Friday, September 27, 2019

Very Brief Blog: CMS Releases Proposed Crosswalk Pricing for CY2020 Lab Tests

On Friday, September 27, 2019, after market close, CMS released its proposed pricing for new CY2020 lab codes and for several codes under appeal in CY2019.

The proposals are under comment until October 27, to:  CLFS_Annual_Public_Meeting@cms.hhs.gov   

CMS priced 90 new codes, disagreeing about half the time with its advisory panel, which met on July 22/23 (pdf).  For more background on prior activities summer, here.  Final pricing will be announced in November.

See the Excel spreadsheet in the cloud at CMS, here.

On Sheet 1 of the Excel, CMS explains there were several codes for which it picked crosswalks that match no suggestion from public or panel.   On Sheet 2, it shows the panel votes, public suggestions, and the rationale for its proposed choice.  Note that CMS slightly updated to Version 2 between 9/27 and 10/2, updating its explanation of its agreement/disagreement with committee minority choices for several rows of the spreadsheet.  And note, I've tallied agreement or disagreement against the panel's majority vote (CMS may have agreed with a panel minority vote.)

(Screenshot, click to enlarge)


Top-Line Analysis: Crosswalk vs Gapfill in the CMS Proposals

Updated 10/2/2019

In a tally of CMS recommendations, I get 64 crosswalk and 26 gapfill = 90 total recommendations.

Now, the expert panel had recommended about the same gross numbers, 62 crosswalk and 28 gapfill.  What's interesting is that there is only 48% agreement between the Panel and CMS decisions.  This 48% includes category changes (from crosswalk to gapfill or vice versa) and other changes (from one crosswalk to a different crosswalk).

Of the 64 CMS crosswalks, 55 were single-code crosswalks and 9 were multiple-code crosswalks.  Of the 9 multiple code CMS crosswalks, just two were to a fraction, one to 0.5X and one to 1.5X.

The one code with a CMS-endorsed 1.5X multiplier was 0070U, Mayo extended version CYP2D6, which had many more variants than the target code 81226 (regular CYP2D6).

Panel vs CMS Decisions  

This current year in 2019, 18 codes are under gapfill.  For 2020, CMS is willing to put 26 codes into gapfill.

As noted earlier, only 48% of the currently proposed CMS recommendations were exactly the same as a the panel majority-vote for that code.

Generally, for any category like "crosswalk" or "gapfill" or even "single code crosswalk," what CMS chose matched the panel close to half the time.  For example, of 64 crosswalk CMS recommendations, 29 or 45% matched panel.   Of 55 single-code CMS crosswalks, 25 or 45% matched panel.  Of 9 CMS multi-code crosswalks, 4 or 44% matched panel.  Finally, of 26 CMS gapfill recommendations, 14 or 54% matched panel.

However, Panel was more likely to make multi-code and fractional code crosswalks in the first place.  Panel made 14 multi-code crosswalks, 8 of which contained a fraction.   CMS made only 9 multi-code crosswalks, and its two that contained a fraction were among the 8 were Panel contained a fraction.  Of the two CMS fractions, one was for a reduced price (0.5X) and one was for a raised price (1.5X).

click to enlarge

Are PLA Codes Different?

73 of the codes are new PLA codes, so whatever the data is for PLA codes, it can't be too different from the full set.   That said, Panel recommended crosswalk for PLA codes 47/73 times, and Panel recommended gapfill for PLA codes 26/73 times.   CMS agreed about half the time. 

If you work from the proposed CMS numbers, overall CMS recommendations added up to almost the same (of 73 PLA codes, 48 for crosswalk and 25 for gapfill.)   So if you are a new PLA code, you stand a 2/3 chance of being quickly crosswalked and the pricing process is finished for you.

However, taken the CMS PLA recommendations as a starting point, the CMS best crosswalk for a PLA code matched the panel majority only 33% of the time (16/48) while the CMS recommendation to gapfill a PLA code matched panel 52% of the time. 

But here's an angle. Of CMS Gapfill recommendations, all were for PLA codes.  Ditto for Panel recommendations.


CMS Trying to be Clear

CMS is trying to be clear in its (brief) rationale descriptions.  It also includes a special table where CMS highlights where its decisions particularly diverged from panel and public comment, for example, if it picked a crosswalk that no one on the panel had suggested.  Here:

CMS highlights several changes; click to enlarge or see Tab 1 of CMS Excel

CMS Makes Apparent Error re BRCA Appeal Codes

Two BRCA sequencing codes are under price appeal, 81163 (BRCA 1,2, Full Seq) and 81165 (BRCA 1 Full Seq).   CMS states that "Other than the initial request for reconsideration, we did not receive public comments on these codes during the June 2019 Annual Laboratory Meeting and prior to the July 2019 FACA CDLT Panel meeting."  This is incorrect.  For example, page 3 of the ACLA public presentation on in June 2019 clearly gave its recommendation and opinion for the repricing of these codes (raise 81163 to 81408 = $200, and raise 81135 to 1/2 of that.)  CAP made the same recommendation.  So did AMP. 

When CMS Changes a Crosswalk, It's Usually to 63% Of Your Suggestion

CMS and Panel agreed that 36 codes should be crosswalked - by method.  But they only agreed that 20 should be crosswalked to the same value, and CMS crosswalked 16/36 to different value.  I sampled every 5th row of this section, and found that CMS crosswalked to values between 52% and 75% of the Panel suggestion.  Every choice by CMS that was different, was cheaper.  Since I checked 6 values, I think there's 2^6 or 1 in 64 chances that CMS was randomly picked different best-crosswalk values it found more accurate, but which could be either higher "or" lower than Panel.  Seems like the dice are weighted. 

click to enlarge


___

For gapfill proposed prices on 18 codes, released August 21, 2019, under public comment until Monday, October 21, 2019, here.

___

For me the spreadsheet version 2 released by CMS opened with a number of hidden rows at different points in the 90 rows, which I had to manually reopen.