{"id":"a8eb4b32-13a0-445d-9cfc-be03a675e44e","arxiv_id":"2505.21132","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":17,"one_line_summary":"A model-based study claims that 3 to 8 month gaps between vaccine doses are cost-optimal, but the result rests on unvalidated assumptions and an inconsistent susceptible-population equation.","lead":"This paper builds a mathematical model of vaccination, immunity waning, and multiple vaccine doses to ask how long to wait between doses. It concludes that intervals of 3 to 8 months minimize a combined cost of infections and vaccination campaigns, but the result depends on many hand-picked parameters and an incompletely specified model.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Comorbid model is not closed: S(t) is set algebraically with no evolution equation, R is double-counted in m1N, and P(t) can go negative, so the simulated cost curves underpinning the 3-8 month optimal interval are not trustworthy.","rationale":"The reader's verdict of REJECT is well supported, and my stress test identifies the same load-bearing weakness: the comorbid model is not a well-posed dynamical system. The missing evolution equation for S(t) is not a minor omission; the algebraic relation (12), when differentiated, is inconsistent with the other flow equations, and the inclusion of both R and m1N double-counts recovered individuals. Because every subsequent result—especially the cost function J(a) in Figure 9—is generated by numerically integrating this system, the central claim about a 3–8 month optimal interval cannot be accepted as reliable. The paper does have strengths: the time-distributed kernels are fitted to clinical/experimental data with reported confidence intervals, and the sensitivity analysis on c3 shows some robustness of qualitative trends. However, those strengths concern parameter estimation, not the structural closure of the model. The manuscript also concedes that parameter values await external validation, which further supports caution. Given the formal gap in the model equations, the appropriate verdict remains REJECT, and my analysis does not change that determination.","tokens_in":13785,"tokens_out":6705,"duration_ms":77720,"concrete_test":"Perform a consistency audit: (i) differentiate Eq. (12) analytically, substitute (13)–(15), and check whether the resulting dS/dt equals −J1(t) − αV′(t) (the physically expected loss of susceptibles from infection and vaccination); the presence of nonzero extra terms indicates the model is ill-posed. (ii) Numerically integrate the full system with a standard DDE solver using (12) to compute S and (13) for P, and record min(S(t)), min(P(t)), and S+I+D+P+R+m1N over [0,2000]; if any compartment goes negative or the total drifts, the system violates conservation. (iii) Recompute Fig. 9 with a corrected closed model—explicit dS/dt with vaccination outflows and P(t) constrained nonnegative—and compare the location of the minimum of J(a); if the optimal interval moves outside the 3–8 month range, the central claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central quantitative claim—that the optimal vaccine dose interval is 3–8 months (Fig. 9, Sec. 4.3)—rests entirely on numerical simulation of the comorbid model (10)–(15). That model is not mathematically closed. After introducing the comorbid class P(t), the paper replaces the mass-balance equation (1) with the algebraic relation (12), S = N − (I + D + P + m1N), and gives no evolution equation for S(t). Differentiating (12) and substituting (13)–(15) yields dS/dt = −J1(t) + (1−b)Rn(t) + N(dm2/dt − dm1/dt), which does not reduce to the expected loss of susceptibles from infection and vaccination (−J1 minus vaccination outflows); the residual terms contain vaccination-rate and recovered-immunity derivatives that are not identically zero. Thus the algebraic constraint is incompatible with the flow equations unless extra, unstated assumptions hold. In addition, Eq. (15) tracks a recovered compartment R while recovered individuals are also folded into m1N in (12), so the population is double-counted and total population is not conserved. Finally, P(t) in (13) is a difference of cumulative quantities and can become negative, which would make J2 and the infection term in (14) nonsensical. Since all epidemic trajectories and the cost function J(a) are generated from this inconsistent system, the central recommendation about dose intervals is not established. The paper's own sensitivity analysis and parameter estimation do not address this structural gap.