TORCH 2007.

Methods	Design: a multi‐centre, randomised, double‐blind, placebo‐controlled, parallel‐group study. Trial duration was 2 week run‐in period and 156 weeks treatment. The trial included 444 centres in 42 countries in North America, Central America and Asia Pacific
Participants	Participants: 3093 patients were randomised (salmeterol: 1542; fluticasone 1551) Baseline characteristics: mean age 65 years; 76% male; mean FEV1 predicted 44%, mean predose FEV1 1.1 L, mean SGRQ score 50 Inclusion criteria: male/female 40‐80 years of age; diagnosis of COPD (ERS); < 10% reversibility of predicted FEV1; FEV1/FVC ratio < 70%; FEV1< 60% predicted; ≥ 10 pack year smoking history Exclusion criteria: Asthma or respiratory diseases other than COPD; lung volume reduction surgery/lung transplant; requirement for > 12 h/day long term oxygen therapy; long term oral corticosteroid therapy; serious uncontrolled disease likely to interfere with medication/cause death in next three years
Interventions	Run‐in: 2 weeks. All maintenance treatment with ICS and LABA ceased, but patients could continue other medications for COPD 1. salmeterol 100 µg per day: 50 µg twice daily 2. fluticasone 1000 µg per day: 500 µg twice daily Inhaler device: DPI (Diskus) Co‐treatment: Patients could continue medications for COPD other than corticosteroids and inhaled long‐acting bronchodilators
Outcomes	All cause mortality; frequency of exacerbations; health status (SGRQ); change in post‐dose FEV1 from baseline to end of study; adverse events
Notes	Exacerbation defined as symptomatic deterioration requiring treatment with antibiotics, systematic corticosteroids, hospitalisation or a combination of these
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Eligible patients were randomly assigned to study treatment in accordance with the randomisation schedule, which was generated using the GW computer program Patient Allocation for Clinical Trials (PACT). Patients were randomised in permuted blocks with stratification according to country and smoking status.
Allocation concealment (selection bias)	Low risk	Medication was allocated by the use of three numbers as follows. • Each subject who was screened was allocated a subject number. This number was unique to each subject and was assigned from a list provided to the site, in chronological order. • Each subject who satisfied the randomisation criteria was assigned a unique treatment number from the Interactive Voice Response (IVR) system which is part of the System for Central Allocation of Drug (SCAD). Once a treatment number had been assigned to a subject, it could not be assigned to any other subject. Neither the subject nor the investigator knew to which treatment arm a subject had been allocated. • At each treatment visit the subject was provided with a treatment pack. Every pack number was unique and corresponded to the study medication pack which was dispensed to the subject at the visit A specialist IVR system company, ClinPhone, managed this system. At the randomisation visit (Visit 2) the principal investigator or designee contacted the IVR system through an automated 24‐hour telephone number; upon providing their unique personal identification number (PIN) and answering a series of questions, the site was provided with the subject’s treatment number as well as a pack number
Blinding (performance bias and detection bias) All outcomes	Low risk	Neither the subject nor the investigator knew to which treatment arm a subject had been allocated. At each treatment visit each subject was issued with a treatment pack containing DISKUS/ACCUHALER inhalers. The inhalers contained either of the four treatments (salmeterol/fluticasone propionate combination product, fluticasone propionate, salmeterol, or placebo) in accordance with the randomisation schedule. The inhalers were labelled in accordance with all applicable regulatory requirements. Each treatment pack and study treatment inhaler was labelled with the protocol number, storage and dosing instructions by GW Research and Development
Incomplete outcome data (attrition bias) All outcomes	Low risk	For patients who withdrew from the study prematurely, all data on exacerbations, health status, and lung function available at the time of a patient’s withdrawal from the study were included in the analysis. All efficacy analyses were performed according to the intention‐to‐treat principle
Selective reporting (reporting bias)	Low risk	All collected data reported