YEAR 03 • CLINICAL TRANSITION

Research & Evidence-Based Medicine

20 learning units • 100 chapter sections • 200 explained questions. Build a question, judge the evidence and connect the answer with the person receiving care.

Original RAHBAR study material for the Year 3 learning pathway. Educational examples support supervised learning; institutional research and clinical requirements still apply. References & further reading ↓

MODULE 18 / 20

Analysis, Reporting & Research Integrity

Choose an analysis that matches the design and make the full path from data to conclusion transparent.

Quick revision ↓ Medical Dictionary

01Match analysis to the question

Begin with the outcome type, comparison, dependence structure and target effect. A paired before-and-after measurement is not an independent two-group sample. Binary, continuous, count and time-to-event outcomes require different approaches. Check model assumptions and seek statistical support before data collection where possible. A software package can calculate a result without knowing whether the chosen analysis answers the clinical question.

02Correlation and regression

Correlation summarises a relationship and does not establish causation. Linear regression commonly models a continuous outcome, logistic regression models log odds for a binary outcome and survival models address event timing. Report what coefficients mean on their actual scale. A statistically adjusted association is still conditional on model assumptions and measured variables; more complicated modelling does not substitute for a credible design.

03Missing data and sensitivity

Describe how much information is missing, for which variables, in which groups and why. Complete-case analysis may be biased or inefficient depending on the missingness mechanism. Imputation methods use assumptions and do not reveal the unknowable true values with certainty. Plan appropriate approaches and examine sensitivity to departures from assumptions. Report the analysis population and distinguish unavailable outcomes from participants who never experienced an event.

04Report for scrutiny

Use a reporting guideline appropriate to the design: CONSORT for randomised trials, STROBE for observational studies, PRISMA for systematic reviews and STARD for diagnostic accuracy. Consult the current checklist and relevant extensions. These guide transparent reporting; they do not certify methodological quality. Present participant flow, planned and exploratory analyses, effect estimates, intervals, harms and limitations, including findings that do not support the initial hypothesis.

05Integrity and reproducibility

Fabrication invents data; falsification manipulates the research record; plagiarism misrepresents another's work as one's own. Acknowledge sources, contributors and conflicts, and follow applicable authorship standards. Preserve analysis code and documented decisions with appropriate access controls. Share data only within consent, ethics and privacy arrangements. If an error is discovered, correct the record transparently rather than concealing it to protect a preferred conclusion.

High-Yield Revision

Paired data require their dependence to be recognised.
Reporting checklists improve transparency, not proof of validity.
Report unfavourable and inconclusive results honestly.

Common Pitfalls

A model cannot repair every flaw in study design.
Imputed values are assumption-based estimates, not recovered facts.

References & Further Reading

Use the full methods and relevant chapters when appraising a paper. These sources support further study; the examples and questions above are original teaching exercises.

Reference review: 25 September 2026. Consult current institutional guidance for project approval and clinical decisions.