KPSGILL policy proposal · model payer standard and comment proposal
Prior authorization: duration, and who actually decided
Two failures, usually discussed as one: approvals that expire on a calendar rather than on a clinical change, and denials whose real decision-maker is not the clinician named on the letter. Seven provisions, drafted.
The problem
Prior authorization has two failures, and they are usually discussed as one. The first is duration: an approval that expires on a calendar rather than on a clinical change forces a physician to re-justify a therapy the payer has already agreed to, for a patient whose condition has not changed. The second is attribution: as denial and pend decisions are increasingly produced or shaped by algorithms, the reviewer named on the letter is often not the decision-maker, and there is no way for the treating physician to tell.
The federal rule now sets timeframes and requires a specific reason for a denial CURRENT LAW · FINAL RULE, and public reporting of authorization metrics begins to expose behaviour that was previously invisible. Neither duty answers either failure. A denial can be timely, specific, algorithmically produced, and clinically wrong.
Documentary baseline on this site: the CMS interoperability and prior authorization final rule with its 2026 and 2027 dates; three state Medicaid managed-care parity audits, one of which found general compliance; and a California bill that would set a minimum authorization duration, unresolved at the end of session. Each carries its own primary source in the record layer.
The recommendation
Three provisions. A duration floor tied to clinical change rather than to the calendar; a named human decision-maker for any adverse determination; and disclosure, to the treating physician, of whether an algorithm participated in the decision and at what stage.
§ 1. Duration. An approval of a prescribed drug, device or service shall remain valid for the lesser of (a) the duration of the prescribed course of treatment, or (b) twelve months, and shall not require reauthorisation absent (i) a material change in the patient’s clinical circumstances, (ii) a change in the prescribed therapy, or (iii) a change in the terms of coverage effective for the plan year.
§ 2. Continuity on transition. An approval issued by a prior payer shall be honoured for ninety days following a change in coverage, absent a material clinical change.
§ 3. Named decision-maker. Every adverse determination shall identify, by name and licence number, the individual clinician who made it, and shall state that clinician’s specialty. A determination bearing no identified clinician is not an adverse determination for the purposes of the plan’s appeal timelines.
§ 4. Algorithmic participation. Where an automated system contributed to an adverse determination, the notice shall state (a) that it did, (b) at what stage — screening, recommendation, or determination — and (c) that the enrollee and the treating clinician may request the criteria applied.
§ 5. No automated final denial. An automated system shall not issue a final adverse determination of medical necessity. A licensed clinician of the relevant specialty shall make that determination and shall be accountable for it.
§ 6. Metrics. A payer shall report, by plan and by service category: approvals, adverse determinations, determinations in which an automated system participated, overturn rate on internal appeal, and overturn rate on independent review.
§ 7. Construction. Nothing in this standard requires the approval of a service that is not a covered benefit, or limits a payer’s authority to review for fraud.
§ 5 is the provision that matters most and is the least likely to survive negotiation. § 6 is the one that would change behaviour fastest, because an overturn rate published beside an automation rate is a self-auditing statistic.
Who bears what
Physician
The recurring administrative tax on stable patients falls away. § 3 restores the ability to have a clinical conversation with an identifiable peer.
Patient
Therapy continuity stops depending on paperwork cycles, and a transition of coverage stops being a treatment interruption.
Payer
Real cost, mostly in review capacity under § 5. The offsetting saving is in appeals, independent review and the utilisation caused by interrupted therapy.
Competition
Metrics under § 6 make authorisation behaviour comparable between plans, which it currently is not.
Equity
Reauthorisation failure falls hardest on patients with the least capacity to chase a plan — which is also the population where interrupted therapy costs the most.
Burden
Shifts from the treating practice to the payer, deliberately. The party imposing the process should carry its cost.
The strongest arguments against
- A twelve-month floor removes the payer’s ability to manage newly available evidence.
- § 5 outlaws automation that is faster and more consistent than human review.
- Naming reviewers exposes them to harassment.
- Metrics will be gamed by reclassifying denials as pends or as administrative closures.
- This raises premiums.
Answers
- § 1(iii) preserves plan-year coverage changes and § 1(i) preserves clinical change. What it removes is reauthorisation for its own sake.
- § 5 does not prohibit automation; it prohibits an automated final denial. Screening and recommendation are expressly permitted under § 4.
- Licence numbers are already public for every treating physician who signs an order. The asymmetry is the anomaly, not the disclosure.
- A serious version of § 6 counts pends and administrative closures as categories. That is a drafting problem, and it is answerable.
- Some. The honest answer is that continuity has a price and interrupted therapy also has one, and only the first currently appears on anyone’s ledger.
Metrics, sunset, open questions
Metrics. Reauthorisations avoided per 1,000 members; median days of therapy interruption at coverage transition; overturn rate on determinations in which an automated system participated, against those where none did; share of adverse determinations naming a same-specialty clinician.
Sunset. Five-year review. §§ 3–5 sunset if published metrics show automated determinations overturned at a rate no higher than clinician determinations — the point is accuracy, not a prohibition on software.
Open questions. Should the duration floor differ for high-cost therapies with genuinely evolving evidence? Should § 4 disclosure extend to the criteria themselves, or only to the right to request them? Does § 5 reach utilisation management performed by a delegated vendor rather than the plan?
Related in the record: the interoperability and prior authorization final rule · algorithmic prior authorization and human accountability · peer-to-peer review · claim denials and the appeals ladder · AB 539, unresolved at the end of session.