Axillary surgery in women with sentinel node-positive operable breast cancer: a systematic review with meta-analyses

The search identified 7436 unique records, of which 7273 were excluded based on the
title and abstract while the full publications of 163 potentially relevant studies
were examined. Of these, 5 trials reported in 13 publications met the inclusion criteria,
two studies were still ongoing (comparing ALND to SLNB [NCT01796444 (Wang 2013]), and ALND or axillary radiotherapy [aRT] + adjuvant treatment versus adjuvant treatment
alone [POSNOC (Goyal 2014a], b])], respectively) while the remaining 147 records were excluded because they were:
not a randomised trial (n = 20), ineligible population (n = 101), unclear intervention
(n = 2) and ineligible intervention (n = 24); See also Additional file 2). The five included studies compared ALND with sentinel lymph node dissection (SLND)
to SLND alone [ACOSOG Z0011 (Lucci et al. 2007]; Olsen and McCall 2008]; Giuliano et al. 2010], 2011]); ATTRM-048-13-2000 (Sola et al. 2013]); IBCSG-23-01 (Galimberti et al. 2011], 2012], 2013])], and ALND to aRT [AMAROS (Straver et al. 2010a], b]; Donker et al. 2014]); OTOASOR (Savolt et al. 2013a], b])]. See Tables 1 and 2 for summary study details and risk-of-bias levels, respectively, and Additional file
3 for full study details and risk-of-bias assessments.

Table 1. Characteristics of the included studies

Table 2. Risk of bias in the included studies

ALND with SLND versus SLND

Figure 1 shows that neither overall survival (summary HR = 0.82, 95 % CI 0.58–1.15; p = 0.25;
I
²
 = 0 %) nor disease-free survival (summary HR = 0.81, 95 % CI 0.63–1.04; p = 0.1;
I
²
 = 0 %) differed between the SLND + ALND and SLND treatment groups overall or in any
of the trials.

Fig. 1. Overall survival and disease-free survival in the studies comparing SLND + ALND to
SLND alone

Meta-analysis of breast cancer recurrence as a dichotomous outcome, rather than as
a time-to-event outcome, was undertaken as the data were not reported as time-to-event
outcomes. However, the length of follow-up for these data was comparable between the
trials (see Table 1). These analyses are illustrated in Fig. 2, which shows that axillary (summary RR = 0.46, 95 % CI 0.14–1.49; p = 0.2; I
²
 = 0 %), local (summary RR = 1.6, 95 % CI 0.86–2.97; p = 0.14; I
²
 = 0 %), regional (summary RR = 0.34, 95 % CI 0.1–1.15; p = 0.08; I
²
 = 0 %) and distant breast cancer recurrence (summary RR = 1.31, 95 % CI 0.8–2.15;
p = 0.28; I
²
 = 0 %) did not differ between the treatment groups.

Fig. 2. Breast cancer recurrence in the studies comparing SLND + ALND to SLND alone. Please
note the following regarding the data included for ACOSOG ZOO11 for regional breast
cancer recurrence: Regional recurrence defined as recurrence in the axillary, supraclavicular
or internal mammary nodes. The authors only report local recurrence, axillary recurrence
and locoregional recurrence. We have subtracted the local recurrence data from locoregional
recurrence data to obtain the regional recurrence data, which is equal to the disease
recurrence in the axilla data, suggesting that no patients recurred in the supraclavicular
or internal mammary nodes, provided all these data only count each patient once. An
entry of 0 in the total number of events column signifies that the study did not report
this outcome

The ATTRM-048-13-2000 trial did not report on short-term adverse events or long term
complications (Table 3) and is therefore at high risk of reporting bias for these outcomes. Inadequate details
were reported on the selection of patients and the outcome assessment, which puts
the results at risk of both patient selection bias and detection bias (Table 2). Moreover, at baseline the tumours were detected by palpation in more ALND than
SLND patients. In IBCSG-23-01 the authors did not report inferential analyses of the
short-term adverse events or long-term complications, but the rates of post-operative
infection, sensory neuropathy (any, grade 3–4), lymphoedema (any, grade 3–4) and motor
neuropathy were all numerically higher in the ALND group (Table 3). No blinding was undertaken of the outcome assessment, which means that the results
are at high risk of detection bias. Moreover, it was unclear whether the results were
subject to attrition bias for short-term adverse events and long-term complications
(Table 2).

Table 3. Morbidity outcomes in the included studies

Inferential analyses of the rates of short-term adverse events were not presented
in the ACOSOG Z0011 trial, but the rates of wound infection, axillary seromas, axillary
paresthesias and objective lymphoma are all numerically higher in the group that received
ALND. The same pattern of results was also observed for the long-term complications
of brachial plexus injury, axillary paresthesias, and objective and subjective lymphoma
at 6 and 12 months; and for subjective lymphoma at 12 months (Table 3). In this trial it was unclear whether outcome assessment was blinded, 30-day short-term
adverse event data were not reported for all the patients, and the outcome data for
long-term complications were missing for progressively larger proportions of patients
in both treatment groups, but possibly more so for the SLND group. This in turn means
that the results must be interpreted with some caution because they are at risk of
detection bias for all outcomes and of attrition bias for the short-term adverse events
outcome; and for the long-term complications the results are at high risk of attrition
bias (Table 2). Moreover, all three trials randomised patients after the results of SLND were known,
which puts these trials at risk of recruitment bias to the extent that patients perceived
at higher risk (e.g., multiple micrometastatic foci) were not invited or chose not
to take part in the studies. This is because any tendency not to recruit patients
perceived to be at higher risk would influence the relative performance of the interventions
in the direction that less extensive surgery (SLND) would appear relatively more beneficial
because the patients who are more likely to benefit from more extensive surgery (ALND),
that is, patients at higher risk, would not be part of the study population. This
could mean that the results are only applicable to the patients seen in clinical practice
who meet the inclusion criteria of these trials, but are also perceived to be at low
risk.

ALND versus aRT

In the AMAROS trial no differences in overall survival, disease-free survival, shoulder
mobility or quality of life were observed between the groups that received ALND and
aRT (Tables 3, 4). However, the rates of (any clinical sign of) lymphoedema were higher in the ALND
group at 1, 3 and 5 years. When lymphoedema was defined as an arm circumference increase
?10 %, the rates only differed significantly at 5 years (Table 3). The trial was open label and did not report short-term adverse events or long-term
complications other than lymphoedema and shoulder mobility for which either progressively
larger or unclear proportions of data were missing, respectively. The results are
therefore at high risk of both detection bias (all outcomes), attrition bias (lymphoedema
and shoulder mobility) and reporting bias (short-term adverse events and long term
complications; Table 2).

Table 4. Overall survival, disease-free survival and breast cancer recurrence in the axilla
in the studies comparing ALND and aRT

The OTOASOR trial did also not find any significant differences between the treatment
groups in overall survival, disease-free survival or axillary recurrence rates (Table 4), however, the OTOASOR trial did not report any morbidity outcomes, which puts the
trial at risk of reporting bias. Moreover, very little information was reported about
patient selection and allocation as well as about potential blinding of outcome assessment,
which exposes the results to risk of both selection bias (all outcomes) and detection
bias (all outcomes) to the extent that these were compromised (Table 2). At baseline, however, more ALND than aRT patients had pT2-3 tumours. On the other
hand, both trials randomised patients before sentinel lymph node biopsy, which suggests
that the study populations are representative of the risk spectrum of those patients
seen in clinical practice that meet the inclusion criteria of these trials.