Co-trimoxazole to reduce mortality, transplant, or unplanned hospitalisation in people with moderate to very severe idiopathic pulmonary fibrosis: the EME-TIPAC RCT

Aetiology of idiopathic pulmonary fibrosis and potential mechanisms of co-trimoxazole

As the pathogenesis of IPF is unknown, the potential mechanisms of action of co-trimoxazole are uncertain. Co-trimoxazole is a broad-spectrum antibiotic with bactericidal effects against respiratory pathogens, with the role of infection in IPF becoming more evident. However, it may have non-antimicrobial effects, targeting cellular processes that have been implicated in the pathogenesis of IPF.

Antimicrobial effects

The role of infection in the pathogenesis of IPF has not been fully evaluated. 8 Infection is common in patients with IPF – even in those not receiving immunosuppression. In the TIPAC trial,21 we found that, of the patients in the placebo group not receiving prednisolone, 62% had an infection during the study. 24 In a meta-analysis25 of patients allocated to the placebo group from clinical trials of patients with IPF, the mean reported rate of pneumonia among studies not permitting immunosuppression was 37.1 per 1000 patient-years, which is higher than in those with chronic obstructive pulmonary disease (COPD). 26 Mortality from IPF increases in the winter even when recognised infection is not considered to be the cause of death. 27

In a study of bacterial culture from bronchial washings,28 more than one-third of patients with IPF were colonised with pathogenic bacteria or Pneumocystis jirovecii,29 the majority of whom were sensitive to co-trimoxazole. Garzoni et al. 30 evaluated the lung microbiota by sequencing bacterial 16S ribosomal ribonucleic acid (rRNA) genes. The study showed that the lungs of patients with ILD are not sterile. Subsequently, two independent groups of researchers, also using sequencing of bacterial 16S rRNA genes, have shown that bacterial load31,32 and lung microbiota profiles enriched with Streptococcus and Staphylococcus spp. 33 predict poor outcomes in IPF. The stability of the lung microbiota and the response to antibiotic therapy are unknown in IPF.

There is also evidence that innate immune responses may be abnormal in IPF, potentially increasing susceptibility to infection. In particular, alveolar macrophages from patients with IPF express a functional deficiency in their ability to kill intracellular bacteria,34 predisposing them to bacterial infection or colonisation. There is evidence that the expression of toll-like receptor (TLR) 2, which has a key role in Gram-positive pathogen sensing, is altered in IPF,35 and TLR9, which stimulates pro-inflammatory cytokine release, is upregulated in biopsies of rapidly deteriorating patients with IPF. 36

Co-trimoxazole has a broad spectrum of activity and is effective against most of the non-anaerobic bacteria in the airways of patients with IPF, including P. jirovecii, Streptococcus spp. and Staphylococcus spp.

Non-antimicrobial effects

Co-trimoxazole has beneficial effects in patients with granulomatous polyarthritis37 that are greater if treatment is started early and is not related to infection. 38 These potentially immunomodulatory effects have been poorly studied. Sulfamethoxazole has a structure similar to other sulfonamides, such as dapsone and sulfapyridine, which are known to have effects on neutrophil chemotaxis39 and superoxide production. 40 In vitro studies have shown that co-trimoxazole and its individual components (trimethoprim and sulfamethoxazole) inhibit neutrophil post-phagocytic myeloperoxidase-mediated protein iodination. 41 In other studies42,43 assessing the effects of different antimicrobial agents on neutrophil respiratory burst, co-trimoxazole and trimethoprim inhibited the chemiluminescence response at therapeutic concentrations.

Oxidant stress has been implicated in alveolar epithelial injury44 and epithelial–mesenchymal transition,45 and IPF patients have increased concentrations of 8-isoprostane in exhaled breath condensate. 46 Neutrophils play an important role in causing oxidant stress in IPF47 and neutrophilic alveolitis features frequently. 48 Furthermore, higher than normal neutrophil counts in sputum are associated with worse lung function49 and the percentage of bronchoalveolar lavage fluid (BALF) neutrophils at diagnosis is an independent predictor of mortality. 50

Thus, co-trimoxazole may inhibit neutrophil activation and reduce neutrophil-derived oxidative stress. These potential non-antimicrobial effects of co-trimoxazole would be predicted to have beneficial effects in IPF independently of and/or in addition to its antimicrobial actions.

Rational for primary outcome

All-cause hospitalisation and death have been recommended by the Pulmonary Fibrosis Foundation summit of end points for Phase III clinical trials in IPF. 52 Others have advocated FVC as an end point, claiming that mortality requires unfeasibly large studies in mild to moderate IPF. 53 The situation is, however, different when evaluating patients with more severe disease. In a meta-analysis of placebo data of all clinical trials of IPF,54 annual mortality in studies selecting only mild–moderate patients was 8%, but in those including moderate to severe patients, annual mortality was 19%, in keeping with the data from the TIPAC trial21 and the epidemiology of the disease. 3 The event rate of mortality and hospitalisations is even higher when selecting patients with severe disease (up to 16% in 3 months55). The outcome measure of all-cause death, non-elective hospitalisation or transplant also meets the European Medicines Agency’s criteria for composite end points, as hospitalisation is an important predictor of mortality. 56 Hospitalisation can be easily and reliably assessed without patient involvement and is the least likely outcome to be influenced by withdrawal from the study, or unintentional or unavoidable unblinding of patients.

Rational for biomarkers

Although the mechanism of IPF is not fully elucidated, it is considered that a trigger or triggers result in alveolar epithelial injury with chaotic epithelial repair, fibroblast proliferation and collagen deposition, which distorts the lung architecture. We assessed markers of infection/inflammation, monocyte activity, neutrophil activity, alveolar epithelial injury, fibro proliferation, pulmonary hypertension and bronchial epithelium (mucus-associated cancer antigens).

The primary objective

To compare the time to death (all causes), lung transplant or first non-elective hospital admission between co-trimoxazole and the placebo in patients with moderate to severe (FVC of ≤ 75% predicted) IPF during a median treatment period of 27 months (range 12–42 months).

Secondary objectives

To compare the clinical efficacy between the co-trimoxazole and placebo in patients with moderate to severe (FVC of ≤ 75% predicted) IPF during a median treatment period of 27 months (range 12–42 months) in terms of:

• time to respiratory-related death, lung transplant or first respiratory-related hospital admission

• time to respiratory-related and all-cause hospital admission

• time to respiratory-related and all-cause death

• QALYs

• health-related quality of life

• cough-related quality of life

• breathlessness and Cough Symptom Scores

• lung function

• oxygen saturations.

Mechanistic objectives

To compare blood biomarkers between the co-trimoxazole and placebo in patients with moderate to severe (FVC of ≤ 75% predicted) IVF at baseline and after 12 months of treatment in terms of markers of:

• infection/inflammation

• monocyte activity

• neutrophil activity

• alveolar epithelial injury

• fibroproliferation

• pulmonary hypertension

• bronchial epithelium.

Study setting

The study was conducted in UK secondary care centres that either met the specifications required for specialist ILD centre status or worked in association with specialist centres.

Sites had to have the facilities for research staff to undertake all of the measurements and store the samples required for the study unless, in exceptional circumstances, approval for the site to be excluded from some aspects of the study was granted by the chief investigator prior to site enrolment. The local principal investigator (PI) was responsible for the conduct of the study at their site in accordance with the protocol and for the safety and medical care of study patients. The investigators had to demonstrate the potential for recruiting the required number of suitable patients within the agreed recruitment period. They were responsible for ensuring that they had adequately trained staff to conduct the trial and for completing delegations of responsibility logs. Both the investigator and the local trust legal representative signed the site agreements. GCP training was required for all staff responsible for trial activities. The frequency of repeat training was dictated by the requirements of their employing institution or was conducted 2-yearly where the institution has no policy. The PI or delegate was required to document and explain any deviation from the approved protocol and to communicate this to the trial team at Norwich Clinical Trials Unit (NCTU).

Patient inclusion criteria

• Male or female, aged ≥ 40 years.

• A diagnosis of IPF based on a multidisciplinary consensus undertaken at a specialist centre (or a MDT otherwise meeting the criteria of a specialist centre) following a review of an appropriate clinical history, characteristic features of UIP on thoracic HRCT and/or UIP histology confirmed by surgical lung biopsy, according to the contemporaneous international guidelines. 8

• Patients could receive oral prednisolone up to a dose of 10 mg per day, antioxidant therapy, pirfenidone, nintedanib or other licensed medication for IPF. Patients were receiving a stable treatment regimen for at least 4 weeks to ensure that baseline values were representative.

• A MRC Dyspnoea Scale score of > 1 to exclude asymptomatic patients.

