Yes, but weakly and only cross-sectionally: states and years with easier healthcare access show fewer preventable deaths — the link mostly reflects poverty, not year-to-year access changes

Summary

Weakly and conditionally, yes — but the relationship is mostly a poverty story, not a pure healthcare-access story. Comparing states and years, places and periods with higher uninsured rates do have higher rates of death from causes considered treatable with timely medical care (diabetes, heart disease, flu/pneumonia, kidney disease, stroke) — even after controlling for the state's poverty rate, a state with a 1-percentage-point-higher uninsured rate has about 0.84 more such deaths per 100,000 people per year (p=0.006, n=459 state-years, 2009-2017, robust to dropping any single state). But this is a between-state effect. When the same relationship is tested the more rigorous way — within a state, over time, controlling for everything fixed about that state and every nationwide year effect — a state's own year-to-year change in its uninsured rate does not significantly predict its own change in preventable mortality (coefficient -0.27, p=0.51). In other words: states that chronically have worse insurance coverage also chronically have more preventable deaths, largely because those same states are also poorer (poverty alone explains far more variance, r=0.63 vs. r=0.30 for insurance) — but a given state improving its own coverage from one year to the next has not been shown here to produce a detectable, short-run drop in its own preventable-death rate. Both the literature and this analysis agree: a raw correlation exists, but it collapses once poverty and fixed state characteristics are controlled — consistent with the published finding that the uninsurance-amenable-mortality association is not statistically significant once poverty and race are accounted for.

Data and method

Coverage constraint, disclosed up front: CDC/NCHS's state-level, cause-specific, age-adjusted annual mortality series (health.cdc_mortality, source_type='annual') only runs 1999-2017 — it does not extend into the actual last ten years (2016-2025). The literal 'last ten years' cannot be tested with this table. The nearest supportable window is the last ten years the annual series actually has data for, 2008-2017; poverty_rate begins in 2009, so the panel used is 2009-2017 (9 years, 51 jurisdictions, 459 state-years). This is stated as a limitation, not glossed over.

Outcome ("preventable/amenable mortality"): health.cdc_mortality carries only 10 leading causes (a condensed NCHS list), not the full clinical amenable-mortality basket used in the peer-reviewed literature (e.g., Commonwealth Fund/Nolte-McKee, which includes conditions like treatable infections, some cancers screened early, and surgical complications). As a proxy, this analysis sums the age-adjusted rates (deaths per 100,000, additive under direct standardization) for the five available causes most widely classified as amenable to healthcare: diabetes, heart disease, influenza/pneumonia, kidney disease, and stroke.

Access proxy: census.sahie_insurance's state-level uninsured_rate (Small Area Health Insurance Estimates). This is a coverage proxy for 'ease of getting healthcare,' not a direct measure of appointment wait times, physician density, or travel distance — those (e.g. health.ahrf_physician_supply) exist in this corpus only as a current single-year snapshot, not a 10-year state panel, so they could not be used for a trend test here.

Confound control: census.poverty_rate (state-year poverty rate) is included because published literature finds the uninsurance-mortality association is confounded by poverty and race; race-specific mortality rates by cause were not available at this grain in the loaded tables, so poverty is the control actually tested.

Models run: (1) uncontrolled pooled OLS; (2) pooled OLS with poverty added, checked with sensitivity_analysis (leave-one-state-out, 51 refits — no sign flip, no crossing of p=0.05); (3) two-way (state + year) fixed-effects panel regression with poverty as a covariate, cluster-robust standard errors on state — the design that isolates the within-state, year-to-year relationship rather than the cross-state one. correlation_matrix confirmed no problematic collinearity between uninsured_rate and poverty_rate_pct (VIF 1.14 and 1.73).

Results

Poverty rate itself is a far stronger predictor than insurance coverage in every specification that includes it (coefficient +7.8 deaths/100k per poverty point, p<0.001; simple correlation with the mortality outcome is r=0.63 vs. r=0.30 for uninsured rate) — consistent with the Commonwealth Fund finding that amenable-mortality variation across states is driven more by demographic/economic factors than by insurance status alone.

Nationally, both the average uninsured rate (16% in 2009 to 9.4% in 2017, largely the ACA Medicaid-expansion/marketplace period) and average amenable-cause mortality (273.8 to 253.1 per 100k) declined together over this window — a real, documented co-movement, but one that a controlled state panel does not attribute to insurance coverage changes specifically, since it could equally reflect the broader downward secular mortality trend, other unmeasured year effects, or slow-moving demographic change the fixed-effects model already absorbs.

What this does and does not show

• Does show: across US states, the ones with chronically worse insurance coverage also have chronically worse preventable-cause mortality, and about a third of that gap survives controlling for poverty.
• Does not show: that a state cutting its own uninsured rate produces a measurable near-term drop in its own preventable death rate — the within-state test found no significant effect over this 9-year window, so this analysis cannot support a causal 'expand coverage, watch preventable deaths fall in your own state' claim from this data alone.
• Consistent with the literature: published research (Commonwealth Fund/Nolte-McKee; NEJM and Lancet Public Health work on Medicaid expansion) finds mixed, state-variable mortality effects of coverage expansion (from roughly -64 to +30 deaths per 100,000 depending on the state and study), and finds the uninsurance-amenable-mortality association weakens or disappears once poverty and race are controlled — exactly the pattern found here.
• Not tested here for lack of a usable table: direct measures of care access difficulty (physician density, travel time, appointment wait) — the one available proxy (HRSA AHRF physician supply) is a current-snapshot-only table with no year dimension, so it cannot support a 10-year trend test.

What This Report Does Not Answer

• Over the last ten years: health.cdc_mortality's age-adjusted, cause-specific annual state series only runs 1999-2017; it does not reach 2016-2025. Used the nearest supportable 9-year window (2009-2017) instead and disclosed this explicitly rather than presenting a shorter or different window as if it were the literal last decade.