Seventeen RCTs (n=2,961) were included.
Three RCTs adequately described an appropriate method of randomisation and 8 RCTs adequately described appropriate allocation concealment. Four RCTs reported strict intention-to-treat analysis. None of the RCTs could be blinded.
Only four regimens were evaluated in more than one trial and only two of the trials used the same comparison regimen.
Only one trial reported a statistically significant difference in median survival for an adjusted analysis of doxorubicin plus cisplatin versus the combination doxorubicin-cisplatin-paclitaxel-granulocyte colony stimulating factor: 12.3 months versus 15.3 months (p=0.032). Three other trials reported survival difference for specific secondary analyses; one of these trials reported poorer survival with the experimental treatment. Differences in survival between treatments were small, with most differences being less than 3.5 months.
More recent studies tended to include a lower proportion of patients with poor performance status (p=0.001) and tended to be larger (p=0.019).
Univariate analyses showed that median survival was statistically significantly associated with the proportion of patients with poor performance status (p<0.001) but not with sample size (p=0.060). The analyses also showed that survival significantly improved over time (1.6 months per decade, p=0.016), although this was explained by the increased proportion of patients with poor performance status in earlier studies. The multivariate analysis showed poorer survival in studies with an increased proportion of patients with performance status 2 or worse (-2.3 months per 10% increase in the proportion of poor status patients, p<0.001), but no improved survival in more recent studies (p=0.99) and no significant association between survival and sample size (p=0.89).