• Able to provide informed consent.

Patient exclusion criteria

• Forced vital capacity of > 75% predicted.

• A recognised significant co-existing respiratory disease, defined as a respiratory condition that exhibits a clinically relevant effect on respiratory symptoms and disease progression, as determined by the PI following a multidisciplinary discussion. For example, patients with bronchiectasis were included only if this was deemed to be traction bronchiectasis as a result of IPF.

• Patients with obstructive airways disease, defined as a forced expiratory volume in 1 second (FEV₁)/FVC of < 60%.

• Patients with a self-reported respiratory tract infection within 4 weeks of screening, defined as two or more of cough, sputum or breathlessness, and requiring antimicrobial therapy, were not eligible because of the difficulty of obtaining reliable baseline lung function.

• Significant medical, surgical or psychiatric disease that, in the opinion of the patient’s attending physician, would affect patient safety or influence the study outcome, including liver [e.g. serum transaminase more than three times upper limit of normal (ULN), bilirubin more than two times ULN (unless the patient has Gilbert’s syndrome) and renal failure (e.g. creatinine clearance rate of < 30 ml/minute/1.73m²)].

• Patients receiving immunosuppressant medication (except low-dose prednisolone) including azathioprine and mycophenolate mofetil. Immunosuppression is not advised for people with IPF. 9 Moreover, combining azathioprine with co-trimoxazole increases the potential for patients to develop neutropenia.

• Female patients of child-bearing potential. Non-child-bearing potential was defined as follows: postmenopausal female patients with at least 12 months of spontaneous amenorrhoea or 6 months of spontaneous amenorrhoea with a serum follicle-stimulating hormone concentration of > 40 mIU/ml, or female patients who had had a hysterectomy, bilateral salpingectomy or bilateral oophorectomy at least 6 weeks prior to enrolment.

• Known allergy or intolerance to trimethoprim or sulfonamides or their combination, for safety reasons.

• Untreated folate or B₁₂ deficiency. This was to ensure that no bone marrow or neurological adverse effects occured with folate therapy to B₁₂-deficient individuals.

• Known glucose-6-phosphate dehydrogenase (G6PD) deficiency or G6PD deficiency measured at screening in males of African, Asian or Mediterranean descent. Sulfonamides are recognised to increase the risk of haemolysis in individuals with G6PD deficiency. The prevalence of G6PD deficiency is higher in males of African, Asian or Mediterranean descent than in those of other ethnic backgrounds. However, the risk of haemolysis is low even in populations with high prevalence. 58

• Receipt of an investigational drug or biological agent within the 4 weeks prior to study entry or five times the drug half-life, whichever is longer.

• Receipt of short-course antibiotic therapy for respiratory and other infections within 4 weeks of screening.

• Patients receiving long-term (defined as > 1 month of therapy) prophylactic antibiotics, as this may have had an impact on lung microbiota. Such patients could enrol in the Efficacy and Mechanism Evaluation of Treating Idiopathic Pulmonary Fibrosis with the addition of Co-trimoxazole (EME-TIPAC) trial, if this was supported by their clinician, after a wash-out period of 3 months.

• Serum potassium level of > 5.0 mmol/l because of the potentially increased risk of hyperkalaemia in patients taking co-trimoxazole in combination with potassium-sparing diuretics (including angiotensin-converting enzyme inhibitors or angiotensin receptor blockers).

No exceptions to the stated eligibility criteria were permitted. Patients could be entered into other observational studies with the prior agreement of the Trial Management Group (TMG) of both studies.

Drug preparation and supply

Europe-wide tendering for investigational medicinal product (IMP) manufacture for the trial was undertaken in June 2014. Following a successful bid, the tender to manufacture IMPs was awarded to Guy’s and St Thomas’ NHS Foundation Trust Pharmacy Manufacturing Unit (GSST PMU) (London, UK). The licence granted to GSST PMU by the MHRA under the requirements of EU Directive 2001/20EC was MA (IMP) 11387.

A dose of 480 mg of co-trimoxazole or matching placebo tablets was packaged in a white opaque high-density polyethylene plastic container that was sealed with a child-resistant/tamper-evident cap and labelled in compliance with applicable regulatory requirements, including EU GMP Annex 13 Investigational Medicinal Products, (Vol. 4, Annex 13; www.gmp-compliance.org/guidelines/gmp-guideline/eu-gmp-annex-13-investigational-medicinal-products; accessed October 2020). Each container contained a 31-day supply, which equated to 124 tablets. In addition, reduced-dose containers were produced containing only 26 tablets for those patients who were reducing to two tablets once a day, three times per week.

The active and placebo IMP containers appeared identical, except that the containers were coded with treatment group 1 or 2, to differentiate between active and placebo packs, on a tear-off strip. When the tear-off strip was removed, the packs appeared identical.

Randomisation was performed centrally with secure database-generated e-mail correspondence by NCTU to research pharmacists only. This enabled both the trial team at NCTU and the local research team (including the PI) at the site to remain fully blinded to the allocation. The ‘semi-blind’ pharmacist receiving the e-mail allocated the IMP according to treatment group 1 or 2 and was unblind to the group (1 or 2), but blind to the intervention.

The IMPs and folic acid were dispensed by the hospital pharmacy to the patient directly or, at baseline, in situations where it would prevent the patient from having to return to the hospital to collect the IMP, by a health-care courier to the patient’s home.

Pharmacies were required to maintain up-to-date accountability, dispensing and destruction logs for inspection by the NCTU and regulatory authorities during the trial. IMPs and folic acid sent by courier (or other signed-for delivery services) to patients required signature on receipt. Patients were advised to store their medication at < 25 °C, but there was no temperature monitoring after dispatch from the third party. Patients were dispensed an initial 3 months’ supply at baseline, with the same amount dispensed 3 months later. They were then given 6 months’ supply, that is six bottles, on each subsequent occasion.

Assignment of intervention

The allocated treatment for a patient was generated by a computer-written code using minimisation under the supervision of the study statistician. Minimisation was performed using Taves’ method with the factors measured at baseline: (1) study site, (2) whether or not the patient had consented to take part in the bronchoscopy substudy (yes/no) and (3) whether or not the patient was receiving licensed antifibrotic medication for IPF at the screening visit (yes/no). To decide on the treatment allocation, the code calculated the number of patients in each group who had the same characteristics as the patient awaiting allocation; patients were allocated to the intervention with the smaller number with a high probability. If the numbers were the same, then simple randomisation was used.

The patients were allocated to the intervention by a process embedded in the web-based data management system. The randomisation code was saved in the study database for later decoding and for emergency unblinding purposes. When a patient was randomised, an e-mail was sent to the appropriate local pharmacy, who prepared the medication pack.

Blinding

This was a double-blind study. The placebo and active treatments appeared identical and were dispensed in identical containers containing the same number of tablets. Other than in instances requiring emergency unblinding or unblinding for safety reporting, all trial patients, care providers, outcome assessors and data analysts remained blind throughout the study.

The decision to unblind a single case was made when knowledge of an individual’s allocated treatment was required to enable treatment of severe adverse event(s) [SAE(s)].

Where possible, requests for emergency or unplanned unblinding of individuals were made through the trial manager and the agreement of the chief investigator was sought. However, in circumstances in which there was insufficient time to make this request or for agreement to be sought, the treating clinician made the decision to unblind immediately. This was done using the study database (local PIs and the chief investigator had special logins that allowed unblinding and that were closely audited within the database management system) or by the chief investigator, who authorised unblinding by the Data Management Team. All instances of unblinding were recorded and reported to NCTU by the local PI, including the identity of all recipients of the unblinding information.

Compliance and adherence

Compliance with study treatment, in the form of returned tablet counts, was monitored as part of drug accountability at each visit.

Concomitant medication

Patients were permitted to receive N-acetylcysteine and antioxidants, prednisolone (up to a dose of 10 mg per day) and licensed treatments for IPF. All concomitant medications were recorded at baseline and any change in concomitant medication was recorded at each visit.

Patients were permitted to receive other medications (e.g. for other conditions), but non-permitted therapies included amiodarone, azathioprine, mycophenolate mofetil, cyclophosphamide, methotrexate, D-penicillamine, colchicine, clozapine, methenamine, dapsone, interferon gamma, ciclosporin, mercaptopurine, repaglinide, pyrimethamine, lamivudine, typhoid vaccination or unlicensed medication.

Therapy requiring caution or increased monitoring included digoxin, warfarin, phenytoin, sulfonylureas and procainamide hydrochloride. Increased monitoring of potassium levels was required if patients commenced medication that increases serum potassium concentration.

