PwC's 9% Group Medical Cost Trend for 2027 and the Rate Filing Benchmark Problem

A pricing actuary should treat PwC's 9% composite group medical cost trend for 2027 (PwC, June 2026), the steepest since 2010, as a credibility-weighted input, not a selection. For a 20,000-member group with two years of data, a Buhlmann-Straub blend lands roughly 60% of the weight on that benchmark and the rest on the plan's own regression.

The "Medical Cost Trend: Behind the Numbers 2027" report, published in June 2026, draws on surveys and interviews with pricing actuaries at 27 U.S. health plans covering more than 103 million employer-sponsored members and 8 million individual ACA marketplace enrollees (PwC, June 2026). The more revealing data point sits alongside the 9% headline: PwC simultaneously revised the 2026 group trend upward from 8.5% to 9.0% and the individual market trend from 7.5% to 8.5% (PwC, June 2026). A benchmark that revises itself upward between publication cycles is signaling that development on 2025 and 2026 accident months is running worse than the prior estimate assumed, which carries directly into how a 2027 experience period should be weighted.

The timing is not academic. Connecticut's 30-day public comment period for preliminary 2027 ACA filings opened June 5, 2026, with four carriers averaging a 15.7% individual market request (ACA Signups, June 2026); Nevada's SERFF deadline fell June 17. Trend selections are being locked against this PwC publication in real time, and what follows is a working method for converting the benchmark into a defensible, component-level, plan-specific selection.

PwC 2027 Group Medical Cost Trend Projection, Highest Since 2010

5th Year

Consecutive Year Group Trend Has Held Above Historical Norms

85%+

Share of Plan Actuaries Expecting 2027 Pharmacy Trend to Outpace Overall Medical Trend

What the 9% Actually Measures

The composite is a weighted average across 27 plans and five distinct inflator categories: AI-enabled provider documentation and coding optimization, GLP-1 and oncology drug spending, behavioral health utilization growth, provider reimbursement pressure from No Surprises Act disputes, and general unit cost inflation from fee schedule renegotiations. Each lands differently depending on network structure and benefit design: a closed physician network with robust prior authorization absorbs the AI coding effect very differently than an open-access PPO with weaker code edits. The 9% is a starting point, not a selection.

The pharmacy signal inside it deserves isolating first. More than 85% of surveyed plan actuaries expect 2027 pharmacy trend to outpace the overall medical composite (PwC, June 2026), driven primarily by GLP-1 prescriptions, whose volumes nearly doubled across the commercially insured population between December 2024 and December 2025, with oral formulations entering market in 2026 below $200 monthly. A plan running a log-linear fit on 36 months of total pharmacy PMPM without separating GLP-1 claims is fitting a curve whose slope changes partway through the window, and the resulting trend understates the forward run rate unless that acceleration is modeled.

Building the Plan-Specific Experience Trend

The credibility exercise starts with a clean experience estimate. Using 24 to 36 months of allowed charge data adjusted for benefit changes and enrollment mix, the actuary runs a log-linear regression of per-member per-month (PMPM) costs against time, producing an annualized exponential rate (e raised to the slope coefficient, minus one). Three data quality checks precede the fit: continuity of the claim runout basis, since switching from paid-through to incurred-through mid-series distorts the slope; removal of large claimants above a chosen threshold if the plan is crediting stop-loss; and confirmation that enrollment has not drifted toward a higher-morbidity population, which would show acceleration that reflects mix rather than cost inflation.

The regression period itself is a judgment call that PwC's mid-year revision sharpens. An equal-weight regression over 36 months smooths the last four to six quarters against three years of lower-trend data, so if 2026 development is adverse, the fit lags reality; a 24-month window with exponential weighting toward recent quarters captures that deterioration and is defensible when documented with its R-squared and residual pattern. Suppose the fitted annual trend comes in at 6.5% against PwC's 9%. The actuary now confronts the credibility question directly: how much of that 2.5-point gap is genuine plan-specific performance, and how much is sampling variance in a mid-size book?

Credibility Weighting the Benchmark

The standard framework for blending plan-specific and industry experience in health pricing is Buhlmann-Straub credibility, which weights the plan's own data by its volume relative to the within-plan versus between-plan variance ratio. The formula is Z = n / (n + k), where n is the plan's exposure in member-months and k is the variance parameter estimated from a multi-plan dataset. The selection is then:

Selected Trend = Z × (Plan Trend) + (1 - Z) × (Industry Benchmark)

For a 20,000-member group with two years of data, n is roughly 480,000 member-months. If the between-plan variance in the reference dataset, whether the PwC 27-plan survey population or a peer database the actuary maintains, implies a k of about 860,000, then Z equals 480,000 / (480,000 + 860,000), or roughly 0.36. The complement, 0.64, is external benchmark weight. The blend gives 0.36 times 6.5% plus 0.64 times 9.0%, or 8.1%, well above the plan's own experience but grounded in the reality that a 20,000-member book cannot reliably tell real low-trend performance from random favorable experience.

The k parameter is the critical judgment here, and it comes from the variance structure of a multi-plan dataset, not the plan's own data. PwC does not publish the between-plan variance directly, so it must be inferred from the dispersion of plan-level trend reports across the survey or from a proprietary peer database; the Milliman Health Cost Guidelines and the SOA Getzen model both offer inputs for the calibration. Because the complement weight exceeds 60%, a swing in k from 600,000 to 1.1 million moves the selection by several tenths of a point, so the value used, its source, and the selection's sensitivity to alternatives all belong in the filing.

