Puhalla 2008.

Methods	Accrual: October 2004 to June 2006 Multicentre, phase II randomised controlled trial conducted in the USA Adjuvant therapy Median follow‐up: not reported
Participants	Age: median 51 years, range 33–75 years 55% postmenopausal Node‐positive non‐metastatic breast cancer Stage II: 72% Median number of positive lymph nodes: 2 (range 1–35) Hormone receptor‐positive: 66% HER2‐positive: 4% (arm 1); 0% (arm B) Excluded: prior chemotherapy or hormone therapy as neoadjuvant or adjuvant therapy
Interventions	Arm 1 ('arm A' in the trial publication): docetaxel (75 mg/m²) every 2 weeks for 4 cycles followed by doxorubicin (60 mg/m²), cyclophosphamide (600 mg/m²) every 2 weeks for 4 cycles Arm 2 ('arm B' in the trial publication): doxorubicin (60 mg/m²), cyclophosphamide (600 mg/m²) every 2 weeks for 4 cycles followed by docetaxel (75 mg/m²) every 2 weeks for 4 cycles For arm 1 and 2: pegfilgrastim (6 mg) on day 2 after each chemotherapy cycle
Outcomes	Primary outcomes Completion of 4 cycles of docetaxel without dose reductions or delays within 10 weeks RDI defined as total cumulative drug doses the participant actually received per time divided by the protocol‐intended cumulative doses per time Secondary outcomes Toxicity, assessed using the NCI CTC version 2 Quality of life
Notes	Clinical trial record: NCT00201708 Funding considerations: supported by research grant from Sanofi‐Aventis; data management support by Bridge Site Clinical Research, Columbus
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote: "A 1:1 fixed block randomisation with a block size of 9 was used. After eligibility was confirmed at the coordinating centre randomisation occurred." Comment: baseline characteristics were similar although there were more stage III participants in the comparator arm.
Allocation concealment (selection bias)	Low risk	Quote: "at the coordinating centre, randomisation occurred and the group assignment was then sent to the outside institution."
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Open‐label study. Participants in both groups received the same drugs though in different order. Therefore, it was judged to be unlikely that this would lead to a material bias in participants' and physicians' behaviours even when unblinded.
Blinding of outcome assessment (detection bias) Toxicity and treatment adherence	Unclear risk	Study used formalised toxicity criteria (NCI CTC version 2) and measured a range of toxicity outcomes, some of which in borderline cases may be affected by unblinding (e.g. neuropathy). Dose‐reductions/delays were based on toxicity assessments. Overall, unblinding may have influenced the physicians' assessments for a subset of outcomes.
Blinding of outcome assessment (detection bias) Quality of life	Low risk	FACT‐B version 4 performed on a small subset of 20 randomly selected participants at 3 time points. Collected for feasibility and not prespecified. Participants were not blinded to treatment allocation.
Incomplete outcome data (attrition bias) All outcomes	Low risk	In the trial publication, there was a mismatch between number of participants in each group in the CONSORT diagram Figure 1 (26 participants in each arm) and Table 1 (28 patients in each arm), with 2 patients unaccounted for. Contacted trial author and received reply confirming that there was an error in the CONSORT diagram and that there were 28 participants in each arm.
Selective reporting (reporting bias)	Low risk	Trial record identified (NCT00201708). Outcomes reported (except for 1 outcome that was related to which regimen to take forward for phase III trial). 2 other outcomes were added in the trial publication that were not in trial registration record – RDI and quality of life but these are considered important outcomes to report.
Other bias	Low risk	None identified