm than in the theophylline arm: a significant difference of £452 (95% CI £132 to £771). This difference was driven by the fact that more participants in the placebo arm than in the theophylline arm received treatment for exacerbations in hospital. The 10 most costly observations (> £10,000) were all in the placebo arm, and were the result of hospital stays of > 40 days. The multiple imputation results mirror the complete-case results with a significant difference in unadjusted costs of £439 (95% CI £32 to £846) higher in the placebo arm.

The difference between arms in total costs is driven solely by the hospital-treated exacerbations and exacerbations treated with oxygen; no other resource group differs significantly between arms. The difference in the number of exacerbations necessitating hospital treatment is likely to be the result of a small number of participants in the placebo arm having very frequent hospital admissions. Therefore, these results should be interpreted with caution.

Exacerbation costs are 22% and 33% of the total costs (theophylline and placebo, respectively), somewhat less than the 60% reported by Britton et al. 27 in 2003, perhaps reflecting differences in management between 2003 and 2015/6, particularly in increased use of preventative drugs, pulmonary rehabilitation and more structured chronic disease management in primary care.

In the unadjusted complete-case analysis, QALY gain was higher in the placebo arm than in the theophylline arm (difference 0.011 QALYs, 95% CI –0.018 to 0.040 QALYs); however, the difference is not significant. Multiple imputation results mirrored the complete-case results; there were no significant differences, with unadjusted results favouring the placebo arm and adjusted results favouring the theophylline arm.

These results reflect the primary outcome of number of exacerbations needing treatment during the 12-month follow-up period; there was no significant difference between arms.

Hettle et al. 100 reported 4-year UK costs in their paper on tiotropium versus usual care in the UK and Belgium. Exacerbation costs ranged from £2295 to £2744 (£574–686 per year) and maintenance costs ranged from £2935 to £3937 (£737–984 per year). This compares to the 1-year costs from this research of exacerbations, ranging from £585 to £1033, and maintenance costs ranging from £2074 to £2101. Although the annual exacerbation costs in the current trial are similar to those of Hettle et al. ,100 the maintenance costs are somewhat higher, reflecting the older average age of participants in the current study (68.4 years vs. 64 years), and the facts that, in the current study, 80% of the participants were prescribed LAMAs (none for the usual care group in the Hettle et al. 100 study), participants with COPD using long-term oxygen were included and participants were more likely to be in the severest GOLD category (14% vs. 8%). Hettle et al. 100 also reported that in Belgium the cost of exacerbations resulting in hospitalisation was 33 times higher than the cost of exacerbations not necessitating hospitalisation, substantiating the increased cost of hospital treatment compared with non-hospital treatment of exacerbations that we found in this research.

The strengths of this research include the fact that there were few participants with no outcome or resource use (low proportion of missing cases: 4.3%), that uncertainty was explored using non-parametric bootstrapping and that, when there was a significant difference in exacerbation costs, this was explored further to identify what was driving this difference.

The two main limitations to the cost-effectiveness analysis are the large number of missing EQ-5D-3L questionnaires (19.1%) and the fact that small numbers of missing data were imputed using naive methods at a disaggregated level.

Strengths and limitations

The main strength of the TWICS trial is that it was a large, pragmatic, predominantly community-based, suitably powered, double-blind, randomised, placebo-controlled, UK multicentre clinical trial with a high follow-up rate for the primary clinical outcome. A total of 1578 individuals were recruited in 121 UK sites; 60% of these were identified in primary care, making it highly likely that the TWICS trial participants reflected normal clinical practice across both primary and secondary care in the UK. The 1-year treatment period allowed for capture of the seasonality of exacerbations. 101

