Tuesday, September 1, 2015

What has AMP recommended for genomic sequencing pricing?

In summer CLFS public meetings of July 2014 and July 2015, the Association for Molecular Pathologists recommended interesting methods by which CMS could price new CPT codes for genomic sequencing procedures, such as 5-50 tumor genes.  AMP also held a webinar on this topic at early 2015 (archived here).

I believe I understand what they've recommended, and helpfully, AMP publicly posted its detailed 12-page public comments to the agency (dated August 6, 2015 (here)).

My sense is that in just a few key places, the AMP document is a bit confusingly worded so the transition from one AMP pricing method to another is not obvious to the reader.   AMP re-presented its methodology to the new CLFS public advisory board which held its first meeting on August 26, 2015 (here).  I haven't yet seen that presentation, the Gray Sheet report on the meeting was a little confusing based on what I already knew about AMP's proposals (Gray Sheet here, subscription).

What follows is, in my own words, my grasp of AMP's proposals.  For the final word, readers should work through AMP's original documentation (here).

I believe AMP provided three quite separate pricing narratives to CMS, and intended them to reinforce each other by being directionally similar.

Lab Direct Cost Survey Method

In the first method, AMP surveyed several labs that did at least one of three chosen codes:  81445, 5-50 tumor genes, 81430, hearing loss, 60 genes, and 81415, single exome.   The labs reported data in broad categories (such as labor versus chemistry cost) with a breakdown of many line items within each category (parts of the lab cost, parts of the chemistry cost).  There is also a small line item called "total validation maintenance overhead cost" but this is quite small, as low as 10%, so it can't represent the full overhead costs of running the laboratory, company, or hospital.  The summed direct costs of the various reporting labs were $577-907 for hotspots in 5-50 tumor genes, and $1890-1949 for 60  genes sequenced for hearing loss, and $1397-3388 for exome sequencing.

These are basically direct costs or what accountants call "costs of good sold."  I have long pointed out that for companies like Labcorp or Quest, with public financials, costs of goods sold are only about half of total costs, so these dollar figures would be less than the revenue required for doing the sequencing as what accountants call "a going concern."   
Procedurally in the current CMS meetings, AMP is appealing that the genomic sequencing procedure codes be "crosswalked" to other CLFS codes, so the raw cost data is not really intrinsic to their crosswalk agenda, just a collateral data point.

20% Rule of Thumb: Sanger to Next Gen Sequencing

In the second AMP pricing method, AMP states briefly that the costs of next-gen sequencing are about 20% of the costs of Sanger sequencing, so they estimated the number of exons in a large genomic procedure and then scaled that estimate down by 80% (multiply by 20%).  They then convert this number to dollars by using gene 81292 which has 21 exons and a CMS price of $644.   For example, if a genomic sequencing procedure such as 81410 has about 240 exons to sequence by NGS, they multiple 240 exons by 20% to get an equivalent amount of "work" by Sanger sequencing of exons.  Here, 240x20%=48.   So the payment for 81410 should be the same "as if" 48 exons were sequenced by Sanger sequencing.

We have a reference price for CMS payments for Sanger sequencing: 81292, 21 exons, $644.  So for every 21 exons in a genomic sequencing procedure (after the 80% mark down in the number of exons), pay $644.  This gives us 48/21 = 2.2, round it down to 2.  Multiply that two times $644 (price for 81292 for sequencing each 21 exons) and to account for the important bioinformatics, add two times 87900 (HIV bioinformatics), or two times $177.   This gives 2x$644 + 2x$177 or $1642.

Log-Natural Method

In the third method, AMP states that a good algebraic fit for the scaling costs of next gene sequencing is the log natural of the number of genes x $644, which is the index price of 81292 sequencing.  Recall that log-natural is a logarithm on a base of about 2.7.   Recall a "normal" log is a power of 10, so that log(100) is 2 and log (1000) is 3.   With a base of 2.7 instead of ten, log-natural (7.3) is 2 and log natural (25) is 3.

For NGS procedures, they apply the log-natural multiplier to both the base code for sequencing, 81292 $644, and the base code for bioinformatics (87900, $177).   So we get: for 60 genes, log-natural (60) is ~4.   So for 60 genes by NGS, pay 4x$644 and 4x$177, which is  $3284.

(This is my math - AMP doesn't dollarize its report).

For 5-50 somatic tumor genes, they take the median typical gene number to be 27, log-natural(27) is 3.3, so they multiple both $644 and $177 by 3 ($2463).   For 51 or more tumor genes, they take the median typical gene number as 100, log-natural(100) is 4.6 so multiply both $644 and $177 by 4.6 ($3776).

Finally for exome sequencing of 20,000 genes, log-natural(20,000) is 9.9, so multiply 9.9x$644 and add 9.9x$177, which is $8127.


CMS Process:
Most of the codes above are in the midst of a gapfill year in CY2015, which represents a CMS decision to gapfill the genomic procedure codes in late CY2014.  However, AMP also appealed that decision, so its comments are an appeal to administratively crosswalk the GSPs to the suggested values, even though at the same time, they are under a gapfill process today at MACs.

Log natural?
What the heck was log natural?  "The natural logarithm of (x) is the power to which (e) would have to be raised to equal (x), if (e) is about 2.718281."  2.718281 is a number that will lead to a wide range of special properties of the logarithm in calculus and trigonometry, just like pi results in a wide range of helpful equations related to circles and spheres.

Why log natural?
AMP states, for aortic dysfunction gene sequencing of nine genes, the 20% Sanger-to-NGS discounting along with AMP's conversion factor related to exon counts results in a genomic procedure pricing multiple of 2 (2 x $644 = 2 x 81292).   Since log natural of nine genes is also 2, one could simply use log-natural of the gene number.

Comparing: Linear, 20% Linear, SqRt, Log Natural, Log Base 10.

The table below looks at 1...20,000 genes with different assumptions - linear Sanger costs (x$644), 20% Sanger costs, SqRt costs (my own curiousity), Log Natural (AMP), and Log-10 (for comparison to log natural).   Table below.