- Classic "Absence of Evidence is not Evidence for Absence" misinterpretation of P-values.  NEJM is making the same error - I'll be visiting their editorial staff to discuss this.

- Matching discarded 80% of subjects and is arbitrary (e.g., you'd get different matches if you
randomly reordered the subjects in the dataset)

- Matching analysis failed to adjust for outcome heterogeneity, biasing the treatment effect 
  towards the null

- Authors apparently failed to do an unbiased survey of subject matter experts to determine what is an adequate set of potential confounders before being biased by what confounders happen to be available in the dataset

On the third point, propensity adjustment adjusts for unmeasured confounders.  It does not do away with the need for adjusting for "big player" prognostic factors.  The non-collapsibility of the hazard ratio is at the root of this problem (same for odds ratios, i.e., a perfectly balanced randomized trial that does not adjust for prognostic factors will get biased effect ratios towards 1.0).