Originally, the TWICS trial aimed to recruit 1424 participants, the sample size being primarily based on the findings of the observational ECLIPSE cohort study of 2138 COPD patients recruited in 46 centres from 12 countries. 21 The ECLIPSE study demonstrated that the best predictor of an exacerbation in a year was a treated exacerbation in the previous year. The ECLIPSE study also identified a frequent exacerbator phenotype, defined as two or more exacerbations in the previous year; moreover, this frequent exacerbator phenotype was relatively stable for 3 years and could be reliably identified by patient report. For the patients experiencing frequent exacerbations recruited into the TWICS trial, data from the ECLIPSE study predicted a mean of 2.22 (SD 1.86) exacerbations in the year of treatment; the sample size for the TWICS trial was based on this. This prediction proved to be remarkably close to what we observed, increasing confidence in the findings, with a mean number of exacerbations in the theophylline arm of 2.24 (SD 1.99) and in the placebo arm of 2.23 (SD 1.97). A notable finding of the TWICS trial was an apparent disparity between the number of exacerbations reported by participants in the year prior to the trial [mean 3.59 (SD 2.15)] and the number of self-reported exacerbations in the treatment year, which was somewhat lower [mean 2.23 (SD 1.99)]. The most likely explanation for this disparity is that we did not ask for dates for the reported exacerbations in the year prior to the trial, whereas, during the trial, we asked for dates and the conventional minimum of 2 weeks between consecutive exacerbation episodes was necessary to consider exacerbations as separate. 68 This resulted in exacerbations that were separated by < 2 weeks being merged. Although further factors contributing to the disparity in exacerbations before and during the trial may include an over-reporting bias by participants and regression to the mean, the exacerbation frequency during the treatment period was remarkably consistent with that predicted by the ECLIPSE study. Although the exacerbation rate observed in the current trial is somewhat higher than in recent explanatory trials,102,103 it is entirely consistent with the recent pragmatic UK Salford Lung Study. 104 The Salford Lung Study, with an inclusion criterion of one or more exacerbations in the previous year, reported exacerbation rates of 1.74–1.90 per year. The slightly higher exacerbation rate in the current trial most probably reflects the participants’ increased propensity to exacerbate (two or more exacerbations in the previous year), as well as the lack of requirement to withhold therapy other than theophylline, meaning that investigators were happier to recruit patients at a higher risk of exacerbation. Although the diagnosis of COPD was confirmed by post-bronchodilator FEV₁/FVC of < 0.7, 18.3% of participants reported a concurrent/previous diagnosis of asthma. This may, in part, reflect a diagnostic bias towards the more socially acceptable diagnosis of asthma in the past, but it is possible that the current trial included up to 18% of participants with asthma–COPD overlap syndrome. Although it is possible that these patients may respond differently to the theophylline, whether or not people with asthma–COPD overlap syndrome responded differently to theophylline was not one of the trial objectives.

By recruiting 1578 individuals, 60% of whom were identified in primary care, the TWICS trial exceeded its original recruitment target of 1424 with at least 50% being recruited in primary care. It was initially envisaged that the TWICS trial would recruit from a limited number (seven) of secondary care sites, with primary care sites acting as PIC sites for these secondary care centres. Recruitment to the TWICS trial was delayed by 5 months because of a worldwide shortage of bottle tops for the drug bottles. Initially, recruitment in 16 primary and six secondary care sites was on target and the TWICS trial achieved its recruitment targets for the feasibility phase by recruiting 100 participants in months 7, 8 and 9, with 55% identified in primary care. Within 4 months, it became apparent that it would not be possible to sustain recruitment, with recruitment falling below the required 59 participants per month to a nadir of 26 in month 15 (October 2014). To address this, a change in recruitment strategy was implemented in month 12 (July 2014), with rapid increases in the number and rate of opening up of primary and secondary care sites. Ultimately, 121 recruiting sites were opened up, comprising 88 primary care sites and 33 secondary care sites. Other primary care practices acted as PICs for primary and secondary care sites. In total, 477 participants were recruited and followed up entirely in primary care, 464 participants were identified in primary care but recruited and followed up in secondary care (this was particularly the case in Scotland) and 637 participants were identified, recruited and followed up in secondary care. This change in recruitment strategy was successful, with monthly recruitment remaining above 50 per month from month 19 and reaching a peak of 81 in month 35.

