California Freight Cleanup → Investigation 6-2

Which 14.7 million Californians do the dose-response studies count — and what does getting it wrong cost?

$0.80B wrong-decision cost — Di vs. Krewski age-threshold split, 50/50 prior

The Di et al. study covers only Medicare enrollees aged 65 and up; the Krewski study covers adults 30 and up. The 14.7M Californians between those ages are the contested population. That gap is not just statistical noise: Di and Krewski select different optimal transport policies, and the expected cost of choosing the wrong one is $0.80B. A meta-analysis at $0.5M could resolve it at a 500:1 return.

Decision context

Earlier investigations split their Monte Carlo draws 50/50 between the Di and Krewski dose-response functions, inheriting the uncertainty rather than resolving it. This investigation makes that disagreement concrete: does the CRF choice actually flip the portfolio recommendation? If it does, the cost of choosing the wrong CRF is not a different benefit estimate — it is a policy error with a specific dollar magnitude. We compute both quantities and translate them into research investment recommendations.

Methodology

Age decomposition. The cell population is rebuilt from the same spatial inputs as all health investigations. Separate mortality burdens are computed for the ≥65 cohort (Di scope) and the ≥30 cohort (Krewski scope). The contested population (ages 30–64) and its annual mortality burden are derived as the difference.

Policy-conditional outcomes. For each CRF, deaths avoided and monetized net benefit are computed across five transport scenarios (T1–T5) relative to baseline, using the MC subset specific to that CRF (Di mask: draws 1–5,015; Krewski mask: draws 5,016–10,000). The optimal policy under each CRF is the scenario maximizing mean net benefit in that CRF’s MC subset.

Wrong-decision cost and EVSI. If Di and Krewski select different optimal policies, the asymmetric wrong-decision costs are computed (cost of following Krewski-optimal when Di is true, and vice versa). Expected wrong-decision cost uses a 50/50 flat prior over which CRF is correct — matching the Investigation 6-1/5 prior. EVSI is anchored on the Investigation 6-3 hierarchical residual: $0.25B represents the remaining decision-relevant CRF uncertainty after pooling Di/Krewski/Turner/IHD cohorts into the hierarchical posterior. Three resolution options are costed: meta-analysis ($0.5M, 6 months), retrospective cohort ($2M, 1 year), and prospective CA cohort ($15M, 5 years).

Headline results

QuantityValue
Population ages 30–64 (contested)14,666,708
Annual mortality burden, ages 30–64 (CA, contested cohort)94,490 deaths/yr
Burden ratio (≥30 / ≥65)4.01×
Di-optimal policyT3_delayed
Krewski-optimal policyT1_baseline
Cost: follow Krewski if Di is true ($B)$0.01B
Cost: follow Di if Krewski is true ($B)$1.59B
Expected wrong-decision cost (50/50, $B)$0.80B
Investigation 6-3 posterior π(Di)1.000
Investigation 6-3 posterior π(Krewski)0.000
Posterior-optimal CRFDi (zero expected regret)
EVSI (Investigation 6-3 hierarchical residual)$0.25B
ROI: meta-analysis ($0.5M)500:1
ROI: retrospective cohort ($2M)125:1

Di and Krewski select different optimal transport policies: T3 (delayed) vs. T1 (baseline ACC II compliance)

Under Di (Medicare ≥65 only), the limited-cohort scope keeps per-scenario deaths avoided small, and net benefit flips sign under program cost: the delayed scenario (T3) ekes out $0.02B net benefit by avoiding program spending. Under Krewski (≥30 cohort, 4× more attributable deaths), the ACC II baseline (T1) dominates because the larger attributable-death count makes the $0 program cost a clear winner. This policy-flip—driven by a 1982–2000 vs. 2000–2016 age-scope difference in two published studies—is the concrete decision stake that makes CRF research non-academic.

Investigation 6-3 posterior resolves to Di with zero expected regret, but $0.80B is the expected wrong-decision cost under a 50/50 prior

The Investigation 6-3 hierarchical posterior puts almost all of its mass above the Di/Krewski midpoint — the posterior mean (β = 0.02439) sits roughly 4σ above the Di/Krewski midpoint, so 1 − Φ((midpoint − μ)/σ) truncates to 1.000 at four-decimal precision. Read π(Di) = 1.000 as “Di is far closer to the Investigation 6-3 posterior than Krewski is” — not as Bayesian certainty (Cromwell’s rule still applies). The Investigation 6-3 posterior is also incommensurable with Di/Krewski because of the region-mean exposure attenuation that biases β upward; a literal posterior probability comparison would require collapsing Investigation 6-3 onto the same exposure resolution as Di/Krewski, which the cascade does not do. Under this posterior, following Di is the zero-regret choice among the Di/Krewski pair. The $0.80B wrong-decision cost (at 50/50 prior) is the expected cost — the minimum-cost realization is $0.01B (cost of following Krewski-optimal if Di is true); the maximum is $1.59B (the reverse). $0.80B is the prior-weighted average: if the Investigation 6-3 posterior is wrong—if the true CRF is Krewski—the cost of having adopted the Di-optimal policy (T3, delayed) is $1.59B. At $0.5M and 500:1 ROI, the meta-analysis is the fastest way to reduce this floor without waiting for the 5-year prospective cohort.

Caveats