Treatment discontinuation

Individual patients stopped treatment early for the following reasons:

• any non-elective hospitalisation or lung transplant (meeting the primary end point)

• a serum potassium level of > 5.5 mmol/l

• co-trimoxazole-related haematological disease (e.g. blood dyscrasia or thrombocytopenia)

• unacceptable treatment toxicity or an AE

• intercurrent illness that prevented further treatment

• any change in the patient’s condition that, in the clinician’s opinion, justified the discontinuation of treatment

• withdrawal of consent for treatment by the patient.

Patients who discontinued protocol treatment for any of the above reasons remained in the trial for the purpose of follow-up and data analysis, unless they requested otherwise. The patients were invited to continue follow-up in the trial, although they no longer took the study drug. However, whenever the patient no longer wished to be followed up either, this view was respected and the patient was withdrawn entirely from the trial. Data that were already collected were kept and included in analyses according to the ITT principle for all patients who stop follow-up early. Patients provided consent so that follow-up information on overall or hospital-free survival could be obtained from medical records using the NHS number, for example through their GP, if required. Research teams were asked to account for the vital status and details of admission to hospital for all patients, regardless of whether they had withdrawn from the intervention or study assessments. Patients who stopped taking the study drug or withdrew from the study were not replaced.

Patients who were admitted to hospital non-electively were deemed to have met the primary end point. From that point onwards, patients ceased to have follow-up measurements taken, but survival status was reported at the end of the study.

Primary outcomes

The primary outcome was the time to death (all causes), transplant or first non-elective hospital admission. This was defined as the time from randomisation to death, lung transplant or first non-elective hospital admission for any reason. Details of patients admitted to hospital or dying were captured by examining SAE reports, hospital patient databases and the tracing of patients missing appointments by contacting their primary care physician, as required. Non-elective admission was defined as a hospital stay lasting more than 24 hours that had not been arranged more than 24 hours prior to admission. Death could not be influenced by unintentional unblinding and we believe it is unlikely that admission to hospital was influenced by this either. Treatment or absence of treatment with co-trimoxazole rarely causes conditions that would require hospitalisation and, given the costs and other implications of admission to hospital, hospitalisation was likely to be because of clinical need.

Secondary outcomes

The individual components of the primary outcome – time to death (all causes), transplant and time to first non-elective hospital admission – were analysed separately as secondary outcomes. In addition, respiratory-related events were analysed separately from non-respiratory-related events.

Extracts from the case report form (CRF) were forwarded to an independent review committee. This committee, chaired by a consultant respiratory physician with experience of undertaking clinical trials, also included a consultant respiratory physician, a nurse consultant specialising in ILD and a primary care physician. Each independently reviewed whether or not the event was respiratory related based on the CRF listing. If no consensus could be achieved, then the chairperson had the casting vote.

The following measurements were undertaken at baseline, 3 and 6 months, and then 6-monthly throughout the study. Every effort was made to collect data at time points within a 2-week window (i.e. 2 weeks before or after the visit schedule) until 6 months, then within a 1-month window thereafter. However, as the study visits were aligned with routine clinical care assessments, this scheduling was not always possible and values obtained outside these windows were captured along with the assessment date. In addition, a final assessment was made at the end of the study if a patient had not had a primary outcome measurement undertaken within 2 months of the end of the study.

Questionnaires

Lung function

Lung function was measured at recruitment and at 6 and 12 months using spirometry performed to American Thoracic Society/European Respiratory Society standards. 66 Spirometry is a routine part of the clinical assessment of people with ILD and is usually measured at all clinical assessments. The FVC and FEV₁ measures were obtained as part of the spirometry assessment. The predicted equations, derived by Crapo et al. ,67 were used to calculate the predicted normal value and percentage predicted values for FVC and FEV₁. Where spirometry was contraindicated or patients were not able to complete spirometry, this was omitted.

Gas transfer measurements were measured at recruitment and at 6 and 12 months using spirometry performed to American Thoracic Society/European Respiratory Society standards. 68 The diffusing capacity of the lung for carbon monoxide (DLCO) was obtained and the percentage predicted values were calculated using the predicted values obtained from the equations derived from the European Coal and Steel Community. 69

Forced vital capacity and DLCO are both components of prognostic modelling algorithms,56,70 are frequently utilised in clinical trials14 and are part of routine care. These were the main lung function outcomes.

Peripheral blood

Peripheral blood was taken at baseline, at 3, 6 and 12 months, and at the end of the study. The peripheral blood samples were stored throughout the study. Blood was collected in ethylenediaminetetraacetic acid (EDTA) and serum Vacutainers^® (Fisher Scientific UK Ltd, Loughborough, UK), centrifuged, and the supernatant aliquoted and stored locally at –70/–80 °C. Periodically, samples were couriered (CitySprint, Loughborough, UK) in a BioTherm 45 (Intelsius UK, York, UK) with dry ice for next-day delivery (category B biological samples) to the Department of Laboratory Medicine at the Norfolk and Norwich University Hospital and stored at –70/–80 °C until the end of the study. Serum of matching baseline and 12-month (± 60 days) samples was analysed for the following biomarkers.

Routine microbiology

Sputum was obtained, where possible, and sent for local microbiological culture and sensitivity testing. For all patients, a nasal swab was sent for viral culture, if clinically indicated. All microbiology laboratories followed a common protocol for sputum processing and susceptibility testing of the bacteria recovered. Assessment of urinary Legionella and pneumococcal antigens or serology for atypical respiratory pathogens were not undertaken routinely as these are not helpful in asymptomatic individuals.

Safety

The following were measured at baseline, 6 weeks, 3, 6, 9 and 12 months, and then 6-monthly for the duration of the study, as well as at the end of the study/at hospitalisation [ideally within a 1- or 2-month window (as above)]:

• full blood count and differential white cell count

• urea and electrolytes

• liver function.

Unless the patient was otherwise due to attend a clinic visit at the 6-week and 9-month time points as part of their standard care, the safety bloods for these visits were performed at the patient’s primary care surgery and the patient followed up by telephone (to check for AEs and any change in concomitant medication). Abnormal routine laboratory values were considered to be AEs if they were outside the normal reference range for the local laboratory.

Measurements during the first non-elective admission

An entry was made on the patient’s medical records and patients were asked to carry a card detailing their involvement in the study, with research nurse and study co-ordinator contact details, to maximise ascertainment of questionnaires, blood samples and routine microbiology in the same manner as undertaken at the routine visits during their admission to hospital.

Bronchoscopy substudy

We had planned a substudy to investigate the effects of co-trimoxazole on BALF in a subset of 50 patients, who volunteered to undergo bronchoscopy at baseline and after 3 months of treatment. Bronchoalveolar lavage (BAL) is the only appropriate method of sampling the lower airways, which are the principal region of disease in IPF. The alternative methods (spontaneous or induced sputum analysis) would have confounded the results by upper airway contamination90 and, to our knowledge, none of the available biomarkers of lung injury and inflammation has been evaluated in sputum. Bronchoscopy with BAL is safe in patients with IPF with a risk of SAEs of < 1 : 1000 at experienced sites. In a study of 281 patients with ILD undergoing BAL, no events necessitated therapy. 91

Bronchoscopies were performed as per current British Thoracic Society guidelines. 92 All bronchoscopies were to be through the oral route to avoid nasal contamination. BAL was undertaken by instilling four 50-ml aliquots of sterile saline through the bronchoscope wedged in a segment of the middle lobe. The material was recovered by gentle suction and aliquots were taken. BALF was placed on ice and strained through sterile gauze prior to centrifugation (at 310 g for 5 minutes at 4 °C) to collect the cell pellet. The supernatant was aspirated and snap frozen at –80 °C in 1-ml aliquots. The cell pellet was washed and resuspended. In addition, bronchoscope washing was stored for sequencing of deoxyribonucleic acid (DNA) derived from bacterial 16S rRNA genes.

Unfortunately, bronchoscopy with BAL was undertaken in only two people for the purposes of the study. We initially planned that this procedure would be undertaken in only two sites (Royal Papworth Hospital NHS Foundation Trust and the Royal Brompton Hospital) to maximise internal consistency as these sites have significant experience with this procedure for the purposes of research. However, despite doubling the number of sites available to offer research bronchoscopy for people interested in participating in this substudy (by adding Aintree University Hospital and University Hospitals Coventry & Warwickshire), and with plans for expansion to another eight trusts, we could still not recruit into the substudy. The Data Monitoring Committee (DMC) recommended abandoning the substudy on 14 June 2017 on the grounds of futility.

The lack of interested patients was thought to be due to the invasive nature of the procedure. Following discussion on 1 August 2017, the Trial Steering Committee (TSC) acknowledged the efforts that had been undertaken to improve recruitment by the chief investigator and the study team, but agreed with the DMC’s recommendation and, as a result, the substudy was closed to recruitment. None of the samples was analysed, as the results would have been meaningless.

