Sunday, March 24, 2019

Have MolDx LCDs Gotten Better? A Look At Some Recent Ones

Summary:  I reviewed a set of recent LCDs from MolDx.   The overall quality of the writing and scientific review has gotten quite good.   The LCDs are detailed, typically over 3500 words, with over 25 citations.  There are elaborate objective tables summarizing test performance and clinical data.  It's a lot of work to produce a pipeline of documents of this quality on a wide range of clinical topics.

Update: See also a post on 8 even newer MolDx LCDs, March 28, 2019, here.


In business school, there is a of emphasis on "learning curves" - factories become more efficient with time, have fewer defects, or at least they should.  This is familiar to anyone with healthcare training - for example, surgeons get faster and better with fewer adverse events with experience.  This is especially prominent with the most complex tasks like robotic surgery.

Is this true for payers as well?

I have tracked pretty much every MolDx LCD since the program's inception, around 2012, and  I lectured on the most recent "block" of LCDs recently at a conference in SF.  (Here).  I'll summarize in this post, but the take-lesson is that yes, MolDx LCDs have gotten impressively well written.  
Part of this is learning curve, and part may be driven by recent legislation and by changes in CMS guidance.   CMS also provides in-service training to medical directors on LCD writing.    
The 21st Century Cures Act of December 2016 required more clarity in LCDs, which I discussed in a blog in November 2017.  Here.  At that time, I already showed a NGS MAC LCD which was characteristic of the improvement evidence discussions.    
More recently, in October 2018, CMS released more widespread changes in the LCD process which we are all still getting used to in CY2019.  Here.   Some of these are good, some may have unintended adverse effects.   Interestingly, some of the manualized changes in LCDs in October 2018, emerging from CMS, captured ideas floated earlier in legislation (not passed) by CAP and others, here.   
LCDs Selected for Review

These have been selected from Noridian's recently posted draft or newly finalized LCDs; here.  I've also put the PDFs in a zip file in the cloud, here.

Basic Statistics

I looked at six recently proposed LCDs.   The average word count is 3636, and the average number of citations 26.8:

3700 words is pretty long.  I wrote a big white paper (27pp, 7500 words) on the history of the CMS NGS NCD, and it was only about twice as long as some of these single-topic LCDs. 

Coverage or Non-Coverage?

All are basically coverage LCDs.  This is because MolDx jurisdictions have a master overlying LCD that says molecular tests are not covered until they are covered by MolDx (L35160, here).   So things that were proposed to MolDx, but not covered, may not get their own LCD.   
MolDx can and does issue specific non-coverage LCDs.   They recently issued an LCD retracting coverage of the CardioDx Corus CAD test, here, which by dollar volume had been amongst the largest proprietary tests paid by Medicare.  That was case where early Medicare coverage was not followed by much private payer coverage.
Overview of the Recent LCDs

Here are some summaries of the LCD features.  Whenever I am asked, "What is the evidence level for an LCD?" or "How many papers  do you need for an LCD?" I tend to roll my eyes, because almost every LCD presents some unique twists or features that are unique to its decision process.  I try to highlight a few of those below.  The opinions are entirely my own.

Structure of LCDs

The LCDs have a systematic structure, usually with a short coverage decision at the top (for patients with symptoms X, Y but not Z, etc.)

This is followed by an introduction summarizing the disease state, and then a detailed summary of the evidence.  

Summary of the evidence forms the body of the LCD, and for diagnostic tests includes both elaborate tables of test performance ("accuracy, PPV, NPV, FP, FN," etc) and tables summarizing the clinical validity or clinical utility studies reviewed in the text.  (E.g. "RCT, 125 patients, +3 months survival, P=0.01").  The overall approach gives an impression of thoroughness and accuracy. 

Professional guidelines are cited and summarized.  A closing section provides criteria for coverage (or non-coverage).   

