Quality of the evidence
No outcomes were supported by high-quality evidence and only
three by moderate-quality evidence (acupuncture for pelvic pain,
exercise for lumbo-pelvic pain and lumbo-pelvic work absenteeism).
Overall, there was low-quality evidence for outcomes because
of high risks of bias and sparse data. Trials were generally
small (range 30 to 855 women; with only three trials including
over 300 women, the GRADE rule of thumb for imprecision/
sparse data; Schünemann 2009). Clinically heterogeneous populations,
interventions, comparisons and outcome measures precluded
pooling the results to arrive at overall estimates of effect in
all but the exercise interventions for LBP and combined pelvic and
back pain. Inclusion criteria were quite different across studies;
women were admitted at different points in their pregnancy, ’diagnoses’
of LBP and pelvic pain ranged from self-report of symptoms
to clinical interpretation of the results of special tests, such as the
posterior pelvic pain provocation test, resulting in a heterogenous
population. Pain and disability were measured in a variety of ways;
pain was measured as intensity, presence, change in pain within
groups, numbers or percentage who reported improvement, and
disability was measured as back-specific function, general function,
ability to perform activities of daily living, change in abilities
within groups, numbers or percentage who reported improved
function, time off work and sleep disturbance. Outcomes were
measured daily, weekly, in the morning, in the evening, over the
course of the pregnancy and during the postpartum period; the
latter outcomes were outside the scope of this review.
Besides the paucity of usable data, the risks of bias contribute to
the lack of confidence we have in the results. Overall, the trial reports
were poorly written and it was difficult to follow some of the
analyses, although more recent trials tended to be more complete.
We only included RCTs (one of which used a cross-over design)
in this review, but in 13 of the trials, the methods of randomisation
were unclear and in 14, the methods of allocation concealment
were unclear. On the other hand, we excluded eight trials
because the techniques they described for randomisation were at
high risk for bias, or allocation procedures were simply unclear.
Current wisdom suggests that randomisation and concealment of
allocation are key study characteristics that reduce the potential
for bias. Blinding of personnel remains difficult in non-pharmaceutical
trials, a reality that increases the risk of bias, especially in
self-reported measures of symptoms. Some of the more recent trials
did attempt to minimise bias by recruiting, for example, only
participants who were naive to acupuncture (Elden 2005; Elden
2008) or by conducting credibility checks (Wang 2009a) to determine
the participants’ expectations of the study interventions
they were offered. In Gil 2011, Martins 2005 and Ekdahl 2010,
baseline pain was different in the two groups. In Wedenberg 2000,
12 of 30 women dropped out of the physiotherapy group, while
none withdrew from the acupuncture group (although two were
excluded from analysis due to receiving both treatments), leading
to potential attrition bias. Based on baseline data, there were no
obvious reasons for the difference in withdrawals between the two
groups.