Sample report · real, unedited output on a published paper · a methods exercise, not a commentary on its authors

The manuscript is Towers et al., Am J Obstet Gynecol 2020;222:83.e1-8 ↗. It has never been retracted, and nothing here suggests it should have been.

This is a methods exercise in a validation context — the way the literature routinely critiques published work. The report below is the engine's unedited output, and nothing about what happened after publication was supplied to it: no later literature, no commentary. We can't claim these models have never encountered a well-known trial — only that none of its aftermath was put in front of them, and that the reading below is argued from the manuscript's own numbers. One section is deliberately not blind to the present: journal compliance is checked against the journal's current requirements, which is why a paper can be asked for something its journal did not require at the time it was written.

Nothing here alleges misconduct. The findings concern how claims are stated and how far the reported results carry them — the ordinary subject of methodological review. Passages from the paper are quoted for criticism and commentary, and remain the work of their authors and publisher.

What this sample shows

The engine read the article as it stood at publication. The paper carries a publisher's banner pointing to a later update; that banner was removed from the input before the run and its absence verified. Nothing about what happened after publication reached the screen — no correction text, no subsequent commentary. What it read was the published article, not the manuscript as submitted: a copyedited paper of record, which differs from a submission in ways the evaluation itself notes.

It took the manuscript at its word about the design, and said so in the strengths. The evaluation credits “prospective data collection with an a priori sample-size calculation”. In June 2023 the journal published a correction stating that the study was listed as a prospective cohort — “which is not the case” — and that the power analysis was performed after four months of data collection had already happened. A screen that reads text cannot know that. The manuscript asserts prospective collection throughout and nothing in it dissents, so the engine believed it, the way a reviewer reading the same pages would have. We are not claiming it caught this.

What it did instead is the part worth judging. It routed the paper — external review may be reasonable with defined major concerns. That route is an aggregate judgment across the whole appraisal, scope and reporting included — but the findings underneath it are read off the paper's own numbers, and they stand whatever the design was called. It named the structure below the headline: the treated group is composed of women who had already achieved full detoxification, so the difference in newborn withdrawal follows from the selection criterion as much as from the drug. It flagged that the headline figure counts only patients who stayed on treatment to delivery, and asked for the intention-to-treat contrast. Among its questions for the authors it asked them to clarify whether the study was registered.

That is a question an editor can act on, and it does not require knowing the design claim was wrong. Registration is the record that would have dated the protocol, and dating the protocol is what the 2023 correction eventually did — after the paper had stood in the literature uncorrected for three years. The engine reached that question from the reported numbers alone.

How firmly it reaches for that document varies between runs, and we would rather say so than show you the best one. An earlier run of this same manuscript asked, in a single question, for the prospective protocol and the registration status — a sharper ask than the one above. Same engine, same prompts, different roll: the appraisal layer is a language model and its questions are not deterministic, which is why the arithmetic layer beneath it is, and why every number in the section below is recomputed rather than written. A sample is one run. Re-running until the strongest wording came back would have made this page a claim about our patience.

Two engines read this manuscript separately, and the report keeps their readings apart. Neither saw the other's work. They named different routes: one returned external review may be reasonable with defined major concerns, the other author clarification or targeted analysis likely needed. The delivered route is external review may be reasonable with defined major concerns. A disclosed split tells an editor where the judgment is genuinely contested; a single confident answer would have hidden it. They also parted on 2 dimensions — audience fit, guideline compliance, and 17 concerns came from one engine alone against 10 both engines reached, so the reconciled report is wider than either reading by itself.

Which way that lands is not fixed, and this page will not pretend otherwise. Every sentence above is read out of the artifact rather than written into the page, because re-running this same manuscript does not reproduce the same split: the appraisal layer is a language model. That is the division of labour the section below rests on — the arithmetic is recomputed and reproducible, the reading of it is a judgement, and the report marks which is which.

