Breaking Down the RP TBWU Schema

After last week’s post about Radiology Partners’ new time-based work units, a reader helpfully shared a spreadsheet with the whole schema: the codes of millions of exams across the portfolio used in its calculation, their RVUs, their TBWUs, and the percentage adjustment for emergency, inpatient, and outpatient contexts. The spreadsheet has 4362 data rows, but it’s interesting, and I think the analysis yields some meaningful insights.

It’s just a huge table, so no methodology or explanation is contained, and we are still left to infer some of its motivations (which I did in the first post, I think correctly). We don’t know, for example, how “time-based” any of this really is. Whether well or poorly conceived, any custom system creates winners and losers, and the RP system is no exception.

Several RP rads have reviewed this data and feel that it’s accurate to their experience so far. I’ve asked around, and seemingly no one, including local practice leadership, had this data. The new system, until now, was essentially totally opaque at this level of detail. I assumed it would take time for some enterprising (or frustrated) individual rads to create even a partial picture from their personal dashboards. Instead, we have a snapshot during its first month in operation (I wouldn’t count on us getting the next iteration). As always, if something here is factually incorrect, I welcome correction from RP directly.

The high-level summary: big bump to plain films at the expense of overall hits to body CT and neuro CT/MRI as well as a general devaluing of ED work. On the breast side, a huge cut to screeners for a not quite commensurate increase for diagnostics. For IR, no delta. GI fluoro still isn’t fair.

The file sums to 47.8 million exams covering the period of 1/1/2025 through 3/9/2026. I’m not sure what if anything the difference between that and the referenced 52M exams on the recent email announcement signifies (I’m guessing nothing, or perhaps some outlier practices were excluded). That overall data foundation is a little less impressive as a baseline for novel weighting when you realize that 9.7M of that is just chest films, but it’s undeniably a big number and a lot of data.

System-wide and volume-weighted, the new TBWU production comes out to 99.7% of 2026 RVU production, so unsurprisingly very close to zero sum on day one, as promised (the only politically survivable way to roll out a proprietary work unit).

However, since the spreadsheet doesn’t break down exam volumes by ED/IP/OP, it’s actually impossible to know if there’s any baked-in reimbursement decline, since we don’t know the relative percentages for how those weightings are applied. It’s possible they net to zero as one would hope/assume, but that doesn’t have to be the case looking back, let alone going forward. Some codes are always valued higher, some are always valued lower, and some depend on the context. (In case you’re curious, the theoretical absolute lower bound using only the worst ED/IP/OP modifier for each code [which isn’t actually possible] would be 95.2% of RVUs.)

Regardless, averages don’t tell stories, because not everyone reads a general equal mix of the 4000+ exam codes.

Plain Films Ate Everything

On the whole, plain films increased +81% from 3.6 million RVUs to 6.6 million TBWUs. That single modality absorbs about 3 million net units of redistributed credit out of a total 5.6 million units increased where TBWU>RVU, more than half of the total upward adjustments.

Some examples:

• XR chest 1-view: 5.4 million exams. RVU 0.18 → TBWU 0.317. +76%.
• XR chest 1-view portable: 1.75 million exams. RVU 0.18 → TBWU 0.317. +76%.
• XR abdomen (KUB): 900k exams. RVU 0.18 → TBWU 0.392. +118%.
• XR lumbar spine 2-3 views: 450k exams. RVU 0.21 → TBWU 0.394. +88%.
• XR fingers (various flavors, ~160k exams combined): RVU 0.13 → TBWU 0.393. +202%.
• XR bone age hand: 21k exams. RVU 0.19 → TBWU 0.869. +357%.

I’m told that after the go-live of the new system, the plain film backlogs evaporated. Generalists and those reading heavy volume plain films are the short-term winners in this system.

Make no mistake, by RP’s own admission vis-à-vis Mosaic, this is temporary. Bumping XR at the expense of CT and MRI allows them to subsequently lower XR thanks to AI-drafting and ultimately skim more revenue off the top.

I won’t pretend to know exactly how they’ll do it. They may just lower TBWU as suggested previously. They also could make practices “internal customers” for their tech and “pay” for those Mosaic drafts as a way to avoid making the TBWU skim look too depressing on the daily productivity meter. By doing so, when they eventually start trying to peddle Mosaic to external customers, they can promise the RP groups that they’re getting a “special deal” or a “preferred rate.”

