National e65 has risen about two years since 2000, but a handful of states show no gain at all — a material reserving risk if LTC pricing assumes one national mortality-improvement trend

Summary

Nationally, life expectancy at age 65 (e65) rose from 17.6 years in 2000 to 19.5 years in 2023 — a gain of about 1.9 years, or roughly one extra month of life per calendar year that passes. The path was not smooth: e65 climbed steadily to 19.6 years by 2019, then COVID-19 cut it to about 18.4 years in 2021 (a two-year setback), before it recovered to 19.5 by 2023 — essentially back to the pre-pandemic trend line.

Yes, the trend differs enough by state to matter for reserving. Comparing NCHS's 1999-2001 decennial state life tables to its 2022 state life tables (the latest full 50-state+DC vintage available), the state-level range widened slightly (3.8-year spread in 1999-2001 vs. 3.9-year spread in 2022, Hawaii highest and West Virginia/Kentucky lowest in both periods) — but the more consequential finding is in the CHANGE, not the level: ten states, including Arizona (-0.68 yr) and Oklahoma (-0.68 yr), show FLAT or slightly NEGATIVE e65 change over roughly two decades, while New Jersey (+1.50 yr), DC (+1.49 yr) and Maryland (+1.34 yr) gained more than the national average of +1.13 years over the same two endpoints. A reserving basis that applies one national mortality-improvement scale to a block of long-term-care business concentrated in a slow-improving or declining state will diverge systematically from one concentrated in a fast-improving state, and the gap (roughly 2 years of e65 divergence between the best- and worst-trending states) is large relative to the ~1.1-year national two-decade improvement itself.

The national trend, year by year

Source: CDC/NCHS National Vital Statistics System, annual United States Life Tables (Health, United States trend table and individual annual NVSR reports).

Two things matter for a reserving actuary reading this series: (1) the two-decade trend is real and positive — roughly 1.9 years over 23 years, a pace of about 0.08 years of e65 per calendar year, well within the range mortality-improvement scales (built from Social Security Administration general-population data, per Academy of Actuaries LTC valuation guidance) are designed to capture; and (2) the trend is NOT monotonic — a reserving basis calibrated naively on 2019-2021 data alone, without adjusting for the pandemic shock, would either overstate the post-pandemic reversion speed or understate the long-run trend depending on which single year got picked.

State variation: the finding that matters for reserving

NCHS publishes full state-by-state life tables only periodically at some vintages and annually at others. The two most complete, directly comparable full-state vintages spanning roughly the "last two decades" are:

• 1999-2001 (U.S. Decennial Life Tables, published as NVSR Vol. 60 No. 9): national e65 = 17.77 years; state range 16.61 (Kentucky) to 20.42 (Hawaii), a 3.81-year spread.
• 2022 (U.S. State Life Tables, 2022, NVSR Vol. 74 No. 12, published December 2025): national e65 = 18.9 years; state range 16.6 (West Virginia) to 20.5 (Hawaii), a 3.9-year spread.

Matching all 51 jurisdictions across these two vintages: the mean state-level change was +0.40 years with a standard deviation of 0.52 years across states — meaning the STANDARD DEVIATION of the state-level change is nearly half again as large as the mean change itself. The correlation between a state's 1999-2001 starting e65 level and its subsequent change was weakly negative (r ≈ -0.16), so this is not simply low-e65 states catching up (convergence) or high-e65 states pulling further ahead (divergence) — it looks more like idiosyncratic, state-specific drivers (opioid mortality, chronic-disease burden, COVID impact, healthcare access) dominating any uniform national trend.

Why this matters for LTC reserving specifically: Academy of Actuaries / SOA guidance for statutory LTC minimum reserves (the 2021 LTC Valuation mortality/lapse basis) builds mortality-improvement scales from Social Security Administration GENERAL POPULATION data applied uniformly, not state-specific improvement rates. LTC blocks are geographically concentrated (by where policies were originally sold, often correlated with where large career agent forces operated historically, e.g. more issue in the Northeast/upper Midwest). If a block is disproportionately weighted toward a state whose e65 trend has been flat-to-negative for two decades (Arizona, Oklahoma, Arkansas, Kansas, West Virginia, Missouri, Mississippi) while the improvement scale assumes the national ~1.1-year gain, the reserving basis is systematically miscalibrated for that block versus one concentrated in New Jersey, DC, Maryland, Vermont, Maine, New York or Massachusetts — the fastest-improving states. Either direction is a real basis-risk exposure that a single national scale cannot capture, and the size of the mismatch (roughly ±0.5 to ±0.7 years of e65 divergence from the national average, i.e. up to 40-60% of the entire two-decade national improvement) is not a rounding error.

Academic confirmation: a 2023 analysis ("Analyzing State-Level Longevity Trends with the U.S. Mortality Database," arXiv 2312.01518) is built on the same premise — that the United States Mortality Database exists specifically because state and regional mortality trends have diverged enough from the national trend to need their own standardized series, a research need the SOA has separately funded for exactly this application.

Data and methodology notes / limitations

• askamerica connector coverage gap: a search_catalog scan of this warehouse (schemas census, health, and others) for "life expectancy at age 65" returned no table carrying U.S. life-expectancy-at-age-65 by state. The closest matches were census tables carrying only 65+ population COUNTS (not life expectancy) and a WHO international life-expectancy-at-birth view (not U.S., not age 65, not state-level). No SQL query was run against askamerica tables for this analysis because none of the corpus's tables cover this measure — this entire analysis is built instead from primary CDC/NCHS publications fetched directly via the Wayback Machine (cdc.gov blocks non-browser fetches from this environment). This is a genuine, confirmed sourcing gap in the connector, not a query that came back empty by mistake.
• 2022 vs. 2023 mismatch: the national year-by-year e65 series above runs through 2023, but the full 50-state+DC comparison is only available through 2022 (the state-life-table series lags the national release by about a year and is not published every year). This means the state comparison table is anchored to a year (2022) that is still running about 0.6 years below the eventual 2023 national level — treat the state CHANGE figures as a lower bound for the true two-decade state-level improvement in states that recovered from COVID faster, and roughly accurate for states where COVID mortality was more persistent.
• "Two decades" span used: 1999-2001 (a 3-year pooled decennial period, the earliest full-state NCHS life table available) to 2022 (23 years) rather than a cleaner 2003-2023 window, because NCHS did not publish a full 50-state life table for any single year close to 2003-2006 — state life tables became an annual product only starting around 2010. This is disclosed rather than smoothed over; the extra 1-2 years at the front end has only a small effect on the reported state-level changes.
• National annual series 2000-2023 is a single consistent NCHS methodology (complete life tables, Medicare-supplemented since 2001) so it is directly comparable year to year; the 1999-2001 decennial and 2022 annual state tables use compatible but not identical estimation methodologies (the decennial tables also incorporate 2000 Census counts and a 5-year birth-count window) — a small, second-order source of noise on top of the state-level changes reported above.