The answer is both yes and no, as I explain below.
1. No, BBM92 is better (or at least, we initially thought so)
E91 was the initial idea which led to the more rigorous BBM92.
In the E91 paper, there is not actual security proof, ant the attack is not really specified: if the CHSH inequality is violated by only a few percents (say $S=2.05$), how many secret bits can Alice and Bob still share? With the techniques known in the early 1990’s there were no way to answer this kind of question. BBM92 was essentially a way to replace the Bell inequality with a more tractable entanglement witness, and to transfer the security proofs of BB84 to entangled state protocols. So in term of security, and practicality, BBM92 is much better than the E91 which is badly specified. And in practice, when a physicist says “E91”, he often actually means “BBM92”...
2. Yes E91 turns out to be better (because it is device independent)
To my knowledge, the first QKD paper actually using Bell inequality violation as an actual security parameter for QKD is the Barett, Hardy Kent paper of 2004 (arXiv/PRL). This was initially seen as a way to to perform QKD even if one does not trust quantum mechanics. This kind of protocols has now expanded to the domain of device-independent quantum cryptography, when one does not need to trust your devices to perform QKD. In this sense, variants of E91 can indeed be safer than BBM92.
Artur Ekert actually said of the latter development that it was “a very good example of when your ideas are more clever than yourself”.