Primary outcome data (number of COPD exacerbations) were collected on 98% of the 1567 participants who commenced the 1-year treatment period (1578 recruited minus 11 post-randomisation exclusions). Several factors contributed to the high follow-up rate. The TWICS trial was designed to be as inclusive as possible by facilitating participation of people with COPD who would normally find it too difficult to participate in a trial because of their ill health. The trial was designed to be relatively ‘light touch’, with three study visits to a local study centre. If participants were unable to attend for assessment, they were visited at home, contacted by telephone or sent the questionnaires to complete at home. Participation and remote follow-up were further facilitated by delivering the trial drug to participants’ homes using a third-party distributor. All participants who ceased taking the trial drug were invited to remain in the trial for follow-up, by face-to-face assessment, telephone assessment or postal questionnaire. If participants could not be followed up directly, for example if they failed to attend follow-up, various methods of follow-up, independent of participant involvement, were used. In the first instance, the participant’s GP was sent a questionnaire enquiring about exacerbations (number, dates, how and where treated); the minimum information requested was the number of exacerbations in the treatment period. Failing this, GP surgeries were contacted by telephone or a request was made for a redacted copy of patient encounter summaries from which the co-chief investigator extracted exacerbation data. The combination of follow-up methods enabled the ITT analysis to include 1489 years of participant follow-up data. Inevitably, there were some participants who did not provide a full 12 months of follow-up data, for example because they died, or for whom 12 months of follow-up data were not available even using a remote follow-up method. A strength of the TWICS trial was that the statistical analytical methods used enabled inclusion of these participants up to the point at which they were lost to follow-up, with their time in study utilised in the offset variable during analysis.

Previous studies investigating the potential anti-inflammatory effects of low-dose theophylline in COPD and asthma (not in conjunction with ICS) have used a ‘one size fits all’ dosing approach, that is all participants received 100 mg bd or 200 mg bd. 43,44,48,99,105–107 In contrast, one of the strengths of the TWICS trial was that theophylline dosing was somewhat personalised, being determined by IBW and smoking status. As noted in the protocol paper,55 population studies have demonstrated that theophylline pharmacokinetics are influenced by weight, COPD disease status (reduced clearance) and smoking (increased clearance). 57–66,108 Smoking increases theophylline clearance by ≈60%, but the plasma theophylline level gradually returns to normal following smoking cessation. This was incorporated into the definition of a non-smoker in the TWICS trial and procedures were implemented to modify, when necessary, the dose of the trial drug in a timely manner if participants changed their smoking status during the treatment period. The use of IBW in preference to actual weight avoided the potential for giving an inappropriately high dose of theophylline to obese participants. In the TWICS trial, theophylline dosing was based on pharmacokinetic modelling, incorporating the major determinants of theophylline C_ss, namely weight, smoking status and clearance of theophylline (low, normal or high), and was designed to achieve a steady-state plasma theophylline concentration of 1–5 mg/l and certainly < 10 mg/l. 55 Theophylline is metabolised in the liver by the enzyme CYP1A2, which is induced by smoking and inhibited by a number of medications, with a consequent increase in plasma theophylline concentration. For this reason, the exclusion criteria included long-term use of drugs with the potential to increase plasma theophylline concentration;94 conversely, concomitant use of drugs with the potential to lower plasma theophylline concentration was permitted in the trial. Reassuringly, the dosing regimen used for the TWICS trial appeared to be effective in establishing low-dose plasma theophylline concentrations of 1–5 mg/l because there was no evidence of the typical sequelae of conventional high-dose theophylline, such as an improvement in FEV₁. In addition, when compared with the placebo group, there was no evidence that participants allocated to low-dose theophylline experienced more SAEs or ARs, nor did they report SAEs or ARs typical of theophylline toxicity, namely gastrointestinal, cardiac, psychological or neurological symptoms. Furthermore, when the reasons for ceasing trial medication were analysed, there were no significant differences between the arms, notably for gastrointestinal, cardiac, psychological or neurological symptoms typical of theophylline toxicity. A consequence of the personalised dosing of the trial drug to achieve a low-dose plasma theophylline concentration well below that associated with typical side effects was that there was no need for blood sampling to monitor plasma theophylline, a necessity that would have greatly increased the complexity of the trial and increased the likelihood of unblinding the participant and/or investigator. The absence of blood testing reduced costs and was extremely popular with primary care sites and contributed to the willingness of many primary care sites to participate in the TWICS trial. The potential limitation of relying on participant-reported smoking status is perhaps less important in this trial, as any smoker claiming to be a non-smoker would have been prescribed the lower dose of theophylline, thereby ensuring that plasma theophylline remained in the low-dose range of 1–5 mg/l.