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This manuscript develops an immuno-epidemiological model with time-since-infection distributed transmission, recovery, and death rates, combined with vaccination-induced and infection-acquired immunity functions that are fitted to clinical data. The model is extended by a comorbid class P(t) and a parameter κ that is claimed to capture multiple strains, and the authors formulate a single-parameter cost function J(a) = c∫I dt + dT/a over the gap a between successive vaccination campaigns. The central conclusions are that coexisting strains increase the number of epidemic peaks, that vaccination gaps of 3–8 months are near-optimal with costs rising sharply beyond 9 months (Figure 9, Section 4.3), and that multiple doses produce a 'hysteresis effect' in immunity that makes frequent vaccination unnecessary. The quantitative claims are generated exclusively by numerical simulation of the comorbid model (10)–(15), so the well-posedness of that model is decisive for the paper's conclusions.","tokens_in":14236,"tokens_out":15737,"duration_ms":156710,"significance":"The modeling direction is sensible: time-distributed kernels are more realistic than classical SIR rates, and the authors fit vaccine-induced immunity (Eqs. 21–22), viral load (Eq. 24), and recovery/death kernels (Eq. 25) to published data, reporting SSE, R², RMSE, and confidence intervals for most fits. The R0 derivation in Section 2.3 for the renewal equation is standard and correct. The qualitative premise that dose interval is a relevant control lever is plausible and practically important. However, the load-bearing quantitative results rest on a comorbid model that is not mathematically closed, a cost function with arbitrary unsensitized weights, and a 'hysteresis' interpretation that is not supported by any mechanism in the model. The authors themselves state in Section 5 that the parameter values in Table 1 have no supporting references and that the predictions await future validation. As it stands, the central recommendation of an optimal 3–8 month dosing interval is not established.","major_comments":[{"comment":"The comorbid model is not closed and is internally inconsistent. After introducing the comorbid class P(t), the paper replaces the mass-balance equation (1) with the algebraic relation (12), S(t) = N − (I + D + P + m1N), and gives no evolution equation for S(t). Differentiating (12) and substituting (13)–(15) yields dS/dt = −J1 + (1−b)Rn + N d(m2 − m1)/dt. The last term contains vaccination-rate contributions (via V_i′(t)) and acquired-immunity convolutions (via ψ and Rn) that do not cancel; specifically N d(m2−m1)/dt = (1−2α)Σφ_i(0)V_i′(t) + [bε − (1−b)]ψ(t)Rn(t), which for the fitted α=0.8, b=0.2, ε=0.7 has coefficients −0.6 and −0.66 rather than the −α and +(1−b) structure expected from a mass balance on S. The algebraic constraint is therefore incompatible with the flow equations unless additional unstated assumptions hold, and the simulated trajectories depend on an unspecified discretization choice. In addition, Eq. (13) defines P(t) as a difference of cumulative quantities, so P can become negative, rendering J2(t) and the κP(t) term in Eq. (14) nonsensical; and recovered individuals are double-counted, since they appear in R via Eq. (15) and simultaneously in m1N inside Eq. (12). Because the cost curves in Figure 9 and the interval recommendation are generated from this system, the central quantitative claim is not supported.","section":"§2.2.3, Eqs. (10)–(15)"},{"comment":"The optimal-gap recommendation depends on the arbitrary cost weights c = 0.01 and d = 5 through their ratio c/d, as J(a) = c∫I dt + dT/a. No sensitivity analysis over c/d is reported, so the flat minimum at 3–8 months and the sharp rise beyond 9 months may shift substantially under other weightings. The symbol c is also used for the transmission proportionality constant (Table 1, Eq. 24), creating ambiguity in the captions of Figures 9–12. Since the 3–8 month window is the paper's central result, the authors should either estimate the cost weights from data or show how the location of the optimum changes as c/d varies.","section":"§4.3, Eq. (27), Fig. 9"},{"comment":"The claimed 'hysteresis effect in immunity levels' is not demonstrated. Immunity in the model is a linear convolution (Eqs. 7–8), and the dynamics (10)–(15) contain no mechanism for rate-independent path dependence; the observation that 4-month and 8-month schedules yield similar epidemic progression (Fig. 8) is at most evidence of diminishing returns from overlapping vaccine responses. The term 'hysteresis' is used in the Abstract and Discussion as if it were an established model output, but no quantitative definition, mechanism, or diagnostic of hysteresis is provided anywhere in the manuscript.","section":"§4.2; Abstract and §5"},{"comment":"Key inputs to the optimization are poorly identified, and the stated robustness is not substantiated. The acquired-immunity fit ψ(t) has an essentially unconstrained location parameter (b3 = −206.6 with 95% CI spanning (−704.3, 291)) and no goodness-of-fit statistics are reported for it, unlike the fits in Eqs. (21), (22), and (24). Table 1 lists no sources for the parameters that most directly shape the model (N, V0, L, k, c, α, κ, b, ε), and Section 5 concedes that 'there are no known references to confirm the parameter values