US Mortality Coverage of the Bayesian Causal Atlas

A deduplicated count of annual US deaths addressed by the 26 atlas analyses
Build 2026-05-28 · CDC NVSR vol. 74 no. 10 (Sep 2025) · FastStats 2024 · Pearl SCM framework
NET UNIQUE annual deaths covered
% of all US deaths (2024)
UNCOVERED residual
26
analyses · 14 clinical domains

Coverage of US Mortality (2024)

Each colored segment is the share of total US deaths owned by one atlas analysis under strict UCOD-priority deduplication. The grey segment is what the atlas does not currently address (COVID-19, septicemia, homicide, infant mortality, congenital, ill-defined causes). Hover any segment for details.

Per-Analysis Breakdown — All 26

Click column headers to sort. Filter by role to isolate primary-attribution rows (12) vs subordinate (14).

Showing 26 rows · Net total:
# Analysis Domain Role Annual Deaths
(NET unique)
% of US UCOD Scope Overlap with

CDC UCOD Reference (2024)

Authoritative 2024 leading-cause data against which atlas attributions are validated. Total 2024 deaths: .

Cause / ICD-10Annual US Deaths% of Total

What's NOT Covered (≈20% of US Deaths)

The 619,078 annual deaths not addressed by any current atlas analysis fall into these categories:

COVID-19 (U07.1, 2024)~50,000
Septicemia (A40-A41)~40,000
Homicide (X85-Y09)~25,000
Flu/pneumonia excess vs #20~30,000
Infant mortality (congenital, SIDS)~20,000
Ill-defined causes (R00-R99)~80,000
HIV, TB, nutritional, other small UCOD~15,000
Residual reconciliation / rounding~360,000

Antithesis — Where This Count Is Vulnerable

1. UCOD undercounts true disease burden. Diabetes is UCOD on ~94K death certificates but appears as a contributing cause on ~270K. CKD: 55K UCOD vs ~250K contributing. Dementia: ~150K UCOD vs estimates as high as 500K under broader definitions. Under contributing-cause attribution the deduplicated total would rise into the 2.7–2.9M range — closely matching the homepage's "2.8 million" claim. The UCOD convention here is more conservative and statistically defensible but mechanically understates disease burden.

2. "Coverage" conflates two different claims. Owning a UCOD chapter ≠ having a complete intervention surface for every death in that chapter. The cancer analysis covers a finite set of variants and stages, not every C-code. The 79.9% figure represents UCOD-chapter coverage, not fraction of deaths the atlas can demonstrably reduce. The latter is bounded above by 79.9% and below by something materially smaller, set by per-analysis Pareto N80 risk reduction (typically 60–80%) compounded with intervention deliverability.

3. Non-fatal endpoints have indirect mortality benefit. Analyses #3, #5, #6, #7, #18 show zero in the NET column because their primary endpoints are non-fatal. This understates their mortality contribution: OA mobility loss increases all-cause mortality via deconditioning; ADHD raises MVC and overdose risk; BPH with AUR can precipitate urosepsis. A more generous count assigning each 2-10K indirect deaths would push the total ~25K higher — still within the 80% range.

4. Homepage claim of "2.8 million" is not internally derived. The homepage figure appears chosen as a round approximation to the pre-pandemic baseline US death total rather than computed from the 26-analysis intervention surface. The bottom-up UCOD count produced here (2.45M) is lower by ~12%. The homepage should arguably be either (a) the rigorous 2.45M with the UCOD-priority caveat, or (b) the broader ~2.8M with the contributing-cause caveat — the two framings answer different questions and the page does not currently distinguish them.