Learn / Epidemiology / Doing epidemiology: surveys, evidence and ethics
Finding and weighing the evidence
14 minute read · 6-question quiz
What you will learn
Frame a question with PICO or PECO and plan a search
Appraise a study using risk-of-bias thinking and design-specific checklists
Summarise a body of evidence with GRADE and translate it into a recommendation
Evidence-based public health combines the best available research evidence with local context, resources and values. The core skills are asking a precise question, finding the research, judging its quality and applying it.
Frame the question
- PICO (interventions): Population, Intervention, Comparator, Outcome. In HIV-positive adults (P), does isoniazid preventive therapy (I) compared with no preventive therapy (C) reduce incident TB (O)?
- PECO (exposures): Population, Exposure, Comparator, Outcome. In children under five (P), does indoor cooking smoke (E) compared with clean fuel (C) increase pneumonia (O)?
- Add time and setting (PICOTS), and decide study designs to include.
A systematic search
- Choose databases: PubMed/MEDLINE, Embase, Cochrane Library, Global Health, CINAHL, African Journals Online, and trial registries; include grey literature (WHO, ministry reports).
- Write search strings combining synonyms with OR and concepts with AND, using controlled vocabulary (MeSH) and free text.
- Record the date, databases and exact strings.
- De-duplicate and screen titles and abstracts, then full texts, ideally with two reviewers, and record reasons for exclusion in a PRISMA flow diagram.
- Extract data into a standard form; appraise risk of bias; synthesise (narrative or meta-analysis).
- Register the protocol (PROSPERO).
Critical appraisal: ask a few good questions
| Design | Key questions |
| RCT | Random sequence? Concealed? Blinded outcome assessment? Complete follow-up? Analysed as randomised? Outcomes as registered? |
| Cohort | Representative? Exposure measured accurately before outcome? Follow-up complete? Confounders measured and adjusted? |
| Case-control | Cases well-defined? Controls from the same source population? Exposure measured the same way? Recall bias? |
| Cross-sectional | Sampling method? Response rate? Valid measurement? Analysis allowing for design? |
| Diagnostic accuracy | Reference standard for all? Spectrum of patients? Blinded interpretation? |
| Qualitative | Clear research question? Appropriate sampling and analysis? Reflexivity? Credible findings? |
Use tools: Cochrane RoB 2 (trials), ROBINS-I (non-randomised interventions), Newcastle-Ottawa Scale (cohort and case-control), QUADAS-2 (diagnostic accuracy), JBI checklists, CASP.
GRADE: certainty of a body of evidence
The GRADE approach rates the certainty of evidence for each important outcome as high, moderate, low or very low. Randomised trials start high and observational studies start low. Certainty is downgraded for risk of bias, inconsistency, indirectness, imprecision and publication bias, and upgraded for large effects, dose-response or plausible confounding that would reduce the effect.
A recommendation also considers the balance of benefits and harms, values and preferences, resource use, equity, acceptability and feasibility. Strong recommendations can be made with moderate evidence, and weak recommendations with high evidence, depending on these factors. GRADE underpins WHO guidelines.
Summary
Frame questions with PICO or PECO; search several databases; document with PRISMA.
Appraise each study with design-appropriate tools (RoB 2, ROBINS-I, NOS, QUADAS-2, CASP).
GRADE rates certainty from high to very low; recommendations also consider harms, values, cost and equity.
Evidence needs local implementation data as well as average effects.
Check registration, funding and conflicts of interest.
Check your understanding
Draft study guide. Spotted an error or something unclear? Tell us at hello@scintria.com. This is educational material, not clinical or regulatory advice.