(Apologies for the delay in replying to you.) I was able to access the paper in question, and of the 71 trials included in their analysis, two trials reported

(Apologies for the delay in replying to you.) I was able to access the paper in question, and of the 71 trials included in their analysis, two trials reported a negative overall survival (OS) outcome (-0.3m and -0.6m) and one trial reported exactly 0m overall survival gain. There were also 23 trials (I think - quick eyeballing) where the overall survival gain wasn't available or reported. Everything else had a positive overall survival benefit reported. In most cases (2/3 cases with a negative or null survival benefit reported, and a majority of the cases where no survival benefit was reported) there was a positive improvement reported in progression-free survival (PFS). So with the exception of a single case (Premetrexed, approved in 2008 for 1L NSCLC with 0m PFS benefit and -0.3 OS benefit - maybe it was better tolerated than the comparator??) there's seems to be a positive benefit to be found somewhere, small as it may have been in some cases. And yes, it's essentially a data-set with a lot of clustering of OS benefit around a low single-digit number of months, with a few higher outliers. EDIT: FWIW, it would also be very uncommon (as borne out by the dataset) for FDA to approve a drug which didn't offer a clear statistically-significant improvement in (usually) OS or (more rarely) PFS.