What was supplied. One document: the published article as a single PDF. An editorial evaluation reads exactly one manuscript document, and that is all this run was given. No protocol, no supplementary files, no statistical analysis plan — and the evaluation says so itself, listing their absence among the limits on what it could assess. A real submission arrives with more than this one did.

This is what the arithmetic layer does with a modern PDF. This manuscript has selectable table text, so its cell values survive into what the engine reads. It transcribed 72 values from four tables and recomputed 66 checks, of which 63 reconciled. Rows whose numbers the engine could not locate back in the manuscript are withheld rather than shown, so the count is what the layer could actually ground, not what it attempted.

Three checks did not reconcile, and none of them is an arithmetic error. Newborn length of stay is summarised as a mean with a standard deviation larger than half the mean, which for a quantity that cannot go below zero implies a skewed distribution that a median and interquartile range would represent more faithfully. The other two are p-values printed as 0.99 where the table they come from recomputes to a slightly different value — the significance call is unaffected in both, and the report says so on each row. The report's own What was read section states what reached the engine and what the layer got out of it.

The journal context is our curated summary of the American Journal of Obstetrics & Gynecology's public author instructions. On a real evaluation a journal supplies its own, including private criteria it does not publish.

RigorMD Editorial Evaluation

Use of naltrexone in treating opioid use disorder in pregnancy

Editorial decision support · Structured manuscript triage and peer-review support. Editors and editorial teams screening manuscripts.

generated 2026-09-06
format editorial-evaluation-report-2026-07-07.1

Structured editorial decision support. Not an editorial decision, and not a replacement for peer review.

Editor's snapshot

Is this worth sending to a reviewer?

External review may be reasonable with defined major concernsConcern

Two engines read this manuscript and named different routes — first engine: External review may be reasonable with defined major concerns; second engine: Author clarification or targeted analysis likely needed. The route above is the one delivered after reconciliation and, where it applies, the numeric layer's protection.

Minor

Audience fit & interest

The obstetrics/OUD treatment topic fits the supplied Obstetrics original-research context, while the novelty placement rests on the authors' literature claim rather than independently supplied evidence.

Concern

Study validity & design

Treatment is assigned by patient choice and the naltrexone arm is opioid-free by design, so the headline NAS difference is confounded by indication and partly definitional.

Concern

Conclusions vs. evidence

The 'viable option' conclusion is appropriately hedged, but the causal framing of lower NAS as a naltrexone effect overreads a comparison that is structurally driven by opioid-free status.

Concern

Statistical reporting & rigor

The multivariate result for 'form of MAT' is reported inconsistently (a 'significant' factor with P>.3), a right-skewed length-of-stay is summarized as mean±SD, and several 2x2 p-values are imprecise.

Concern

Guideline & checklist compliancefixable

STROBE documentation and several current journal-required administrative statements are not visible in the supplied PDF/context and should be requested or clarified.

Clarity & communication

Not assessed: no grounded writing or communication signal in this read. The dimension is reported only when one is.

Contribution typewhere it sits in the literature — not a concern rating

ConfirmatoryEvolutionaryNovelLandmark

Authors claim the first and largest prospective study of naltrexone MAT in pregnancy; novelty rests on the authors' own statement and is not independently verified here.

Journal and manuscript metadata

Manuscript: AJOG-2020-83e1
Article type: Original research
Journal section: Obstetrics
Prompt version: editorial-evaluation-prompt-2026-08-29.1

Topic and audience fitClear

Well aligned obstetrics original research

The manuscript is a prospective obstetric cohort comparing naltrexone MAT with methadone/buprenorphine MAT in pregnant patients with opioid use disorder, matching the declared original_research article type and Obstetrics section.

Grounding

Use of naltrexone in treating opioid use disorder in pregnancy

Design, data, and conclusions fitConcern

Confounded by treatment choice

Group membership is determined by patient choice after successful detoxification, and the naltrexone arm is opioid-free by design, so the lower NAS rate is substantially confounded by indication and is close to definitional rather than a demonstrated pharmacologic effect. The authors acknowledge non-randomization but the conclusions still narrate a treatment benefit.

