Institutions Act Fast and Take Back Slowly

In 2008 the FDA gave bevacizumab, sold as Avastin, accelerated approval for metastatic breast cancer.1 The approval rested on one trial in which tumors took longer to progress. Accelerated approval works like a loan against future evidence: the drug reaches patients early, and later trials are supposed to show that it helps them live longer or better. For breast cancer the later trials did not. On June 29, 2011, after a two-day hearing, the FDA’s oncology advisory committee voted 6 to 0 to uphold its finding that the drug’s risks outweighed its benefits.2

Two weeks later the National Comprehensive Cancer Network’s breast cancer panel met in Philadelphia and kept its recommendation. Bevacizumab with paclitaxel remained “an appropriate therapeutic option,” at category 2A.3 On October 3, Blue Shield of California said it would stop covering the drug for breast cancer from October 17, except for women already on it and requests it approved case by case.4 On November 18 the FDA commissioner revoked the indication, saying “there is no proof of benefit to justify the risks.”1 Three days later Medicare said it would keep paying.5

All of these institutions had the same trials in front of them. They acted in a different order and months apart, and by the end of the year two of them had reached opposite answers. One of those answers rested on a rule. Medicare treats an off-label use of a cancer drug as medically accepted when the compendia the Secretary of Health and Human Services recognizes support it, and NCCN’s is one of them.6 Once the panel kept its recommendation, the use stayed eligible, and CMS chose not to change its coverage. Its local contractors could still deny claims, and one of them, Palmetto GBA, had moved to stop paying that January before reversing itself.5 Blue Shield, bound by no such rule, had already stopped for new patients.

One finding, many clocks

Records Reviewed on a Calendar Are Always Late argued that a record is wrong for exactly as long as its review interval. The world moves on events, and the record moves when someone is scheduled to look. Bevacizumab shows what happens when one event has to move several records. Each institution asks for different evidence before it acts and keeps its own calendar. The regulator runs a withdrawal process with a hearing and a comment period. The guideline panel meets when it meets. Each insurer decides for its own members, and Medicare inherits the compendium.

So the delay between a finding and a change in what patients get belongs to the chain, and no single institution’s clock shows it. I call it institutional latency: the time between evidence becoming available to an institution and that institution acting on it, measured at each institution in a chain that acts on the same evidence.

Some of it is already counted. Every year the European pharmaceutical industry’s W.A.I.T. indicator, compiled by IQVIA, counts the days between European approval of a new medicine and its availability to patients in each country: 578 on average in the 2024 edition.7 It’s a useful number, and it covers one link, in one direction.

The way out

The other direction is where the chain gets interesting. Two recent studies looked at how NCCN handled cancer drugs with accelerated approval. Of 36 indications the FDA later converted to regular approval, the guidelines matched the FDA on 30, and had updated 20 of them before the FDA acted. Of 12 indications the FDA withdrew, the guidelines matched it on 7.8 The second study found 8 of 21 withdrawn indications still in the guidelines, 6 of them at category 2A.9

Read together, the guidelines caught up with the regulator’s confirmations far more often than with its withdrawals: five conversions in six, against seven withdrawals in twelve. These are rates of catching up. In the first study every indication had been recommended from the start, and both studies count whether a guideline caught up. How long it took is the open question. It’s one kind of institution in one field, and it establishes no law. It does fit what each direction has going for it. A sponsor has every reason to speed adoption and none to speed withdrawal. An error of acting is visible and has a name attached, while the cost of waiting is spread across patients who rarely learn they were waiting. Where that cost has been made visible, the clock has moved. The AIDS activists Steven Epstein studied in Impure Science changed how the FDA weighed evidence and how fast it approved drugs, by making the cost of its waiting impossible to ignore.10 Withdrawal seldom has anyone organized to do the same.

Who carries the interval

In the fall of 2011, whether a woman with metastatic breast cancer could get bevacizumab paid for depended on which institution she answered to. A Blue Shield member starting the drug after October 17 needed an exception, while one already on it kept coverage. A Medicare patient kept coverage after the FDA had said there was no proof of benefit, for as long as her local contractor kept paying. Neither of them chose the clock she was on.

Clinicians carry the interval too. When a guideline says one thing and a coverage policy says another, the reconciliation happens one patient at a time, in appeals, peer-to-peer calls and explanations. That work doesn’t appear in any institution’s accounting, which is part of why each institution can call its own timing reasonable.

Which delay earns its keep

Some latency is the system working. The confirmatory trial is the whole point of accelerated approval, and waiting for it is the price of early access. A guideline panel that declines to move on a single trial is sometimes right. The useful question is which delay earns its keep, and four tests separate the two kinds:

1. Does the interval produce information? Waiting for a trial to read out does. Waiting for the next scheduled meeting usually doesn’t.
2. Who carries it, and could they refuse? A delay whose cost falls on the people deciding is a different thing from a delay whose cost falls on people with no say.
3. Is it symmetric? A body that takes a drug up on one clock and takes it back on a slower one is letting the delay work for someone.
4. Is anyone answerable for it? An interval that no institution owns is the clearest sign of harm.

Medicare’s payment after the revocation is the instructive case for the fourth test. The statute lets the compendium decide which off-label uses count as accepted, CMS declined to override it, and contractors decided claim by claim. Answerability for the interval was spread across three bodies, and the one that set its length, the guideline panel, paid nothing for it.

Timing the whole chain

An institution that times only its own link can call itself fast while the chain is slow. Counting one link has already changed the argument about it: a European health system can see where its own number sits against its neighbors’. There’s nothing comparable for the chain as a whole, or for the way out. Most of the dates are public: advisory committee votes, label changes, guideline versions, coverage policies. No one’s job is to put them on one line and ask who paid for the gaps.