7  Chapter 7 — Choosing a Study Design

Applied Epidemiology in Aquaculture — Part III: The Investigator’s Toolkit

Where this sits — Anticipate (primordial/primary prevention). Before you can anticipate risk you have to gather evidence the right way. The design you choose decides which measure of association (Chapter 5) you can compute, how strong the evidence is, and how much it costs.

7.1 Overview

The same question — “does river water cause BKD?” — can be studied in several ways, and the choice is not cosmetic: it determines whether you can measure risk at all, whether you can establish time order, and how many fish (and how much money) you need. This chapter maps the main observational designs, what each can and cannot deliver, and how to pick one. No new dataset: we reason with the farms you already know and a short simulation on the BKD cohort from Chapter 5.

7.2 7.1 Descriptive vs analytic

Descriptive studies (case reports, case series, surveys) characterise disease by animal, place and time and generate hypotheses — the register in Chapter 2 is descriptive. Analytic studies test a hypothesis by comparing groups, and come in three workhorse forms, distinguished by where you start.

7.3 7.2 The three workhorse designs

7.3.1 Cross-sectional

A snapshot: measure exposure and outcome at the same time in a sample of the population (our Chapter 2 register, extended with risk factors). Cheap and fast, and the natural source of prevalence. Its weakness is temporality — you see exposure and disease together and usually cannot tell which came first, so it is strongest for stable exposures (water source, species) and weak for causal claims.

7.3.2 Cohort

Start from exposure: follow exposed and unexposed groups forward and count new cases. This is the only observational design that measures incidence and relative risk directly, and it respects time order (exposure precedes outcome). The national SRS study is a cohort — 1,546 production cycles followed for weekly mortality (Diethelm-Varela et al., 2025). Cohorts are powerful but expensive, and inefficient for rare diseases (you follow many fish to see few cases).

7.3.3 Case-control

Start from the outcome: take a group of cases and a comparable group of controls, then look backward at their exposures. You cannot compute risk (you chose how many cases to include), so the measure is the odds ratio. Case-control is extremely efficient for rare diseases and fast outbreaks — and a well-drawn sample recovers the cohort’s answer. From the Chapter 5 BKD population, taking all cases plus a random 4,000 controls gives:

# OR from a case-control sample of the BKD population
# (all 3,975 cases + 4,000 random controls)
case_control_OR   # 2.38   -- cohort OR was 2.41, RR 2.25

Almost the same OR, from a fraction of the controls. The catch is selection and recall of exposure: controls must come from the same population that produced the cases, or the OR is biased.

7.4 7.3 Group-level and hybrid designs

  • Ecological studies use groups, not individuals as the unit — e.g. correlating antimicrobial use with SRS mortality across 58 neighbourhoods (Montt et al., in press). Cheap and good for policy-scale drivers, but prone to the ecological fallacy: a pattern across farms need not hold within a fish.
  • Nested case-control draws cases and controls from within an existing cohort — the efficiency of case-control with the rigour of cohort sampling.
  • Case-crossover uses each case as its own control at an earlier time — useful for transient triggers (did a delousing bath precede the SRS onset?).
  • Panel / repeated cross-sectional designs track populations over time, as surveillance systems do (Chapter 12).

7.5 7.4 Choosing — a quick guide

Design Starts from Measure Best for Main weakness
Cross-sectional population snapshot prevalence, OR quick burden & stable exposures no time order
Cohort exposure incidence, RR strong causal evidence, common outcomes cost; poor for rare disease
Case-control outcome OR rare disease, outbreaks, speed selection & exposure bias
Ecological groups correlation policy-scale drivers ecological fallacy

The decision follows the question: a rare, acute problem (an AHPND crash) suits case-control; a common exposure with a measurable incidence (water source and BKD) suits a cohort; a fast “how much is out there right now” suits cross-sectional.

7.6 7.5 Interpretation on the farm

Design is where rigour is won or lost — no analysis rescues a biased sample. For Anticipate work, the practical sequence is usually: a cross-sectional survey to size the problem, a cohort or nested case-control to pin down the drivers, and ecological analysis to see how those drivers play out at neighbourhood scale. Matching the design to the question (and the budget) is the difference between an answer you can act on and a number you cannot trust.

7.7 7.6 Exercises

Try it yourself (7.1). For each question, name the most appropriate design and the measure it yields: (a) “What fraction of our ponds currently carry Streptococcus?” (b) “Do vaccinated cages have lower SRS mortality over the cycle?” (c) “What distinguished the three ponds that crashed with AHPND from those that didn’t?”

7.2 Explain why a case-control study cannot report a relative risk, and what it reports instead.

7.3 A colleague finds that neighbourhoods with higher antimicrobial use have higher SRS mortality, and concludes that antibiotics cause SRS. Name the fallacy and explain it.

7.4 You must investigate a rare, fast-killing syndrome affecting a handful of cages. Which design, and why would a cohort be impractical?

7.8 Summary & what’s next

  • The design decides the measure: cohort → RR/incidence; case-control → OR; cross-sectional → prevalence.
  • Cohort gives the strongest observational evidence but is costly; case-control is efficient for rare or fast disease; cross-sectional is a quick snapshot without time order.
  • Ecological designs answer policy-scale questions but risk the ecological fallacy.
  • A good case-control sample recovers the cohort’s odds ratio at a fraction of the cost.

In Chapter 8 — How Many Fish Should I Sample? we stay in the investigator’s toolkit and make these designs executable: sampling methods, sample-size calculation, and cost-effective, risk-based surveillance.

7.9 References & further reading

  • Dohoo I, Martin W, Stryhn H. Veterinary Epidemiologic Research. 2nd ed. VER Inc.; 2009. — study design chapters.
  • Thrusfield M, Christley R. Veterinary Epidemiology. 4th ed. Wiley-Blackwell; 2018.
  • Diethelm-Varela B, et al., Mardones FO. Epidemiology of SRS in farmed salmon. J Fish Dis. 2025;e70097. — a national cohort study.
  • Montt R, Mardones FO, Reyes P, Rojas H. Spatial clustering of SRS mortality and antimicrobial use. Aquaculture, in press. — an ecological design.
  • Salgado-Caxito M, et al., Mardones FO. Qualitative risk assessment. 2022. — structuring evidence for anticipation.

Files for this chapter: none new (reuses bkd_watersource.csv).