The ED / IP / OP Question

System-wide, volume-weighted, the multipliers land here:

• ED multiplier: 1.007
• IP multiplier: 1.143
• OP multiplier: 1.010

Inpatient reads are worth about 13% more than ER or outpatient reads on average, which presumably accounts for sicker patients, extra comparisons to pull, etc. That sounds fine.

But averages don’t tell the story here, because ED is discounted relative to OP for most of the highest-volume codes:

• XR chest 1 view (5.4M exams): ED 0.277, IP 0.393, OP 0.337
• CT head without contrast (2.8M exams): ED 0.654, IP 0.846, OP 0.665
• XR chest 2 views (2.5M exams): ED 0.266, IP 0.387, OP 0.259
• CT abdomen/pelvis with contrast (2.1M exams): ED 1.428, IP 1.829, OP 1.518
• XR chest 1 view portable (1.75M exams): ED 0.277, IP 0.393, OP 0.337
• CT abdomen/pelvis without contrast (1.0M exams): ED 1.380, IP 1.720, OP 1.397
• CT cervical spine without contrast (951k exams): ED 0.747, IP 0.930, OP 1.136
• XR abdomen 1 view (KUB) (901k exams): ED 0.351, IP 0.411, OP 0.393
• CT chest without contrast (703k exams): ED 1.437, IP 1.503, OP 1.486
• CT chest angiography with contrast (697k exams): ED 1.363, IP 1.510, OP 1.588

The logic for an ED penalty is presumably a shorter “average read time” in part based on the overscanning of normal people. But “average time to read” and “actual cognitive effort per read” are not the same thing. Part of the issue with the premise of this time-based weighting for an ER reader is that part of the faster read-time for ER cases is not the simplicity but the turnaround time pressure. It’s often stressful, rapid-fire work by necessity, not by default.

For any per-click ED and teleradiology contractors that RP uses to cover nights and weekends, a meaningful chunk of the ED worklist just got repriced downward relative to its RVU value and its IP/OP counterparts. If productivity thresholds don’t follow, individual carveouts made, and/or bonuses adjusted, then  effective compensation falls. I’ve already heard from ED and neuro readers that they’re looking elsewhere.

Modalities Paying for the Bump

The modalities that funded the plain-film bump, volume-weighted:

• CT: −10.5% (net −2.0M units)
• Mammography: −14% (net −650k units)
• MR: −7% (net −410k units)
• US: −2% (net −110k units)
• DEXA: −46% (net −70k units)
• Nuclear medicine: −19% (net −55k units)

Several big-volume and high-dollar CT/MRI workhorse codes got trimmed 15–25%:

• CT head without contrast (2.8M exams): RVU 0.83 → TBWU 0.680. −18%.
• CT abdomen/pelvis with contrast (2.1M exams): RVU 1.77 → TBWU 1.463. −17%.
• CT cervical spine without contrast (950k exams): RVU 0.98 → TBWU 0.767. −22%.
• CT chest angiography (700k exams): RVU 1.77 → TBWU 1.402. −21%.
• MR brain without contrast (500k exams): RVU 1.44 → TBWU 1.120. −22%.
• MR brain with and without contrast (360k exams): RVU 2.23 → TBWU 1.670. −25%.

You can read this two ways. The charitable reading is that the RVU system universally undercompensates plain films relative to a plethora of more desirable CT/MRI exams, and TBWUs attempt to correct for that, maybe even based on actual measured read times. The less charitable reading is that RP has looked at which study types Mosaic will make cheaper to produce and is setting itself up to reprice the work accordingly. These are not entirely mutually exclusive.

A subspecialized neuroradiologist takes a real hit in this system.

The Mammography Trick

Mammo has been extremely popular with trainees in recent years thanks to many flexible, no-call job openings and high-income opportunities, in large part due to the growth and generous reimbursement for screening tomosynthesis.

