Thomas 2012.

Methods	A multicentre, randomised, double‐blind, placebo‐controlled trial
Participants	1. Setting: the trial recruited participants from 20 hospitals in the UK and Ireland, either in hospital setting or retrospectively via medical coding in records or poster adverts 2. Number of participants randomised: 123 3. Sex (men/women): 42/81 4. Mean age (range): treatment group‐ 56.7 (18 ‐ 81), placebo group‐ 59.5 (23 ‐ 84) 5. Area of body involved: leg 6. Number of episodes of cellulitis prior to intervention: ≥ 1
Interventions	Study groups: Oral penicillin V (phenoxymethylpenicillin) 250 mg X 2/d Oral Placebo tablets 250 mg X 2/d Duration of treatment: 6 months Follow‐up: on‐prophylaxis phase‐ telephone calls from co‐ordinating centre every 3 months , post‐prophylaxis phase‐ phone calls at 6‐month intervals. In addition participants were asked to complete a study diary and to immediately inform the trial staff of a repeat episode
Outcomes	Primary outcomes: 1. Time to next episode of cellulitis Secondary outcomes: 1. The proportion of participants with repeat episodes of cellulitis at the end of the treatment phase, and at the end of the non‐intervention follow‐up phase 2. The number of repeat episodes of cellulitis 3. The proportion of participants with new oedema and/or ulceration at the end of the treatment phase, and at the end of the non‐intervention follow‐up phase 4. The number of nights in hospital for the treatment of repeat episodes of cellulitis 5. The number of adverse drug reactions and/or adverse events of interest (death, nausea, diarrhoea, thrush, rash) 6. Cost effectiveness 7. Predictors of response model to explore the impact of known risk factors in predicting the efficacy of prophylaxis (stated to be conducted only if a significant treatment effect was found)
Notes	1. Criteria for diagnosis of cellulitis: ‐ Criteria for an episode of cellulitis for study eligibility (index episode): a confirmed diagnosis of cellulitis by the recruiting dermatologist; if the patient was not seen by the recruiting clinician, validation of diagnosis was sought from the medical records and interview with the patient. In this case specific criteria were required consistent with clinical signs and symptoms of cellulitis (specified in the report). Any doubt over the certainty of the diagnosis was grounds for exclusion ‐ Criteria for a repeat episode of cellulitis during treatment or follow‐up phases: next episode of cellulitis in either leg that had been reported by the participant and confirmed by a medical practitioner (sensitivity analysis was performed for patient‐reported cases that required antibiotic treatment and were not confirmed as stated) 2. The paper reported good and balanced adherence in both groups: "From self‐reported tablet counts, 97 (79%) patients fully adhered to treatment, defined as at least 75% of tablets taken.This was similar across treatment groups". Funding source and Declaration of interest: NC (The BUPA Health Foundation) and no conflicts of interests. The Thomas 2012 and Thomas 2013 report on 2 studies that were led by the same group of researchers under the title of: Prophylactic Antibiotics for the Treatment of Cellulitis at Home ‐ The PATCH trials (The PATCH II and PATCH I studies, respectively). Similar study designs were reported for both trials including randomisation process, allocation, blinding, definitions of outcomes, monitoring and analysis.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote: "participants were randomised by staff at the coordinating centre using a web‐based randomisation service provided by the Clinical Trials Unit (CTU)...". Comment: investigators used computer‐generated random list
Allocation concealment (selection bias)	Low risk	Quote: "Treatment allocations were concealed from all members of the trial team". Comments: central block randomisation was conducted with varying block sizes
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Quote: "Participants and all members of the study team were blinded to treatment allocation throughout the trial, and analysis was performed prior to breaking of the randomization code...Although the treatments were packaged in an identical way, and the placebo tablets were of the same size and shape as penicillin V, the tablets were not identical due to difficulties in obtaining a matched placebo product. Nevertheless, there was a low risk of unblinding...". Comment: The trial included placebo, randomisation list was held by the CTU, breaking of the allocation code was allowed according to decisions by the data monitoring committee and as prespecified in the protocol of the trial
Blinding of outcome assessment (detection bias) All outcomes	Low risk	As detailed for blinding of the participants and personnel: there was blinding, and it was unlikely that the blinding could have been broken
Incomplete outcome data (attrition bias) All outcomes	Low risk	Quote: "Of the 123 participants randomized, 20 (16%) participants (11 penicillin V and nine placebo) did not reach the end of the study... Participants in both groups had a similar study time experience, and approximately 80% had at least 2 years of follow‐up". Comment: most of participants completed follow‐up (84%); their number is balanced between groups; the reasons for withdrawal from study are probably not related to treatment or outcome;and ITT was performed
Selective reporting (reporting bias)	Low risk	Changes to outcomes, as prespecified in the protocol, were explained. Other outcomes were reported as mentioned in the protocol, that had been registered and available online (via ongoing trials registries and the trial website ‐Thomas 2012)
Similarity of groups at baseline (baseline imbalance bias)	Low risk	Quote:"...these stratification factors and other baseline variables were broadly similar across the two treatment groups". Comment: Baseline characteristics were reported and balanced
Early termination (early stopping bias)	Low risk	Quote:"...the identification of suitable participants was much slower than anticipated, and recruitment was therefore stopped after 2 years due to funding limitations. The possible reasons for this failure to recruit have been reported elsewhere" (Thomas 2010). Comment: Sample size calculation was stated (a sample of 400 participants) and the goal of recruitment was not achieved but carefully examined. Nevertheless, the study ended according to the protocol (see Thomas 2012)
Other bias	Low risk	No other sources of potential bias were found