{"id":"20219eab-4b71-4b7a-bb7f-38537a341b67","arxiv_id":"2509.04939","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Pre-treatment blood levels of cell-free DNA fragments longer than 1650 base pairs predict early progression and progression-free survival in advanced cancer patients treated with immune checkpoint inhibitors.","lead":"A French multicenter study measured cell-free DNA fragment sizes in 126 patients before starting immunotherapy and found that higher levels of very long DNA fragments predicted better treatment response. If confirmed, this blood test could help decide which cancer patients benefit from immune checkpoint inhibitors.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Primary predictor R>1650 is measured outside the device's calibrated size range; the paper explicitly says size determination is unreliable above 1650 bp, so the headline association may reflect non-specific signal rather than true long-fragment abundance.","rationale":"The reader identified the same load-bearing concern: R>1650 is measured in a region where the BIABooster device cannot reliably determine fragment size. The paper's own Methods confirm that size determination is unreliable above 1650 bp, and the 'relative arbitrary units' are not a calibrated concentration. This is not merely a limitation; it undermines the construct validity of the primary predictor. If R>1650 is not a true measure of long cfDNA fragments, the central claim of a fragment-size-specific biomarker is unsupported. The paper also contains an internal inconsistency in the stated calibrated range (100–1500 bp vs. 75–1650 bp), which further weakens confidence. The concrete test would settle whether the association survives calibrated measurement and adjustment for total concentration. Since the reader's verdict was already CONDITIONAL, our stress-test does not change the verdict; it reinforces the need for external validation with a method that can accurately size long fragments.","tokens_in":17044,"tokens_out":3806,"duration_ms":35395,"concrete_test":"Use archived plasma from the same cohort and re-measure fragment size distribution with a method accurate above 1650 bp (e.g., pulsed-field capillary electrophoresis with a validated ladder, or whole-genome sequencing with fragment-length calling). Compute the true fraction of fragments >1650 bp and compare with the BIABooster R>1650. Then refit the logistic/Cox models using the calibrated fraction. Additionally, perform a spike-in experiment: add known concentrations of 2–10 kb DNA to a constant cfDNA sample and check whether R>1650 changes linearly and independently of C_TOT. Also test whether R>1650 remains associated with EP/PFS after adjusting for C_TOT in a bivariate model. If R>1650 does not track true long-fragment abundance, or if the association disappears after C_TOT adjustment, the central claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline predictor R>1650 is derived from the region above 1650 bp, which the Methods state is outside the device's calibrated range: 'Above 1650 bp, all fragments migrate at similar speeds, making size determination unreliable.' The manuscript also gives inconsistent upper limits for reliable sizing: the instrument description says the size distribution is 100–1500 bp, while the later quality-control paragraph claims reliable measurements between 75 and 1650 bp. R>1650 is expressed in relative arbitrary units and is normalized by C_TOT, which itself is integrated only up to 1650 bp. If the signal above 1650 bp is influenced by injection overloading, slow-migrating contaminants, or baseline drift, then R>1650 may be a proxy for total cfDNA concentration or sample quality rather than a true length-specific fraction. Because R>1650 is the strongest predictor and the central claim of the paper, the entire conclusion rests on an uncalibrated measurement.