Analysing Epidemiological Data — Pattern Recognition and Risk Factor Quantification

Q1 — Sample Band 6 response (8 marks), annotated

Calculations. In the meta-analysis population: ARR = 1.4% − 1.06% = 0.34 percentage points (0.0034). NNT = 1 ÷ 0.0034 ≈ 294. This means that in a population with the same baseline risk as the trial participants (~1.4% over 7 years), approximately 294 people must take daily aspirin for 7 years to prevent one colorectal cancer case. For this 55-year-old patient with a personal prior risk of 3.5% over 10 years: applying the same 24% RRR, aspirin reduces her risk to ≈ 3.5% × (1 − 0.24) ≈ 2.66%. ARR = 3.5% − 2.66% ≈ 0.84%. NNT ≈ 1 ÷ 0.0084 ≈ 119. Her NNT is considerably lower than the population average because her higher baseline risk means the absolute benefit is larger. [2 marks — correct population ARR/NNT AND patient-specific NNT with reasoning]

Evidence quality. A patient-level meta-analysis of 3 large, double-blind, placebo-controlled RCTs sits at Level 1 of the evidence hierarchy — the highest possible level. This substantially increases confidence compared to a single study. Specific features that strengthen confidence: (1) double-blind design eliminates assessment and performance bias in outcome recording; (2) patient-level pooling (rather than study-level) allows more precise estimates and subgroup analyses. Specific features that limit confidence: (1) median 7-year follow-up — CRC can have a latency of decades; it is possible that aspirin effects differ beyond 7 years; (2) the meta-analysis pools only 3 trials — replication in more diverse populations would be needed for the highest confidence. [2 marks — correct evidence hierarchy position AND 2 design features with reasoning]

Harm-benefit evaluation. The known harms of daily aspirin are non-trivial. GI bleeding risk of 2–4% per year means that over 7 years, the cumulative risk of at least one GI bleeding event may exceed the absolute risk of CRC prevention. A haemorrhagic stroke, while rarer, is a severe outcome. For the average-risk population (NNT ≈ 294), the harm-benefit balance is unfavourable — preventing 1 CRC case requires treating 294 people, many of whom will experience GI bleeding over that period. For this high-risk patient (NNT ≈ 119), the balance is better but still requires careful individual consideration: her personal gastrointestinal health history, concurrent medications (anticoagulants), and preference all bear on the decision. [2 marks — uses NNT to frame harm-benefit and applies it to both population and individual contexts]

Justified recommendation. The evidence from this Level 1 meta-analysis is sufficient to support a recommendation that high-risk individuals consider daily low-dose aspirin for CRC prevention — but not an unconditional "yes". For this patient, the recommendation should be: aspirin is a reasonable option given her elevated prior risk (NNT ≈ 119 is clinically meaningful for a serious outcome like CRC), but the decision must weigh her individual GI health history, any anticoagulant use, and personal preferences. This is not a universal recommendation — for average-risk individuals (NNT ≈ 294), the evidence does not support routine aspirin for CRC prevention alone. The GP should use shared decision-making. [2 marks — nuanced individual recommendation that rejects a simple yes/no, acknowledges the population vs individual distinction, and uses the calculation results]

Marking criteria.

• 1 mark — Correctly calculates population ARR (= 0.34%) and NNT (≈ 294) with working shown.
• 1 mark — Correctly calculates patient-specific NNT (≈ 119) by applying 24% RRR to her 3.5% personal risk.
• 1 mark — Correctly identifies the meta-analysis as Level 1 evidence and explains why.
• 1 mark — Identifies and explains at least 2 specific features that strengthen OR limit confidence in the evidence (design quality, follow-up duration, number of pooled trials, generalisability of trial populations).
• 1 mark — Uses the NNT values to explicitly evaluate the harm-benefit trade-off with reference to aspirin's GI bleeding risk.
• 1 mark — Distinguishes between population-level NNT (≈ 294, less favourable) and individual-patient NNT (≈ 119, more favourable) in the harm-benefit discussion.
• 1 mark — Reaches a nuanced recommendation that is neither a blanket yes nor a blanket no — references the patient's individual risk profile and limitations of applying population statistics to one person.
• 1 mark — Uses precise epidemiological language throughout (ARR, NNT, RRR, evidence hierarchy, statistical vs clinical significance, confounders, absolute vs relative risk).