Pharmacovigilance

Study monitoring

Prior to commencement of the trial, a quality management and monitoring plan (QMMP) was produced, which detailed all planned and systematic actions established to ensure that the EME-TIPAC trial was performed and that the data were generated, documented and reported in accordance with the principles of GCP and applicable regulatory requirements. The QMMP was reviewed annually during the trial by the NCTU Quality Management Group and updated when appropriate.

Any findings identified during monitoring that caused concern were to be discussed with the chief investigator and/or the TMG, with discussions recorded and stored in the trial master file.

Central quality control procedures included a formal, documented site assessment procedure; the signing of a PI statement agreeing to the responsibilities of the role; the review of delegation logs, PI curricula vitae and GCP certificates; regular trial team meetings to review data and recruitment; and a review of anonymised screening logs, trial drug accountability and dispensing logs at the NCTU at periodic intervals. The trial database was also programmed to prevent the randomisation of ineligible patients, with liver function, renal function, G6PD and lung function tests validated in real time.

Quality control procedures at clinical sites included formal site initiation training (either in person or by teleconference), electronic CRF review, and the periodic checking of essential documents, investigator site files (ISFs) and pharmacy site files (PSFs).

In addition, central monitoring was performed throughout the trial and documented by the trial team in an annual monitoring report that was provided to the trial sponsor and the TMG. Central monitoring including the following actions:

• Collection of dispensing and accountability logs from all participating pharmacies and the subsequent reconciliation of these with data contained in the CRF.

• Protocol compliance checks, for example, checking that no dose re-escalations occurred (through a review of accountability logs) or checking adherence to the protocol-defined non-IMP regimen while patients were on active treatment.

• An additional centralised eligibility review of patients’ age, FVC per cent predicted, and folate, B₁₂ and serum potassium levels to identify any ineligible patients.

• Collection of delegation logs for review and cross-referencing against consent forms and the database to ensure that only appropriately delegated members of staff were performing trial-related activities.

• Collection of ISF and PSF checklists to ensure that sites were working to current trial documentation.

• Ongoing review of CRF data for errors, inconsistencies and missing key data points.

• A review of overall data accuracy and completeness for each site to flag issues and escalate where applicable.

• The cross-referencing of visit dates against expected visit dates to ensure that sites were carrying these out in accordance with the protocol visit schedule.

• A review of all AE data to ensure that these were coded using Medical Dictionary for Regulatory Activities (MedDRA) terminology and to identify any under-reported SAEs.

• Patients provided consented to enable the NCTU to hold a copy of the consent form to ensure that the correct version had been used, the correct staff members had undertaken consent and the consent form had been completed appropriately.

• Out-of-hours contact details provided to patients were tested by the trial team outside working hours for the top three recruiting sites during the trial.

On-site monitoring was performed at the top 10 recruiting sites between 2017 and 2018. The on-site monitoring visits, in addition to activities performed during central monitoring, involved source data verification of a sample of patients at the site, checks to ensure that documentation was completed according to GCP, a review of the clinic notes to check for unreported notable or serious events, a pharmacy inspection, and ISF and PSF review.

Risk-based central statistical monitoring (CSM) was performed by the trial statistician, or a delegate, prior to each DMC meeting, with the aim of identifying potential recording and entry errors, procedural errors and possible fraud. Blinded CSM results were to be discussed with the trial manager prior to each DMC meeting to enable any issues to be resolved prior to the meeting, if possible, or escalated to the chief investigator.

Direct access to patient records

Participating investigators agreed to allow trial-related monitoring, including audits, Research Ethics Committee review and regulatory inspections, by providing access to source data and other trial-related documentation, as required. Patient consent for this was obtained as part of the informed consent process for the trial.

Sample size

The primary outcome measure was unplanned hospitalisation-free survival, which is a composite end point of the time to death (all causes) or first non-elective (all-cause) hospital admission. The study duration was estimated to be 30 months’ recruitment phase and an additional 12 months’ follow-up after the last patient was recruited (a total of 42 months after the first patient was enrolled), which approximated to a median patient study duration of 27 months. The trial was designed to have 80% power (two-sided test, significance level of 5%) to show a change in hospitalisation-free survival from a median value of 28.8 months in the placebo group to 51.1 months in the co-trimoxazole group [hazard ratio (HR) of 0.56] over this study period, assuming that 264 patients were randomised. This was based on a sensitivity analysis of patients from the TIPAC trial21 with reduced lung function (i.e. FVC< 70% predicted) using an ITT analysis.

With regard to the power of the mechanistic studies, we assumed that 264 patients would provide data for the mechanistic aspect. This would provide 80% power to detect a difference of 6.7 mg/l in CRP concentration based on a SD of 19.38 mg/dl,71 of 0.51 ng/ml in MMP-7 concentration based on a SD of 1.48 ng/ml82 and of 99 ng/ml in SP-D concentration based on a SD of 212 ng/ml. 94 It was not possible to undertake a power calculation for the change in the microbiota. However, co-trimoxazole is effective against many of the organisms detected in BAL from routine culture and genotyping techniques and, therefore, we expected, within a proposed group of 50 patients, to be able to detect a change in the flora.

Statistical analysis

A statistical analysis plan (SAP) was produced and agreed with the TSC and DMC prior to analysis (see Appendix 1):

• primary outcome –

  • time from randomisation to death (all causes), lung transplant or first non-elective hospital admission for any reason

• secondary efficacy outcomes –

  • time from randomisation to death (all causes)

  • time from randomisation to first non-elective hospital admission for any reason

  • time from randomisation to lung transplantation

  • the K-BILD health-related quality-of-life questionnaire score, the MRC Breathlessness Score, the EuroQol-5 Dimensions (EQ-5D) quality-adjusted life-years assessment, CSS and quality-of-life LCQ score

  • lung function, including assessment by spirometry (FVC) and total DLCO

• secondary outcome measures for safety (measured at local hospital laboratories) –

  • full blood count

  • urea and electrolytes

  • liver function

  • AEs including SAEs.

Trial oversight committees

A TSC with independent members oversaw the conduct and progress of the trial. The committee met by teleconference every 6 months for the duration of the study and comprised the following individuals:

• Professor Ron du Bois (chairperson) – no affiliation – retired

• Dr Kim Harrison – Swansea University

• Dr Sanjay Agrawal – University of Leicester

• Professor Ann Millar – University of Bristol.

On 3 August 2016, University Hospitals of Leicester NHS Trust was added as a recruiting site with Dr Felix Woodhead as PI. As Dr Agrawal held a substantive contract with the same trust, to meet the National Institute for Health Research (NIHR)’s definition of independence (i.e. that the TSC member is not part of an institution acting as a recruiting centre), Professor Ann Millar joined the TSC to maintain the presence of three independent members on the TSC.

An independent DMC oversaw the safety of patients within the trial. The committee met by teleconference at least annually for the duration of the study and comprised the following individuals:

• Dr Nik Hirani – University of Edinburgh

• Professor Sarah Pett – University College London

• Dr Jack Bowden – University of Bristol.

All TSC and DMC members were required to complete a Terms of Reference form and declare any potential competing interests.

Breaches and protocol deviations

Breaches of trial protocol or GCP were recorded and reported to the trial sponsor. Protocol deviations were recorded on the NCTU non-conformance database and were included in TMG, DMC and TSC reports.

A summary of breaches and protocol deviations is given in Appendix 1, Table 24. Patients who were the subject of a protocol deviation remained in the ITT population, the safety population and the PP population (if compliance criteria were met).

Eligibility violations

One patient was randomised in error. The patient had a lung function that was too high and violated the inclusion criteria. The decision was made to exclude this person from the analysis.

Patient flow

The Consolidated Standards of Reporting Trials (CONSORT) flow chart for this trial, in respect of the primary outcome, is provided in Figure 3. In summary, a total of 1305 patients were screened, of whom 420 were assessed for eligibility, 349 met the criteria and 342 were randomised. Of the 342 randomised patients, 172 were randomised to the placebo group and 170 were randomised to the active treatment group. One patient, randomised to the active treatment group, was randomised in error and their data were not analysed. A total of 58 randomised patients dropped out of the study (the reasons are given in Appendix 1, Table 25). A total follow-up of 394.6 person-years was observed with 164 events.

FIGURE 3.

The CONSORT flow chart for the trial. Reproduced with permission from Wilson et al. 95

Reproduced with permission from Wilson et al. 95

Patient withdrawal

A total of 58 (17%) patients withdrew from the study: 32 (19%) from the active treatment group and 26 (15%) from the placebo group. The reasons for withdrawal are given in Appendix 1, Table 25.