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The SChISM study reports a prospective, multi-center, real-world cohort of 126 patients with advanced NSCLC, HNSCC, UC, and ccRCC treated with immune-checkpoint inhibitors, for whom baseline plasma cfDNA size profiles were quantified using the BIABooster capillary electrophoresis device. The authors derive twelve cfDNA features (concentrations by size bins, peak positions, peak width, and total concentration) and test their association with early progression (EP) and progression-free survival (PFS). They report that higher total cfDNA and short fragments are associated with worse outcome, while long fragments, especially the relative quantity of fragments >1650 bp (R>1650), are associated with lower odds of EP and longer PFS, with AUC=0.73 and C-index=0.69. Associations persist after adjustment for age, sex, ECOG, tumor type, and NLR in the multi-cancer cohort, and in NSCLC and HNSCC subgroups. Bootstrap resampling provides out-of-bag estimates of accuracy and predictive values. The authors conclude that cfDNA size profiles, particularly R>1650, outperform PD-L1 and NLR and may reflect cGAS-STING-mediated immune activation.","tokens_in":17326,"tokens_out":2965,"duration_ms":33027,"significance":"If the central finding is valid, it would constitute a non-invasive, multi-cancer predictive biomarker for ICI response that is independent of tumor genotype and could be measured without DNA extraction. The study has several strengths: a prospective design with trial registration, inclusion of multiple tumor types, adjustment for established confounders, and an internal bootstrap validation with out-of-bag performance metrics. The subpopulation analyses, although small, provide some consistency checks. However, the headline predictor R>1650 is measured in a size range that the Methods explicitly state is outside the device's calibrated/validated range, and the statistical analysis involves in-sample variable selection and threshold optimization without multiplicity correction. These issues are load-bearing for the central claim and require substantive revision.","major_comments":[{"comment":"The primary predictor R>1650 is derived from fluorescence signal above 1650 bp, a region where the Methods state 'all fragments migrate at similar speeds, making size determination unreliable' and where values 'don't represent true sizes or concentrations.' The manuscript also gives inconsistent calibrated ranges: the instrument description says 100-1500 bp, while the later paragraph claims reliable measurements between 75 and 1650 bp. Because R>1650 is the strongest predictor and the central claim of the paper, the entire conclusion rests on an uncalibrated signal. If this signal reflects total cfDNA concentration, injection overloading, or electrophoretic artifacts rather than true long-fragment abundance, the reported AUC=0.73 and C-index=0.69 are not interpretable as a biological association with fragment size. Please provide calibration/validation data for the >1650 bp region or rea","section":"Methods, 'BIABooster DNA analysis' and 'cfDNA variables'"},{"comment":"Twelve cfDNA variables were derived, and additional ratios were explored; the variable with the highest in-sample AUC (R>1650) was then reported as the best predictor. No correction for multiple testing is applied to the reported ORs, HRs, or p-values. Similarly, the Kaplan-Meier analyses and log-rank tests use thresholds optimized on the same data via surv_cutpoint, so the associated p-values are optimistic. The bootstrap analysis resamples performance after the variable is chosen, but it does not account for the selection step. This inflates the strength of evidence. Please report the number of tests performed, apply or justify a multiple-testing correction for the main analyses, and describe the threshold selection procedure transparently in the interpretation of p-values.","section":"Methods, 'Statistical analysis'; Results, Table 2A"},{"comment":"The conclusion that cfDNA size profiling 'outperforms PD-L1' is not directly supported by the analyses. PD-L1 (TPS/CPS) is not compared in the multi-cancer cohort in Table 2A; it is only available as a confounder in a subset of NSCLC patients, and CPS is available for only 10 HNSCC patients. The comparison in the Discussion uses pooled literature AUCs for TMB and PD-L1, which is not a head-to-head comparison in the same patient cohort. Please either provide a direct comparison with PD-L1 in the relevant subpopulations or soften the claim to 'compares favorably with literature benchmarks.'","section":"Abstract and Discussion"},{"comment":"In the NSCLC first-line subgroup, the univariable associations of P2, P2-P1, R>1650, and C_TOT with EP do not remain significant in multivariable analysis, despite AUCs above 0.83. The PFS associations remain significant, but the EP association is a central component of the paper's claim. The authors should discuss the instability of the EP association after adjustment, and avoid presenting the NSCLC subgroup as confirming the EP result.","section":"Results, 'In NSCLC first-line subgroup' and Table 2B"}],"minor_comments":[{"comment":"The size range is stated as 