Disaggregating the Composite by Service Line

An 8.1% blend applied uniformly across inpatient, outpatient, professional, and pharmacy will be wrong in every category, because the 2027 cost structure is unusually dispersed by service line. Pharmacy trend for plans with significant GLP-1 utilization is likely running 12% to 15% gross of rebates, while inpatient is more plausibly 5% to 7%. A flat 8.1% overstates inpatient and understates pharmacy, and where pharmacy already represents 30% of total allowed costs, that misallocation compounds into pricing inadequacy. Running the same credibility framework at the service-category level, using component benchmarks from PwC's supplemental tables, the 2026 Milliman Medical Index, or a peer database, fixes this; the PMPM-weighted sum of component selections must then reconcile to the 8.1% composite.

Pharmacy needs a further split. GLP-1 trend should be isolated and projected from the plan's NDC-level claims; as our GLP-1 trend factor analysis describes, a logistic adoption curve captures the non-linear acceleration better than a linear regression. The remaining specialty pharmacy trend (oncology biologics, rare disease agents) and traditional drug trend (generics, branded non-specialty) carry different slopes and rebate offsets, so a composite pharmacy selection that blends these three sub-components will be incoherent while GLP-1 adoption is still in the steep part of the S-curve.

Stripping the AI Coding Step-Change Out of the Slope

Of the five inflators, the AI-enabled documentation and coding effect demands the most careful regression treatment. Roughly 70% of surveyed plans ranked AI coding tools among their top three inflators, and about 20% named it the single largest driver (PwC, June 2026). Revenue cycle vendors have deployed machine learning that optimizes evaluation-and-management (E&M) code level selection, lifting average allowed per claim without a proportionate rise in care intensity. Where networks adopted these tools during the estimation window, the historical regression mistakes a one-time code-level migration for a permanent trend rate.

The test uses E&M level distribution data. Pull the plan's professional-services claims and examine the split across CPT codes 99213, 99214, and 99215, the three most common office visit levels, before and after the adoption period. If the share of 99215 visits rose from 18% to 28% over 18 months while encounter frequency held flat and the contracted fee schedule increased only 3%, that 10-point migration is a coding precision effect, not an acuity shift. A normalization factor on the professional-service PMPM before fitting removes the step change from the slope: if the shift raises average allowed per encounter by 8%, the factor on post-adoption claims is about 0.926 (1 divided by 1.08). Because this is a judgmental modification of the data, the filing memorandum should disclose the adoption period, the pre- and post-period E&M distributions, and the adjustment.

The Individual Market Needs a Morbidity Loading on Top

The 8.5% individual market trend explicitly excludes adverse selection from the expiration of enhanced premium tax credits. The Congressional Budget Office estimated that ePTC expiration adds 4.3% to gross benchmark premiums in 2026 and 7.7% in 2027 (CBO), and the Georgetown Center on Health Insurance Reforms' early filing analysis points to a second consecutive year of double-digit marketplace increases (Georgetown CHIR, 2026) as carriers price the enrollment composition shift. The 8.5% addresses cost per member; a separate morbidity loading addresses the change in who the member is as subsidy-sensitive enrollees exit.

That loading requires an assumption about the elasticity of enrollment with respect to net premium. Plans with data from the 2018 to 2019 ePTC reduction can estimate a plan-specific elasticity, though the 2026 environment differs. Wakely's morbidity shift analysis, covered in our Wakely morbidity methodology article, frames adjustment factors of roughly 2.9% to 6.5% above baseline trend depending on enrollment assumptions (Wakely), and the structure is additive rather than multiplicative on the cost trend.

No Surprises Act Disputes Are a Separate Out-of-Network Trend

The 2026 clarification of the NSA independent dispute resolution payment methodology produced a sharp rise in IDR dispute volumes, which PwC's survey named as a contributor to the 9% composite. As arbitrators apply the qualifying payment amount framework, effective out-of-network cost allowances are ratcheting upward, particularly in emergency medicine and anesthesiology. A plan that does not track OON cost trend separately from in-network medical trend risks understating the prospective allowance; the fix is to run OON as a discrete component on plan-specific cost-sharing data, fitting separate trends for emergency facility, non-emergency facility, and professional OON services. In markets where IDR adoption has been highest, OON trend may be running 15% to 25%, well above the in-network composite, and folding it into a single medical line obscures the exposure concentration.

The Behavioral Health Saturation Question

Behavioral health utilization has grown 62% since 2018 as mental health parity enforcement under MHPAEA expands reimbursable services (PwC, June 2026), and the 2027 selection hinges on whether the plan sits in the acceleration phase or near a plateau. A plan that adopted telehealth parity in 2022 and saw utilization spike 40% in the following year may now run a lower incremental trend as that level stabilizes, while one that has not yet implemented parity across all covered services may still be accelerating. The two call for materially different forward selections, so the filing should document the factual basis for the actuary's placement on this curve. State regulators processing 2027 filings this summer increasingly request service-category documentation: a submission with a disaggregated build, a credibility blend showing sourced k and n inputs, explicit treatment of the AI coding effect, and a parity-utilization rationale will withstand review, where "trend assumption: 8.5%, consistent with industry benchmarks" will not.

What the Selection Has to Withstand

The mid-year revision to 2026 trend is the most actionable signal in the release: when a survey covering 103 million members lifts the current year by 50 basis points, it is external evidence that recent quarters should carry more weight than a flat regression gives them. And none of the drivers behind the 9%, the highest projection in 17 years, looks like a one-year cyclical bump. GLP-1 adoption is still climbing its logistic curve, provider coding optimization tools are not being unwound, parity enforcement is expanding, and IDR volume will stay elevated as the payment methodology settles. A 2027 selection built on mean reversion toward a historical 5% to 6% norm is pricing against the cost drivers actually running in the data, while a Buhlmann-Straub blend with appropriate weight on the PwC benchmark and proper component-level disaggregation produces a number that can be defended to a regulator, explained to a client, and revisited with next year's data without embarrassment.