Potential for improvement with current dataConcern

Reframing and reanalysis feasible

With the data in hand the authors can reframe the NAS finding as reflecting opioid-free status rather than a naltrexone effect, present length-of-stay as median (IQR), and clarify the multivariate model reporting; the core observational limitation cannot be removed but can be handled transparently.

Statistical and methodological alignmentConcern

The section written for the reader a clinical reviewer is not: threats to the inference that survive clinical peer review — unadjusted confounding and its direction, multiplicity without correction, immortal time, an odds ratio narrated as a risk, a stateable missing-data mechanism, unmodelled clustering, competing risks treated as censoring, a precision claim the sample cannot support, ITT and per-protocol conflated. Where one applies it is named here; the instance it rests on, and the reviewer competence that would settle it, are below.

Confounding and reporting inconsistencies

The design does not address confounding by indication in the primary comparison, the multivariate statement is internally contradictory ('only significant factor' yet reported as P>.3), and a skewed continuous outcome is summarized as mean±SD. A methods/biostatistics reviewer is needed to check the model specification and effect-estimate presentation.

Grounding

Length of hospital stay, da 5.5 6.1 20.8 6.0 < .0001

Statistical/methods follow-up needConcern

Biostatistics review advised

Editorial follow-up should route the multivariate modelling, denominator handling for the NAS comparison, and the contradictory significance statement to a biostatistician before or alongside external review.

Grounding

10/119 [8.4%] vs 79/105 [75.2%]

What was read

This evaluation read one manuscript document, which is what an editorial evaluation takes — 38,224 characters of extracted text.

The manuscript refers to Table 1, Table 2, Table 3 and Table 4. The deterministic layer transcribed 72 structured values from the text and recomputed 66 checks.

Where a table is an image rather than selectable text, its cells cannot be transcribed and the checks that would depend on them do not run. This report does not distinguish a table with nothing to flag from a table it could not read — so a short list of checks is not, on its own, evidence that the numbers are sound.

Recomputed checks

Statistics recomputed from the manuscript's own printed numbers — arithmetic only, run independently of both appraisers. A consistent check means the printed value is reproducible from the values beside it; it does not assess the underlying data. The arithmetic on any one number is exact and repeats; which numbers this layer could pull out of the manuscript does not, so the count below is what this run reached rather than a fixed property of the paper. 66 checks recomputed · 63 consistent · 3 flagged.

Check

Two-proportion p-value recomputation

every applicable test contradicts reported p = 0.99 (Fisher's exact p=1.0000, chi-square p=0.8951, Yates-corrected chi-square p=1.0000); the significance call itself stands

Reported vs recomputed

test: unstated
recomputed_p_fisher: 1
recomputed_p_chi2: 0.8951
recomputed_p_chi2_yates: 1
reported_p: 0.99
table: 115, 6 · 104, 5
significance_flip: no
test_dependent: no

Inconsistent — flagged in findings

Check

Two-proportion p-value recomputation

every applicable test contradicts reported p = 0.99 (Fisher's exact p=1.0000, chi-square p=0.8567, Yates-corrected chi-square p=1.0000); the significance call itself stands

Reported vs recomputed

test: unstated
recomputed_p_fisher: 1
recomputed_p_chi2: 0.8567
recomputed_p_chi2_yates: 1
reported_p: 0.99
table: 5, 116 · 4, 105
significance_flip: no
test_dependent: no

Inconsistent — flagged in findings

Check

Mean vs SD skew screen

presentation (mild): 1 of 10 mean±SD summaries of non-negative quantities have mean − 2·SD < 0, so those distributions are likely skewed (Length of hospital stay (d)) — median (IQR) would represent these data better