Per-protocol analysis

A total of 123 events occurred for the primary outcome: 64 in the placebo group and 59 in the active treatment group. The total exposure time was 264.6 years, roughly an average of 1.08 person-years. The rate of events was 0.44 (59/132.6) per person-year in the active treatment group and 0.48 per person-year (64/132) in the placebo group. The estimated unadjusted hazard ratio was 0.9 (95% CI 0.7 to 1.3) and when adjusting for the factors used in the minimisation algorithm it was virtually unchanged at 0.9 (95% CI 0.7 to 1.3), as shown in Table 7. The time to event is displayed graphically in Figure 5. This figure demonstrates some crossing of the survival curves, but overall the proportional hazards assumption was found not to be violated (p = 0.9908).

FIGURE 5.

Kaplan–Meier estimate of time to event: PP.

Modified per-protocol analysis

A total of 106 events occurred for the primary outcome: 44 in the placebo group and 62 in the active treatment group. The total exposure time was 216.2 years, roughly an average of 1.02 person-years. The rate of events was 0.48 (44/92.5) per person-year in the active treatment group and 0.50 per person-year (62/123.7) in the placebo group. The estimated unadjusted hazard ratio was 0.9 (95% CI 0.6 to 1.4) and when adjusting for the factors used in the minimisation algorithm it was virtually unchanged at 0.9 (95% CI 0.6 to 1.4), as shown in Table 7. The time to event is displayed graphically in Figure 6. This demonstrates some crossing of the survival curves, but overall the proportional hazards assumption was found not to be violated (p = 0.3193).

FIGURE 6.

Kaplan–Meier estimate of time to event: mPP.

Time-to-event outcomes

Intention to treat

The individual component events of the primary outcome are given in Table 8 in the same format as that of the primary outcome. The number of events is reduced compared with the primary outcome so that the CIs are correspondingly wider. The total exposure time is the same for all outcomes as once one event had occurred the follow-up was censored for all other events. The number of deaths was higher in the active treatment group than in the placebo group, with 24 and 18 deaths, respectively, but the difference was not significant (HR 1.5, 95% CI 0.8 to 2.8; p = 0.167). Respiratory-related deaths were, similarly, higher in the active treatment group than in the placebo group, with 20 and 17 deaths, respectively, but the difference was not significant (HR 1.4, 95% CI 0.7 to 2.6; p = 0.343). Hospital admissions were roughly equal in both the active treatment and placebo groups, 59 and 61, respectively; but the difference was not significant (HR 1.1, 95% CI 0.7 to 1.5; p = 0.754). Respiratory-related hospitalisations were roughly equal in both the active treatment and placebo groups, 40 and 42, respectively, but the difference was not significant (HR 1.0, 95% CI 0.7 to 1.6; p = 0.827). In all cases, the unadjusted and adjusted analyses were practically the same. As the Kaplan–Meier plot would provide a biased estimate due to informative censoring, a cumulative incidence plot is provided for each outcome in Figures 7 and 8. A separate analysis was not carried out for lung transplant as only one event occurred in each treatment group.

TABLE 8 - Secondary outcome results for individual components of primary outcome

Adjusted for site and baseline antifibrotic therapy.

Adjusted for site, baseline antifibrotic therapy and baseline value.

Secondary outcome	Treatment group				Hazard ratio (active treatment vs. placebo)
	Active treatment (N = 169)		Placebo (N = 172)
	Total exposure time (years)	Number of events to date (incidence)	Total exposure time (years)	Number of events to date (incidence)
ITT
Deaths censored at date of primary event (if primary event not death)	185.6	24	209.1	18	Unadjusted:a 1.5 (95% CI 0.8 to 2.8); p = 0.167
					Adjusted:b 1.5 (95% CI 0.8 to 2.8); p = 0.169
Respiratory-related death (censored at time of primary event)	185.6	20	209.1	17	Unadjusted:a 1.4 (95% CI 0.7 to 2.6); p = 0.343
					Adjusted:b 1.4 (95% CI 0.7 to 2.6); p = 0.352
Non-elective hospital admissions (all cause)	185.6	59	209.1	61	Unadjusted:a 1.1 (95% CI 0.7 to 1.5); p = 0.754
					Adjusted:b 1.1 (95% CI 0.7 to 1.3); p = 0.731
Respiratory-related hospitalisation (censored at time of primary event)	185.6	40	209.1	42	Unadjusted:a 1.0 (95% CI 0.7 to 1.6); p = 0.857
					Adjusted:b 1.0 (95% 0.7 to 1.6); p = 0.827
PP
Deaths censored at date of primary event (if primary event not death)	132.6	13	132.0	12	Unadjusted:a 1.1 (95% CI 0.5 to 2.4); p = 0.824
					Adjusted:b 1.1 (95% CI 0.5 to 2.4); p = 0.812
Respiratory-related death (censored at time of primary event)	132.7	12	132	11	Unadjusted:a 1.1 (95% CI 0.5 to 2.5); p = 0.821
					Adjusted:b 1.1 (95% CI 0.5 to 2.5); p = 0.821
Non-elective hospital admissions (all cause)	132.6	45	132.0	51	Unadjusted:a 1.1 (95% CI 0.7 to 1.7); p = 0.581
					Adjusted:b 0.9 (95% CI 0.6 to 1.4); p = 0.644
Respiratory-related hospitalisation (censored at time of primary event)	132.6	29	132.0	35	Unadjusted:a 0.8 (95% CI 0.5 to 1.3); p = 0.451
					Adjusted:b 0.8 (95% CI 0.5 to 1.4); p = 0.487
mPP
Deaths censored at date of primary event (if primary event not death)	92.5	11	123.7	12	Unadjusted:a 1.2 (95% CI 0.5 to 2.8); p = 0.597
					Adjusted:b 1.3 (95% CI 0.6 to 2.9); p = 0.585
Respiratory-related death (censored at time of primary event)	92.5	10	123.7	11	Unadjusted:a 1.2 (95% CI 0.5 to 2.9); p = 0.630
					Adjusted:b 1.2 (95% CI 0.5 to 2.9); p = 0.624
Non-elective hospital admissions (all cause)	92.5	32	123.7	49	Unadjusted:a 0.9 (95% CI 0.5 to 1.3); p = 0.487
					Adjusted:b 0.9 (95% CI 0.6 to 1.3); p = 0.511
Respiratory-related hospitalisation (censored at time of primary event)	92.5	23	123.7	34	Unadjusted:a 0.9 (95% CI 0.5 to 1.5); p = 0.635
					Adjusted:b 0.9 (95% 0.5 to 1.5); p = 0.657

FIGURE 7.

Cumulative incidence function: death only.

FIGURE 8.

Cumulative incidence function: hospitalisation only.

Per protocol

The individual component events of the primary outcome are given in Table 8 in the same format as that of the primary outcome. The number of events is reduced compared with the primary outcome so that the CIs are correspondingly wider. The total exposure time is the same for all outcomes as once one event had occurred the follow-up was censored for all other events. The number of deaths was higher in the active treatment group than in the placebo group, with 13 and 12 deaths, respectively, but the difference was not significant (HR 1.1, 95% CI 0.5 to 2.4; p = 0.824). Respiratory-related deaths were similarly higher in the active treatment group than in the placebo group, with 12 and 11 deaths, respectively; however, the difference was not significant (HR 1.1, 95% CI 0.5 to 2.5; p = 0.821). Hospital admissions were roughly equal in both the active treatment and placebo groups, with 45 and 51 hospitalisations, respectively, but the difference was not significant (HR 1.1, 95% CI 0.7 to 1.7; p = 0.581). Respiratory-related hospitalisations were roughly equal in both the active treatment and placebo groups, 29 and 35, respectively, but the difference was not significant (HR 0.8, 95% CI 0.5 to 1.3; p = 0.451). In all cases, the unadjusted and adjusted analyses were practically the same. As the Kaplan–Meier plot would provide a biased estimate as a result of informative censoring, a cumulative incidence plot is provided for each outcome in Figures 9 and 10. A separate analysis was not done for lung transplant as only one event occurred in each treatment group.

FIGURE 9.

Cumulative incidence function: death only – PP.

FIGURE 10.

Cumulative incidence function: hospitalisation only – PP.

