This paper presents an analytic decision-theoretic threshold for deploying population MCED screening, derived from Bayes' rule and expected-utility theory. The core contribution is an odds-form inequality with an explicit "actionable fraction" m that separates detection from mortality benefit, treating overdiagnosis as a first-class harm inside the benefit term. No trial is run; all numeric inputs are labeled as illustrative parameters.
Correctness of the derivation. I verified the algebra independently. PPV = Se·π / (Se·π + (1−Sp)(1−π)) is standard. The expected utility E[dU] = π·Se·(m·B − (1−m)·H_od) − (1−π)(1−Sp)·H_fp − c is correct four-outcome bookkeeping given the stated assumptions. The rearrangement to odds-form threshold π/(1−π) > (1−Sp)·H_fp / [Se·(m·B − (1−m)·H_od)] follows algebraically. The worked example with Sp=0.995, Se=0.5, π=0.006, m=0.6, B=1, H_od=0.3, H_fp=0.05 checks: PPV ≈ 0.38, RHS = 0.00104, prior odds ≈ 0.00604, so the inequality is satisfied; when m drops to 0.25 the RHS becomes 0.02 and screening is net-harmful. These numbers are internally consistent.
Novelty — 5. The expected-utility threshold approach to screening decisions is not new. Decision curve analysis (Vickers & Elkin, Med Decis Making 2006) already frames screening in terms of net benefit with an explicit harm/benefit trade-off, and the Pauker–Kassirer threshold model (1975, 1980) established Bayesian decision-theoretic thresholds for diagnostic testing decades ago. The paper does not cite or engage with this large prior literature, presenting the framework as if it were originating the idea that expected utility determines the deploy/no-deploy boundary. The genuine novelty is the explicit parameter m — the actionable fraction that subtracts overdiagnosis harm inside the benefit term and can flip the sign of the decision even when raw detection accuracy is high. This is a useful reparameterization, and the paper's exposition of how m dominates the inequality is clear. But the algebraic structure is a straightforward rearrangement of net-benefit bookkeeping. A paper that acknowledged the DCA tradition and showed how m extends it would be stronger; as it stands, the contribution is a tidy conceptual note rather than a new method.
Rigour — 7. The paper is pure analysis and makes no fabricated empirical claims — this is to its credit. Every numeric input is explicitly flagged as illustrative, not a measurement. I verified the references: Welch & Black (2010, JNCI, overdiagnosis), Etzioni et al. (2003, Nat Rev Cancer), Croswell et al. (2009, Ann Fam Med, false-positive cumulative incidence), Klein et al. (2021, Ann Oncol, CCGA MCED validation), Pepe et al. (2001, JNCI, biomarker phases), and Cover & Thomas (2006, Elements of Information Theory) all resolve as real publications, though the citation keys in the manuscript are sloppy (some keys differ from standard DOI formats). The paper explicitly lists its limitations in Section 6: single-round screening assumption, aggregate parameters masking tumour-type heterogeneity, utilities treated as known and commensurable, omission of cumulative false-positive dynamics, and — crucially — the model cannot generate m, B, or H_od. No causal claims are drawn from anecdote. Claims are proportionate to the evidence base (which is algebraic, not empirical). The gap is the absence of engagement with prior decision-theoretic work, which is a scholarship deficit rather than a rigour error per se.
Significance — 5. If prospectively validated, the framework could sharpen trial design by making explicit what must be measured (m, B, H_od, H_fp, per-tumour-type Se, Sp) and by providing a pre-registrable falsification target. Section 7 does a reasonable job of sketching what a confirmatory trial would need. However, the framework's practical utility is severely limited by the difficulty of estimating m prospectively — it is counterfactual, confounded by lead-time and length-time bias, and requires long-term follow-up of screen-detected versus clinically-detected cases. The paper acknowledges this but does not solve it. Without a feasible estimation strategy for m, the threshold remains a conceptual tool with an uncertain path to changing clinical practice. The paper would not, as it stands, change screening policy.
Clarity — 7. The derivation is laid out stepwise, the parameter definitions are explicit, the illustrative calculation is clearly separated from any empirical claim, and the limitations section is honest. The prose is readable. The main weakness is that the paper does not situate itself in the existing decision-theoretic screening literature — a reader unfamiliar with DCA might think the entire expected-utility approach is novel, which is misleading.
Overall assessment. This is a competent analytic note that correctly identifies m as a load-bearing parameter and derives a clean threshold inequality. The derivation is sound, no data are fabricated, and limitations are acknowledged. However, the paper overstates its novelty by failing to engage with decades of decision-theoretic work on screening thresholds (DCA, Pauker–Kassirer), and the practical significance is constrained by the unavailability of m in the absence of long-term mortality follow-up. The paper is a useful conceptual contribution but does not advance the field substantially beyond existing net-benefit frameworks.