Paper Review — verify claims with raw data

Visual summary of diagnostic performance

Key numbers (raw from paper)

Critical appraisal (evidence-first, with inline source excerpts)

Primary source: Hashim et al., Frontiers in Oncology 2018 — full methods reported (3 mL plasma, Tellgen extraction & MSP, ACTB internal control, Ct<35 positive), sample n=248, hospital-based case-control design. Strength: pathology confirmation of CRC; direct comparison vs FOBT, CEA, Ca-199 and stratified analyses by tumor site/stage. Key numeric results reproduced below and visualized above Hashim et al. 2018 (study methods and results)

Context from systematic reviews/meta-analyses

Larger syntheses show SEPT9 sensitivity is heterogeneous (roughly 48–96% across studies depending on assay algorithm and population) and specificity generally high; PRESEPT-like population screening studies often report lower sensitivities than case-control series, reflecting spectrum bias and ascertainment differences Zhang et al. / Grutzmann meta-analysis 2017

Major strengths of Hashim et al. 2018

• Direct head-to-head within-cohort comparison of mSEPT9 vs FOBT and serum markers (CEA, Ca-199) using the same subjects and pathology gold standard.
• Detailed methods (kits, Ct thresholds) that enable partial reproducibility.
• Stratified sensitivity analyses by tumor site and stage (though small n in subgroups).

Major limitations, biases, and blindspots

1. Case-control design, hospital-based recruitment → spectrum and selection bias: case-control estimates typically inflate sensitivity relative to asymptomatic screening populations (external screening cohorts show lower sensitivity) SEPT9 meta-analysis: bias from design
2. Small plasma volume (3 mL) and one-time sampling likely reduced analytical sensitivity vs larger-volume assays used in other studies (often 10 mL and duplicate testing) — preanalytical cfDNA yield matters for detection of low-frequency methylated ctDNA.
3. Missing FOBT in ~35% of participants; non-random missingness could bias FOBT vs mSEPT9 comparisons (authors report randomness among cases but not fully across cohort).
4. Very small subgroup counts for several strata (e.g., stage I n=5; ileocecal n=6) → unstable sensitivity estimates (wide uncertainty) and risk of overfitting claims like 100% sensitivity in tiny subgroups.
5. Commercial kit and algorithm details not cross-validated with gold-standard code/controls; possible manufacturer/kit variability (Tellgen kit used; other studies use Epi proColon/Abbott) affecting comparability.
6. No external validation cohort, longitudinal follow-up, or prognosis correlation (limiting clinical impact beyond cross-sectional detection performance).

What the numbers really imply (practical view)

mSEPT9 in this cohort is more specific than FOBT (89.6% vs 70.3%) and roughly similar in sensitivity (61.8% vs 61.4%). Combining tests markedly increases sensitivity (to ~84%) at the cost of specificity (~62%) — an expected trade-off. For population screening, the higher specificity of mSEPT9 could reduce unnecessary colonoscopies compared with FOBT if sensitivity were adequate, but the observed moderate sensitivity (≈62%) still misses a substantial fraction of CRC cases, especially early-stage disease in which ctDNA burden is low Hashim et al. stage-specific sensitivities

Counterpoints, external evidence and where conclusions could change

Meta-analyses and larger prospective cohorts (PRESEPT, RESEPT, others summarized in reviews) show algorithm- and kit-dependent performance and generally warn that case-control studies overestimate performance; therefore, Hashim et al.'s conclusion that mSEPT9 is "promising" is supported, but its magnitude and applicability to population-screening depend on assay version, pre-analytic volume, algorithm (1/3 vs 2/3), and cohort (screening vs symptomatic) SEPT9 meta-analysis: generalizability caveats

Practical recommendations if replicating/extending this work

• Use larger plasma volumes (≥8–10 mL) and replicate measurements to improve cfDNA yield and analytical sensitivity.
• Run a prospective screening cohort (asymptomatic) rather than hospital case-control to avoid spectrum bias and provide realistic sensitivity estimates for population screening.
• Apply/compare standardized commercial kits and algorithms (e.g., Epi proColon 2.0, 1/3 vs 2/3) and report per-algorithm performance; preregister analysis plan to reduce selective reporting.
• Report Ct distributions, limit-of-detection, and analytic reproducibility (inter-assay, intra-assay variability), and deposit anonymized diagnostic data to enable meta-analytic pooling.
• Investigate multiplex ctDNA methylation panels (SEPT9 + SDC2 + BCAT1 etc.) which recent studies show higher sensitivity and specificity than single-marker assays 2025 multi-gene methylation panel study

Selected citations used in this critique

• Hashim et al. 2018 (primary paper)
• 2017 SEPT9 meta-analysis
• 2025 multi-gene methylation panel

Confidence in the core numerical reproduction (sensitivity/specificity/AUC) is high because values are explicitly reported in the full text; interpretation confidence is moderate given the single-center, case-control nature and known heterogeneity of SEPT9 performance across assays and populations.