Modified per protocol

The individual component events of the primary outcome are given in Table 8 in the same format as that of the primary outcome. The number of events is reduced compared with the primary outcome so that the CIs are correspondingly wider. The total exposure time is the same for all outcomes as once one event had occurred the follow-up was censored for all other events. The number of deaths was about the same in the active treatment group and the placebo group, 11 and 12, respectively, but the difference was not significant (HR 1.2, 95% CI 0.5 to 2.8; p = 0.597). Respiratory-related deaths were similarly higher in the active treatment group than in the placebo group, with 10 and 11, respectively, but the difference was not significant (HR 1.2, 95% CI 0.5 to 2.9; p = 0.630). Hospital admissions were roughly equal in both the active treatment and placebo groups, 32 and 49, respectively, but the difference was not significant (HR 0.9, 95% CI 0.5 to 1.3; p = 0.487). Respiratory-related hospitalisations were roughly equal in both the active treatment and placebo groups, 23 and 34, respectively, but the difference was not significant (HR 0.9, 95% CI 0.5 to 1.5; p = 0.635). As the Kaplan–Meier plot would provide a biased estimate as a result of informative censoring, a cumulative incidence plot is provided for each outcome in Figures 11 and 12. A separate analysis was not done for lung transplant as only one event occurred in each treatment group.

FIGURE 11.

Cumulative incidence function: death only – mPP.

FIGURE 12.

Cumulative incidence function: hospitalisation only – mPP.

Intention-to-treat results

The questionnaire data were available for approximately 160 individuals at 12 months post randomisation, with more data available in the placebo group than in the active treatment group. For the LCQ, the mean score in the active treatment group was 15.37, compared with 14.59 in the placebo group. This mean difference of –0.75 was in favour of the active treatment group, but was not significant (p = 0.267). After accounting for baseline score, the difference decreased slightly (–0.60, 95% CI –1.56 to 0.36) and was still non-significant (p = 0.219). None of the scores for components of the LCQ was significantly different between the treatment groups, but all of the mean differences were in favour of the active treatment group.

The MRC score had the same median value in both the active treatment and placebo groups and neither the difference between groups (p = 0.9359) nor the change from baseline was significant (p = 0.2932). The average CSS was 44.74 in the active treatment group, compared with 49.69 in the placebo group; this difference was not significant (p = 0.243), nor did this change after adjusting for baseline (p = 0.570). The K-BILD questionnaire had average total scores of 50.32 in the active treatment group and 50.74 in the placebo group; the score difference was not significant (p = 0.834), nor did this change after adjusting for baseline (p = 0.932). None of the scores for components of the K-BILD questionnaire was significant. The results are shown in Table 9.

Per-protocol results

The questionnaire data were available for approximately 110 individuals at 12 months post randomisation, with more data available in the placebo group than in the active treatment group. The LCQ total score was borderline statistically significantly different between the active treatment and placebo groups. The mean score in the active treatment group was 15.91, compared with 14.30 in the placebo group; this mean difference of –1.53 (95% CI –3.11 to 0.04) was in favour of the active treatment group and was of borderline statistical significance (p = 0.057). After accounting for baseline score, the difference decreased slightly to –1.24 (95% CI –2.37 to –0.11), but was significant (p = 0.032). None of the components of the LCQ was significantly different between the active treatment and placebo groups in the unadjusted analyses, but all of the mean differences were in favour of the active treatment group. In the adjusted analyses, the physiological score was significantly different in favour of the active treatment group, with a mean difference of –0.44 (95% CI –0.85 to –0.03; p = 0.037).

The MRC score had the same median value in both treatment groups and was not significantly different (p = 0.8363); the change from baseline was also not significantly different (p = 0.4482). The average CSS was 44.80 in the active treatment group, compared with 49.56 in the placebo group; this difference was not significant (p = 0.402), nor did this alter after adjusting for baseline (p = 0.855). The K-BILD questionnaire had average total scores of 51.47 in the active treatment group and 50.82 in the placebo group; this difference was not significant (p = 0.759), nor did this alter after adjusting for baseline (p = 0.881). Of the components of the K-BILD questionnaire, only the chest component was significant, with a mean score of 62.46 in the active treatment group compared with 54.62 in the placebo group; this mean difference of –8.41 (95% CI –16.05 to –0.76; p = 0.031) was in favour of the active treatment group and, although it reduced to –6.85 (95% CI –13.29 to –0.41) after adjusting for baseline, it was still significant (p = 0.037). The results are shown in Table 10.

Modified per-protocol results

The questionnaire data were available for approximately 90 individuals at 12 months post randomisation, with more data available in the placebo group than in the active treatment group (Appendix 1, Table 28). For the LCQ total score, the mean in the active treatment group was 15.67 compared with 14.13 in the placebo group; this mean difference of –1.47 (95% CI –3.20 to 0.26) was in favour of the active treatment group, but was not significant (p = 0.096). After accounting for baseline score, the difference decreased slightly to –1.43 (95% CI –2.72 to –0.14), but was significant (p = 0.029). None of the components of the LCQ was significantly different between the active treatment and placebo groups in the unadjusted analysis, but all of the mean differences were in favour of the active treatment group. In the adjusted analysis, all of the components were close to significance, but only the physiological score was significantly different, in favour of the active treatment group, with a mean difference of –0.54 (95% CI –1.04 to –0.05; p = 0.030).

The MRC score had the same median value in both the active treatment and placebo groups and was not significantly different (p = 0.6585). In addition, the change from baseline was not significantly different (p = 0.4859). The average CSS was 43.13 in the active treatment group, compared with 49.83 in the placebo group; this difference was not significant (p = 0.287), nor did this change after adjusting for baseline (p = 0.668).

The K-BILD questionnaire had average total scores of 51.49 in the active group and 50.77 in the placebo group; this difference was not significant (p = 0.805), nor did this change after adjusting for baseline (p = 0.485). None of the scores for components of the K-BILD questionnaire was significant.

Clinical measurement outcomes at 12 months

Modified per-protocol results

The lung function data were available for approximately 80 individuals at 12 months post randomisation, with more data available in the placebo group than in the active treatment group. The results are shown in Appendix 1, Table 29.

The absolute FVC had a mean value of 2.23 l in the active treatment group, compared with 2.26 l in the placebo group. This difference was not significant in either the unadjusted (p = 0.83) or the adjusted (p = 0.30) analysis. The per cent predicted FVC was 52.49% in the active treatment group and 53.98% in the placebo group; this was not significant in either the unadjusted (p = 0.47) or the adjusted (p = 0.39) analysis.

The absolute FEV₁ had a mean value of 1.84 l in the active treatment group, compared with 1.89 l in the placebo group. This difference was not significant in either the unadjusted (p = 0.58) or the adjusted (p = 0.70) analysis. The per cent predicted FEV₁ was 56.16% in the active treatment group and 59.20% in the placebo group; this was not significant in either the unadjusted (p = 0.18) or the adjusted (p = 0.29) analysis.

The absolute DLCO had a mean value of 3.71 mmol/minute/kPa in the active treatment group, compared with 3.69 mmol/minute/kPa in the placebo group. This difference was not significant in either the unadjusted (p = 0.99) or the adjusted (p = 0.37) analysis. The per cent predicted DLCO was 41.44% in the active treatment group and 42.22% in the placebo group; this was not significant in either the unadjusted (p = 0.77) or the adjusted (p = 0.21) analysis.

Repeated measures analysis

Ancillary analysis

Imputed results

The imputed results are shown in Table 20. These results show no statistically significant difference in any of the comparisons.

Biomarker results

Intention-to-treat analysis

The biomarker data were available for 157 patients at baseline. The data are presented in Appendix 1, Table 38; the data were heavily skewed, with extreme values, so a non-parametric testing approach was taken. Overall, the groups were well balanced given the high variability of the measures.

At outcome, differences were observed in CRP levels (p = 0.016), with a small difference in median, but, as seen in Figure 13, the difference was mainly due to the few extreme values in the active treatment group. A significant difference (p = 0.019) was also seen in CA19-9 levels; again, a small difference in median values, but the difference was due to extreme values in the active treatment group, as shown in Figure 14. There was also a difference in the change from baseline in SP-D levels (p < 0.001), as shown in Figure 15; the difference was due to the greater spread of change in placebo group. A significant difference in the change from baseline in CRP levels (p = 0.005) was also observed and was due to some large changes in a few individuals in the active treatment group – this is shown in Figure 16. A significant difference in the change from baseline in CA-125 levels (p = 0.032) was also observed and was due to some large changes in a few individuals in the active treatment group, as shown in Figure 17.

The results are given in Appendix 1, Table 39.

FIGURE 13.

Histogram of CRP levels at 12 months in the ITT sample. (a) Active treatment; and (b) placebo.

FIGURE 14.

Histogram of CA19-9 levels at 12 months in the ITT sample. (a) Active treatment; and (b) placebo.

FIGURE 15.

Histogram of change in SP-D levels from baseline to 12 months in the ITT sample. (a) Active treatment; and (b) placebo.

