Pfizer: proportional and functional risks

On Wednesday, July 19 shortly after noon local-time an EF-3 tornado did extensive damage north of Rocky Mount, North Carolina. Included in this target zone is a very large Pfizer manufacturing facility (see photo below).

Several initial reports indicated that one-quarter of sterile injectables consumed by US hospitals were produced at this facility. This is not correct. Recently the facility has been the source for one-quarter of Pfizer’s US production of these products — or, perhaps, eight percent of national consumption.

US Pharmacopeia has identified 150 individual medications sourced from Rocky Mount. Pfizer has identified over sixty products sourced from Rocky Mount “with less than 3 months’ inventory in the distribution chain.” (More and more.) On Friday, July 21 the FDA Commissioner said, “Our initial analysis has identified less than 10 drugs for which Pfizer’s North Carolina plant is the sole source for the U.S. market.” Pfizer also explained:

Most of the damage was caused to the warehouse facility, which stores raw materials, packaging supplies, and finished medicines awaiting release by quality assurance. Pfizer is working diligently to move product to other nearby sites for storage and to identify sources to replace damaged raw materials and supplies. Pfizer is also exploring alternative manufacturing locations for production across our significant manufacturing presence in the U.S. and internationally and across the company’s partner network. After an initial assessment, there does not appear to be any major damage to the medicine production areas.

In the hours after the tornado strike Pfizer and several pharmaceutical wholesalers placed a wide range of medications on allocation, restricting new purchases to preexisting patterns.

For the usual purposes of Supply Chain Resilience there is a substantive distinction between Tracking Flows and Tracing Products. Flow impediments almost always threaten time-extended, wide-area, high population consequences. Product problems — in the case of most products — have much more constrained consequences. Product tracing exceptions to this principle (such as water and fuel) tend to prove the rule (exceptio probat regulam). Water, fuel, and similar products involve wide-area, time-extended flows serving high populations.

On Thursday morning last week, some credible information incorrectly signaled that several different products within a similar category (sterile injectables) originated from a single source for one-quarter of national flows. By late Friday afternoon corrected information prompted an understanding that, instead, Rocky Mount was the single source for less than one-tenth of national flows. By Saturday morning — looking at an early list of products available only or mostly from Rocky Mount — my risk-assessment plummeted from flood-level to something closer to a leaking pipe.

As a matter of strict volume, exposed risk was reduced by about two-thirds (8 percent vs. 25 percent). But as a matter of priority, my personal sense of strategic risk fell much more. Does this shift simply represent an exchange of cognitive fallacies? Or are there circumstances where the difference between a one-fourth proportion and a one-twelfth proportion totally — and rationally — transforms the strategic context?

My very fallible intuition is informed by a (potentially) rigorous empiricism regarding a network’s degree distribution. In network science terms my flow orientation is especially sensitive to the presence or absence of a hub (aka Giant Component or Giant Island or Strongly Connected Component and more). Demand and supply networks consist of nodes and links (aka vertices and edges). How connected are these nodes and links? How dependent and interdependent are these nodes and links? Are flows (of information, money, material, or more) more evenly distributed or more concentrated among these nodes and links? What portion of hospitals (demand nodes) are dependent on outputs from any single production and/or distribution node?

One of my favorite network scientists explains:

In most real networks, the degree distribution is highly asymmetric (or skewed): most of the nodes (the trivial many) have low degrees while a small but significant fraction of nodes (the vital few) have an extraordinarily high degree. A highly connected node, a node with remarkably high degree, is called hub… This asymmetric shape of the degree distribution has important consequences for the processes taking place on networks. The highly connected nodes, the hubs of the network, are generally responsible for keeping the network connected. In other words, the network falls apart if the hubs are removed from the network. On the other hand, since hubs are rare, a randomly chosen node is most likely not a hub, and hence the removal of random nodes from the network has a negligible effect on the network cohesion. Substantially, networks with long tail degree distributions are resilient to random removal of nodes (failure) but vulnerable to removal of the the hub nodes (attack).

The Pfizer facility at Rocky Mount is clearly non-trivial, but its degree distribution has been much less than initially reported. Rocky Mount is an important network intersection, but something less than a hub. Temporary loss of flow from Rocky Mount does create more — and noticeable — network disequilibrium for several medications. But crucially this network has been able to preserve its cohesion (remain substantially connected) even with failure of the Rocky Mount node.

What are proportional characteristics that distinguish a hub from a non-hub? Come back soon. I will try out a possible answer for your consideration.