As with all studies, there are limitations associated with the TWICS trial. The primary outcome for the trial was the number of participant-reported exacerbations during the 1-year treatment period; to facilitate recall, participants were given a diary card to make notes on exacerbations, treatment and health-care usage. The definition of an exacerbation was the widely used ATS/ERS guideline recommendation of a worsening of a patient’s dyspnoea, cough or sputum beyond day-to-day variability sufficient to warrant a change in management. 68 The minimum management change was treatment with antibiotics or OCSs; consequently, the TWICS trial quantified only moderate and severe exacerbations. However, these exacerbations are the ones that are the most burdensome to patients and health-care services. A limitation of the TWICS trial is that the relatively conservative definition of exacerbation probably underestimates the frequency of symptom-defined mild exacerbations that are short-lived and treated by a patient with a temporary increase in bronchodilator therapy. 109 The identification of such mild exacerbations would have required participants to complete daily symptom diary cards, adding to the intrusiveness of the trial and considerably adding to the data-entry burden of research staff. Although the TWICS trial did not quantify mild exacerbations, there were no significant differences between the treatment and placebo arms in QoL/impact on health status as quantified by EQ-5D-3L/CAT, suggesting that either low-dose theophylline had no effect on mild exacerbations or, if there was an effect, it did not impact on health status/health-care usage.

A possible limitation of participant-reported exacerbations is the accuracy of such a report over a 6-month period. Although it would have been possible to obtain such exacerbation data from health-care records, it is well documented that people with COPD do not report all of their exacerbations to health-care professionals. 18,110–112 Patient recall of COPD exacerbations has been shown to be highly reliable over a year: in the London COPD cohort study,112 there was no significant difference between the number of exacerbations recorded on diary cards and patient estimates of their exacerbation number over the same 1-year period [mean 2.4 (SD 2.2) vs. mean 2.3 (SD 2.1)] and there was 93% agreement between patient-recalled and diary-recorded exacerbations. There was, however, a difference between the number of treated exacerbations recorded on diary cards and the number of treated exacerbations remembered by the patient over the same 1-year period [mean 2.3 (SD 2.1) vs. mean 1.8 (SD 1.8)] and there was 88.6% agreement between patient-recalled and diary-recorded treated exacerbations. 112 The patient representatives helping with the TWICS trial were adamant that it was fairly straightforward to recall the number of exacerbations over a 6-month period. A small validation exercise was conducted at two of the largest sites (Aberdeen and Aintree) during the TWICS trial to confirm that participant recall was indeed valid. The validation was done by requesting a care/encounter summary from the GP and comparing this with the participant report. In Aberdeen, 43 records (a 20% sample) were checked; in 37 there was complete agreement between the participant and GP reports. In Aintree, 24 records were checked and in 16 there was complete agreement between the patient and GP reports. Therefore, in a 4% sample of participants, there was 80% agreement. This rate of agreement was slightly lower than that reported by Quint et al. ;112 however, current GP records may not be as reliable a source of exacerbation data as in the past, given that patients have rescue packs at home and can access help for their exacerbations through many non-GP sources, for example pharmacies, emergency and walk-in centres, and accident and emergency departments.