100-1500 bp in the instrument description and 75-1650 bp in the DNA analysis paragraph. Please harmonize these values and clarify which range is used for calibration.","section":"Methods, 'Instrument and capillary assembly'"},{"comment":"The legend uses 'R>1660' where it should be 'R>1650'.","section":"Figure 4A legend"},{"comment":"The label 'positive predive value' contains a typo: 'predictive.'","section":"Figure 2D"},{"comment":"The phrase 'avoid the unnecessary continuation of ICI therapy in sensitive patients' appears to say the opposite of what is intended; likely 'insensitive' or 'non-responsive' patients.","section":"Discussion"},{"comment":"The text says 'aross' (typo for 'across') in the sentence 'All tumor types were represented aross the three clusters.'","section":"Results, 'Unsupervised hierarchical clustering'"}],"recommendation":"major_revision","confidential_remarks":"The central measurement issue (R>1650 outside calibrated range) is not a routine statistical concern; it questions the biological interpretability of the primary variable. However, the authors may be able to address this by demonstrating device calibration for long fragments, or by refocusing the claim on calibrated variables such as R[580,1650]. I would not reject outright because the prospective design and internal validation are valuable, but the current manuscript overstates the evidence."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know two things about this paper. First, it reports a genuinely new and clinically useful association: in a prospective multi-cancer cohort, higher pre-treatment levels of very long cfDNA fragments (>1650 bp) were linked to better ICI response and longer PFS. The direction is new — most cfDNA literature emphasizes short fragments as the bad actors, and no prior study has pointed to long fragments as protective. Second, the primary predictor is measured outside the instrument's calibrated range. The Methods explicitly state that above 1650 bp, all fragments migrate at similar speeds, so size determination is unreliable. The authors say the values still reflect relative amounts, but that assertion is not backed by calibration or comparison with an independent sizing method. If that signal is actually a proxy for sample quality, total concentration, or some electrophoresis artifact, the headline result falls apart.\n\nWhat the paper does well: the study is prospective, the cohort is real-world and multi-cancer, and the statistical methods are mostly appropriate. They adjust for key confounders (age, sex, ECOG, tumor type, NLR), run bootstrap internal validation, and report consistent associations in two homogeneous subpopulations. The optimal threshold for R>1650 was similar across cohorts (0.036–0.039), which is a good sign. The authors also openly call for external validation and acknowledge that size ranges were arbitrary. That honesty is worth something.\n\nThe soft spots follow from the measurement problem. The paper treats an uncalibrated region as a biological quantity. In addition, twelve candidate variables were tested with no multiple-testing correction, and thresholds were optimized on the same data that generated the hazard ratios and AUCs. The cohort shrank from 337 to 126 after exclusions, which could introduce selection bias, though the exclusions are described. The claim of outperforming PD-L1 is not directly tested — PD-L1 is not included in the main multivariable model, only in subgroup analyses with many missing values. The cGAS-STING mechanistic story is plausible but speculative.\n\nBottom line: this is a serious paper with a real hypothesis, but the core measurement needs to be verified. If the above-1650 signal can be anchored to true fragment lengths — for instance by spiking controls or orthogonal sequencing — the biomarker could have legs. As it stands, the result is intriguing but not solid. I would send it to peer review, because the question matters and the limitations are addressable, but I would push hard for either a justification of the measurement or a reframing that does not rest on uncalibrated values. A good referee could help the authors make that case.","headline":"A clinically promising cfDNA biomarker for ICI response, but the headline predictor R>1650 is measured in a size range the device itself says is unreliable, so the central claim needs stronger support before it can be trusted.","tokens_in":17833,"tokens_out":1766,"would_cite":false,"duration_ms":21312,"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":"Pre-treatment