predicted.' The sensitivity analysis in Section 4.1 perturbs only c3 over a narrow interval and measures only I(t), never the optimized quantity J(a), so the Discussion's assertion that the model is 'robust against changes in parameter values' is unsupported. These identification gaps propagate directly into the 3–8 month recommendation.","section":"§3.3, Eq. (23); §4.1, Fig. 7; §5"}],"minor_comments":[{"comment":"The note 'assuming a reproduction substituting of the form S(t) = N − ϵaλt' is garbled; presumably S(t) = N − εa e^{λt} is intended, and the exponent should be written without ambiguity.","section":"§2.3, Note"},{"comment":"The sentence 'ϕj ≡ ϕ and ψj ≡ ψ, for j = 1, 2, · · ·, N' uses N, the population size, where K, the number of doses from Eq. (8), is clearly intended.","section":"§3.3, p. 14"},{"comment":"R0 is defined by truncating β at an assumed disease duration τ, but the fitted viral-load kernel (24) is a Gaussian with support on all positive times; the truncation is never reconciled with the fitted β, and no computed value of R0 from the estimated parameters is reported.","section":"§2.3, Eq. (19) vs §3.4, Eq. (24)"},{"comment":"The constraint '0 ≤ m(t;n) ≤ 1' is stated in the minimization problem but is never enforced, checked, or discussed in the results; the definition of J(n) as the result of a minimization over n is circular notation, and the passage from (26) to (27) via T = an is stated imprecisely.","section":"§4.3, Eqs. (26)–(27)"},{"comment":"The table entry 'r(t), d(t) and death rates and death rates' contains a duplicated phrase, and the survival probability p0 = 0.9975 used in Eq. (25) does not appear in the table.","section":"Table 1"},{"comment":"No goodness-of-fit statistics (R², SSE, RMSE) are given for the ψ(t) fit, in contrast to the fits in Figures 2 and 4; given the extremely wide confidence intervals reported in the text, this omission is material to assessing the fit.","section":"§3.3, Fig. 3"}],"recommendation":"reject","confidential_remarks":"The manuscript is not ready for publication in its current form. The comorbid model (10)–(15) is not closed: no evolution equation for S is given, the algebraic constraint (12) is incompatible with the flow equations, P(t) can become negative, and recovered individuals are double-counted. Because all figures supporting the central 3–8 month dosing recommendation are produced by simulating this system, the core results would need to be re-derived under a correctly specified model, which amounts to a substantial reformulation and re-running of the entire numerical study. The authors also state in Section 5 that the Table 1 parameters lack supporting references and that validation awaits future experiments, and no code or data deposit is provided despite the Data Availability statement. The topic fits the journal's scope, but the quantitative claims as presented are not trustworthy."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"One thing you should know: the headline result—cost-minimizing dose interval of 3–8 months—is a numerical output of a model that is not mathematically closed. After introducing the comorbid class P(t), the authors replace the mass-balance for S with the algebraic relation S = N − (I + D + P + m1N) and never give dS/dt. Differentiating that relation and substituting the other equations leaves residual terms from vaccination and immunity that do not match the infection outflow. The simulated trajectories and cost curves are therefore not trustworthy. The stress-test note checks out on reading.\n\nWhat's genuinely new: the model combines multiple vaccine doses, waning immunity, and a comorbid class in an integro-differential framework, and the authors fit immunity, viral load, and recovery/death kernels to published data. The R0 derivation is standard and fine. The sensitivity analysis around c3 is reasonable, though it only varies one parameter.\n\nThe soft spots are structural. As above, S is not evolved; P can become negative; recovered individuals are double-counted in R and in m1N, so total population is not conserved. The cost function J(a) is minimized over a single variable with arbitrary weights c and d; the flat 3–8 month minimum surely shifts with those weights. The \"hysteresis\" is just a label for the flat cost plateau—no mechanism is given. Finally, nothing is validated against an independent epidemic dataset; the authors acknowledge this, but it means the central recommendation is a model artifact until the model is fixed and tested.\n\nThe paper is not sloppy in tone; it engages the literature and is candid about limitations. But the central quantitative claim does not hold. A serious referee could help the authors fix the model, but as it stands this is not a publishable result.