Reported vs recomputed

screened: 10
flagged: 1
quantities: Length of hospital stay (d)
rows: quantity: Length of hospital stay (d) · mean: 5.5 · sd: 6.1 · spread label: sd · n: 121 · implied from sem: no

Inconsistent — flagged in findings

Consistent, nothing to report: Two-proportion p-value recomputation (21); Subgroup counts vs analytic N; Percentage recomputation (26); GRIM test — mean/N consistency (2); Impossible p-values; Threshold-only p-value reporting; Over-precise p-values; Unlabeled dispersion (SD vs SEM); Cross-location value consistency (2); Percentage precision vs sample size; Two-group mean ± SD p-value recomputation (5); Fragility index vs loss to follow-up.

Non-obvious weaknessesConcern

Near tautological primary outcome

NAS arises from in-utero opioid dependence; the naltrexone arm is opioid-free by protocol, and the 10 NAS cases in that arm came from patients who discontinued naltrexone and resumed opioids. The 'significantly lower NAS' finding therefore largely reflects opioid-free status at delivery rather than a naltrexone treatment effect. A perinatal epidemiologist or addiction-medicine reviewer should assess whether the causal framing is defensible.

Confounding by indication

Assignment is by patient choice after successful detoxification, so the naltrexone group is a self-selected, more motivated and already opioid-free population; baseline balance on demographics does not address this selection, and multivariate adjustment cannot recover the counterfactual.

Contradictory multivariate reporting

The text states the form of MAT was the only significant factor for NAS yet reports it as P>.3, an internally inconsistent result that a reader could not reconcile; the model list and p-values need reconciliation.

Skewed continuous summary

Length of hospital stay is presented as mean±SD (5.5±6.1) where mean minus two SD is negative, indicating right-skew; median (IQR) would better represent the distribution and the recomputation flags this.

Grounding

Length of hospital stay, da 5.5 6.1 20.8 6.0 < .0001

Imprecise 2x2 pvalues

Reported p=0.99 for balanced 2x2 comparisons does not match exact/chi-square recomputation (which give ~1.0/0.86); the non-significance call is unaffected but the reported values are imprecise.

Grounding

Cocaine 5 (4.1) 4 (3.7) .99
White, n (%) 115 (95) 104 (95.4) .99

Confounding by detoxification success and engagement

The comparison is vulnerable to a confounder that ordinary clinical review may underweight: patients able and willing to detoxify and choose naltrexone may differ in motivation, addiction severity, and care engagement, likely biasing relapse and NAS outcomes toward the naltrexone group. A perinatal epidemiologist or addiction-methods reviewer should assess what measured proxies can be adjusted or stratified.

Full-cohort and continuation-to-delivery estimands blur

The manuscript highlights no NAS among patients receiving naltrexone to delivery, but that is a post-baseline continuation subgroup rather than the same estimand as initial treatment-group comparison. A methods reviewer should ask for clearly separated intention-to-treat-like, as-treated, and per-protocol analyses.

NAS ascertainment may be exposure-aware

NAS requiring treatment is clinically meaningful but depends on serial Finnegan scoring and treatment thresholds; the supplied text does not state whether scorers or treatment teams were blinded to maternal MAT. A neonatology reviewer should check whether exposure awareness could have influenced treatment decisions.

Grounding

consecutive Finnegan scores of 10
single Finnegan score of

Adjusted model underreported

The multivariable result is summarized mainly as significance/non-significance, without model family, coefficients, confidence intervals, or an explicit rationale for covariate selection and events-per-variable adequacy. Statistical review should determine whether the model supports the comparative claim.

Multiplicity across secondary outcomes

Multiple obstetric, neonatal, exposure-timing, and blood-level comparisons are presented with a general .05 threshold and no stated multiplicity plan; secondary findings, especially head-circumference subgroup trends, should be treated as exploratory. A statistical reviewer should check whether any conclusion relies on unadjusted secondary testing.

Journal-specific publishing criteriaConcern

Ajog at a glance present

The required 'AJOG at a Glance' section for original research is present with the three prescribed elements.