A limitation of the TWICS trial was the proportion of participants who ceased taking their trial drugs (26%) was higher than anticipated (6%), although this was somewhat offset by 10% over-recruitment (n = 154). There was no evidence of bias in ceasing trial medication, with the proportion and the reasons given for ceasing trial medication being equally distributed between those allocated to low-dose theophylline and those allocated to placebo. The original sample size for the TWICS trial (n = 1424) accounted for 6% of participants ceasing their trial medication based on the four high-quality studies reported in a Cochrane review83 of theophylline in COPD, in which 3 out of 51 (6%) participants dropped out. In reality, 413 of the 1578 participants either never started/initiated medication (post-randomisation exclusions, n = 11; non-initiation, n = 8) or ceased taking the trial medication (non-persistence, n = 393). This 26% rate of ceasing trial medication is greater than anticipated, but in keeping with the ASSET trial of low-dose theophylline, which reported a 34% rate of ceasing trial medication. 44 The higher than anticipated rate of ceasing trial medication in the TWICS trial was most likely the consequence of the relatively high rates of comorbidities in participants, giving rise to symptoms that were attributed to the trial medication and a heightened awareness of ARs listed in the PIL and the package insert accompanying the trial medication. This is consistent with 46% of participants reporting ARs typical of high-dose theophylline (but equally distributed between the two trial arms) and why 20% of those ceasing trial medication gave gastrointestinal symptoms as the reason for ceasing trial medication, although there was no significant difference in the incidence of such symptoms in those ceasing low-dose theophylline and those ceasing placebo. Some participants were asked to discontinue trial medication because they had stopped taking an ICS. During the trial, there was an emergent change in prescribing practice away from ICS-containing preparations to LABA/LAMA inhalers; however, this had minimal impact on the trial (certainly < 20 participants), most probably because the participants in this trial were at high risk of exacerbation and were participants in whom there is still a role for ICSs. Although 413 participants ceased trial medication during the TWICS trial, a review of the medication returns indicated that 66 of these participants had > 70% adherence while taking the trial medication when averaged over the 12-month treatment period; for example, they ceased trial medication at 11 months. These individuals were included in the per-protocol analysis. Although per-protocol analyses are biased by their very nature, the per-protocol analysis for this trial included 1142 years of participant data (85% of the 1338 years indicated by the power calculation); therefore, it is not surprising that the results of the per-protocol analysis were almost identical to those of the ITT analysis. Although adherence to the trial medication was quantified through pill counting, it was not practicable to assess adherence to the ICS, as this would have entailed use of non-routine care methodologies such as diaries cards, metered inhalers, etc. The rationale for the use of low-dose theophylline is as an adjunct to ICS therapy; that we were unable to verify adherence to ICS therapy is a limitation of this trial.

Generalisability

This trial has good external validity as it was of a pragmatic design that reflected normal clinical practice across both primary and secondary care in the UK. Participants remained on their existing COPD medications, they were managed in the normal way by their usual health-care teams and the trial recruited from 121 sites (88 primary care and 33 secondary care) that spanned the UK; many of the secondary care sites were district general hospitals. We consider it to be highly probable that the TWICS trial participants are typical of COPD patients in normal clinical practice across both primary and secondary care in the UK and that the findings are generalisable to clinical practice in the UK.

The TWICS trial recruited participants who were highly likely to experience an exacerbation during the 1-year treatment period, as evidenced by the fact that they had experienced two or more treated exacerbations in the previous year. In contrast to many COPD trials, we did not exclude potential participants with mild COPD, as evidenced by FEV₁ of > 80% predicted; 9% of the TWICS trial participants had mild COPD based on spirometry criteria, but fulfilled the frequent exacerbator phenotype,21 enhancing the generalisability of the trial. Recruitment to the TWICS trial was limited to frequent exacerbators because, in clinical practice, these are the patients who are usually commenced on this ‘third-line’ therapy;24 moreover, a trial of participants less likely to experience exacerbation (e.g. those experiencing one exacerbation in the previous year) would have been much larger (n = ≈3000) and somewhat more costly. Although we did not test whether or not the addition of low-dose theophylline to ICSs had an effect on people who experienced less frequent exacerbations, there is no scientific or clinical reason why low-dose theophylline should have a differential effect according to the frequency of exacerbations. Therefore, it would seem reasonable to extend the findings of the current study to people with COPD who are at a low risk of experiencing an exacerbation.

