A case study of time-series regression modeling: Risk factors for pond-level mortality of farmed grass carp (Ctenopharyngodon idella) on a southern Chinese farm

Abstract

Limited research has been done using multivariable statistical methods to assess factors associated with fish mortality in warm-water finfish aquaculture in China. We carried out a case study to test the hypothesized association between pond-level daily mortality of farmed grass carp and predisposing environmental and husbandry factors. Based on logbook data from a single farm in Guangdong province (China) in 2013, two-stage time-series regression (TSR) analyses were conducted to estimate the lagged effect of these predisposing factors on grass carp mortality. Factors assessed included temperature fluctuations, movement of fish into and out of ponds, and 3 types of treatments (antibiotics-antiparasitics, traditional Chinese medicine-probiotics, and chemicals to improve water quality). First, coefficients were estimated using a generalized linear negative-binomial model for each pond, and these coefficient estimates were combined using meta-analytic techniques. Sensitivity analyses were done to compare effects of changes in the 3 modeling components: distributional forms, number of spline knots, and types of autocorrelation terms. Model results in the case study indicated 2 risk factors might be associated with increased mortality of grass carp: (1) movements-in of new fish during the previous 14 days; and (2) increasing water temperature during the previous 7 days. Sensitivity analyses indicated good consistency of the estimates with different modeling components. Our findings highlight the utility of assessing daily farm records using TSR to develop hypotheses about potential risk factors for grass carp mortality in China.

A 2-stage hierarchical interrupted time-series analysis to quantify the long-term effect of subclinical bacterial kidney disease on performance of farmed Atlantic salmon (Salmo salar L.)

“…To deal with the hierarchical structure, we used one equation for a longterm pattern for all fish groups within a site cycle. A similar approach was performed in (Jia et al, 2018), where different ponds with fish were modeled separately. The average numbers of fish groups per site cycle was 10 -22 (Table 1), which provided sufficient observations to model the site cycles separately.…”