Medical Research: Data Collection, Analysis and Reporting

The second half of the medical research process for MD, MS and PhD researchers, covering instrument design, piloting, data cleaning, missing data, statistical analysis, effect size interpretation, reporting guidelines and peer review.

An approved protocol is a plan, not a result. This guide covers the ten steps that turn it into a defensible paper, from building the data collection instrument through to responding to peer review. The recurring theme is that quality is built in during collection and cannot be added during analysis. A dataset with undocumented decisions in it will produce a paper with undefendable claims in it.

Step 9: Build the Instrument Around the Data Dictionary

Your data dictionary from the protocol phase is the specification. The collection form implements it, and every field on the form should map to a dictionary entry. Fields that do not map are fields nobody agreed to collect, which raises both a governance problem and an analysis problem.

• Collect raw, derive later. Record date of birth and record date, then compute age. Recording age alone destroys information you cannot recover.
• Constrain at entry. Dropdowns and permitted ranges prevent the free-text chaos that consumes weeks of cleaning.
• Distinguish types of blank. Not asked, asked and refused, and not applicable are three different states and must be coded differently from a plain missing value.
• Use validated instruments where they exist, and cite the validation study. A modified instrument is an unvalidated one, so report any modification.

Step 10: Pilot Before You Launch

Run the full process end to end on a small number of participants or records, including the analysis on that pilot data. Piloting collection without piloting analysis is the standard mistake, because the fields you discover are missing only become obvious when you try to compute something with them.

What a pilot should tell you: how long collection actually takes per participant, which questions get misread, whether the recruitment rate supports your timeline, and whether the data export lands in your analysis software in a usable shape.

Step 11: Collect With an Audit Trail

Every entry should be traceable to a source and a time. In regulated trial settings this is formalised, but the principle applies to any study, including retrospective chart reviews.

That last point is the one that saves careers. If a reviewer asks how a value changed, a script answers in seconds and a manually edited spreadsheet cannot answer at all.

Step 12: Clean, Then Lock the Dataset

Cleaning is a documented process with four passes. Structural checks, that every record has an identifier and no duplicates exist. Range and logic checks, that values are plausible and internally consistent, so no pregnancies in male participants and no discharge dates before admission. Consistency checks against source, on a sample. And derived variable checks, confirming that computed fields recompute correctly.

Then lock it. Freeze the analysis dataset, record the date and version, and do the analysis on that frozen copy. Analysing a moving dataset means your results are not reproducible even by you.

Step 13: Handle Missing Data Honestly

Missingness is not a nuisance to be deleted, it is a finding to be reported. Start by reporting how much is missing, in which variables, and whether it differs between groups.

You cannot test which mechanism applies from the data alone, so argue it from what you know about the collection process. Whatever you choose, state it in the methods rather than leaving readers to infer it from a shrinking denominator.

Step 14: Describe Before You Infer

The first table of almost every clinical paper is the baseline characteristics table, and it does real work. It shows who was actually studied, it lets readers judge generalisability, and in a randomised trial it shows whether randomisation balanced the groups.

Match the summary to the distribution. Mean and standard deviation for roughly symmetric continuous data, median and interquartile range for skewed data, counts and percentages for categorical data. Reporting a mean for a skewed variable such as length of stay is a small error that signals a larger inattention.

Do not put significance tests in the baseline table of a randomised trial. Any imbalance there is by definition due to chance, so a p-value tests a hypothesis nobody holds.

Step 15: Follow the Pre-Specified Analysis Plan

Check assumptions and report that you did. Normality of residuals rather than of the raw outcome, equality of variances, independence of observations, and for regression, linearity and the absence of severe collinearity. Clustered data, such as patients within wards, breaks independence and needs a mixed model or a clustered standard error rather than a simple test.

Declare any analysis you did not pre-specify as post hoc. A well-labelled exploratory finding is publishable. A post hoc finding presented as though it were planned is not defensible.

Step 16: Interpret the Effect, Not Just the P-Value

A p-value tells you how compatible the data are with the null hypothesis. It does not tell you the size of the effect, its clinical importance, or the probability that your hypothesis is true. Report the point estimate with its confidence interval as the primary result, and treat the p-value as supporting information.

