11 Chapter 11 — Antibiotics and AMR Stewardship
Applied Epidemiology in Aquaculture — Part IV: Interventions
Where this sits — Respond (tertiary) and Safeguard (quaternary). Antibiotics are the main Respond tool once an outbreak is established. But their overuse is the leading Safeguard concern — the unintended consequence that quaternary prevention exists to minimise. This chapter measures both the treatment and its cost.
11.1 Overview
When SRS takes hold, antibiotics are the response of last resort — and in Chile the dominant one: SRS drives about 94% of the salmon industry’s antimicrobial use. This chapter does two jobs. First, Respond: how to measure whether treatment works, and why naïve “treated vs untreated” comparisons mislead. Second, Safeguard: how to read antibiograms and MICs, track resistance over time, and apply the stewardship that keeps these drugs useful — because the current treatment-centred model is a self-reinforcing cycle of reduced effectiveness and rising resistance (Diethelm-Varela et al., in press).
We use amr_isolates.csv: 427 P. salmonis florfenicol susceptibility results, 2017–2025, with MIC, S/I/R and genogroup.
11.2 11.1 Preparation
pacman::p_load(rio, here, tidyverse, janitor)
amr <- import(here("data", "amr_isolates.csv"))11.3 11.2 Measuring treatment effect — and its trap
The instinct is to compare mortality in treated vs untreated cages and call the difference the treatment effect. Don’t — it is wrecked by confounding by indication: the cages that get treated are the sick ones, so a naïve comparison can make an effective antibiotic look useless or even harmful. Valid evidence needs either a before/after contrast within treated units, or adjustment for severity (the Chapter 9 toolkit).
Even with a fair comparison, SRS treatment effectiveness is limited for a reason that is not resistance at all: roughly 40% of florfenicol-treated fish never reach therapeutic tissue concentrations under field conditions, because dose is delivered in feed to animals of varying weight and appetite (Price et al., 2018, 2019). Many “treatment failures” are delivery failures. The practical implication — dose for weight and temperature, start early, prefer injectable where justified — is a Respond lesson with direct Safeguard consequences, because sub-therapeutic exposure selects for resistance.
11.4 11.3 Reading the antibiogram
Susceptibility is summarised by the minimum inhibitory concentration (MIC) — the lowest drug concentration that stops growth — classified against breakpoints into S/I/R (susceptible / intermediate / resistant), or against the wild-type cut-off into wild-type vs non-wild-type:
tabyl(amr, sir) # distribution of S / I / R
amr |> group_by(genogroup) |>
summarise(pct_R = mean(sir == "R") * 100)Two rules keep antibiogram reading honest. First, MIC is only comparable within one method and lab — the same caution as Ct values in Chapter 6. Second, summarise the distribution, not just a mean: a bimodal MIC histogram (a susceptible cluster and a resistant cluster) tells you a resistant subpopulation is emerging long before the median moves.
11.5 11.4 Tracking resistance over time
Resistance is a moving target, so surveillance of susceptibility is itself a Safeguard activity. The share of isolates with reduced florfenicol susceptibility climbed over the study period:
amr <- amr |> mutate(nwt = as.integer(mic >= 4))
amr |> group_by(year) |> summarise(pct_nwt = mean(nwt) * 100)
# 2017 ~37% ... 2025 ~85%
# test the trend with logistic regression
glm(nwt ~ year, data = amr, family = binomial) |> broom::tidy(exponentiate = TRUE)
# year OR 1.23 (odds of non-wild-type rise ~23% per year, p < 0.001)Year-to-year values are noisy — which is exactly why you test the trend with a model rather than compare two endpoints. Here the odds of an isolate being non-wild-type rise about 23% per year (OR 1.23, p < 0.001): an unambiguous upward march, consistent with the national picture of rising reduced susceptibility concentrated in the dominant LF-89-like genogroup (Rozas-Serri, 2026; keynote AMR surveillance).
11.6 11.5 The Safeguard view: AMR as an unintended consequence
Quaternary prevention asks what harm the control measures themselves cause. For SRS the answer is stark and self-reinforcing:
- Single-drug dependence — florfenicol is ~97.6% of all antimicrobials prescribed, so resistance to one drug threatens the whole response capacity.