The MolDx LCDs close with an extremely brief section called, "Analysis."   "Analysis" provides three words, Quality, Strength, and Weight, with a modifier for each (e.g. Quality:Moderate, Strength: Moderate, Weight: Moderate.")   CMS requires the LCD to include "Analysis" but I don't think CMS specifies the form.   This is the only part I would criticize.  (For more detail, see footnote.*) 

Update: New LCDs released by MolDx on March 28 still have the cryptic "Analysis" words like "Strength vs Weight" but now, they ALSO have a several paragraph analysis summary, that weighs the evidence strategically vis-a-vis the particular coverage decision. Good.

Veracyte Envisia Test for Idiopathic Pulmonary Fibrosis (L37891)

This is a test that helps resolve the diagnosis of idiopathic pulmonary fibrosis (IPF) in cases where a prior high resolution CT scan was non-diagnostic.  Unlike many of these LCDs, the coverage rules section is fairly long.  However, it follows a simple principle: If the CT is definitive for IPF, or unlikely for IPF, do not use the molecular test.  Only use the test when the CT is probable but indeterminate for IPF.  Guidelines and rules for this call, are provided.   Overall, the LCD flow and text are very clear. 

This is one of several LCDs in this short set they resolve an ambiguous situation.   Since the situation is ambiguous, it depends on the call of the reader of the high resolution CT scan.  It would be difficult for Medicare to audit such a call, without bringing in an equally qualified specialty radiologist (pretty thin on the ground at CMS or OIG), and having a copy of the original scan.  

Inivata InvisionFirst (LBx) (DL37899)

This is a gene panel liquid biopsy test for patients with advanced lung cancer.  Coverage is closely defined, for advanced lung cancer at progression, or advanced lung cancer where a tissue biopsy has failed or is not medically feasable for the patient.   The coverage maps closely to that provided in July 2018 for the Guardant G360 test.   The use of liquid biopsy in lung cancer has been increasingly recommended, with a number of high profile publications in recent months.   (Guardant, publicly held, has a market cap in the $7B range.)   

Coverage in some cases hinges on whether the treating clinicians view the tumor location and patient status as viable for biopsy.   The LCD also deals with cancer patients getting an NGS test, so it falls under the domain of the recent CMS NGS for uses of NGS in cancer.   Audit would involve determining whether Medicare's auditing physician also viewed the anatomy and patient as indicating a difficult to biopsy case.   

Myriad MyPath Melanoma (DL37881)

This LCD was recently finalized exactly as proposed.   This test is used on paraffin block biopsy tissue after a board-certified dermatopathologist has determined that the specimen is melanocytic, but equivocal for cancer, AND, the patient would otherwise have further surgery (surgical lesion wide excision or sentinel node surgery).   

Once again, as mentioned Veracyte Invisia, this LCD involves a human decision that a diagnostic test is ambiguous.  Here, it is a glass slide reading of the biopsy that is uncertain, and could only be audited by another board-certified dermatopathologist looking through the original glass slides and special stains of a case one-by-one, a difficult task.

Decipher For Prostate/Biopsy (DL37820), Biotheranostics BCI Expansion (DL37824)

Both of these involve expansion in coverage of a test.   Both are MAAA-type tests which are gene expression profiles of the likely aggressiveness of a cancer and the need for further intervention.   The Decipher test had originally been approved for use post-prostatectomy, to assess need for additional surgery; here, it is approved for use in decision-making on biopsy.   The BCI test is a breast cancer gene expression test, which also received an expanded indication for a larger population of women who are post-lumpectomy and making therapy intervention decisions such as regarding adjuvant chemotherapy.   

Usually, health technology assessments (including CMS NCDs) discuss the evidence first (1) objectively ("summary of evidence") and then (2) more qualitatively, meaning they turn to discussing in paragraph form the pro's and con's or biases ("Analysis").  

At MolDx currently, "Analysis" means these pop-up two-word phrases like "Quality:Moderate" which are terse and not very enlightening.  None of the three categories (Quality, Strength, Weight) are defined, nor the possible modifiers to each.  For example, what would be a "High Quality" body of evidence that is "Low Strength"?  Or how would you arrive at a body of evidence having"High Weight" but "Low Quality?"  We don't know.