• Screening mammography with tomosynthesis is cut from RVU 1.33 → TBWU 0.778 (−41%) across 2.7 million exams.
• Whole diagnostic mammography bucket (622k exams, every diagnostic variant): RVU 1.412 → TBWU 2.571, +82%
  • Diagnostic tomosynthesis subgroup (550k exams, the dominant slice): RVU 1.487 → TBWU 2.586, +74%

Breast imaging as a modality loses about 650k units, or −14% volume-weighted. There are plenty of people, even breast imagers, who would agree that diagnostics are unpaid relative to screeners, but making this change is a serious adjustment to breast compensation, and it especially devalues a portion of the telemammo workforce, cutting the reimbursement for highly popular screening moonlighting by almost half, and only propping up the nonnegotiable must-staff-every-day part that deals with real-time patients.

This is exactly the kind of cut-then-compensate pattern you would expect for AI-assisted work. Screening mammo has always been one of the biggest targets for AI: the exam quality is guaranteed, the training sets are massive, outcomes data available, and the reporting is completely standardized. Drafts are coming very soon, but screeners are already fast. I wouldn’t be surprised if, after a couple of years of data collection, a very small number of people (maybe just the head of breast) are putting their name on a ton of basically autonomous negative screener reports for a couple of bucks a pop with humans only meaningfully reviewing the positives.

GI Fluoro is Still a Terrible Deal

I guess fair accounting for time doesn’t apply to fluoroscopy, since there’s no way anyone would consider these bumps to sufficiently account for the effort and time spent on these examinations unless most RP fluoro is performed by PA/NP/RAs?

• FL esophagus barium swallow with video (95k exams): RVU 0.52 → TBWU 0.610. +17%.
• FL esophagus barium swallow (63k exams): RVU 0.59 → TBWU 1.023. +73%.
• FL upper GI single contrast (28k exams): RVU 0.78 → TBWU 1.034. +33%.
• FL upper GI double contrast (12k exams): RVU 0.88 → TBWU 1.096. +25%.
• FL barium enema single contrast (5.6k exams): RVU 1.01 → TBWU 1.507. +49%.
• FL barium enema double contrast (1.5k exams): RVU 1.23 → TBWU 1.496. +22%.
• RF double contrast esophagram (7.2k exams): RVU 0.68 → TBWU 0.897. +32%.

Not sure how one justifies this except that seemingly no radiologists choose to do fluoro anymore; fluoro chooses you. You don’t need to incentivize it if you don’t have a choice: it’s not readable remotely, moonlightable, and no one will ever cherry-pick it.

Some other random highlights:

• Everyone’s favorite ultrasound was ruined: US lower extremity veins bilateral (275k exams): RVU 0.68 → TBWU 0.343. −50%.
• Nuclear cardiology was destroyed? SPECT perfusion changes range from −23 to −84% (but the majority are NM heart perfusion SPECT multiple, which was reduced 71.6%).
• A bilateral renal ultrasound (RVU 0.72 → TBWU 0.628) went down and is worth less than a unilateral, which went up (RVU 0.57 → TBWU 0.703).
• There are some random changes to very low-volume exams. For example, a rare unilateral breast MRI (~240 exams) was decreased 12% (RVU 2.05 → TBWU 1.807), whereas bilateral breast MRI (91k exams) was increased 19% (RVU 2.24 → TBWU 2.665). There are other examples of low volume as well as similar (but differently named) exam types presumably slipping through the cracks and being valued in somewhat arbitrary-seeming ways.
• “RECON” ordereables are valued wildly different than the same de novo scans, ranging from −41% to +45%. Not sure how much that makes sense, especially when they are sometimes read by the same person and sometimes not.
• IR was completely untouched (probably because it would cost too much to pay them fairly in a time-based system).

Takeaways

The study types most obviously in Mosaic’s near-term pipeline—plain films, screening mammography—are the ones with the most aggressive repricing. Plain films up now, to be cut later once the AI is more helpfully reading them. Screening mammography cut now while diagnostics are up, which is the distribution you’d expect in a world where AI handles the screening triage and the human reads the flagged cases; it presumably means RP is okay losing a significant fraction of their remote telemammo workforce. I wonder by next year how many screeners the head of RP breast and/or chief of breast AI will have their name on.

Time-based clearly doesn’t include procedures, as IR was left untouched. GI fluoro was bumped, but clearly “time-based” also doesn’t really reflect a simple weighted average of time spent.

The ED discounting on several high-volume codes and the contractor-hostile implications of the ED repricing are all consistent with a system optimized for one thing: extracting efficiency from the diagnostic read workflow as AI takes on more of it, while giving RP maximum flexibility to recapture that efficiency at the enterprise level.