15  Chapter 15 — PICO, Systematic Review and Meta-analysis

Applied Epidemiology in Aquaculture — Part VI: Evidence Synthesis

Where this sits — Safeguard / Revisit (quaternary prevention). The final prevention level is about learning — standing back to weigh all the evidence, find the gaps, and feed the result into policy. This chapter is the method behind that loop, and behind this book’s own case-study papers.

15.1 Overview

No single study settles a question. Evidence synthesis — asking a precise question, finding all the relevant studies, and combining them — is how a field knows what it knows. This chapter covers the three tools: PICO to frame an answerable question, the systematic review (with PRISMA) to assemble the evidence without cherry-picking, and meta-analysis to pool it into a single estimate with a measure of disagreement. We reproduce, on a small scale, the author’s own synthesis of the sea-lice–SRS effect.

We use lice_srs_studies.csv: six studies (challenge and field) estimating the effect of sea lice on SRS mortality.

15.2 15.1 Preparation

pacman::p_load(rio, here, tidyverse, meta, metafor)   # PRISMA2020 for the flow diagram
st <- import(here("data", "lice_srs_studies.csv"))

15.3 15.2 PICO — framing an answerable question

A vague question (“do lice matter for SRS?”) cannot be searched or answered. PICO forces precision:

  • Population — farmed Atlantic salmon in seawater
  • Intervention/exposure — high sea-lice burden (egg-bearing females per fish)
  • Comparator — low or no lice burden
  • Outcome — SRS mortality

PICO becomes the search strategy (the terms), the inclusion criteria (which studies count), and the structure of the extraction table. It is the small discipline that makes a review reproducible instead of an opinion.

15.4 15.3 The systematic review and PRISMA

A systematic review finds and appraises all studies meeting the PICO criteria, documenting the process so others can repeat it. The PRISMA flow diagram records how many records were found, screened, excluded and included — the audit trail that separates a systematic review from a selective one. The author’s own SRS review illustrates the power and the sobering findings: of 315 records screened, only 70 were field-based and met inclusion; 87% of field research covers epidemiology, pathogen genetics and diagnostics, while only 13% addresses vaccine performance, pharmacokinetics and host immunity — and there are just two field vaccine-effectiveness studies for more than 30 licensed products (Diethelm-Varela et al., in press). A systematic review does not only pool evidence; it maps the gaps, which is the Safeguard loop in action.

15.5 15.4 Meta-analysis — pooling the evidence

When studies are comparable, meta-analysis combines their effect sizes, weighting each by its precision. We work on the log scale and use a random-effects model (which allows the true effect to vary between studies — almost always the right choice in field biology):

m <- metagen(TE = log(st$rr), seTE = st$se_log, studlab = st$study,
             sm = "RR", method.tau = "DL")
forest(m)        # forest plot

The pooled result:

Random-effects pooled RR = 1.44 (95% CI 1.24–1.67)

Sea lice raise SRS mortality by about 44%, and the interval excludes 1 — a real effect, closely matching the ~1.40 reported in the author’s full analysis (Mardones et al., in review). The forest plot shows each study’s estimate and weight, with the pooled diamond at the bottom; it is the single most informative figure in evidence synthesis.

15.6 15.5 Heterogeneity and publication bias

Pooling is only honest if you report how much the studies disagree. The I² statistic gives the share of variation due to real between-study differences rather than chance:

# Q = 6.1 (df = 5), I^2 = 18%, tau^2 = 0.007

Here I² = 18% — low-to-moderate heterogeneity. Much of it is explained by design: challenge experiments pooled to RR 1.69, field studies to 1.33. That is not a flaw but a finding — the effect is larger under controlled infection and smaller, more variable, in the field (as your keynote notes, it depends on lice level and farm context). A funnel plot then checks for publication bias (the tendency for small negative studies to go unpublished); asymmetry is a warning that the pooled estimate may be inflated — especially relevant when, as here, the field evidence base is thin.

15.7 15.6 Interpretation on the farm

Evidence synthesis is the Safeguard phase made concrete. It converts a scattered literature into one defensible number with its uncertainty and its gaps — the basis for a treatment guideline, a vaccine-registration standard, or a research-funding priority. The lesson from the author’s SRS review is pointed: the field over-studies what is easy (epidemiology, genetics, diagnostics) and under-studies what decides outcomes (vaccine effectiveness, pharmacokinetics). Naming that imbalance, and feeding it back into policy and research design, is how the five-level framework closes its loop and improves the next cycle of Anticipate–Prevent–Detect–Respond.

15.8 15.7 Exercises

Try it yourself (15.1). Write a PICO question for the effect of SRS vaccination on cumulative mortality, and list two inclusion and two exclusion criteria that follow from it.

15.2 Run the random-effects meta-analysis and report the pooled RR and I². Re-run it as a fixed-effect model — which gives a narrower interval, and why is random-effects usually more honest?

15.3 Pool the challenge studies and the field studies separately. Interpret the difference as a source of heterogeneity rather than noise.

15.4 Explain, using the funnel plot, why a meta-analysis of a field with few small studies (like SRS vaccine effectiveness) should be read cautiously.

15.5 Using the author’s review figures (70/315 field-based; 13% on vaccines/PK/immunity), state one research priority that the evidence map implies.

15.9 Summary & what’s next

  • PICO turns a vague question into a searchable, answerable one.
  • A systematic review (with PRISMA) assembles all the evidence reproducibly and maps the gaps.
  • Meta-analysis pools effect sizes (random-effects RR 1.44 here), displayed as a forest plot.
  • Always report heterogeneity (I²) and check publication bias; here, design (challenge vs field) explained most disagreement.
  • Evidence synthesis is the Safeguard/Revisit loop: it feeds findings and gaps back into the framework.

In Chapter 16 — Genetics of Disease Resistance we close the book with the cross-cutting layer: breeding fish that resist disease, and the molecular epidemiology that tracks the pathogen itself.

15.10 References & further reading

  • Diethelm-Varela B, Atero N, Gelcich S, Gallardo A, Mardones FO. Field-based evidence on P. salmonis management: a systematic review and Swiss-Cheese evidence map. Reviews in Aquaculture, in press.
  • Mardones FO, Córdova-Bührle F, Diethelm-Varela B, et al. Meta-analysis of sea-lice effects on SRS mortality and R₀. In review.
  • Lhorente JP 2014; Figueroa 2017; Meyer 2019; Arriagada 2019; Hillman 2020; Diethelm-Varela 2025 — the studies pooled here.
  • Page MJ, et al. The PRISMA 2020 statement. BMJ. 2021. — reporting standard.
  • Dohoo I, Martin W, Stryhn H. Veterinary Epidemiologic Research. 2nd ed. VER Inc.; 2009.

Files for this chapter: data/lice_srs_studies.csv and scripts/make_meta_studies.R (new).