Milan.

Methods	Study design: randomised controlled trial Country: Italy Study period: 1998 to 1999 Inclusion criteria: women aged 40‐75 years with invasive primary breast cancer ≤ 2 cm, treated with breast‐conserving surgery Exclusion criteria: history of other cancer (except non‐melanoma skin cancer), multi‐centric breast cancer and previous excisional biopsy Length of follow‐up ( median and range): 102 months (1‐120 months)
Participants	No. in trial arm: ALND: N = 257; SLNB: N = 259 Age: ALND: median (range) = 56 (40‐75) years; SLNB: median (range) = 55 (40‐75) years Stage distribution: not reported Proportion node positive: ALND: 83/259; SLNB: 92/259 Histological type of breast cancer: ALND: ductal infiltrating, N = 212; lobular infiltrating, N = 20; other, N = 25. SLNB: ductal infiltrating, N = 209; lobular infiltrating, N = 18; other, N = 32
Interventions	Sentinel lymph node biopsy (SLNB) plus axillary lymph node dissection (ALND) vs SLNB followed by ALND only if metastases were found in the SLN. Both groups also received breast‐conserving surgery.
Outcomes	Overall survival, breast cancer‐related events (axillary metastases, supraclavicular metastases, intrabreast tumour reappearance, distant metastases), contralateral breast cancer, axillary pain, numbness or paraesthesia on operated side, arm mobility, aesthetic appearance of axillary scar, arm swelling (difference between circumference of treated and untreated arms)
Axillary node surgery	Minimum no. nodes to be removed according to protocol: not reported, but at least 1 sentinel node should have been removed Nodes removed ALND arm: 429 SLN from 257 participants (mean = 1.66 SLN/participant; mean non‐sentinel lymph nodes/participant = 24) Nodes removed SLNB arm: 424 SLN from 259 participants (mean = 1.63 SLN/participant; mean non‐sentinel lymph nodes/participant = 24) Method of node pathological analysis: Each sentinel node was bisected along major axis, embedded in optimal‐cutting‐temperature compound, then frozen in isopentane cooled with liquid nitrogen (SLNs < 5 mm diameter were embedded and frozen whole). 15 pairs of 4 µm thick sections were cut at 50 µm intervals, from each half node (60 sections/node). Any remaining tissue was sectioned at 100 µm intervals. If more than 1 sentinel node was found, all were analysed in this way. One section of each pair was hematoxylin and eosin stained; if this was ambiguous, the other section of the pair was stained for cytokeratins. Further treatment for node‐positive cases: yes (ALND)
Radiotherapy	RT ALND arm: 50 Gy to ipsilateral breast over 8 weeks, with 10 Gy boost to skin surrounding the surgical scar RT SLND arm: 50 Gy to ipsilateral breast over 8 weeks, with 10 Gy boost to skin surrounding the surgical scar RT same in all trial arms?yes
Hormone and chemotherapy	ALND: hormonal therapy: N = 133; chemotherapy: N = 21; both hormonal and chemotherapy: N = 99; neither: N = 4   ALND: Hormonal therapy: N = 126; chemotherapy: N = 16; both hormonal and chemotherapy: N = 106; neither: N = 11   Significantly more women in ALND arm had chemotherapy than in SLNB arm, but rates of hormone therapy ‐ both hormone and chemotherapy and no hormone or chemotherapy ‐ did not differ between groups.
Notes	Baseline differences? Groups appear comparable. Intention‐to‐treat analyses? Survival, disease control in axilla and breast cancer recurrence: per‐protocol analysis employed, but few protocol violations (7/264 ALDN participants and 9/268 SLNB participants were excluded from analyses). Long‐term adverse events: no intention‐to‐treat analyses undertaken. Only women with negative sentinel nodes (who did not go on to have ALND) were included in the SLND group for long‐term adverse events analysis.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated permuted blocks
Allocation concealment (selection bias)	Low risk	Randomised after resection of tumour. Data centre telephoned surgeon with treatment group information.
Blinding of outcome assessment (detection bias) Disease control in the axilla	Unclear risk	No information was provided.
Blinding of outcome assessment (detection bias) Breast cancer recurrence	Unclear risk	No information was provided.
Blinding of outcome assessment (detection bias) Short term adverse events	Unclear risk	Outcome was not reported.
Blinding of outcome assessment (detection bias) Long term adverse events	Unclear risk	No information was provided.
Incomplete outcome data (attrition bias) Survival	Low risk	Participants are accounted for at 10‐year follow‐up (Veronesi 2010).
Incomplete outcome data (attrition bias) Axillary recurrence	Low risk	Participants are accounted for at 10‐year follow‐up (Veronesi 2010).
Incomplete outcome data (attrition bias) Breast cancer recurrence	Low risk	Participants are accounted for at 10‐year follow‐up (Veronesi 2010).
Incomplete outcome data (attrition bias) Short term adverse events	Unclear risk	Outcome was not reported.
Incomplete outcome data (attrition bias) Long term adverse events	High risk	Only a sample of 100 women from each group was included in this analysis. The SLND group sample was biased – see below.
Selective reporting (reporting bias)	Unclear risk	Short‐term adverse events were not reported.