MedPubStatistical analysis and biostatistics support for medical research
Statistics is where many medical manuscripts lose reviewers: the wrong test for the data, p-values without effect sizes, or a methods paragraph that does not say what was done. We analyse your data with you and document every decision so the results stand up to statistical review.
Support at every stage of a study
Statistical help is most valuable before data collection, when the sample size and analysis plan can still change. But most requests arrive after data are collected, often from residents completing an MD/MS thesis or faculty converting a dataset into a paper. We work at either stage:
- Planning: sample-size calculation, choice of primary outcome and analysis plan for a synopsis, protocol or ethics committee submission.
- Analysis: data cleaning, descriptive statistics, hypothesis testing, regression and survival models.
- Reporting: publication-ready tables and figures, a statistical methods paragraph and results text that report effect sizes with confidence intervals.
- Peer review: help answering statistical comments from reviewers, including re-analysis where it is justified.
Analyses we commonly perform
| Question | Typical methods |
|---|---|
| Compare two groups | Independent t-test or Mann–Whitney U; chi-square or Fisher's exact test for categorical outcomes |
| Compare before and after in the same patients | Paired t-test or Wilcoxon signed-rank; McNemar's test for paired proportions |
| Compare three or more groups | One-way ANOVA or Kruskal–Wallis, with appropriate post-hoc tests |
| Association between variables | Pearson or Spearman correlation; simple and multiple linear regression |
| Risk factors for a binary outcome | Logistic regression with odds ratios and 95% confidence intervals |
| Time to an event | Kaplan–Meier curves, log-rank test, Cox proportional hazards regression |
| Accuracy of a diagnostic test | Sensitivity, specificity, predictive values, likelihood ratios, ROC curves and AUC |
| Agreement between raters or methods | Cohen's kappa, intraclass correlation, Bland–Altman analysis |
Reporting that survives statistical review
Journals increasingly send manuscripts to a statistical reviewer, and the SAMPL guidelines and ICMJE recommendations set out what they expect. In practice that means: naming each test and the software version; stating how assumptions such as normality were checked; reporting the effect size (mean difference, odds ratio, hazard ratio) with a 95% confidence interval rather than a p-value alone; giving exact p-values rather than "p < 0.05"; explaining how missing data were handled; and labelling unplanned analyses as exploratory. Every analysis we deliver is written up to that standard.
Sample size for theses and protocols
An undersized study is a common reason for thesis and protocol queries from ethics committees, and for reviewers to dismiss a negative result as underpowered. We calculate sample size from your primary outcome using standard formulas or G*Power, and document the assumptions and source study. Our guide to sample size calculation for medical research shows the formulas with worked examples.
What we will not do
We do not fabricate or impute data without disclosure, drop participants to change a result, run tests until one is significant, or present exploratory findings as confirmatory. These practices are research misconduct under COPE guidance, and statistical reviewers are trained to detect them. If the honest result is null, we help you report it well; well-reported null results are publishable.
Related services and guides
Have a dataset that needs analysis?
Upload de-identified data with your objectives. We reply within one working day with the analysis plan, a fixed quote and turnaround.
Upload Your DataStatistical analysis: common questions
What do you need from me to start the analysis?
The dataset (Excel, CSV, SPSS or similar) with a short codebook explaining each variable, your objectives or hypotheses, the study protocol or thesis synopsis if available, and the target journal. If data are already analysed, send the existing output too, so that we can check it rather than start over.
Which software do you use?
SPSS or R, depending on the analysis and on what you or your department use. You receive the output files and, for R, the script, so the analysis can be reproduced and you can answer reviewer questions.
Can you calculate the sample size for my thesis synopsis?
Yes. We need the primary outcome, the expected value from a previous study or pilot data, the difference you consider clinically important, and the significance level and power your institution requires. We document the formula, assumptions and reference so the justification can go straight into the synopsis or methods section.
Will you change the analysis to get a significant p-value?
No. Analyses follow the study design and the pre-specified objectives. Selectively reporting tests until something is significant (p-hacking) is a form of research misconduct and reviewers increasingly look for it. We report effect sizes and confidence intervals, not just p-values, and we flag post-hoc analyses as exploratory.
Do you handle de-identified data only?
Please send de-identified data: remove names, hospital numbers, phone numbers and other direct identifiers before upload. Data are used only for the agreed analysis and handled under our privacy policy.