FIGURE 16.

Histogram of change in CRP levels from baseline to 12 months in the ITT sample. (a) Active treatment; and (b) placebo.

FIGURE 17.

Histogram of change in CA-125 levels from baseline to 12 months in the ITT sample. (a) Active treatment; and (b) placebo.

Adverse events and serious adverse events

A total of 1336 AEs occurred during the follow-up period: 696 in the active treatment group and 640 in the placebo group. This was roughly equal in both treatment groups. The AEs occurred in 288 individuals: 146 (in 86.4% of individuals randomised) in the active treatment group and 142 (in 82.6% of individuals randomised) in the placebo group. There were 37 SAEs: 20 in the active treatment and 17 in the placebo groups. There were 16 (9.5%) and 12 (6.8%) patients with one or more SAE in the active treatment and placebo groups, respectively.

The classification of AEs is given in Table 21 and differences were seen in three categories, with rates of general disorders, investigations, and metabolism and nutrition disorders all higher in the treatment group than in the placebo group.

Blood measures

The summary statistics of the safety blood measures collected at various time points are given in Table 22. Further data on safety blood measures are given in Appendix 1, Tables 44–50. Consistent significant differences were observed at 6 weeks and 3 and 6 months in haemoglobin levels, with mean values approximately 6 g/dl higher in the placebo group than in the active treatment group; red cell count (RCC), with mean values approximately 0.2 × 10¹²/l higher in the placebo group than in the active treatment group; haematocrit, with mean values approximately 0.01% higher in the placebo group than in the active treatment group; sodium levels, with mean values approximately 2 mmol/l higher in the placebo group than in the active treatment group; creatinine levels, with values approximately 12 µmol/l higher in the active treatment group than in the placebo group; bilirubin levels, with mean values approximately 1.5 µmol/l higher in the placebo group; and alkaline phosphatase levels, with mean values approximately 7 IU/l higher in the active treatment group. At 12 months, there was a significant difference in creatinine levels only. Beyond 12 months, the number of patients is limited and no significant differences are observed.

Microbiology

We obtained 17 sputum samples and one nasal swab in total for all patient visits. Of these, only three grew a relevant microbiological agent on culture: Staphylococcus aureus (n = 1), Haemophilus influenzae (n = 1) and yeasts (n = 1).

Co-trimoxazole

To our knowledge, there have been only two previous trials of prophylactic therapy with co-trimoxazole in people with ILD. 20,21 In a trial of 20 patients with advanced idiopathic fibrotic lung disease, Varney et al. 20 showed an increase in FVC from a median of 1.9 to 2.3 l (95% CI 1.3 to 3.0 l; p = 0.05) and an increase in the shuttle walk test distance from 255 to 355 m (95% CI 200 to 450 m; p = 0.002) after 3 months of treatment. Seven patients had a usual interstitial pneumonia pattern on HRCT and, therefore, would be classified as having IPF; the rest had a HRCT pattern in keeping with a combination of UIP or non-specific interstitial pneumonia, or had unclassifiable fibrotic ILD. Of note is the fact that seven participants in the active treatment group and four in the placebo group were receiving prednisolone at a median dose of 10 mg per day.

The subsequent two trials found that there was no change in FVC measurements with 12 months of co-trimoxazole therapy (from 2.3 to 2.18 l in the TIPAC trial21 and from 2.3 to 2.26 l in the EME-TIPAC trial) and no difference in the change for the placebo and co-trimoxazole treatment arms [0.00 l (95% CI –0.11 to 0.11 l) and –0.01 l (95% CI –0.09 to 0.07 l), respectively]. The lack of benefit identified in the two larger studies is not likely to be caused by the increased variability of measurement as part of a multicentre study. The values were mostly taken from clinical data, as the trials were designed to collect data obtained as part of routine care and, therefore, will have been undertaken to a clinical standard by qualified pulmonary function technologists.

Given the remarkable improvement in these physiological measurements with co-trimoxazole in the initial study, it is possible that the patients were not stable at baseline and co-trimoxazole was treating an unrecognised bacterial lower respiratory tract infection. This is a plausible explanation because clinical stability was not required as part of the entry criteria for the study of Varney et al. 20 and the beneficial effects of FVC were seen after 3 months of treatment, with no further improvement after pulmonary rehabilitation or 1 year of open-label treatment. Interestingly, the shuttle walk test distance did not improve after pulmonary rehabilitation in either the placebo or active treatment groups, which is in contrast to recent systematic review data of the effects of pulmonary rehabilitation in people with IPF. 96

The second trial exhibited a benefit in terms of quality of life and, in the case of those adhering to treatment, all-cause mortality. 21 The symptoms domain score of the SGRQ was reduced in those receiving co-trimoxazole, but increased in the group receiving the placebo, indicating worse quality of life in this group. The difference between the two treatment groups was 6.88 (95% CI 1.7 to 12.06) units when the results were adjusted for baseline, but there was no difference in other domains of the SGRQ or the total SGRQ score. The 6.88-unit difference is larger than the minimum important difference for the SGRQ. 97 The symptom domain of the SGRQ is derived from the first eight questions of this tool, which cover issues relating to cough, sputum, breathlessness, wheeze and attacks of breathlessness. 98 The improvement in these symptoms may reflect the changes seen in the current study in terms of CSS, LCQ and the chest symptom domain of the K-BILD questionnaire in the PP analysis. The symptom domain of the K-BILD questionnaire asks about chest tightness, air hunger and wheeze, and, therefore, is similar to the symptom domain of the SGRQ. This study indicated a significant difference in CSS, a visual analogue scale rating patients’ overall cough severity, after 18 months of treatment, and a significant overall effect on CSS over the duration of the study.

Cough is an important problem for people with IPF. 99 Patients report cough to be particularly bothersome, and cough significantly contributes to the burden of disease. 100 It is an important problem in end-of-life care, but is also present at an early stage of the disease, with one-third of people with IPF having at least one consultation for cough in the year before diagnosis. 101 Unfortunately, it is difficult to manage cough in people with IPF as there are no recognised treatments. 100 The suggestion of improvement in cough quality of life and cough-related symptoms in two separate clinical trials (this trial and the TIPAC trial21) requires further evaluation given the lack of available treatment options.

In the analysis confined to those patients adhering to the treatment and protocol, there was a reduction in deaths and an improvement in QALYs in the TIPAC trial. 21 The reduction in QALYs is likely to reflect mortality, at least in part, because death represents a QALY of zero. Unfortunately, and in contradiction to our hypothesis, we were not able to identify a significant treatment effect, in terms of our composite score of non-elective hospitalisation, all-cause death or transplant rates with co-trimoxazole in people with IPF in the current study.

The discordance between the two results can largely be explained by the change in standard practice between the first and second studies, that is, the discontinuation of immunosuppression and the introduction of antifibrotic therapy. This resulted in a much better survival than in the previous study. 20 In the TIPAC trial,21 the median hospital-free survival in the placebo group was 12.8 months, whereas in the EME-TIPAC trial the median hospital-free survival was 23.3 months.

When the first study was undertaken (i.e. January 2008–December 2010), standard treatment for IPF was immunosuppression with prednisolone, with or without azathioprine or mycophenolate and N-acetylcysteine. However, the results of the Raghu et al. 9 study became available in 2012 and indicated that immunosuppression resulted in higher mortality and hospitalisation rates than no immunosuppression; these findings were considered to be mostly due to respiratory issues, presumably caused by infection and changes in prescribing practices.

In the TIPAC trial,21 60% and 57% of individuals were receiving prednisolone in the placebo and active treatment groups, respectively, with approximately 30% receiving azathioprine and 4% receiving mycophenolate in each group. The majority of those receiving prednisolone were taking a moderate to high dose, with only 13% (placebo group) and 19% (active treatment group) taking < 10 mg per day. This contrasts with the current study, which excluded all those receiving > 10 mg of prednisolone (as part of the entry criteria), and in which only 6% of individuals were taking a low dose (< 10 g/day) of prednisolone.

A systematic review of participants randomised to the placebo group of eight multicentre large-scale randomised controlled trials, totalling 1631 patients with 2067 patient-years’ follow-up, reported that prednisolone therapy results in a 31% increase in all-cause mortality. 25 In a separate analysis,56 two trials totalling 1156 individuals suggested that low-dose prednisolone increased mortality by 54% and a high dose increased mortality more than twofold compared with no prednisolone. To our knowledge, there are no reliable data to evaluate the increase in hospitalisation rates with prednisolone in IPF, but prednisolone had a greater effect on mortality than hospitalisation rates in the TIPAC trial. 21 Prednisolone has been known to result in a 39% increase in pneumonia and a 3.6-fold increase in lower respiratory tract infections. 54 Importantly, prednisolone therapy, particularly at higher doses, is frequently prescribed to patients with severe disease and, therefore, any effect of prednisolone on outcomes from observational data is likely to be an overestimation because of confounding biases. It is clear, therefore, that prednisolone and other immunotherapy increase mortality in people with IPF, possibly because of the increased incidence of respiratory tract infections. Undiagnosed or unrecognised respiratory tract infection may have been prevented or treated with co-trimoxazole in the first study, but not in the current study, resulting in no difference in mortality.