- A vicious cycle — variable dosing leaves ~40% of fish sub-therapeutic, which both reduces effectiveness and selects for resistance, which drives more treatment (Diethelm-Varela et al., in press).
- Spillover beyond the farm — extended antibiotic treatment selects multiresistant gut bacteria in salmon (Higuera-Llantén et al., 2018), and resistance genes accumulate in farm sediments — the One Health stake that makes this everyone’s problem.
Stewardship is the Safeguard response: confirm before treating (Chapter 6), MIC-guided drug choice, correct dosing for weight and temperature, injectable over in-feed where pharmacokinetically justified, genomic AMR surveillance, and — most powerfully — strengthening the upstream Anticipate and Prevent layers so that fewer outbreaks need treating at all. AMR is not a side-issue to disease control; it is the constraint that determines whether today’s control strategy survives the decade.
11.7 11.6 Interpretation on the farm
The two halves of this chapter are inseparable. A Respond decision — reach for florfenicol — is simultaneously a Safeguard decision that nudges the resistance curve in §11.4 upward. The disciplined path is to treat early, correctly dosed, and confirmed, to track susceptibility as a routine output, and to judge every antibiotic course against its contribution to the resistance trend, not just this cycle’s mortality. The ultimate stewardship lever is the rest of this book: every outbreak prevented is a course of antibiotics never given.
11.8 11.7 Exercises
Try it yourself (11.1). Produce the S/I/R distribution and the percentage resistant by genogroup. Which genogroup carries more reduced susceptibility, and why does that matter for choosing treatment?
11.2 Fit the logistic trend of non-wild-type on year and report the odds ratio. Explain why this is more trustworthy than comparing the 2017 and 2025 percentages directly.
11.3 A manager compares mortality in treated vs untreated cages, finds treated cages did worse, and concludes florfenicol is useless. Name the bias and propose a fairer analysis.
11.4 List three stewardship actions that reduce AMR selection without reducing the care of sick fish, and classify each as Respond or Safeguard.
11.5 Explain how strengthening Prevent (Chapter 10) and Anticipate (Chapter 5) lowers antibiotic use — the quaternary-prevention argument in one paragraph.
11.9 Summary & what’s next
- Naïve treated-vs-untreated comparisons suffer confounding by indication; many SRS “treatment failures” are delivery failures (~40% sub-therapeutic).
- MIC and S/I/R summarise susceptibility, comparable only within a method/lab; read the distribution, not the mean.
- Resistance trends are noisy — test them with a model; here reduced florfenicol susceptibility rose ~23% per year.
- Quaternary prevention (Safeguard) targets AMR as the unintended consequence of treatment; stewardship plus stronger upstream prevention is the answer.
In Chapter 12 — Surveillance and Early Warning we return to Detect at the system level: how sensitive a surveillance system really is, demonstrating freedom from disease, and statistical early-warning signals.
11.10 References & further reading
- Diethelm-Varela B, Atero N, Gelcich S, Gallardo A, Mardones FO. Field-based evidence on P. salmonis management: a systematic review (quaternary prevention). Reviews in Aquaculture, in press.
- Higuera-Llantén S, et al. Extended antibiotic treatment selects multiresistant gut bacteria in salmon. PLoS ONE. 2018.
- Price D, et al. Florfenicol pharmacokinetics and sub-therapeutic exposure. Prev Vet Med. 2018;156:16–21; Aquaculture. 2019.
- Montt R, Mardones FO, Reyes P, Rojas H. Spatial clustering of antimicrobial use and SRS mortality. Aquaculture, in press.
- Rozas-Serri M. Salmonid rickettsial septicaemia: a review. Reviews in Aquaculture. 2026. — AMR surveillance and florfenicol dependence.
- Dohoo I, Martin W, Stryhn H. Veterinary Epidemiologic Research. 2nd ed. VER Inc.; 2009.
Files for this chapter: data/amr_isolates.csv and scripts/make_amr_isolates.R (new).