Q2 — Sample Band 6 response (7 marks), annotated

Study design and causation. The study is a prospective cohort study — researchers followed 8,000 Japanese adults and tracked who developed diabetes, comparing those who drank ≥4 cups of green tea per day to those who drank less. This is an observational study: participants were not randomly assigned to drink or not drink green tea. Because it is not an RCT, it cannot establish that green tea prevents (causes a reduction in) diabetes. The study can only show an association — that green tea consumption and lower diabetes incidence occurred together. [1 mark — correctly identifies observational design and explains why causation cannot be claimed]

Error 1: "Scientists have proved." Observational studies do not prove causation. Proof of prevention would require at least a well-designed RCT randomly assigning participants to drink green tea or a control beverage and measuring diabetes incidence. Green tea drinkers may differ from non-drinkers in many ways (diet, exercise, sleep, socioeconomic status) — these confounders can explain the association. A cohort study, however large, cannot rule out confounding. [1 mark]

Error 2: Misinterpretation of the 18% figure. "18% less likely" is almost certainly the relative risk reduction — not the absolute risk reduction. Without knowing the baseline incidence of diabetes in non-drinkers, we cannot evaluate clinical importance. If the 10-year incidence in non-drinkers was, say, 8%, an 18% RRR translates to an ARR of only 1.44% and an NNT of ~69 — modest. If baseline incidence was 2%, the ARR would be only 0.36% and NNT ≈ 278. The media report presents a relative figure as if it directly tells you how much better off you personally will be, which it does not. [1 mark]

Error 3: Misinterpretation of p = 0.02. A p-value of 0.02 means that if there were truly no association between green tea and diabetes risk, there is a 2% probability of observing a result at least as extreme as this by chance alone. It means the result is statistically unlikely to be due to chance — but it does NOT mean the effect is large, clinically important, or that green tea has been "proved" to prevent diabetes. It also does not control for confounders, which remain the dominant threat to validity in a cohort study. [1 mark]

Error 4: "Replace medications with green tea." This claim is dangerously unsupported. The study observed healthy adults — it did not study whether green tea reverses or treats established diabetes in patients already using medication. Recommending discontinuation of diabetes medications based on a single observational study of prevention would be clinically negligent and is not supported by any evidence presented. [1 mark]

Additional evidence required. Before any clinical recommendation: (1) a well-powered double-blind RCT randomly assigning participants to daily green tea extract vs placebo, with diabetes incidence as the primary outcome and adequate follow-up; (2) identification of a plausible biological mechanism (e.g. epigallocatechin gallate improving insulin sensitivity — some preclinical evidence exists); (3) replication across diverse (non-Japanese) populations; (4) evidence that benefits outweigh harms including the caffeine content and potential interactions with medications. [1 mark]

Defensible reformulation. "A 10-year cohort study of 8,000 Japanese adults found that drinking four or more cups of green tea per day was associated with approximately 18% lower relative risk of developing type 2 diabetes. While this statistically significant association is consistent with some preclinical evidence of benefit, the observational design cannot rule out confounding — green tea drinkers may differ from non-drinkers in diet and lifestyle in ways that explain the association. Randomised controlled trials are required to determine whether green tea prevents diabetes before any clinical recommendation can be made." [1 mark]

Marking criteria.

• 1 mark — Correctly identifies the study as a prospective cohort study and explains why cohort studies cannot establish causation (observational design, cannot control for confounding, no randomisation).
• 1 mark — Identifies and correctly explains Error 1: the word "proved" is inappropriate — a cohort study can only establish association, not proof of causation; potential confounders (diet, exercise, socioeconomic factors) remain uncontrolled.
• 1 mark — Identifies and correctly explains Error 2: "18% less likely" is a relative measure (RRR), not an absolute figure; without knowing the baseline incidence, clinical significance cannot be assessed; quotes RRR vs ARR distinction.
• 1 mark — Identifies and correctly explains Error 3: p = 0.02 means the result is statistically unlikely to be due to chance, NOT that the effect is large or that prevention has been proved.
• 1 mark — Identifies Error 4: "replace medications with green tea" is wholly unsupported; the study observed healthy adults, not patients with established diabetes; discontinuing effective treatment on this basis would be clinically dangerous.
• 1 mark — States the type(s) of additional evidence required (RCT, biological mechanism, replication across populations) before any clinical recommendation could be made.
• 1 mark — Reformulates the claim into a biologically defensible statement using correct epidemiological language (association not causation, relative risk, observational design, need for RCT, potential confounders).