blood DNA size profiles predict early progression and survival on immunotherapy in a prospective multi-cancer study.","keywords":["cell-free DNA","fragmentomics","liquid biopsy","immunotherapy","early progression","progression-free survival","biomarker","BIABooster"],"falsifier":"Measure the >1650 bp fraction in the same baseline plasma samples with an orthogonal high-resolution method that can truly resolve long fragments, such as nanopore sequencing or pulsed-field electrophoresis; if the orthogonal long-fragment fraction does not correlate with R>1650 or does not reproduce the progression-free survival association, the central claim is an artifact of the uncalibrated BIABooster range.","tokens_in":16991,"feed_emoji":"🧬","tokens_out":4355,"duration_ms":44167,"temperature":0.7,"pith_summary":"This paper claims that the size distribution of cell-free DNA in blood drawn before starting immune-checkpoint inhibitor therapy predicts whether advanced cancer patients will progress early. In 126 patients across four carcinoma types, the relative amount of very long fragments (>1650 bp) showed the strongest association: patients with more of these fragments were less likely to progress at first imaging and had longer progression-free survival. These associations held after adjustment for age, sex, performance status, tumor type, and neutrophil-to-lymphocyte ratio, and in two homogeneous subgroups, suggesting a multi-cancer, treatment-independent signal. If true, a simple blood test could complement or outperform PD-L1 and NLR for treatment stratification.","feed_headline":"Long DNA fragments predict immunotherapy responders","feed_subtitle":"A simple blood test for DNA fragment size beat PD-L1 and NLR at spotting early progression.","key_machinery":"The central object is the cfDNA fragment size profile, obtained by capillary electrophoresis and summarized as the relative concentration of fragments in size bins aligned to nucleosome multiples. The key predictive variable, R>1650, is the relative quantity of fragments longer than 1650 base pairs; it carries the argument because patients with more of these very long fragments show markedly lower early-progression risk and longer progression-free survival, and the signal remains after multivariable adjustment.","core_discovery":"On the paper's own terms, the central discovery is that baseline cfDNA size profiles, measured with the BIABooster device, contain a predictive signal for immune-checkpoint inhibitor response. The proportion of fragments longer than 1650 bp (R>1650) was the single best discriminator: it was associated with lower odds of early progression (OR = 0.39) and longer progression-free survival (HR = 0.54), with an AUC of 0.73 and a C-index of 0.69. Longer dinucleosomal fragments and a wider gap between the first two nucleosomal peaks also predicted better outcomes, while high total cfDNA concentration and an abundance of short mononucleosomal fragments predicted worse outcomes. The paper interprets","pith_inferences":["Because R>1650 is measured in a range where the device cannot reliably determine fragment size, the result may partly reflect total high-molecular-weight DNA or an electrophoretic artifact; orthogonal sizing could show whether the effect is truly length-specific.","The paper's proposed cGAS-STING mechanism is a hypothesis, not a demonstrated causal chain; if length-dependent STING activation holds, R>1650 could become a pharmacodynamic biomarker for STING-directed therapies.","The arbitrary size-bin boundaries leave information on the table; a data-driven continuous model of the fragment size curve might improve prediction beyond the single R>1650 variable.","An external prospective validation in an independent cohort, ideally with treatment-specific stratification, would be needed before this becomes a clinical decision tool."],"forward_implications":["Baseline cfDNA size profiling could identify patients unlikely to progress early on ICI therapy, supporting earlier treatment adaptation or avoiding ineffective cycles.","R>1650 may serve as a multi-cancer predictive biomarker that outperforms PD-L1 and NLR, using a standardized noninvasive assay.","Combining cfDNA size features with NLR could catch progressors missed by either marker alone, since the two signals are weakly correlated.","If long cfDNA fragments indeed activate cGAS-STING, R>1650 could also guide patient selection for combination therapies involving STING agonists.","The consistency of the optimal R>1650 threshold across the overall