Although the results of this trial are generalisable to the UK and probably to other high-income countries, the findings may not be applicable to low- or medium-income countries, with differing pharmacogenetic profiles, where theophylline remains a frequently used therapy in COPD, most probably because it is inexpensive compared with inhaled therapies. 113–116 The randomised, double-blind, placebo-controlled trial of Zhou et al. 117 raises the possibility that, in China at least, there is a therapeutic response to low-dose theophylline in the absence of ICS. In that trial,117 the addition of low-dose theophylline to usual COPD treatment in 110 people with COPD (theophylline, n = 57; placebo, n = 53) for a year significantly reduced the frequency of exacerbations when compared with placebo [mean 0.79 (SD 1.16) vs. 1.70 (SD 2.61); p = 0.047]. The participants in that trial117 differed considerably from those taking part in the TWICS trial: only 30% were taking regular medication prior to the trial, and this was restricted to inhaled salbutamol; use of ICS, LABA and LAMA was excluded; and the target plasma theophylline concentration (5–10 mg/l) was also somewhat higher than the target range (1–5 mg/l) identified for optimum synergistic interaction between corticosteroids and low-dose theophylline. The use of low-dose theophylline in conjunction with corticosteroids in China is being addressed by the ongoing Theophylline And Steroids in COPD Study (TASCS), which is recruiting 2400 people with COPD in China. 118 They are being randomly allocated to low-dose prednisolone (5 mg od) or low-dose theophylline (100 mg bd) with low-dose prednisolone (5 mg od) for 48 weeks. The primary outcome is exacerbation rate over the 48-week treatment period. This study has been presented in abstract form119 and demonstrated that the combination of low-dose theophylline and prednisolone did not reduce the annualised rate of exacerbations. It will be interesting to compare the results of TASCS with the TWICS trial, although it should be noted that the routine use of OCSs as a maintenance treatment for COPD, even though they are cheaper than ICS, would never be contemplated in developed countries for clinical and ethical reasons.

Patient and public involvement

Patient and public involvement in this trial was limited, but effective, and lessons were learnt that have been implemented in a subsequent study funded by the NIHR Health Technology Assessment (HTA) programme (reference 15/130/20),120 for example a person with COPD is a joint grant holder in that study.

A patient with COPD was a voting member of the TWICS trial TSC; recruitment and retention of a patient representative was hindered by ill health. The first patient approached declined because of ill health. The patient representative nominated by CHSS as part of their Voices Scotland initiative had to resign because of ill health. A third patient representative was identified and he made an active contribution to the TWICS trial. Support of the TSC patient representative was actively undertaken by several members of the local study team. In the subsequent trial funded by the NIHR HTA programme,120 we have a patient representative who is supported by CHSS’s Voices Scotland lead who is not only a voting member of the TSC, but also co-ordinator and representative of a panel of 15 COPD patients (as they like to be called).

A representative of the British Lung Foundation and a person with COPD made important contributions to trial design procedures [what was acceptable (e.g. spirometry) and what was not acceptable (e.g. daily diary cards)]. Perhaps the most important suggestions were to deliver trial medication to home addresses and to facilitate follow-up for ill participants by way of home visits and telephone and postal questionnaires. PPI resulted in many changes to the design and content of the ‘short’ PIL (a one-page summary PIL) and the ‘long’ PIL (a more detailed PIL). The importance of these changes is evidenced by the success in recruitment, and there were no changes to the PIL throughout the trial. PPI was particularly insightful during TSC deliberations concerning the validity of patient recall of COPD exacerbations.

