{"id":"6f3cc60c-8c93-4148-952e-55cb3a7a67c7","arxiv_id":"2505.16696","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Variations in His-Purkinje structure have little individual effect on simulated QRS morphology, but parameter interactions can produce abnormal and premature QRS complexes.","lead":"This computational study varied nine structural parameters of the simulated His-Purkinje system in a single heart model and measured how ECG QRS waveforms changed across 22,528 trials. It found that most individual parameter changes barely affect QRS shape, but certain parameter combinations can trigger abnormal or premature QRS complexes.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The reported low QRS sensitivity may be an artifact of the 10 Hz low-pass ECG filter: QRS bandwidth extends well above 10 Hz, so amplitudes, durations, and outlier morphologies are filter-limited (Sec. 2.2).","rationale":"The reader's weakest assumption concerned external generalization from a single heart mesh. My concern is more fundamental: the internal measurement pipeline may invalidate the QRS features before any generalization is considered. A 10 Hz low-pass filter is below the frequency content of the QRS complex, so the extracted QOIs and all downstream Sobol statistics are suspect. This is a concrete, testable issue: rerunning a subset with a higher cutoff or no filter would reveal whether the reported insensitivity is a property of the heart model or an artifact of signal processing. If the result is filter-dependent, the paper's headline claim and its clinical implications are unsupported in the current version. This critique is directed at the method, not the authors. I also acknowledge the paper's strengths: the large Saltelli sample, the open-source network generator, and the independent ECG forward-calculation benchmark. The concern is not about computational effort or statistical rigor within the stated pipeline; it is about whether that pipeline measures what the paper claims it measures.","tokens_in":19578,"tokens_out":5745,"duration_ms":54783,"concrete_test":"Re-run a representative subset (e.g., 500 Saltelli samples plus the identified outlier trial) through the same pipeline with the ECG-Deli low-pass cutoff changed from 10 Hz to 100 Hz, or with the filter disabled, and re-extract the ten QOIs using the same MATLAB algorithm. Then recompute the first-order and total Sobol indices for these trials. If QRS amplitude or duration distributions widen materially, or if Sobol indices change beyond sampling error (for example, S1 for median branch length or branch angle rises above 0.05, or total-order indices no longer cluster near 1), the low-sensitivity conclusion is filter-dependent and the manuscript must be revised accordingly.","verdict_should_be":"REJECT","load_bearing_attack":"Section 2.2 states that raw extracellular potentials were low-pass filtered with ECG-Deli at a 10 Hz cutoff. Diagnostic QRS complexes contain substantial spectral content above 10 Hz, and a 10 Hz low-pass filter severely attenuates sharp QRS peaks and broadens wave boundaries. Because every one of the 22,528 trials passes through the same filter, the measured Q-wave, R-wave, and S-wave durations and amplitudes, and the Sobol indices built on them, describe a heavily filtered signal rather than the electrophysiologically meaningful QRS. The central conclusion that individual HPS parameters have low impact on QRS morphology, and the specific outlier morphologies that motivate the interaction claims, could therefore be filter artifacts. The paper provides no validation of this filter choice; the ECG forward-calculation benchmark validates the potential computation, not the filtering or delineation pipeline. A 10 Hz low-pass cutoff is far below standard diagnostic ECG low-pass settings (typically 100-150 Hz), so this is an internal correctness risk, not merely an external-generalization caveat. If 10 Hz is a typo, the manuscript must state the intended cutoff; if it is not, the QRS metrics are not clinically representative.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper performs a global, variance-based sensitivity analysis of His-Purkinje system (HPS) structural parameters on QRS-morphology metrics in a computational human heart model. Nine HPS generation parameters are varied with Saltelli sampling across 22,528 simulations, and ten QRS-related quantities of interest (Q-, R-, S-wave durations, amplitudes, peak times, and overall QRS duration) are extracted from simulated 12-lead ECG traces. First- and total-order Sobol indices are computed with SALib. The authors report low first-order sensitivity for wave amplitudes and durations, large total Sobol indices interpreted as interaction-driven variability, and a small number of outlier trials, including one trial with markedly abnormal QRS morphology and a cluster of trials with premature QRS formation. They conclude that minor individual HPS structural differences are unlikely to affect model fidelity or clinical interpretation in physiologically normal hearts, while certain parameter combinations can produce clinically relevant abnormalities.","tokens_in":19821,"tokens_out":4291,"duration_ms":36726,"significance":"If the conclusions are correct, the study would be a valuable contribution to cardiac digital twin personalization: it is a large-scale application of global sensitivity analysis to a full 3D cardiac electrophysiology model, uses a standard and reproducible sampling and sensitivity pipeline (Saltelli sampling, SALib), and directly targets an understudied source of model uncertainty (HPS structure). However, the central