A woman I was helping last year needed a specialist and her plan's directory listed fourteen in-network practices accepting new patients within twenty miles. She got through eleven of them over two weeks, in the gaps around a job that does not permit personal calls. Four had never taken the plan. Two had stopped. One was a physician who retired in 2021. Two were not accepting new patients. One number rang out across three attempts on three days.
Two were correct. She is now a patient of one of them, eleven calls and about five hours later.
That the directories are wrong is not a discovery. It is one of the most replicated findings in American health policy, the error rates are enormous, and every study for a decade has said so.
What keeps me arguing about it is the second-order thing, which gets much less attention: the directory is not only a service that fails a person with a phone. It is the measurement instrument.
Network adequacy — whether a plan has enough doctors within enough distance for the people it covers — is a regulatory requirement, and it is assessed by counting the providers in the network. The count comes from the plan's own directory. So a plan whose directory contains four practices that never took it, two that stopped, and a retired physician has fourteen providers for regulatory purposes and two for a person with a phone.
The error is not noise around the measurement. It is on the numerator, in one direction, and nobody has an incentive to clean it.
I want to state the defence properly because I have heard it made by people doing the work and it is not an excuse. Keeping a directory current is genuinely hard: physicians move, retire, change what they accept, join and leave groups, and the plan finds out when somebody tells it, which is often nobody. Practices are asked to attest quarterly and many do not. There is real effort going into this and the people doing it are not the problem.
But the difficulty explains the errors, and it does not explain why the count still counts. Those are separate questions and only the first ever gets discussed.
And notice where the difficulty is currently resolved. The plan cannot keep the record current, so the record stays stale, so the work of establishing which of fourteen entries is true is done by the enrollee, one call at a time, on her own hours. Five of them, in her case. That labour is a real cost of running a network this way, it is incurred every time somebody needs a specialist, and it appears on no budget line anywhere, because the only ledger it could appear on belongs to a person who is not a party to the adequacy determination.
And it recurs on a schedule, which is the part that made me recognise it. Networks are re-contracted for each plan year. A practice that took the plan in November may not in January, and the directory is republished with the new year on it and much of the old content inside it. So the enrollee who spent five hours establishing which two of fourteen entries were true has established it for one plan year. At renewal her coverage continues, her eligibility is unchanged, nothing about her has moved — and the document she has to work from is new again.
That is the same shape as the renewal interval and it lands on the same people for the same reason: a periodic administrative event that costs the institution nothing, costs the enrollee hours, and is counted by nobody because the hours are not on a ledger anyone keeps.
The fix is not a better directory, which has been attempted repeatedly at considerable expense. It is to measure adequacy on something that cannot be inflated by a stale record — the plan's own claims data. A provider who submitted claims for that plan in the last six months is demonstrably in the network. That number exists, it is exact, it requires no attestation from anybody, and it is already sitting in the system that pays the bills.
I have put that in two comment letters. The response both times was that claims-based measurement would understate the network, because a provider who has not billed recently may still be available.
That is true. It would understate. The current method overstates by a factor I could measure with a telephone in one afternoon, and no one has ever asked me to justify a preference for the error that leaves somebody a doctor.
