Sunday, May 5, 2024

Nerdy Blog: Gapfill 2024 & Algebra-AI for Genomic Sequencing Procedures (GSP)

 A few days ago, I produced a fairly standard assessment of the 29 codes under the "Gapfill" process in 2024.  Find it here:

Key takeaways are that the gapfill process for new genomic sequencing procedures (GSP) will be impactful, and that extra attention will probably be paid to the first-ever liquid biopsy GSP codes.  

This is the first shake-up to the GSP coding and pricing system in a decade.  Generally, the old system was simply with 5-50 tumor genes paying about $600 and 51 or more genes paying about $2900.

I also noted that there is probably more discrepancies among the MACs than ever before.  But regardless of how low or high other MACs set gapfill prices, the only prices that matter are MolDx's, because CMS sets final prices to the median, and that is controlled by MolDx.  Still, for the six genomic sequencing procedure codes at the top of the report, the MolDx prices add to $10,762, the Novitas prices to $19,741, and the NGS MAC prices to S11,203.


In this new blog, we use Chat GPT to analyze the MAC prices across the GSP codes.

First, I took the GSP codes under gapfill, and put them in a single table with pricing offered by MolDx, by Novitas, and by NGS MAC.  

Next, I added columns for the key variables (features) of the new codes, copying this from an AMA CPT table.   For example, some codes include Tumor Mutation Burden, some don't.

Next, I assigned algebraic letters to each test feature, as follows.

  • A, DNA only.
  • B, DNA-RNA.
  • C, SEQ
  • D, copy number variation.
  • E, MSI.   
  • F, TMB,   
  • G, rearrangements.
  • H, liquid biopsy format.

That results in the following equations:

  • A+C+E=897 81457
  • A+C+D+E=1046 81458
  • B+C+D+E+F+G=$2990 81459
  • H+B+C+D=$1196 81462
  • H+A+C+D+E=$1345 81463
  • H+B+C+D+E+F+G=$3290 81464
click to enlarge

This is 6 equations in 8 unknowns, which does not have a general solution.  However, you can still solve for some variables, or elucidate contradictions.  Basically, what CHATGPT did is exactly the same as a kid solving A+B=10 and B=6 therefore A=4.  In the larger system just shown, ChatGPT comes to this impasse:  It appears that the system of equations is overdetermined, meaning that it has more constraints than variables to solve for. This could imply either a contradiction in the equations or the need to remove redundant equations.

Find Google Cloud Spreadsheet here [grid]:

Find Chat GPT Algebraic Analysis here [text]:

Here is a summary lifted the extensive CHATGPT4 analysis to try and parse the MolDx prices vs the AMA CPT features.  (Read, or skim, the GPT4 analysis at the prior link.)


GPT Says:

Having more variables than equations generally leads to an underdetermined system, which often has infinitely many solutions or might not have any. However, even in such cases, as you mentioned, it's possible to make inferences about the relationships between the variables and detect inconsistencies if they exist.

Key Learnings from the Equations Today:

Finding the Relationships:

We were able to deduce that 

𝐷 = $149 (CNV)

A + E = $149 also. (DNA anal. plus MSI)

H= $300.  (LBX)

Spotting Redundancies:

We identified that some equations were redundant, which happens when one equation can be derived from another or from a combination of others.

Highlighting Inconsistencies:

By simplifying the equations, we found that some results were contradictory, indicating an inconsistency in the given system.

For the Student: Practical Approach:

In cases where a system of linear equations has more variables than equations, it's useful to:

  • Solve for Dependencies:
  • As we did, find relationships between variables to reduce the number of unknowns.
  • Check for Redundancies:
  • Test whether some equations merely repeat information already expressed in other equations.
  • Look for Inconsistencies:
  • Simplify and combine the equations to see if contradictions arise, which can highlight any errors or issues.
  • Conclusion
    • Even without explicit solutions, you can gain insights into the relationships between variables and test for inconsistencies or redundancies, which is a valuable approach when faced with similar problems in the future.

click to enlarge


The equations with MolDx prices, could be replaced with NGS MAC prices, but they are generally very similar (not identical, though) to MolDx prices.   the MolDx prices could be replaced with Novitas prices, but the latter show very little "spread" among the 6 codes.