quantitative claims currently rest on QRS features extracted from signals processed with a 10 Hz low-pass filter, which is not representative of diagnostic ECG bandwidth and could compromise every reported amplitude, duration, and sensitivity index. The single-geometry design and arbitrary ±30% parameter ranges also limit the stated clinical generalization. The computational scale and the careful reporting of distributions are strengths, but the filter issue must be resolved before the paper's main claims can be accepted.","major_comments":[{"comment":"Raw extracellular potentials were low-pass filtered with a 10 Hz cutoff before QRS delineation. A 10 Hz cutoff is far below the diagnostic ECG bandwidth (typically 0.05-150 Hz or at least 100 Hz); it severely attenuates sharp QRS peaks and broadens the apparent wave boundaries. Because all 22,528 trials and all ten QOIs are derived from this filtered signal, the reported low sensitivity and the outlier morphologies used to claim interaction-driven clinical changes describe a 10 Hz-filtered signal rather than a clinically representative QRS. The benchmark in the Supplementary validates the forward extracellular-potential computation, not the filter or delineation pipeline, and no validation of the 10 Hz choice is provided. The authors should either correct this if it is a typo, justify the cutoff, or re-run the analysis at a diagnostic bandwidth and show that the conclusions are unchanged.","section":"Section 2.2"},{"comment":"The text reports that all total Sobol indices 'either equaled one or contained one in their confidence intervals,' but no confidence intervals or estimator diagnostics are reported anywhere in the paper or supplementary materials. Reporting uncertainty on Sobol indices is necessary because ST near one for every parameter is an extreme claim (it implies every parameter participates in interactions that collectively explain essentially all output variance) and can be an artifact of estimator bias or finite sample size. Please report bootstrap confidence intervals, replicate-based estimates, or convergence checks for both S1 and ST, and temper the interaction-driven interpretation accordingly.","section":"Section 3.3"},{"comment":"The conclusion that 'minor structural differences between a healthy patient's HPS and that of a generic model are unlikely to significantly impact model fidelity or clinical interpretation' is an inference about human populations, but the evidence comes from a single bi-ventricular mesh derived from one de-identified CT scan and from parameter ranges chosen as ±30% of nominal values because 'little is known about the true distribution of HPS structure among human populations.' This is not an internal inconsistency, but it is an overreach: the sensitivity indices and outlier rates could change with ventricular geometry, torso anatomy, or population-informed parameter ranges. The conclusion should be restricted to the studied mesh and ranges, or supported by additional geometries and population-informed distributions.","section":"Sections 2.2, 2.3, and 5"}],"minor_comments":[{"comment":"The text says 'Table 4 presents the summary statistics for the peak time distributions,' but Table 4 reports QRS duration statistics; the peak-time statistics are in Table 5. Please correct the cross-reference.","section":"Section 3.2"},{"comment":"The claim that premature QRS formation 'obscured P-wave formation' is based on visual inspection of Figure S10, while the paper explicitly states that P waves were not characterized. This should be framed as a qualitative observation or supported by quantitative P-wave analysis.","section":"Section 4"},{"comment":"There are minor unit and notation inconsistencies: '2.16mS cm' should be '2.16 mS/cm'; Table S1 uses 'mv' rather than 'mV'; and Eq. (1) uses a scalar σ_b although the text refers to a conductivity tensor.","section":"Section 2.2 and Table S1"},{"comment":"The sentence 'The number of outlier trials for these QOI was relatively small compared to the total number of trials conducted was very low (<1%)' contains a duplicated predicate and should be rephrased for clarity.","section":"Section 3.1"},{"comment":"The right panel's y-axis label 'Percent Difference from Nominal Trial (%)' is unclear; please clarify whether the plot is a scatter plot, rug plot, or histogram of the 25 outlier trials, and define the parameter color mapping in the caption.","section":"Figure 6"}],"recommendation":"major_revision","confidential_remarks":"The 10 Hz filter issue is decisive: if the cutoff is genuine, the reported QRS amplitudes, durations, and sensitivity indices are not clinically representative, and the central conclusion cannot be accepted without re-running the analysis at a diagnostic bandwidth. I do not see evidence of misconduct; the statistical pipeline is standard and the computational effort is substantial. Please ask the editor to verify whether the authors have a data/code availability statement, since the paper mentions open-source libraries but does not include a repository link."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The thing to know: this is a real computational campaign, not a toy. Twenty-two thousand full 3D heart simulations with a complete His-Purkinje system, standard Saltelli sampling, Sobol indices, and a validated ECG forward calculation. That scale alone makes it worth a serious look. The paper is also honest about what it did not do: no fitted parameters folded into the statistics, no overclaiming of mechanistic novelty. Prior HPS sensitivity work was local or partial; this is the first global variance-based treatment of complete HPS structure on a whole heart with QRS outcomes. Credit where due: the low main effects for amplitudes and durations, the interaction-driven outliers, and the branch-count effect on peak timing are all plausible within the simulated model, and the reported low coefficients of variation support the claim that isolated parameter changes are not the main story.