How the numbers are made.

1  The model & the exact formulas

No black box. Each indication carries a hazard ratio (HR) for death per 4 weeks (28 days) of treatment delay, taken directly from a peer-reviewed meta-analysis — most centrally Hanna et al., BMJ 20201. For a delay of d days we assume the hazard scales smoothly with time and compute three quantities:

A worked example: head & neck radical radiotherapy has HR = 1.09 per 4 weeks. A 14-day delay gives RR = 1.09^(14/28) = 1.09^0.5 ≈ 1.044, i.e. an estimated +4.4% relative increase in mortality risk. Applied to a baseline mortality of 20% across 1,000 delayed patients, that is 0.20 × 0.044 × 1,000 ≈ 9 estimated excess deaths.

2  From four-week to per-day risk

The published evidence reports mortality hazard ratios per four weeks (28 days) of delay, but a real-world denial or appeal can stall treatment by any number of days. To bridge that gap, the calculator converts the per-4-week hazard ratio to any delay length using a log-linear interpolation:

Equivalently, the log of the relative risk is assumed to grow linearly with the number of days delayed: ln(RR) = (days/28) × ln(HR). This is a constant-relative-hazard assumption — each additional day of delay multiplies the hazard by the same factor — and it is the assumption made explicitly by Hanna et al., BMJ 20201, whose meta-analysis modeled mortality as rising in a continuous, per-unit-time fashion with each four-week increment of delay.

3  The source data: every indication

Every coefficient the calculator can apply is listed below, with its hazard ratio per 4 weeks, the published 95% confidence interval, whether the interval excludes 1.0 (statistical significance), and a link to its source study. Nothing is hidden — including one indication whose effect is not statistically significant, kept in deliberately so the tool cannot be accused of cherry-picking.

Source data loaded from data/indications.json. If this list looks empty, the static fallback table is shown instead.

4  How the confidence intervals are derived

We do not invent uncertainty bounds — we carry through the ones each source study published. The lower and upper 95% confidence limits of the hazard ratio are run through the same exponential formula as the point estimate:

Because x(d/28) is monotonic in x for a fixed positive delay, applying the transform to the published HR bounds yields a valid 95% interval for the modeled relative risk at that delay. This is a deterministic re-expression of the source interval, not a re-estimation: we add no statistical power and claim none.

5  Dashboard data provenance (CMS-0057-F)

The Insurer Dashboard reports prior-authorization metrics — denial rates, appeal-overturn rates, decision turnaround — drawn from insurers' own public regulatory filings, not from any private or estimated figure of ours. The backbone is the CMS Interoperability and Prior Authorization Final Rule, CMS-0057-F4, which requires Medicare Advantage organizations, Medicaid/CHIP fee-for-service and managed care, and Federally-Facilitated Exchange QHP issuers to publicly post their prior-authorization metrics annually beginning March 31, 2026 for calendar year 2025.

How each figure is sourced and what varies

• Aggregate Medicare Advantage figures come from KFF's analysis of CMS-reported MA contract data (published Jan 28, 2026, covering CY2024).5
• Plan rows marked Verified come from payer-hosted prior-authorization metric disclosures such as Cigna, Humana, UnitedHealthcare, and Aetna pages/PDFs. Third-party aggregates, including AuthDenied, are used only as discovery or benchmarking when explicitly labeled.6
• Program-integrity findings (e.g. that 13% of a 2019 sample of denied requests met Medicare coverage rules) come from HHS-OIG reports.7

6  What the dashboard data can & cannot show

The dashboard is built on payer transparency disclosures, which are powerful but genuinely limited. We separate what these data legitimately support from what they cannot.

What CMS-0057-F establishes

• Mandatory decision timeframes. Under CMS-0057-F4, affected payers must decide expedited (urgent) prior-auth requests within 72 hours and standard requests within 7 calendar days (effective Jan 1, 2026).
• A denial-reason requirement. Payers must provide a specific reason for denials, which over time creates a more auditable record than today's opaque practice.
• Public annual reporting. The first public prior-auth metrics are due March 31, 2026 (covering CY2025), giving the first standardized, regulator-mandated dataset to compare across plans.

What these disclosures cannot yet show (KFF limitations)

KFF's analysis of existing Medicare Advantage prior-auth data5 documents how far short current disclosures fall:

• Disclosures are hard to find and not consistently located, making systematic collection laborious.
• Reporting is inconsistent across plans and years — differing definitions, denominators, and time periods mean raw plan-to-plan comparison is imperfect.
• Figures are aggregated across services, so a high-stakes oncology denial is pooled with routine, low-risk requests and cannot be isolated.
• Prescription drugs (Part D) are excluded from these prior-auth counts.
• Denial reasons are largely absent in the legacy data, so the disclosures cannot say why care was withheld or whether it was clinically appropriate.

7  A two-layer model & research roadmap

This project is deliberately built in two layers. The first is what you see today; the second is the only credible path to insurer-specific harm estimates.

