Ghost networks get the attention. What about “hidden networks”?

Last week I shared a finding that surprised me: a provider removal from a hospital directory is almost twice as likely to propagate to payer directories as an addition. The classic directory fear is the inflated "ghost network", padded with doctors who've stopped practicing or moved away. This is the opposite. I'll call it the "hidden network."

This week, I tested four hypotheses for what might drive hidden networks. None was wholly explanatory. I'm sharing the headlines below. If you have other ideas, or would like access to some of J2 Health 's data to run your own analysis, send me a message. I'd be very curious to further understand what's going on!

Here's what I tested.

Credentialing. The idea: once a hospital's roster lists a new provider, payers still have to credential them — a step that doesn't exist for removals. Credentialing can take a while, but if it were the cause of hidden networks, we'd expect the dynamic to be temporary. Most new providers do end up credentialed, and so would reach payer directories eventually, just with a lag. The result: over the year of history we have access to, system adds don’t show up with a lag; most adds on system rosters simply never flow through. (I also checked whether providers on the OIG sanctions list were disproportionately filtered out. 97% of sanctioned providers are being excluded by most payers, but there aren't enough of them to explain the broader phenomenon.)

Availability. The idea: payers might decline to list providers who can't actually see new patients, to avoid confusing or frustrating members looking for a new doctor. The result: providers flagged as not accepting new patients on a hospital directory are ~10% more likely to be suppressed on the payer directory. The effect holds for directories with a direct MyChart scheduling integration, where live availability data is more likely to be flowing. This is an arguably laudable move (more on that next week), though one that's out of line with most state regulatory guidance. Either way, the effect is too small to explain most of the hidden network effect.

Cost steerage. The idea: to keep costs under control, payers might decline to add expensive doctors to public directories. This is a more extreme version of a common practice I've discussed with a few network leaders, whereby pricier providers are listed lower on a directory rather than dropped. The result: this is hard to measure, but using academic systems as a proxy for expensive providers, the thesis doesn't have legs. Adds from academic systems actually flow through at a higher rate than adds from presumably cheaper systems.

Network sculpting. The idea: maybe systems aren't contracting in full. If a system roster is made up of employed providers plus several affiliated ACOs, CINs, or medical groups that can opt in or out of a payer contract, hidden networks might just reflect partial contracting. The result: this had some explanatory power — providers in an ACO/CIN were 17% less likely to flow through to a payer site than providers in the system's employed medical group — but still nowhere near enough to explain the phenomenon. The outcome could also point to a different, mechanical, dynamic: ACOs/CINs may simply do a worse job getting roster updates to payers than employed groups, rather than being legitimately carved out.

Credentialing, availability, steerage, and network sculpting are all major network pain points where a lot of novel work is being done, including by fellow Oscar Health Network team alumni at companies like Pearl Health , Garner Health , and CertifyOS. As regulators sharpen their focus on directory accuracy, they'll be aiming at a moving target: a successful regulatory scheme has to distinguish "bad" network leakage — directories that don't update when they should — from legitimate network management aimed at accuracy, usability, and cost efficiency. And that's on top of the bad upstream data from systems, which remains rampant. I'm both excited and anxious to see how the REAL Providers Act and future data-accuracy initiatives get implemented. A hard-learned lesson in network: the devil is in the details, and a dozen dynamics are always occurring at once.