\n\nThe soft spots are real, and one is load-bearing. Section 2.2 says the raw extracellular potentials were low-pass filtered with a 10 Hz cutoff. Diagnostic QRS content extends well above 10 Hz, and a 10 Hz filter will attenuate sharp peaks, broaden wave boundaries, and alter durations and amplitudes. Since every one of the 22,528 trials passes through the same filter, the Sobol indices and the distributions describe a heavily smoothed signal, not the electrophysiologically meaningful QRS. That is an internal correctness risk, not just an external generalization caveat. If the 10 Hz is a typo, the paper must say what the cutoff really was. If it is not a typo, the authors need to redo the analysis at a diagnostic bandwidth or at least demonstrate that the specific conclusions are filter-invariant. The ECG forward-calculation benchmark validates the potential computation, not the filtering or the peak-picking.\n\nOther issues are more minor but worth naming: one heart geometry, one stochastic realization per parameter set, total Sobol indices near one reported without confidence intervals, and a ±30% parameter range chosen without knowing the true human HPS distribution. The conclusion that minor HPS differences are unlikely to affect clinical interpretation goes beyond what a single-anatomy, filtered simulation can support. Tone that down and show multi-anatomy sensitivity or repeated seeds.\n\nWho is this for? People building cardiac digital twins and anyone doing global sensitivity analysis on expensive biophysical models. It deserves a serious referee, but the referee should ask for a filter correction or justification, confidence intervals on the Sobol indices, and a more careful generalization statement. I would not cite the current version until the filter issue is resolved.","headline":"A serious, expensive sensitivity study whose central clinical claim is undermined by a 10 Hz ECG low-pass filter that likely distorts the very QRS metrics being analyzed.","tokens_in":20366,"tokens_out":2414,"would_cite":false,"duration_ms":24333,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Minor variations in the heart's His-Purkinje conduction structure barely change most QRS features on their own, but certain parameter combinations can distort QRS timing or morphology and even obscure the P-wave.","keywords":["His-Purkinje system","QRS complex","ECG simulation","Sobol sensitivity analysis","cardiac digital twin","monodomain model","ventricular depolarization"],"falsifier":"Run the same Saltelli–Sobol sampling on two or more additional ventricular geometries from different patients; if a single HPS parameter within the stated ranges shifts a QRS amplitude, duration, or peak time beyond the normal ECG thresholds used here, the claim that minor HPS differences are clinically negligible would be contradicted.","tokens_in":19371,"feed_emoji":"🫀","tokens_out":9851,"duration_ms":74183,"temperature":0.7,"pith_summary":"This paper asks whether the fine details of the His-Purkinje system, the heart's specialized conduction network, matter for the QRS complex seen on an electrocardiogram. The authors generate thousands of plausible Purkinje networks on a single heart mesh, varying nine structural parameters such as branch number, branch length, fascicle angles, and repulsivity, and measure ten QRS features across four leads. They find that changing any one parameter alone rarely moves wave amplitudes, durations, or timing outside normal clinical bounds. The exceptions arise from parameter combinations: joint settings can produce morphologically abnormal QRS complexes, and some networks trigger premature QRS formation that obscures the P-wave. The paper concludes that minor structural differences between a healthy patient's conduction system and a generic model are unlikely to undermine digital-twin fidelity when both are physiologically normal.","feed_headline":"Minor heart-conduction tweaks barely change QRS; combos can distort","feed_subtitle":"A 22,528-simulation study finds single parameters matter little, but combinations can alter timing and mask P-waves.","key_machinery":"The central machinery is Sobol sensitivity analysis applied to a monodomain reaction-diffusion heart model. Purkinje networks are built by a rule-based fractal-tree algorithm controlled by nine structural parameters; Saltelli sampling generates 22,528 joint parameter combinations; and the ECG is reconstructed from extracellular potentials recovered via the monodomain source model. First-order Sobol indices separate each parameter's direct contribution to variance in a QRS quantity of interest, while total-order Sobol indices capture its contribution through interactions, so the gap between the two reveals how much of the variability is interaction-driven.","core_discovery":"The paper's central claim is that QRS morphology is mostly insensitive to individual HPS structural parameters, while interaction effects among parameters carry the clinically visible variability. Sobol first-order indices show that no single parameter exceeds roughly a 10% share of the observed variance for any QRS feature, and most main effects fall below the study's 5% significance threshold; the one consistent exception is the number of branches, which directly affects QRS