The CDC's National Center for Health Statistics put the 2025 US age-adjusted death rate at 689.2 per 100,000, down 4.6% from 722.1 in 2024 and the lowest figure on record (CDC/NCHS, July 2026).

For life, annuity and pension actuaries the number that matters is not the record. It is that this is the second consecutive annual decline, arriving on top of a projection scale most reserving models have not moved.

Key Takeaways

  • 689.2 deaths per 100,000, down 4.6% from 722.1 and the lowest age-adjusted rate on record, following 2024's own 3.8% fall from 750.5 in 2023.
  • An estimated 70,000 overdose deaths, down 14% and the third straight annual decline, doing a disproportionate share of the work inside the headline figure.
  • 0.78% is the annual improvement cap SECURE 2.0 places on the pension mortality basis, roughly one sixth of the single-year population move, with a zero-improvement window still held for 2020 through 2023.
  • Understated improvement means understated liabilities. An annuity or defined benefit promise projected on a scale slower than realized gains carries too short a duration and too low a present value.
  • Overdose deaths are 25% to 30% of all deaths at ages 25 to 39, so the gain lands unevenly: a young term book benefits where an older permanent book, paced by cardiovascular and cancer mortality, does not.

The Record, Decomposed

Vital Statistics Rapid Release Report No. 44, based on 99.9% of 2025 death records processed as of May 10, 2026, counted 3,094,593 total US deaths, with rates falling for every age group and both sexes. Farida Ahmad, a CDC health scientist and co-author, credited a continued decline in fatal drug overdoses as a primary driver.

Men died at 811.1 per 100,000 against 582.9 for women, a 228.2-point gap that narrowed only slightly from 2024 and remains the widest sex differential among the major mortality inputs actuaries track. Heart disease stayed the leading cause at 694,708, cancer at 622,832 and unintentional injuries at 184,265, while influenza and pneumonia climbed from eleventh to eighth at 56,511.

Cause of death2025 deaths (provisional)Rank
Heart disease694,7081
Cancer622,8322
Unintentional injuries184,2653
Influenza and pneumonia56,5118 (up from 11)

The overdose component is the clearest driver. An estimated 70,000 Americans died of drug overdoses in 2025, a 14% drop and the third consecutive annual decline, with fentanyl, cocaine and methamphetamine deaths all falling even as Arizona, Colorado and New Mexico ran against the national trend.

That is a continuation rather than a turn. The same trend already lifted US life expectancy to 79.0 years in 2024, up 0.6 year from 78.4 in 2023, with the male-female gap narrowing to 4.9 years (NCHS Data Brief 548, January 2026). The 2024 age-adjusted rate itself fell 3.8% from 750.5. Two back-to-back improvements of comparable size, not a single post-pandemic rebound followed by a plateau.

A Capped Scale Against Two Years of Population Improvement

That pattern lands on an open question in improvement modeling. The SOA's 2026 Mortality Improvement Model extended the underlying dataset through 2023 but left the MP-2021 projection scale unchanged, on the view that post-pandemic data was insufficient to justify a new scale. The 2025 release supplies a data point that decision did not have.

The regulatory consequence is mechanical. IRS Notice 2026-27 sets the mortality tables qualified defined benefit plans must use for 2027 minimum funding and lump sums under IRC section 430(h)(3)(A). That framework, covered in the 2027 DB mortality tables analysis, builds from the Pri-2012 base tables, applies MP-2021, caps annual improvement at 0.78% under SECURE 2.0, and holds a zero-improvement window for 2020 through 2023.

The zero window was a reasonable response to a pandemic that killed disproportionately among older adults. It was not designed to persist once population mortality resumed improving faster than the capped assumption two years running. A 4.6% single-year drop in the age-adjusted rate is roughly six times the 0.78% the pension basis may reflect, before compounding a second comparable decline.

For annuity and pension liabilities the direction of that gap runs one way. A single-premium immediate annuity or a defined benefit promise pays for as long as the annuitant lives, so a scale assuming slower gains than are being delivered projects the liability's expected duration too short and its present value too low.

Selection sharpens it rather than softening it. The 2020-2024 Individual Payout Annuity Mortality Experience Study found the same tension in insured annuitant data. An insured annuitant pool is healthier and wealthier than the general population by underwriting, so if the general population is outrunning MP-2021, the subset the scale is meant to project has every reason to be doing so too.

Pricing sits on the same assumption. LIMRA recorded single-premium pension risk transfer sales jumping 132% in the fourth quarter of 2025 to $28 billion, then falling 47% year over year to roughly $3.8 billion across 102 contracts in the first quarter of 2026. A carrier pricing a 2027 buyout off a table that has not absorbed two years of confirmed improvement is underpricing the longevity risk it assumes, and the shortfall only emerges as the block runs longer than reserved.

What the Population Number Is Not

A population death rate and an actuarial improvement scale are not the same measurement, and the distinction limits how much weight the CDC figure can carry directly. VSRR No. 44 reports a crude age-adjusted rate across the whole population, blending true cohort-level decline with compositional shifts. The overdose drop is concentrated among ages 25 to 54, which changes which age groups are dying at what rate rather than slowing mortality uniformly at every age.

MP-2021 projects age-and-sex-specific factors precisely to isolate the cohort effect from that cause-mix noise. No single population statistic is a ready-made replacement improvement factor. What is harder to dismiss is the order of magnitude: a capped scale sitting that far below two years of delivered population improvement surfaces in an experience study eventually rather than staying a modeling footnote.

The gain is also not uniform across products. Overdose deaths account for roughly 25% to 30% of all deaths among ages 25 to 39, with the average age at overdose death about 30 years younger than all-cause mortality (SOA, 2025). RGA claims data in the same report showed insured overdose claims peaking in 2021 and declining since, concentrated among smokers, policies under $100,000 face and later durations.

A term book weighted to that age-and-face profile has real improvement upside from the overdose decline. A book weighted toward older, larger-face permanent business does not, because cardiovascular and cancer mortality set its pace and both fell more modestly than the overdose-specific number. The 2025 Individual Life Mortality Improvement Scale found the picture "mostly, but not entirely, normalized," with residual excess mortality still visible at ages 65 and older.

The regulatory calibration inherits the same ambiguity. The NAIC's C-2 longevity risk RBC framework is a scenario-based stress meant to capture exactly the case where realized improvement outpaces the pricing assumption for several years running, targeting year-end 2027. Its field test is the first occasion to see whether that stress is calibrated against the pattern the population data now shows or against the baseline still embedded in MP-2021.

Further Reading

Sources