The directory knows doctors. It barely knows nurses, dentists and pharmacists

Some numbers here have moved. This update measures the 2026-05-08 release. Every profession improved at 2026-08-20, and the widest gap narrowed sharply: pharmacy workers with a workplace went from 1 to 526, and dentists from 4.7% to 13.3%. The pattern below still holds, at about a fifth of its former size. See the 2026-08-21 update.

The US government keeps a big list of everyone who provides health care. Doctors, nurses, dentists, therapists, counselors, pharmacists. It is called the National Provider Directory.

For each person, the list is supposed to say where they work. That one fact is what makes the rest useful. If the list does not say where you work, it cannot say which hospital holds your records, and software cannot look you up.

We already knew a lot of people were missing that fact. What nobody had checked is who.

So we checked. All 227,727 health workers the list says are active in Pennsylvania. Here is what we found.

Some jobs are in the list. Some are barely in it at all

Read that last row again. Pennsylvania lists 12,419 pharmacists and 576 pharmacy technicians. Exactly one of the 12,995 has a workplace recorded.

We expected some difference between jobs. This is not some difference. The top of the list is nearly eight in ten. The bottom is one in thirteen thousand.

Why it splits this way

Look at which jobs are near the top. Doctors, nurse practitioners, eye doctors, foot doctors. Now look at the bottom. Dentists, nurses, pharmacists, counselors.

The top group bills Medicare. The bottom group mostly does not. Dentists are barely in Medicare at all. Pharmacists bill through the pharmacy, not as themselves. Counselors are usually paid by Medicaid or private insurance.

So the list is not really describing who provides care in Pennsylvania. It is describing who bills Medicare. Those are different groups of people, and nothing in the list tells you which one you are looking at.

Why you should care about this

Any time someone says "X% of providers are in the directory", that number is mostly about Medicare-billing jobs. It sounds like a statement about all health workers. It is not.

There is a second problem, and it is the one that worries us more.

If you look up a dentist and find no workplace, you cannot tell which of these is true:

• the dentist genuinely has no way for software to reach them, or
• the dentist does, and the list simply never wrote it down.

Those need completely different fixes. Right now they look identical.

We guessed wrong, and we are saying so

Before we ran this, we wrote down what we expected to find. We expected the gap was mostly filler. Student IDs, equipment suppliers, medical transport, billing offices. Real ID numbers, but not people who hold your medical records.

There is some of that. Pennsylvania has 9,804 ID numbers belonging to students in training, 708 belonging to nurse's aides and 84 belonging to doulas.

But all of it added together is 11,858 out of 227,727. That is 5.2%. Taking it out moves the overall number by about one point.

So our guess was wrong. We are leaving it written down anyway, rather than quietly swapping it for the answer we found. A guess you can check is the whole point of writing one down first.

We tried the obvious fix. It barely helped

Medicare keeps its own records of where people work, and publishes them free. We loaded all of it. Four government files, no login needed.

It found workplaces for 95,054 Pennsylvania health workers. That sounds like a lot.

It closed 1.9% of the gap.

Almost everyone Medicare knows about was already in the directory with a workplace listed. It found the same people twice.

That is annoying, and it is also the strongest evidence we have that the pattern above is real. If the gap were random, Medicare records would have filled a random chunk of it. Instead they filled almost none, because the people missing from the directory are the people who do not bill Medicare.

It does mean the remaining fix has to come from somewhere else. Private insurers and state Medicaid programs know where these people work. Those lists are public. That is where we are going next.

A new map

We also built a map of Pennsylvania, county by county. You can switch what it shows, zoom in and click anything.

https://ainpi.dev/states/pa/connectivity

Three things we did on purpose, because maps mislead very easily.

It shows shares, not totals. A map of totals is just a map of where people live. Philadelphia would win every time and you would learn nothing.

Empty counties are striped, not pale. A county with nobody in it is a different fact from a county where nobody is reachable. If both were pale grey you could not tell them apart.

The color steps are uneven on purpose. One county, Montour, has 7,146 health workers and only 17,860 residents. That is because Geisinger is headquartered there. With even color steps, Montour takes the whole range and the other 66 counties come out the same shade.

That last one is worth knowing whoever you are. One unusual place can quietly flatten an entire map.

Check our work

Please do not take our word for any of this.

• Every person we counted, one row each: https://ainpi.dev/api/v1/findings/role-gap-composition-pa.csv
• How we counted: https://ainpi.dev/findings/role-gap-composition
• The Pennsylvania map: https://ainpi.dev/states/pa/connectivity

The job categories are not ours. They come from the standard code set every provider already uses. We did not invent a single grouping.

All of it is public data. Repeating our work is free.

— Eugene Vestel, FHIR IQ