peak timing. Total-order indices near one indicate that the rare outlier trials, including a QRS with a deep, long S-wave and a cluster of early-formed QRS complexes, come from specific parameter combinations rather than isolated changes. The paper therefore argues that a generic but physiologically normal HPS is adequate for most digital-twin ECG interpretation, while warning that certain joint parameter settings can generate abnormal morphology or premature QRS complexes that future models should account for.","pith_inferences":["Because the study samples parameters on a single ventricular mesh, the sensitivity indices likely mix HPS effects with that geometry's own response; repeating the design on several patient-derived meshes would test how far the 'healthy patient' conclusion generalizes.","The ±30% parameter ranges were chosen without population data on HPS variability, so wider natural variation, if it exists, could re-scale the low-sensitivity conclusion even without changing the interaction structure.","The premature-QRS result suggests a clinical hypothesis the paper does not pursue: unexplained early QRS complexes with apparently missing P-waves on real ECGs could reflect unusual Purkinje architecture rather than primary atrial disease.","A targeted follow-up could map the interaction surface by fixing all but two of the influential parameters (branch angle, second fascicle angle, repulsivity) and locating the boundary where QRS timing jumps to the early secondary point."],"forward_implications":["Cardiac digital twins built from a generic healthy HPS should reproduce normal QRS amplitudes and wave durations even if their branching geometry differs slightly from the patient's.","Reproducing an abnormal QRS in a model will likely require matching combinations of HPS parameters, not adjusting one branch or fascicle value alone.","Because the number of branches has a direct effect on QRS peak timing, timing-sensitive simulations should prioritize getting Purkinje density right.","Some HPS configurations produce premature QRS complexes that merge with or obscure the P-wave, which could mislead arrhythmia simulations that depend on P-wave presence.","The near-total interaction indices for durations and amplitudes imply that future sensitivity studies of cardiac models should include interaction terms rather than main effects alone."],"supporting_citations":[{"why":"Supplies the rule-based fractal-tree algorithm used to generate every HPS network in the study.","marker":"[17]"},{"why":"Provides the monodomain ECG benchmark used to validate the extracellular potential recovery.","marker":"[32]"},{"why":"Supplies the standardized ECG interpretation guidelines that define normal QRS morphology and the clinical boundaries used for the QOIs.","marker":"[14]"},{"why":"Introduces the Saltelli sampling scheme that produces the 22,528 joint parameter trials.","marker":"[40]"},{"why":"Defines the first-order and total Sobol sensitivity indices that separate direct from interaction-driven variability.","marker":"[41]"},{"why":"Provides the variance-decomposition framework and estimators used to compute the sensitivity indices.","marker":"[42]"},{"why":"Implements the Sobol index calculations in the open-source sensitivity-analysis library used by the authors.","marker":"[44]"}],"fun_headline_variants":["Single heart-wiring tweaks barely alter QRS, combos do","QRS shape mostly ignores single His-Purkinje changes, interactions matter","Minor wiring tweaks barely affect ECG QRS, but combos can distort","Single HPS parameters do little to QRS; interactions drive odd shapes","Heart model: single wiring tweaks barely matter, combos can distort QRS"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The conclusions assume that results from one fixed heart mesh, with HPS parameters varied only within ±30% of a single nominal set, represent how a healthy patient's His-Purkinje system varies across the human population.","fun_headline_variants_meta":{"raw":{"variants":["Single heart-wiring tweaks barely alter QRS, combos do","QRS shape mostly ignores single His-Purkinje changes, interactions matter","Minor wiring tweaks barely affect ECG QRS, but combos can distort","Single HPS parameters do little to QRS; interactions drive odd shapes","Heart model: single wiring tweaks barely matter, combos can distort QRS"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000718,"raw_usage":{"total_tokens":3271,"prompt_tokens":1035,"completion_tokens":2236,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":651,"completion_tokens_details":{"reasoning_tokens":2136}},"tokens_in":651,"tokens_out":2236,"duration_ms":12716,"temperature":1.0,"reasoning_tokens":2136,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T14:56:25.298808+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same Saltelli–Sobol sampling on two or more additional ventricular geometries from different patients; if a single HPS parameter within the stated ranges shifts a QRS amplitude, duration, or peak time beyond the normal ECG thresholds used here, the claim that minor HPS differences are clinically negligible would be contradicted.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the rule-based fractal-tree algorithm used to generate every HPS network in the study."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the first-order and total Sobol sensitivity indices that separate direct from interaction-driven variability."},{"cited_title":"Saltelli, M","cited_arxiv_id":null,"evidence_quote":"Provides the variance-decomposition framework and estimators used to compute the sensitivity indices."}],"review_version":1}