On 2015 Aug 31, Wichor Bramer commented:

In my opinion the conclusions of this article are rather overdrawn. The authors have determined that the coverage of PubMed is high enough to be used in reviews. However, they have not performed searches for a systematic review, they have performed multiple searches for known items. If recall would be ideal PubMed would retrieve enough relevant articles to be used as a single database. However, recall is never ideal in any database. My recent observations (which have not yet been published) show that Medline retrieved 937 out of 1192 included references for 38 published reviews, so only 79%. Embase and medline together retrieved 1110 references (94%). So overall recall in embase is much better than in medline alone. In my opinion overall recall is not the best measure for database usefulness. When deciding a strategy for a systematic review, one decides for only one review, and not for 50 of 38. A better parameter is the minimum number observed. In medline the minimum was 53%, compared to 76% in embase/medline. Neither is acceptable in my opinion. The authors of this article also found that for some reviews, the coverage in pubmed was much lower than the total coverage. One database cannot be used to replace all other databases in the search for a systematic review.

On 2015 Aug 31, Wichor Bramer commented:

In my opinion the conclusions of this article are rather overdrawn. The authors have determined that the coverage of PubMed is high enough to be used in reviews. However, they have not performed searches for a systematic review, they have performed multiple searches for known items. If recall would be ideal PubMed would retrieve enough relevant articles to be used as a single database. However, recall is never ideal in any database. My recent observations (which have not yet been published) show that Medline retrieved 937 out of 1192 included references for 38 published reviews, so only 79%. Embase and medline together retrieved 1110 references (94%). So overall recall in embase is much better than in medline alone. In my opinion overall recall is not the best measure for database usefulness. When deciding a strategy for a systematic review, one decides for only one review, and not for 50 of 38. A better parameter is the minimum number observed. In medline the minimum was 53%, compared to 76% in embase/medline. Neither is acceptable in my opinion. The authors of this article also found that for some reviews, the coverage in pubmed was much lower than the total coverage. One database cannot be used to replace all other databases in the search for a systematic review.