Write the limitations section as though a hostile reviewer wrote it, then answer each point. Naming a weakness before a reviewer does converts an objection into evidence of judgement.

Step 17: Report Against the Right Guideline

The EQUATOR Network indexes the reporting checklists that journals expect. Find yours before drafting, not after, because the checklist tells you which methods details to have recorded.

Submit the completed checklist with the manuscript where the journal asks for it, and make sure the page numbers on it are real. Editors do check.

Step 18: Authorship, Submission and Peer Review

Settle authorship early using the ICMJE criteria, which require substantial contribution, drafting or critical revision, final approval and accountability. All four must be met. Acquiring funding or supervising a department does not by itself qualify, and agreeing this at protocol stage prevents a dispute at submission.

Choose the journal by scope and audience rather than by impact factor alone, read the aims page and two recent papers of your type, and check indexing. Then write the response to reviewers as a numbered document that quotes each comment, states what you changed and gives the new line reference. Where you disagree, say so with a reason rather than complying reluctantly, because reviewers are not always right and editors know it.

Suggested Images, Free to Reuse

These are real Wikimedia Commons files with the licence stated on each file page. Verify the page and attribute as the licence requires.

Where Ethical Support Fits

The analysis and the interpretation are yours, because they are what your examiners and your readers are assessing. Support that sits alongside that work is legitimate and useful. A PhD expert writer can review whether your methods section is reproducible from the text alone, check that your results tables match the numbers in your prose, walk you through what a confidence interval is actually saying so you can write the discussion yourself, tighten a manuscript to a journal word limit, and format references to the NLM style most biomedical journals use. What should never be outsourced is the analysis you will be asked to defend, the interpretation of your own findings, or any part of the data itself.

Frequently Asked Questions

How much missing data is too much?

There is no universal threshold. What matters is whether missingness is related to the outcome and whether your handling changes the conclusion. Run the analysis both ways and report both.

Should I use SPSS, R or Stata?

Use what your department supports and what you can script. Scripted analysis in any of the three is reproducible, and point-and-click analysis in any of them is not.

What if my results are negative?

Report them. A well-powered negative result is evidence, and non-publication of negative findings distorts the literature. Registries exist partly to make selective publication visible.

How do I handle a reviewer asking for an unplanned analysis?

Do it, label it as post hoc and reviewer-requested in the manuscript, and keep the pre-specified analysis as the primary result.

Can I publish my thesis chapter as a paper?

Usually yes, but check your institution's policy on prior publication and the journal's position on theses in repositories. Declare it in the cover letter rather than hoping it goes unnoticed.

Who should be first author?

Conventionally the person who did most of the work, with the senior supervisor last. Agree it in writing at protocol stage, because retrospective negotiation of authorship is where collaborations break.

Your Next Step Today

Open your dataset and write the script that turns the raw export into your analysis file, even if you have only collected ten records so far. If that script exists from the beginning, every cleaning decision you make is documented automatically, and the reproducibility question that stops many vivas becomes a file you can simply hand over.

Trusted Sources

• EQUATOR Network, the index of health research reporting guidelines. Accessed 13 August 2026.
• STROBE for observational studies and CONSORT for randomised trials. Accessed 13 August 2026.
• PRISMA statement for systematic reviews and meta-analyses. Accessed 13 August 2026.
• COREQ and SRQR for qualitative research. Accessed 13 August 2026.
• ICMJE authorship criteria. Accessed 13 August 2026.
• NLM sample references for biomedical citation format. Accessed 13 August 2026.
• US Office of Research Integrity, guidance on research misconduct. Accessed 13 August 2026.
• Cochrane Handbook for Systematic Reviews of Interventions. Accessed 13 August 2026.

Journal requirements and institutional policies change. Where this guide and your target journal's instructions differ, follow the journal.

Related services

• Statistics assignment help
• Nursing assignment help

Related guides

• How to Do Medical Research: Topic Selection to Approved Protocol
• Choosing the Right Statistical Test
• Reporting Statistics in APA Style: Numbers, Tables and p-Values
• Writing a Clinical Audit Report

All academic writing guides or browse our academic writing services.