The support of the British Lung Foundation and CHSS has been invaluable throughout the trial, identifying volunteers for PPI and publicising the trial.

Conclusions

Staff at recruitment sites

Contributions of authors

Graham Devereux contributed to the conception and design of the trial, conduct of the trial, recruitment and follow-up of participants, interpretation of results and writing/editing the report.

Seonaidh Cotton contributed to the design of the trial, was responsible for the day-to-day management of the trial and contributed to the interpretation of results and writing/editing the report.

Shona Fielding contributed to the design of the trial, was responsible for statistical analysis and contributed to the interpretation of results and writing/editing the report.

Nicola McMeekin was responsible for the health economic analysis and contributed to the interpretation of results and writing/editing the report.

Peter J Barnes contributed to the conception and design of the trial, the interpretation of results and writing/editing the report.

Andy Briggs contributed to the conception and design of the trial, oversaw the health economic analysis and contributed to the interpretation of results and writing/editing the report.

Graham Burns contributed to the conception and design of the trial, the interpretation of results and writing/editing the report.

Rekha Chaudhuri contributed to the conception and design of the trial, conduct of the trial, recruitment and follow-up of participants, the interpretation of results and writing/editing the report.

Henry Chrystyn contributed to the conception and design of the trial, the interpretation of results and writing/editing the report.

Lisa Davies contributed to the conception and design of the trial, conduct of the trial, recruitment and follow-up of participants, the interpretation of results and writing/editing the report.

Anthony De Soyza contributed to the conception and design of the trial, conduct of the trial, recruitment and follow-up of participants, the interpretation of results and writing/editing the report.

Simon Gompertz contributed to the conception and design of the trial, conduct of the trial, recruitment and follow-up of participants, the interpretation of results and writing/editing the report.

John Haughney contributed to the conception and design of the trial, conduct of the trial, recruitment and follow-up of participants, the interpretation of results and writing/editing the report.

Karen Innes was responsible for aspects of the day-to-day management of the trial and contributed to the interpretation of results and writing/editing the report.

Joanna Kaniewska was responsible for aspects of the day-to-day management of the trial and contributed to the interpretation of results and writing/editing the report.

Amanda Lee contributed to the conception and design of the trial, oversaw the statistical analysis and contributed to the interpretation of results and writing/editing the report.

Alyn Morice contributed to the conception and design of the trial, conduct of the trial, recruitment and follow-up of participants, the interpretation of results and writing/editing the report.

John Norrie contributed to the conception and the design of the trial, the conduct of the trial, the interpretation of results and writing/editing the report.

Anita Sullivan contributed to the conception and design of the trial, conduct of the trial, recruitment and follow-up of participants, the interpretation of results and writing/editing the report.

Andrew Wilson contributed to the conception and design of the trial, conduct of the trial, recruitment and follow-up of participants, the interpretation of results and writing/editing the report.

David Price contributed to the conception and design of the trial, the conduct of the trial, the interpretation of the results and writing/editing the report.

Publication

Devereux G, Cotton S, Barnes P, Briggs A, Burns G, Chaudhuri R, et al. Use of low-dose oral theophylline as an adjunct to inhaled corticosteroids in preventing exacerbations of chronic obstructive pulmonary disease: study protocol for a randomised controlled trial. Trials 2015;16:267.

Data-sharing statement

All data requests should be submitted to the corresponding author for consideration. Access to available anonymised data may be granted following review.

Patient data

This work uses data provided by patients and collected by the NHS as part of their care and support. Using patient data is vital to improve health and care for everyone. There is huge potential to make better use of information from people’s patient records, to understand more about disease, develop new treatments, monitor safety, and plan NHS services. Patient data should be kept safe and secure, to protect everyone’s privacy, and it’s important that there are safeguards to make sure that it is stored and used responsibly. Everyone should be able to find out about how patient data are used. #datasaveslives You can find out more about the background to this citation here: https://understandingpatientdata.org.uk/data-citation.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health and Social Care.