Grounding

AJOG at a Glance
Why was this study conducted?
BACKGROUND:
OBJECTIVE:
STUDY DESIGN:
RESULTS:
CONCLUSION:

Structured abstract present

A structured abstract with Background/Objective/Study Design/Results/Conclusion is present, consistent with the journal requirement; exact word count could not be verified from the PDF.

Grounding

OBJECTIVE: Our study objective was to evaluate prospectively obstetric

Required statements partially verifiable

IRB, COI and funding statements are present, but AI disclosure, author contributions, a condensation/short title, and study/trial registration required for original research are not evident in the supplied file.

Grounding

The authors report no conflict of interest.

IRB statement present

The manuscript includes an institutional review board approval statement, satisfying the visible ethics-statement requirement in the supplied PDF.

Author contributions and AI disclosure not visible

Current supplied journal instructions require author-contribution and AI-disclosure statements, but these are not visible in the supplied PDF text. The office should request the submission metadata or author clarification rather than assuming absence from all files.

Trial registration status unclear

The current supplied journal instructions list trial registration as required for original research, and this prospective treatment cohort does not show a registration number or an explicit exemption/non-trial statement in the supplied PDF. Author clarification is needed on whether registration was applicable and, if not, why.

Condensation and short-title elements not assessable

The clean article PDF does not reliably show all submission-title-page elements, including any required condensation line. The title page or submission system metadata would resolve this.

Reference style appears journal-formatted

The references are numbered and presented in journal style in the supplied PDF. Full cited-in-order verification was not performed from the extracted text.

Grounding

1. Scholl L, Seth P, Kariisa M, Wilson N,

Author-instruction and reporting-guideline complianceConcern

Strobe cohort not referenced

As an observational cohort the manuscript falls under STROBE, but no STROBE adherence or flow accounting beyond the Figure is stated; a reviewer should confirm complete reporting of eligibility, confounder handling, and losses.

Registration not shown

Trial/study registration is a required statement for original research in the supplied instructions and is not visible in the PDF for this prospectively conducted study.

STROBE cohort-reporting package incomplete

Because this is a prospective cohort study, STROBE is the relevant public reporting framework. The manuscript reports several cohort elements, but the supplied files do not include a STROBE checklist and the statistical reporting gaps around model specification, adjusted estimates, and sensitivity analyses should be reconciled against STROBE items.

Place in the literatureMinor

First prospective extension of retrospective series

The work extends prior retrospective Australian naltrexone-implant series with a prospective oral-naltrexone cohort and novel paired maternal/cord blood levels, positioning it as an incremental-to-novel contribution against existing evidence and a task-force statement advising against naltrexone.

Grounding

naltrexone should not be used as a MAT for the management of OUD in pregnancy because of insufficient data
This is the first prospective study

Strengths worth preserving

Prospective data collection with a prespecified sample size calculation and a clearly described clinic protocol.
Novel paired maternal/umbilical-cord naltrexone and 6-beta-naltrexol levels with time-since-discontinuation, informing placental transfer and clearance.
Transparent acknowledgement of non-randomization and of possible type II error for secondary outcomes.
Directly engages a contrary professional-society recommendation, giving the findings clear clinical context.
Prospective data collection from a designated obstetric OUD clinic with a contemporaneous comparison group from the same clinical setting.
Objective urine toxicology was reported during prenatal care and at delivery.
Maternal and umbilical cord naltrexone/metabolite levels provide useful pharmacologic information in pregnancy.
The authors explicitly acknowledge nonrandomization and possible selection bias.
The most defensible conclusion language is already partly hedged toward selected patients and should be preserved rather than broadened.