The line between the two layers is the line we never cross casually: Layer 1 reports association-level risk and transparency metrics; only Layer 2, with consented patient-level data and ethical oversight, could responsibly support claims about a specific payer's contribution to specific harm. Independent work by Johns Hopkins Medicine (2025)9 finding measurable patient harm linked to prior authorization is what motivates building Layer 2 properly.

8  Build order

The sequence is chosen so that credibility is established before reach. Each step depends on the one before it:

1. Methodology first. Publish the full, checkable model and its limitations before any number is shown — this page.
2. The oncology calculator, built on Hanna et al., BMJ 20201 and condition-specific delay studies.
3. A clean CMS prior-authorization database, normalizing the CMS-0057-F4 disclosures and existing KFF5 aggregations into a consistent, sourced dataset.
4. A plan transparency score — rating how openly and consistently each plan discloses its prior-auth metrics. This is a measure of disclosure quality, explicitly not a "deaths caused" ranking.
5. Consented patient and clinician story intake, capturing lived experience with appropriate consent.
6. Practice partnerships with IRB approval and data-use agreements — the Layer 2 research step that links real delays to real outcomes.

9  Assumptions & limitations

Stated plainly, so a reader, journalist, or critic can weigh every claim:

• Association, not individual causation. The model estimates a population-level association between delay and mortality risk. It is not a prediction about any individual patient and does not establish that a delay caused any specific death.
• The source studies are observational. The underlying meta-analyses pool observational cohorts that adjust for confounders to varying degrees; residual confounding cannot be excluded. Hazard ratios are read as published.
• Baseline mortality varies by stage and patient. The estimated-excess-deaths figure is only as good as the baseline mortality entered; that baseline differs enormously by cancer stage, patient, and outcome window.
• Smooth scaling is an assumption. We assume the hazard scales smoothly as HR(d/28) across the delay window. The source measured effects per 4-week block, so long extrapolations beyond the studied range are less reliable and are treated as extrapolations.
• Evidence is strongest in oncology. These coefficients come from cancer-treatment-delay studies and should not be exported to unrelated conditions.
• One non-significant indication is kept for honesty. Breast adjuvant radiotherapy (HR 0.98, 95% CI 0.88–1.09) shows no statistically significant effect and is retained deliberately, so the dataset is not cherry-picked toward harm.

10  Legal & ethical framing

Prior Authorization Accountability is an independent, public-interest data-journalism project. Three rules govern everything we publish:

• Estimates are labelled as estimates. Every modeled figure is presented with its confidence interval and full methodology. Statistically non-significant indications are flagged; extrapolations beyond the studied conditions are marked; any figure without a working source is labelled "[illustrative — replace with sourced data]."
• Built on insurers' own published data. Accountability claims rest on metrics the plans themselves publish under CMS-0057-F and on peer-reviewed dose-response studies — characterizing risk based on those disclosures, framed as our opinion grounded in the disclosed, sourced facts shown alongside it.
• We do not name individual deaths or assert a company "caused" them. Nothing here asserts that any named company, plan, or person "killed," "murdered," or is legally responsible for any specific death. We characterize estimated excess mortality risk, not blame for individuals.

How the generated report should be used

• Peer-to-peer: use the read-aloud statement as an urgency script for the medical reviewer; keep the references at the end of the packet.
• Appeal / expedited review: pair the documented treatment timeline with the plan's own published prior-authorization metrics and the selected oncology delay literature.
• Legal review: use the packet as an evidence map and record-preservation aid, not as standalone proof of individual causation.
• Policy / lawmaking: use the memo to show why plan-wide PA metrics are insufficient unless regulators also require oncology service-level timestamps, denial reasons, appeal outcomes, and treatment-start delay reporting.

Calculator disclaimer

11  Full reference list

• Hanna TP, King WD, Thibodeau S, et al. Mortality due to cancer treatment delay: systematic review and meta-analysis. BMJ 2020;371:m4087. doi:10.1136/bmj.m4087
• Ungvari Z, et al. Treatment delay significantly increases mortality in colorectal cancer: a meta-analysis. GeroScience 2025;47(3):5337–5353. doi:10.1007/s11357-025-01648-z
• Ungvari A, et al. Quantifying the impact of treatment delays on breast cancer survival outcomes: a comprehensive meta-analysis. GeroScience 2025. doi:10.1007/s11357-025-01719-1
• CMS. CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F) — fact sheet and public prior-auth reporting requirements. cms.gov
• KFF. Medicare Advantage Insurers Made Nearly 53 Million Prior Authorization Determinations in 2024 (Jan 28, 2026; CY2024 data). kff.org
• Cigna. CMS Annual Prior Authorization Statistics: 2025. cigna.com PDF
• HHS-OIG. Some Medicare Advantage Organization Denials of Prior Authorization Requests Raise Concerns (OEI-09-18-00260, Apr 2022). oig.hhs.gov
• American Medical Association. 2024 Prior Authorization Physician Survey (8% of physicians reported a prior-auth-related patient death or permanent disability). ama-assn.org
• Johns Hopkins Medicine. Researchers Find Measurable Patient Harm Linked to Prior Authorization (systematic review, 2025). hopkinsmedicine.org