Data Interpretation for MRCOG Part 1

Reading a Confidence Interval

A 95% confidence interval is the range within which the true population value would fall in 95% of identically-conducted studies. It is not a 95% probability that the true value lies in this particular interval.

The examinable rule is where the null value sits:

• For a difference (mean difference, risk difference), the null is 0.
• For a ratio (RR, OR, HR), the null is 1.

An interval crossing its null is non-significant at p = 0.05. A wide interval means an imprecise estimate — usually a small sample — and a non-significant wide interval is not evidence of no effect.

Effect Measures

| Measure | Definition | Use | |---|---|---| | Relative risk (RR) | Risk in exposed / risk in unexposed | Cohort, RCT | | Odds ratio (OR) | Odds in exposed / odds in unexposed | Case-control; approximates RR when the outcome is rare | | Absolute risk reduction (ARR) | Control risk − treatment risk | What the patient experiences | | Number needed to treat (NNT) | 1 / ARR | Always round up | | Hazard ratio (HR) | Instantaneous event rate ratio | Time-to-event, survival |

The distinction that matters clinically: a relative risk reduction of 50% sounds identical whether the baseline risk is 2 in 100 or 2 in 1,000,000. The absolute numbers differ by four orders of magnitude. Consent conversations use absolute risk.

When the outcome is common, the odds ratio exaggerates the relative risk. This is why an OR of 3 for a 40%-prevalence outcome does not mean triple the risk.

Diagnostic Tests

| | Disease + | Disease − | |---|---|---| | Test + | TP | FP | | Test − | FN | TN |

• Sensitivity = TP / (TP + FN) — a highly Snsitive test, when Negative, rules OUT (SnNOUT)
• Specificity = TN / (TN + FP) — a highly Specific test, when Positive, rules IN (SpPIN)
• PPV = TP / (TP + FP)
• NPV = TN / (TN + FN)

Sensitivity and specificity are properties of the test. PPV and NPV depend on prevalence — the same test performs worse in a low-prevalence population, which is the whole difficulty of population screening for rare conditions.

Likelihood ratios are prevalence-independent: LR+ = sensitivity / (1 − specificity); LR− = (1 − sensitivity) / specificity. An LR+ above 10 or an LR− below 0.1 substantially shifts post-test probability.

Screening

Wilson and Jungner's criteria: the condition should be an important health problem with a recognisable latent stage, there should be an accepted treatment, facilities for diagnosis and treatment, a suitable and acceptable test, an agreed policy on whom to treat, and the cost should be economically balanced.

Biases specific to screening evaluation:

• Lead-time bias — earlier diagnosis lengthens apparent survival without postponing death.
• Length-time bias — screening preferentially detects slow-growing, better-prognosis disease.
• Overdiagnosis — detecting disease that would never have caused harm.

Mortality, not survival, is the honest endpoint for a screening programme.

Study Designs, in Order of Susceptibility to Bias

1. Systematic review / meta-analysis of RCTs
2. RCT — randomisation addresses confounding, including unmeasured confounders. Allocation concealment and blinding address selection and ascertainment bias.
3. Cohort — follows exposure forward. Gives incidence and relative risk. Vulnerable to loss to follow-up.
4. Case-control — starts from outcome and looks back. Efficient for rare outcomes; gives an odds ratio, not a risk. Vulnerable to recall bias.
5. Cross-sectional — prevalence; cannot establish temporality.
6. Case series / report — hypothesis-generating.

Intention-to-treat analyses participants in the group to which they were randomised, regardless of what they received. It preserves randomisation and gives the more conservative, more realistic estimate. Per-protocol analysis answers a different question and reintroduces selection bias.

Reading a Forest Plot

Each square is a study estimate, its size proportional to weight; the horizontal line is its confidence interval. The diamond is the pooled estimate, its width the pooled interval. A vertical line at the null (1 for ratios) shows which studies reached significance.

Heterogeneity is quantified by I² — roughly, the percentage of variation attributable to between-study differences rather than chance. Above about 50% is substantial, and a pooled estimate over highly heterogeneous studies may be meaningless however tight its interval. A funnel plot asymmetry suggests publication bias.

High-Yield Exam Points

• Ratios are null at 1; differences are null at 0.
• Sensitivity and specificity are fixed properties; PPV and NPV move with prevalence.
• NNT rounds up, always.
• The odds ratio overstates the relative risk when the outcome is common.
• Intention-to-treat preserves randomisation and is the conservative analysis.
• Screening programmes are judged on mortality, because survival is contaminated by lead-time bias.

Clinically reviewed by NHS obstetricians and gynaecologists

MRCOG AI is an independent educational tool. It is not affiliated with, endorsed by, or connected to the Royal College of Obstetricians and Gynaecologists (RCOG).