cohort and two subgroups suggests the signal may be robust across tumor types and treatment lines."],"supporting_citations":[{"why":"Supplies the BIABooster device and the 10 fg/µL sensitivity on which the cfDNA size profiling is based.","marker":"[19]"},{"why":"Shows cfDNA concentration and size can be measured directly from blood plasma without prior DNA extraction, the workflow used here.","marker":"[20]"},{"why":"Prior evidence that cancer patients are enriched in short cfDNA fragments, against which the paper's long-fragment finding is framed.","marker":"[17]"},{"why":"Provides the apoptosis and nucleosome release mechanism that motivates interpreting fragment sizes as biological signals.","marker":"[9]"},{"why":"Meta-analysis giving pooled PD-L1 and TMB AUC/PPV values that the cfDNA classifier is compared against.","marker":"[27]"},{"why":"Supports the proposed cGAS-STING pathway link between cytosolic DNA and immune activation.","marker":"[42]"},{"why":"Shows cGAS is activated by DNA in a length-dependent manner, underpinning the mechanistic interpretation of long fragments as immunostimulatory.","marker":"[43]"}],"fun_headline_variants":["Long DNA fragments predict better checkpoint inhibitor outcomes","cfDNA length profile anticipates immunotherapy response","Long DNA fragments predict longer survival on ICIs","Fragment size profile outperforms PD-L1 for ICI response"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The key predictor is a relative signal from fragments longer than 1650 bp, a range where the device's own methods say size determination is unreliable; if that signal just tracks total cfDNA concentration or an instrument artifact, the association is not about long fragments at all.","fun_headline_variants_meta":{"raw":{"variants":["Long DNA fragments predict better checkpoint inhibitor outcomes","cfDNA length profile anticipates immunotherapy response","Long DNA fragments predict longer survival on ICIs","Fragment size profile outperforms PD-L1 for ICI response"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000832,"raw_usage":{"total_tokens":3579,"prompt_tokens":961,"completion_tokens":2618,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":705,"completion_tokens_details":{"reasoning_tokens":2571}},"tokens_in":705,"tokens_out":2618,"duration_ms":20295,"temperature":1.0,"reasoning_tokens":2571,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T05:45:27.881871+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the >1650 bp fraction in the same baseline plasma samples with an orthogonal high-resolution method that can truly resolve long fragments, such as nanopore sequencing or pulsed-field electrophoresis; if the orthogonal long-fragment fraction does not correlate with R>1650 or does not reproduce the progression-free survival association, the central claim is an artifact of the uncalibrated BIABooster range.","supporting_citations":[{"cited_title":"BIABooster: Online DNA Concentration and Size Profiling with a Limit of Detection of 10 fg/μL and Application to High -Sensitivity Characterization of Circulating Cell -Free DNA","cited_arxiv_id":null,"evidence_quote":"Supplies the BIABooster device and the 10 fg/µL sensitivity on which the cfDNA size profiling is based."},{"cited_title":"Size and Concentration of Cell -Free DNA Measured Directly from Blood Plasma, without Prior DNA Extraction","cited_arxiv_id":null,"evidence_quote":"Shows cfDNA concentration and size can be measured directly from blood plasma without prior DNA extraction, the workflow used here."},{"cited_title":"Enhanced detection of circulating tumor DNA by fragment size analysis","cited_arxiv_id":null,"evidence_quote":"Prior evidence that cancer patients are enriched in short cfDNA fragments, against which the paper's long-fragment finding is framed."},{"cited_title":"Cell -Free DNA and Apoptosis: How Dead Cells Inform About the Living","cited_arxiv_id":null,"evidence_quote":"Provides the apoptosis and nucleosome release mechanism that motivates interpreting fragment sizes as biological signals."},{"cited_title":"The cytosolic DNA -sensing cGAS-STING pathway in cancer","cited_arxiv_id":null,"evidence_quote":"Supports the proposed cGAS-STING pathway link between cytosolic DNA and immune activation."},{"cited_title":"cGAS is activated by DNA in a length‐ dependent manner","cited_arxiv_id":null,"evidence_quote":"Shows cGAS is activated by DNA in a length-dependent manner, underpinning the mechanistic interpretation of long fragments as immunostimulatory."}],"review_version":1}