\n\nVerdict: send to peer review only if the editor wants to salvage a promising idea; otherwise reject. I would tell the authors to close the model, conserve population, and re-run the cost analysis.","headline":"A useful operational question buried in a model that does not close: the 3–8 month optimal interval is not supported as written.","tokens_in":14742,"tokens_out":2868,"would_cite":false,"duration_ms":33548,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["92D30","92C60"],"pacs":[],"model":"deepseek-v4-flash","headline":"The interval between vaccine doses is a tunable control parameter: 3–8 month gaps minimize infection-plus-vaccination cost, while gaps beyond 9 months drive costs up sharply.","keywords":["vaccine dose interval","immuno-epidemic model","hysteresis effect","comorbidity","vaccine efficacy","immunity waning","basic reproduction number","optimal vaccination schedule"],"falsifier":"Re-run the model with an explicit equation updating the susceptible population once infections are split into the healthy and higher-risk streams, and see whether the minimum of $J(a)$ still lies at 3–8 months; if it moves or disappears, the recommended window depends on an unstated bookkeeping choice.","tokens_in":13531,"feed_emoji":"💉","tokens_out":8160,"duration_ms":82659,"temperature":0.7,"pith_summary":"This paper argues that the timing between vaccine doses is not a logistics detail but a control parameter that determines how much disease spreads and how much a vaccination program costs. Using an epidemic model in which infectivity, recovery, death, and immunity all depend on time since infection or vaccination, the authors compare campaigns spaced 4, 8, and 12 months apart and then minimize a cost function that balances infections against the number of campaigns. Their central quantitative result is that the cost stays near its minimum when successive doses are 3 to 8 months apart, and rises abruptly when the gap exceeds 9 months. They interpret this as a hysteresis effect: frequent boosters add little immunity once earlier doses have shaped the immune response, so fewer, well-spaced campaigns can achieve the same control at lower cost and with less intrusion.","feed_headline":"Vaccine gaps of 3–8 months keep epidemic costs lowest","feed_subtitle":"Wider spacing beats frequent boosters; beyond 9 months, waning immunity lets costs and new peaks return.","key_machinery":"The machinery is an integro-differential immuno-epidemic model with time-since-infection distributed infectivity, recovery, and death, plus a vaccination-driven immunity variable. New infections obey $J(t) = \\frac{S(t)}{N}\\int_0^t \\beta(t-\\eta)J(\\eta)\\,d\\eta$; immunity is a convolution $m(t) = \\frac{1}{N}\\sum_i \\int_0^t \\phi_i(t-\\eta)V_i'(\\eta)\\,d\\eta$; and the optimization is carried by the cost function $J(a) = c\\int_0^T I(t;a)\\,dt + d\\,T/a$, where $a$ is the uniform gap between campaigns. The dose-gap $a$ is the control that carries the argument, and the reproduction number is derived as $R_0 = \\int_0^{\\tau} \\beta(x)\\,dx$ in the no-immunity limit.","core_discovery":"The paper's central claim is that the gap between successive vaccine doses is a first-order epidemic-control parameter. For a fixed time horizon and a cost that charges both infections and each vaccination campaign, the combined cost $J(a)$ stays near its minimum for gaps between 3 and 8 months and climbs abruptly once the gap exceeds 9 months. The mechanism offered is hysteresis: each dose builds an immune memory that makes later doses, if given too soon, add little extra protection, so frequent vaccination is not only more expensive but largely redundant. The model also predicts that coexisting strains—represented through a higher infectivity multiplier $\\kappa$ on a comorbid (higher-risk) class—produce more, smaller epidemic peaks and shrink the range of safe gaps, while higher vaccine efficacy and faster vaccination rates stretch that range.","pith_inferences":["A natural extension is to make vaccine efficacy dose-dependent: real-world primary and booster doses differ, and the paper's assumption of equal efficacy across doses may shorten or lengthen the optimal window.","The 3–8 month window is derived from a particular fitted immunity-waning curve; re-running the cost minimization with the alternative Gaussian fit reported in the paper would show how sensitive the window is to the choice of $\\phi(t)$.","The model treats campaigns as evenly spaced; an adaptive schedule that spaces early doses more tightly and later doses more loosely might push the cost minimum even lower than the fixed-gap optimum.","A policy-facing test would compare model-predicted infection costs under 4-, 8-, and 12-month booster schedules against real-world data from a country that adopted each schedule."],"forward_implications":["Campaigns can be spaced 3–8 months apart without sacrificing epidemic control, so the number of vaccine drives—and their direct and indirect costs—can be cut.","Waiting 12 months or more between doses is predicted to let immunity wane enough that new epidemic peaks appear, so overly long gaps are not a free saving.","Higher vaccine efficacy and faster vaccination rates