Suggested reviewer or statistical-review focus

Whether the primary NAS comparison is confounded/definitional given the naltrexone arm is opioid-free by design, and how this should reshape the causal claims.
Specification and reconciliation of the multivariate NAS model, including the contradictory 'only significant factor' with P>.3 statement.
Handling of changing denominators (10/119 vs 79/105) after excluding <34-week gestations from the NAS comparison.
Appropriateness of mean±SD for skewed outcomes (length of stay) versus median (IQR).
Adequacy of powering for secondary outcomes (birthweight, head circumference) and interpretation of borderline trends in Table 3.
Perinatal epidemiology/statistics: define the primary estimand and separate choice-to-naltrexone, as-treated, and continuation-to-delivery analyses.
Assess residual confounding by detoxification success, motivation, addiction severity, care engagement, timing of opioid discontinuation, and traditional MAT dose.
Request adjusted effect estimates with 95% confidence intervals, model family, covariate-selection rationale, and events-per-variable assessment.
Neonatology/NAS reviewer: verify Finnegan scoring standardization, treatment thresholds, and whether assessors/treating teams were blinded to maternal treatment.
Check whether secondary endpoint and subgroup claims are framed as exploratory given no stated multiplicity plan.
Confirm and correct the two minor P=.99 inconsistencies and summarize hospital length of stay with median/IQR.
Editorial office: obtain STROBE checklist, registration/exemption statement, author-contribution statement, and AI disclosure if not already in submission metadata.

Highest-yield author queries

Please clarify the multivariate result for 'form of MAT': is it significant, and what is the exact estimate and p-value?
How do you distinguish a naltrexone pharmacologic effect on NAS from the fact that the naltrexone arm is opioid-free at delivery?
Was this prospective study registered, and can you provide the registration identifier?
Please provide AI-use disclosure, author contributions, and a condensation/short title per journal requirements.
Can length of hospital stay and other skewed outcomes be reported as median (IQR)?
What was the prespecified primary estimand: initial clinical pathway choice, actual treatment received, or naltrexone maintained through delivery?
Please provide intention-to-treat-like, as-treated, and per-protocol/continued-to-delivery analyses for NAS, with denominators reconciled across exclusions under 34 weeks.
Please provide adjusted risk differences or risk ratios with 95% confidence intervals and full model specification for NAS and other key outcomes.
What measured variables capture OUD severity, treatment engagement, prior relapse/overdose, opioid dose, timing of detoxification, and adherence, and can these be adjusted or stratified?
Were neonatal Finnegan scorers and treatment decision-makers blinded to maternal MAT exposure; if not, how was ascertainment bias minimized?
Was this prospective treatment cohort registered, and if not, what is the rationale for non-registration or exemption?
Please supply the STROBE checklist and any current journal-required author-contribution and AI-disclosure statements not visible in the PDF.

Limitations of this evaluation

Assessment is based solely on the submitted PDF; supplementary files, cover letter, and any registration or reporting checklists were not available.
Exact abstract word count, condensation, short title, author-contribution and AI-disclosure statements could not be confirmed from the extracted text.
Table values were extracted from imperfect PDF text, so some numeric quotes may not exactly match the typeset table and were treated cautiously.
No access to the analytic dataset or statistical code prevented independent verification of the multivariate model and p-values beyond the recomputed 2x2 checks.
This evaluation used only the supplied clean PDF text, journal metadata, private-guideline block, and deterministic numeric checks; no protocol, statistical analysis plan, code, data set, supplement, or peer-review history was available.
Separate submission-system materials may contain title-page items, trial-registration explanations, author contributions, AI disclosure, or reporting checklists that are not visible in the supplied PDF.
No independent literature search or clinical-trial registry search was performed, so novelty and registration observations are based only on the supplied context.
The deterministic numeric checks covered selected arithmetic issues only; this evaluation did not recompute every table value or every statistical test de novo.
Exact abstract word count and full reference cited-in-order compliance could not be reliably verified from the extracted PDF text.

Scope Legal and scope limits

Scope. Structured editorial decision support; not an editorial decision and not a replacement for peer review. Not medical, legal, or regulatory advice. Limited to supplied materials. Terms | Privacy | Security