An officer running radar on one side of a highway slows traffic on the other side too. Those drivers are not being clocked. Yet they brake anyway, because the officer is a reminder that somewhere, sometime, someone may be watching. Traffic researchers call this a halo effect. Field studies have found it miles beyond a visible police unit and for weeks after intensive enforcement ends. The officer’s radar will never show that slowing. Only a radar pointed the other way could.
Critics judge prior authorization (PA) by what the review catches — the denials, appeals, and overturns that get counted and reported. But PA can also work as a deterrent, discouraging overtreatment and inappropriate or low-value care before a provider ever submits an authorization request for a drug, test, procedure, or other health care item or service. The value of a deterrent is in what never happens — and what never happens leaves no record.
The numbers critics count
Critics of PA point to the visible record. A KFF analysis of CMS data found that in 2024, Medicare Advantage insurers made nearly 53 million PA determinations and denied about 4.1 million in full or in part. Just 11.5% of the denials were appealed, and of those appeals, 80.7% were partially or fully overturned. The reading is straightforward — more than 9 in 10 requests are approved, and most challenged denials collapse, so the requirement is friction that rarely changes the outcome.
A second set of numbers comes from the ACA marketplace, where a KFF analysis of CMS transparency data found that in 2024, HealthCare.gov insurers denied 19% of in-network claims — claims for care already delivered — and consumers appealed fewer than 1% of those denials. Critics cite these figures as evidence that denial is routine and appeals are too burdensome or futile to pursue. Paired with the Medicare Advantage numbers, they feed a broader story of a denial machine.
What those numbers cannot show
A high overturn rate is consistent with several explanations. The initial decisions may have been poor. Information may have arrived only on appeal. Providers may have appealed only their strongest cases. Aggregate data cannot say how much each explanation contributes to the overturn rate, so the 80.7% overturn rate cannot establish that a similar share of all denials was wrong — and it cannot show the initial reviews were right, either.
The ACA marketplace numbers describe something different altogether. They are claims denials for care already delivered, not PA determinations. Lack of PA or referral accounted for just 9% of reported denial reasons in 2024, and lack of medical necessity just 5%; the largest categories were administrative reasons at 25% and an unexplained “other” at 36% — and because a claim can generate more than one reported reason, these percentages are not mutually exclusive shares of distinct claims. Numbers like these cannot measure how accurately PA operates, and a low rate of consumer appeals against them is a claims statistic, not a PA statistic.
What happens when the watcher steps away
An approval rate measures the authorization requests that were submitted. It says nothing about care that was never ordered or prescribed because the requirement exists. Imaging that guidelines do not support, a brand drug before the generic was tried, or another marginal service may never generate a PA request at all. A more appropriate alternative may be chosen instead. Those decisions appear in no denial file or appeals report.
Payers believe that PA signals that someone is watching, and this is a useful deterrent of inappropriate care. Milliman’s actuaries call this deterrence the sentinel effect — authorization requests may never be submitted because providers expect a denial or find the administrative effort not worth it.
Milliman has modeled what removing PA could do. In a 2023 study commissioned by the Blue Cross Blue Shield Association, it estimated that eliminating PA in the commercial market could increase premiums by $43 billion to $63 billion annually — a mid-range increase of 4.8% for a broadly scoped program, with the full range running from 1.5% to 9.5%. The model does not quantify several effects that could move the estimate in either direction. Removing PA would remove its sentinel effect — the deterrence of some care before an authorization request is submitted — which, in turn, could make utilization and premiums rise more than what the model captures. But the increase could also be smaller: PA sometimes redirects care to a lower-cost alternative rather than eliminating it, and removing PA would reduce administrative costs. Because these effects pull in different directions, the estimate suggests that eliminating PA could materially increase premiums, but not exactly how much. The concern is not limited to payer-sponsored analysis: KFF’s Drew Altman likewise wrote in 2025 that eliminating PA could increase use of services that PA currently deters.
The most direct way to test a deterrent is to remove it — and Massachusetts has begun doing exactly that. Effective June 5, 2026, Massachusetts eliminated PA in fully insured plans for a wide range of services, including inpatient admissions, physical, occupational, and speech therapy; imaging to stage cancer or track its treatment; and medications for serious mental illness and certain chronic conditions. The regulations also prohibit PA for emergency, urgent, primary, and maternity care — areas where PA was already uncommon in practice.
The state’s purpose was to relieve burden, not to run a study. But the change creates an opportunity to measure what happens when PA is removed. The regulations anticipate the possibility that utilization could rise after removal. For several of the removed categories — including the therapies and certain chronic-condition medications — a carrier that can demonstrate a significant, risk-adjusted increase in utilization for two consecutive quarters may apply to reinstate limited PA for a designated period.
Changes in utilization in the affected services will be the earlier and more direct evidence of the removal’s effects. If the now-prohibited requirements were mostly friction, utilization may change little. If they were deterring care, utilization should rise. And if it does, the important question is what the increase consists of — appropriate care the requirements had prevented, or marginal and inappropriate care they had deterred. Distinguishing the two will require assessing the care that increased, not simply whether utilization rose. Premium effects, if any, will come later and be harder to isolate.
The costs are real
The visible costs of PA are real. Providers spend hours on submissions and appeals. Patients wait. And some denials are wrong. A 2022 review by the HHS Office of Inspector General found that 13% of a sample of denied Medicare Advantage requests met Medicare coverage rules — requests the plans should have approved. Bad PA can cause real harm, and nothing in the sentinel effect excuses a criterion that is stricter than the evidence. The corrective mechanisms — appeals, overturn tracking, criterion revision — are intended to identify and correct those criteria. A poorly aimed deterrent deters the wrong things. That is a reason for better PA — not, by itself, a reason to remove it.
A serious objection to the sentinel effect is that it deters good care along with bad — and it can. Delayed care, abandoned treatment, and serious adverse events are real harms. The AMA’s 2025 Prior Authorization Physician Survey shows how commonly physicians report these harms: 95% of physicians said PA delays access to necessary care; 79% said it can at least sometimes lead patients to abandon treatment; and 26% reported that it has led to a serious adverse event for a patient in their care. But those percentages measure the share of physicians reporting these experiences, not the rate of harm per authorization or per patient. The survey also does not capture potentially inappropriate care that was never delivered.
What honest measurement requires
The critics’ ledger counts the burden of every authorization request submitted and the harm of every wrongful denial, and counts nothing for any inappropriate or low-value care the requirement may have prevented, any premium increases that may have been avoided, or any more appropriate therapy that may have been tried first. Half a ledger is not an accounting.
Measuring PA honestly means measuring what it prevents. That requires counterfactuals — utilization before and after requirements are removed, markets with and without them, programs that exempt high-performing providers from review. It means netting the administrative savings of removing review against any utilization and premium increases that follow. None of this proves that review produces net savings. The claim that survives is more modest: approval rates, denial rates, and appeal outcomes, standing alone, cannot show that PA accomplishes little, because those measures cannot capture care that never reached the authorization process.
Every dropped requirement, in Massachusetts or wherever comes next, is a chance to measure what it had been doing. When officers leave a stretch of road, what the traffic does next becomes evidence — but only if someone measures it.