widen the window of acceptable gaps, while the presence of multiple strains (larger $\\kappa$) narrows it.","The same model structure can be re-fitted to other diseases by replacing the immunity, viral-load, recovery, and death curves, making the dosing-interval question testable outside COVID-19.","If adopted, the hysteresis view implies that 'boost as often as possible' is not just wasteful but may be counterproductive, because extra doses add little once the immune plateau is reached."],"supporting_citations":[{"why":"Supplies the vaccine-induced immunity data and the time-distributed recovery and death rate functions that the model fits and uses.","marker":"[35]"},{"why":"Provides the distributed recovery/death modeling framework on which the basic model is built.","marker":"[33]"},{"why":"Supplies real-life protection data used to fit the infection-acquired immunity function $\\psi(t)$.","marker":"[38]"},{"why":"Supplies viral-load kinetics data for SARS-CoV-2 variants used to fit the transmission rate $\\beta(\\tau)$.","marker":"[39]"},{"why":"Provides clinical evidence that dosing interval affects vaccine efficacy, motivating the cost optimization.","marker":"[30]"},{"why":"Supports the modeling premise that optimized dosing schedules can prolong immunity and suppress variant emergence.","marker":"[31]"},{"why":"Provides observed COVID-19 infection kinetics and multiple-wave patterns used to interpret the model's peak behavior.","marker":"[9]"},{"why":"Supplies evidence that vaccine-induced immunity wanes and booster potency can decline with variants, a factor in the model.","marker":"[32]"}],"fun_headline_variants":["Vaccine dose gaps of 3-8 months cut epidemic costs","Optimal vaccine spacing: 3-8 months, beyond 9 fails","Wider dose intervals beat frequent shots for control","For lowest cost, space vaccine doses 3-8 months apart"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The recommendation rests on the model's bookkeeping that keeps the susceptible population as a leftover of the other compartments after infections are split into healthy and higher-risk streams, even though no evolution equation for that leftover is stated.","fun_headline_variants_meta":{"raw":{"variants":["Vaccine dose gaps of 3-8 months cut epidemic costs","Optimal vaccine spacing: 3-8 months, beyond 9 fails","Wider dose intervals beat frequent shots for control","For lowest cost, space vaccine doses 3-8 months apart"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000293,"raw_usage":{"total_tokens":1667,"prompt_tokens":862,"completion_tokens":805,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":478,"completion_tokens_details":{"reasoning_tokens":731}},"tokens_in":478,"tokens_out":805,"duration_ms":8683,"temperature":1.0,"reasoning_tokens":731,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T13:34:16.433914+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the model with an explicit equation updating the susceptible population once infections are split into the healthy and higher-risk streams, and see whether the minimum of $J(a)$ still lies at 3–8 months; if it moves or disappears, the recommended window depends on an unstated bookkeeping choice.","supporting_citations":[{"cited_title":"Immuno- epidemiological model-based prediction of further covid-19 epidemic outbreaks due to immunity waning","cited_arxiv_id":null,"evidence_quote":"Supplies the vaccine-induced immunity data and the time-distributed recovery and death rate functions that the model fits and uses."},{"cited_title":"An epidemic model with time-distributed recovery and death rates.Bulletin of Math- ematical Biology, 84(8):78, 2022","cited_arxiv_id":null,"evidence_quote":"Provides the distributed recovery/death modeling framework on which the basic model is built."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies real-life protection data used to fit the infection-acquired immunity function $\\psi(t)$."},{"cited_title":"Sars-cov-2 omicron variant replication in human respiratory tract ex vivo","cited_arxiv_id":null,"evidence_quote":"Supplies viral-load kinetics data for SARS-CoV-2 variants used to fit the transmission rate $\\beta(\\tau)$."},{"cited_title":"et al Liu","cited_arxiv_id":null,"evidence_quote":"Provides clinical evidence that dosing interval affects vaccine efficacy, motivating the cost optimization."},{"cited_title":"et al Dogra","cited_arxiv_id":null,"evidence_quote":"Supports the modeling premise that optimized dosing schedules can prolong immunity and suppress variant emergence."},{"cited_title":"Infection kinetics of covid-19 and con- tainment strategy","cited_arxiv_id":null,"evidence_quote":"Provides observed COVID-19 infection kinetics and multiple-wave patterns used to interpret the model's peak behavior."},{"cited_title":"et al Menegale","cited_arxiv_id":null,"evidence_quote":"Supplies evidence that vaccine-induced immunity wanes and booster potency can decline with variants, a factor in the model."}],"review_version":1}