How much is a great stats department worth?

I’ve used this before, but it remains a great case study. Celebrating that you won the first quarter, but then lost the game by a massive margin is more of a sports tale than a pharma one, typically.

Opdivo was better positioned at the start, and its magnitude of victory in that first quarter was huge. Keytruda and Opdivo had very similar label footprints early on.

The pivotal change came when Merck focused on a PD-L1 subset in 1L NSCLC while BMS did all-comers (remember 2L NSCLC was all-comers). The former succeeded; the latter failed. This made a big difference at the time. In addition, Merck drove label expansion very hard, and also did a better job of designing their studies.

Merck’s stats unit looked at how they could most productively get signals in early phase, headed by Bob Beckman and Con Chen. Not many people know about their work, but they made a big difference in how they did development with Keytruda.

Keytruda has many indications (more than Opdivo); their success rate was higher; and they also have the advantage of ‘bulk use’: if you are only going to stock on one PD-1/L1 in your pharmacy, you pick the one with the most indications. So for example, Bavencio has a label for RCC, but is outgunned by Keytruda 4:1. This is in part because Keytruda has OS (not just PFS) data, but mainly due to supply issues.

This is a perfect illustration of asymmetric learning: searching for signals, finding them and exploiting them is a great example of setting up for eventual success, rather than the quick win. The $40bn cumulative difference between the two winners (so far), or the $15bn per year, is a healthy reward for learning harder and better than your direct competitor.

(Much of this shamelessly ‘borrowed’ from my smarter colleague, Alexander Gray .)