The other major change in prescribing practice was the introduction of antifibrotic therapy. The American Thoracic Society/European Respiratory Society 2011 guidelines8 expressed some caution about using pirfenidone; however, the relevant national boardsfinalised approval of the decision late in 2010, before the results of the CAPACITY study14 were published in May 2011, indicating a reduction in FVC with pirfenidone compared with placebo. Pirfenidone was licensed by the European Medicines Agency on 28 February 2011 for the treatment of adults with mild to moderate IPF and the NICE IPF pirfenidone Technology Appraisal number 282 (TA282) was published in April 2013. 102 The INPULSIS study was published in May 2014103 and the nintedanib Technology Appraisal number 379 (TA379) was published in January 2016. 104 Recruitment into the current trial began in April 2015. Despite the narrow window of eligibility of antifibrotic therapy in the UK (FVC between 50% and 80% predicted), a large proportion of people in the EME-TIPAC trial were eligible for this form of treatment because our entry criteria excluded those with mild disease. Indeed, 75% of people were taking antifibrotic therapy: 137 (40%) patients were taking pirfenidone and 116 (34%) were taking nintedanib.

Although data from randomised controlled trials have not shown a survival benefit for either pirfenidone or nintedanib, there is sufficient evidence to suggest that antifibrotic therapy prolongs life. FVC is considered a good surrogate marker for survival and, therefore, it is reasonable to expect that the improvement in FVC decline with these drugs equates to a survival benefit. Using combined data from the CAPACITY14 and ASCEND74 studies, the relative risk for all-cause mortality in the active treatment groups was half that in the placebo groups (HR 0.52, 95% CI 0.31 to 0.87; p = 0.0107) at 1 year, with maintenance of improved mortality at 120 weeks. 17 In another similar combined analysis using the CAPACITY14 data, an open-label extension study (RECAP105) and people meeting the inclusion criteria for these studies from the Inova Fairfax Hospital database, a survival analysis has shown life expectancy to be 8.72 years (95% CI 7.65 to 10.15 years) with pirfenidone and 6.24 years (95% CI 5.38 to 7.18 years) with standard care. 106 Pirfenidone was estimated to improve life expectancy by 2.47 years (95% CI 1.26 to 4.17 years), which equated to 25% of the expected years of life lost due to IPF. 106 Likewise, a combined post hoc analysis from trials of nintedanib also showed a significant reduction in mortality. It is possible, therefore, that the effects of these drugs will mask any survival benefit from co-trimoxazole. 107

Another difference between the TIPAC trial21 and EME-TIPAC trial is the severity of illness of the patients enrolled; the EME-TIPAC trial recruited people with FVC < 75% predicted. It is known that patients with IPF have greater bacterial burden than healthy individuals,31 but it is not known whether or not disease severity itself influences the microbiological flora. It is known that the treatment response to pirfenidone is similar for people with advanced disease (FVC of < 50%) and those with mild disease. 14,108 Patients not receiving prednisolone in the TIPAC trial21 were recruited only if the PI felt that their disease was declining based on evidence of a reduction in lung function. As a reduction in FVC has been shown to relate to reduced survival,109 this may have also contributed to the higher mortality rates in the TIPAC trial. 21 We do not believe that co-trimoxazole is likely to have been more effective in people with mild disease and that restricting the entry criteria to those with reduced FVC values is responsible for the negative findings of the study; those with more severe disease were likely to benefit from co-trimoxazole, as suggested by the results of a post hoc analysis of the TIPAC trial. 21

In the EME-TIPAC trial, there was an option for people with ARs to high-dose (960 mg twice daily) co-trimoxazole to reduce the dose (to 960 mg three times weekly), whereas, in the TIPAC trial,21 this was not permitted. However, fewer people stepped down to the lower dose (18%) in the EME-TIPAC trial than withdrew from treatment (40%) in the TIPAC trial. 21

In a retrospective review of the Japanese Diagnosis Procedure Combination database,23 293 people with IPF who had an acute exacerbation of IPF, received invasive mechanical ventilation for respiratory failure and were treated with high- (at least 2.88 g/day) or low-dose (480 mg to 2.4 g/day) co-trimoxazole were compared with those who received no treatment. A significant dose-related improvement was seen with co-trimoxazole in terms of survival, with high-dose treatment patients surviving > 100 days and low-dose treatment or no co-trimoxazole patients surviving < 50 days. Like the TIPAC trial,21 but unlike the EME-TIPAC trial, the benefit was seen in people receiving immunosuppression, as all patients were receiving high-dose corticosteroid therapy. This adds weight to the suggestion that co-trimoxazole has a role for people who are immunosuppressed and to the rationale for the discrepancy between the two TIPAC studies. 21

Other antibiotics

Doxycycline may have effects on matrix metalloproteinases110 and has been evaluated in an open-label study that found no physiological improvement, but an improvement in chest X-ray involvement;111 however, there are no data on this drug from placebo-controlled trials. The effects of doxycycline are unknown, but are currently being examined in the Clean-Up IPF study (https://clinicaltrials.gov/ct2/show/NCT02759120; last accessed 20 January 2021). This is a prospective observational study of either co-trimoxazole or doxycycline therapy in over 500 people with IPF over a 12- to 36-month follow-up period, capturing the same outcomes as the EME-TIPAC trial.

Azithromycin attenuates myofibroblast differentiation112 and had beneficial effects in a bleomycin mouse model of pulmonary fibrosis. 113 To our knowledge, there have been no randomised controlled trials with macrolides, but two retrospective studies114,115 have investigated the role of azithromycin in IPF. One was a retrospective review of low-dose macrolide (250 mg three times per week) given to people who had three or more lower respiratory tract infections or courses of antibiotics over the preceding 12 months. The primary end point was hospitalisations, which reduced from 0.29 per patient per year in the 12 months before treatment to 0.08 per patient per year in the 12 months with macrolide treatment. Recurrent chest infections was not an entry criterion for the EME-TIPAC trial, and it is not known if co-trimoxazole had any beneficial effect on this phenotype. Like the TIPAC trial21 and the EME-TIAPC trial, macrolide therapy did not have any effect on the rate of decline of lung function as measured by FVC or DLCO. 114 In another (single-centre) study,115 mortality reduced after the treatment regime changed from quinolones to macrolides for acute exacerbations of IPF. However, as this was not controlled for time of entry into the study, there are likely to be other confounding factors and the findings may be open to question.

Adherence to medication

We showed that the overall adherence was 81% and 86% in the active treatment and placebo groups, respectively, with 72% of patients in both groups complying with treatment by > 80%. In the TIPAC trial,21 overall adherence to the study medication was better, with 96% of patients in the active treatment group and 90% in the placebo group receiving > 80% of the scheduled study drug doses; however, we do not believe that the difference in adherence is responsible for the difference in the findings of the study.

Implications for clinical practice

Our results suggest that the prophylactic use of co-trimoxazole for the treatment of IPF ought not to be recommended. This study does not make any recommendation regarding the use of co-trimoxazole or other antibiotics in the situation of acute exacerbations or concomitant respiratory infections in IPF.

Recommendations for research

We found a consistent beneficial effect of co-trimoxazole on different measures of cough (CSS, cough quality of life and symptoms of disease-related quality of life) in different analyses. Although it is possible that this is a chance finding, we found similar benefits in the previous TIPAC trial,21 suggesting that these effects are real. A further study to evaluate the effect of co-trimoxazole in terms of cough should be considered.

We examined the effects of sensitivity analysis in terms of adherence to treatment and treatment regime. However, our SAP did not provide provision to undertake exploratory analyses to identify whether or not there are groups of individuals who may benefit from co-trimoxazole (e.g., those with recurrent chest infections, significant traction bronchiectasis or high bacterial burden). Additional studies of antibiotic therapy in those who may benefit most may be warranted.

Although this study rules out a role for co-trimoxazole in unselected individuals with moderate to severe IPF, and it is unlikely that other broad-spectrum antibiotics will be beneficial, we cannot exclude the possibility that other therapies that alter the lung microbiota will improve outcomes in IPF. Other studies of antibiotics, possibly with a more targeted approach, should be considered.