Journal of Clinical Pathology Submission Guide

State the intended use and represented population before presenting performance. Name the reference standard and how disagreements were resolved. Report sample selection, prevalence or spectrum, missing and indeterminate results, thresholds, blinding, precision, reproducibility, calibration, and uncertainty as applicable.

For image-based or AI studies, explain dataset provenance, labeling, pathologist adjudication, scanner or platform variation, external validation, leakage controls, subgroup performance, and failure cases. A high aggregate score does not establish safe use.

• Use scale bars, stain or assay details, magnification, labels, and accessible legends.
• Explain specimen type, handling, fixation, processing, quality thresholds, and exclusions.
• Put the reference-standard comparison and clinically meaningful errors where readers can find them.
• Distinguish exploratory cutoffs from prespecified or independently validated thresholds.
• Reconcile the abstract, figures, tables, supplement, data statement, and conclusion.

Match reporting to the study job

Use the reporting guideline that fits the design and intended use. Diagnostic-accuracy studies often need STARD; prediction models may need TRIPOD or a current successor; observational work may use STROBE; systematic reviews may use PRISMA. The checklist is not a substitute for reasoning, but it exposes missing information about sampling, reference standards, analysis, and participant flow.

For laboratory-method comparisons, explain calibration, precision, analytical range, interference, carryover, lot variation, and quality-control rules as relevant. For prognostic markers, separate association, discrimination, calibration, incremental value, and clinical utility. A statistically independent association does not automatically improve a decision.

Build an error-consequence table

Create one artifact with error type, example case, likely cause, frequency or uncertainty, clinical or workflow consequence, detectability, and mitigation. Include indeterminate and excluded cases. This forces the manuscript to represent performance as a decision system rather than a headline metric.

Check patient flow, specimen counts, exclusions, reference-standard results, analysis denominators, figures, repositories, and abstract values against one another. Why this guide exists: official JCP instructions define the contract, but they cannot determine whether validation supports intended use. At Manusights, the intended-use and error-consequence tests expose that gap.

1. Intake: article type, files, reporting material, ethics, consent, and declarations.
2. Editorial assessment: pathology ownership, novelty, evidence sufficiency, and readership value.
3. Review: sampling, reference standard, validation, interpretation, reproducibility, and clinical use.
4. Revision: every response remains consistent with the data, images, code, limitations, and claims.

Association becomes diagnosis. State what the study can classify or predict and what validation is still missing.

The reference standard is opaque. Explain who labeled cases, which information they saw, and how disagreement was resolved.

Representative failures disappear. Show indeterminate cases, discordance, artifacts, and settings where performance degrades.

We use an intended-use test because diagnostic papers often optimize a metric before defining the decision. Write who will use the test, on which specimen, at which point in the pathway, against which alternative, and with what consequence of a false positive, false negative, or indeterminate result. Only then choose the performance measures. A metric is clinically meaningful when it answers that use case.

We also separate development from evaluation. Threshold selection, feature engineering, case exclusion, and model tuning belong to development. Performance used to support adoption should come from data not used to make those choices, ideally representing the variation the test will encounter. If independent evaluation is not available, say so and narrow the claim.

Example: a diagnostic biomarker

Define the reference standard, prevalence or spectrum, specimen handling, assay reproducibility, threshold, and handling of indeterminate values. Report sensitivity and specificity with uncertainty, but also show how predictive interpretation depends on the represented setting. Explain whether the biomarker adds information beyond current practice rather than comparing it only with no test.

Example: computational pathology

Report slide and patient splitting, scanner and site variation, label provenance, pathologist adjudication, external evaluation, subgroup performance, and failure cases. Show representative errors. A model can perform well while exploiting leakage or acquisition artifacts, so validation design is part of the claim, not a technical appendix.

The same assay can require very different evidence depending on what a laboratory is supposed to do with it. A screening tool needs sensitivity and a credible plan for false negatives. A confirmatory test needs specificity, reference-method agreement, and a clear account of discordant cases. A prognostic marker needs a defined time horizon, censoring strategy, calibration, and evidence that it adds information beyond existing clinical variables. State the intended use before presenting the headline performance number.

Do not let a high area under the curve substitute for this decision chain. Show how the threshold was chosen, what happens on either side of it, and whether performance holds outside the development data. If the paper cannot answer those questions, describe it as an early analytical or proof-of-concept study rather than a ready clinical tool.

Pathology evidence depends on the route from specimen to label. Describe pre-analytical handling, inclusion and exclusion, tissue or sample adequacy, assay failure, missing data, and the expertise used to establish the reference label. For imaging or computational pathology, say whether cases, patients, slides, and sites were separated correctly during model development and evaluation. A random image split can leak patient or site information even when the code reports a clean train-test split.

The figures should let a pathologist inspect both success and failure. Include representative difficult cases, not only visually persuasive examples. A confusion matrix or performance summary becomes more useful when paired with error categories, uncertainty, and the consequences of a wrong classification. If disagreement between pathologists is part of the problem, show how it was measured and resolved instead of treating the final label as ground truth without qualification.

The abstract, main tables, supplementary files, ethics statement, data-access language, and cover letter should describe the same cohort and intended use. Check sample counts after every exclusion, confirm that model names and thresholds agree, and make sure the limitations do not disappear from the cover letter. That record-level consistency is easy to verify before upload and difficult to repair after reviewers find a mismatch.