POPH90068 Chap.5 Surveillance and Evidence
Surveillance and Evidence
Surveillance can reveal detected burden and trends, while comparative studies test whether an intervention produced an outcome. Observational studies are valuable but can be distorted by confounding, selection and different testing behaviour. Randomisation strengthens causal inference within the studied population, yet precision, adherence and generalisability still matter.
This chapter builds a disciplined conclusion from design, estimate, confidence interval, population and uncertainty. Conflicting findings are compared through population, comparator, outcome and follow-up rather than resolved by preference. The goal is a bounded recommendation that identifies what the evidence permits, where transport requires caution and which future result would change the decision.
What this chapter covers
- 01
Routine notification systems
- 02
Observational associations
- 03
Randomised comparisons
- 04
Confounding and selection
- 05
Effect estimates and precision
- 06
Generalisability and communication
Worked example · free
Reconcile conflicting studies
- 1Treat the observational findings as plausible but vulnerable to confounding.
- 1State that the trial did not establish benefit in its enrolled population.
- 1Interpret interval width and study limitations without declaring proof of no effect.
- 1Limit conclusions to represented populations and specify further evidence needed.
Key terms
- Confounding
- Distortion caused when another factor relates to both exposure and outcome.
- Randomisation
- Allocation intended to create comparable groups for estimating an intervention effect.
- Confidence interval
- A range expressing the precision and compatible values around an estimate under stated assumptions.
- Null value
- The effect value corresponding to no difference in the selected measure.
- Generalisability
- The extent to which findings apply beyond the studied participants and setting.
Surveillance and Evidence FAQ
Why can observational and randomised results differ?
People receiving an intervention in routine care may differ in healthcare engagement, testing, behaviour and baseline risk. Randomisation aims to balance those factors, although trial adherence, sample size and population selection still shape the estimate.
Does an interval including the null prove no effect?
No. It means the study did not demonstrate a statistically significant effect at the stated threshold. The estimate, interval width, design and clinical importance must be interpreted together.
When is surveillance data useful?
Surveillance is useful for tracking diagnosed disease, distribution and population trends. Its interpretation must account for testing, reporting delays, available variables and changes in who enters the system.
Assessment move
Use an evidence card for every study: design, population, comparison, outcome, effect direction, precision, limitations and transportability. When studies conflict, compare these fields rather than choosing the result you prefer. Practise writing a conclusion that is useful but bounded: what the evidence supports, what it does not establish and what additional data would change the recommendation.
For surveillance, add testing and reporting context. For trials, add adherence and represented population. This routine prepares evidence appraisal for both short answers and the proposal justification. Read estimates in a fixed order. Name the outcome and comparison, report the direction and magnitude, inspect the interval for precision and compatibility with the null, then return to design.
Avoid replacing that sequence with a significant or not significant label. A wide interval can leave important benefit and little effect compatible with the data, while a narrow interval around a small estimate supports a different decision. State what remains uncertain and whether the uncertainty concerns existence, size, population, delivery or all four.
For observational evidence, sketch a causal map before judging the result. Place intervention exposure and outcome on the page, then add factors that could influence both, such as healthcare engagement, baseline risk, testing frequency or other preventive practices. Ask how each factor was measured or controlled and what residual difference could remain.
For a randomised trial, shift the audit to allocation, adherence, loss to follow-up, outcome ascertainment, precision and who was enrolled. This keeps the critique appropriate to the design instead of reciting a generic limitations list. Reconcile apparently conflicting studies with a comparison grid. Put population, intervention definition, adherence, comparator, outcome, follow-up and estimate in rows.
Differences between cells may explain why results diverge without declaring one study correct and the other useless. Then write a synthesis sentence that identifies the population and outcome for which confidence is greatest. Follow with a boundary sentence naming where evidence is absent or transport requires caution. Do not use biological plausibility to erase an inconvenient empirical result.
Treat surveillance as a measurement system, not a simple count. Draw the route from infection through symptoms or screening, attendance, testing, diagnosis, notification and database inclusion. Mark where policy, service disruption or reporting practice could alter observed numbers. Compare notifications with testing volume, positivity and population coverage where available.
If a program changes testing, anticipate that detection may rise before any reduction in transmission is visible. End revision by making a decision statement. Choose among further research, monitored pilot, limited recommendation or broader implementation, based only on the evidence in the case. Specify a review trigger and what new information would change the decision.
This final step turns appraisal into public-health reasoning while preserving uncertainty rather than using uncertainty as a reason to say nothing.
