The Medicare Date of Service rule defines "date of specimen collection" as the "date of service" - for inpatients, unless the test is ordered over 14 days after the hospital discharge.
The way it works is this: The patient is admitted July 1, has surgery July 2, is discharged July 6. A molecular test is ordered ten days later (July 16). This July 16 is less than 14 days after July 6, the "DoS" is July 2 - the date of specimen collection. Since "the date of service" (!) is during the inpatient stay, between the inpatient date of admission and date of discharge - the service is not payable; it is included in the DRG which covers all services from date of admission to date of discharge.
But: Does this make sense? What if Medicare law defines inpatient stay or inpatient services as those which occur between the time of admission and the time of discharge? Is it legitimate then for the little DOS regulation to throw an event (the molecular test) weeks into the past, before the test was even ordered?
CMS has touched on this in Medicare Advantage policy. CMS found that Medicare Advantage plans were rejected good prior auth paper work because it was signed after the time of testing (e.g. that past date of service fiction). CMS reprimanded such plans for bad behavior, and noted the DOS regulation does not actually move real events into the past. (91 FR 20014, 4/14/2026, Some plans may improperly rely on the Part B laboratory date of service policy...the MA plan may deny the laboratory's prior authorization on the grunds that the DOS has already occurred, citing the DOS policy...it is wholly inconsistent with 42 CFR 422.138(b) to apply the DOS policy ...to deny a request...for services that have not yet been performed." Similar, it is a fiction that the molecular lab test weeks in the future, took place, weeks in the past, during the span of time that legally defines the inpatient interval.
How shaky are the foundations on which DoS billing has been built? I asked Chat GPT to write a detailed policy memo on the topic. PDF link provided.
Report >> here.
The main research run took AI 1m46sec. I then reviewed for feasability, problems, surprises.