Risk factors associated with mortalities attributed to infectious salmon anaemia virus in New Brunswick, Canada
Abstract
Outbreaks of unexplained mortalities attributed to infectious salmon anaemia (ISA) were examined in the 1996 year class of Atlantic salmon in three regions of New Brunswick, Canada. A total of 218 net pens at 14 sites deemed to have been exposed to ISA virus (ISAV) were surveyed for mortality records and management, environmental and host characteristics. Based on definitions of mortality patterns, clinical ISA disease outbreaks occurred in 106 net pens. There were eight sites in which >50% of net pens experienced ISA outbreaks during the study period. Factors related to their potential role in transmission of virus to new sites or new net pens at the same site were identified as sea lice vectors, divers visiting multiple sites, sites belonging to companies with more than one site, exposure to other year classes at the site, and proximity to other infected net pens. Host resistance factors associated with greater risk of outbreaks were identified as larger groupings, general health following smolt transfer, stressful husbandry procedures during growout, and health or productivity during colder water periods. Despite very close proximity between sites, modification of these management factors would probably influence the severity of mortalities caused by ISAV.
Introduction
Infectious salmon anaemia (ISA) was first identified in Norway as a clinical entity in 1984 (Thorud & Djupvik 1988) and subsequently shown to be caused by a virus (Dannevig, Falk & Namork 1995) classified as an orthomyxovirus (Falk, Namork, Rimstad, Mjaaland & Dannevig 1997). ISA is a notifiable disease under the Office International des Epizooties (OIE) and in recent years it has been identified in eastern North America (Lovely, Dannevig, Falk, Hutchin, MacKinnon, Melville, Rimstad & Griffiths 1999), UK (Rodger, Turnbull, Muir, Millar & Richards 1998), Chile (Kibenge, Gárate, Johnson, Arriagada, Kibenge & Wadowska 2001) and the Faroe Islands (Anonymous 2000). Losses from clinical ISA and slaughter measures to control its spread can have severe economic impacts on salmon farms. Transmission of ISA virus (ISAV) between salmon farms can occur through human activities (Munro, Murray, Fraser & Peeler 2003) and possibly by wild carrier fish (Nylund & Jacobsen 1995; Raynard, Murray & Gregory 2001).
This study was initiated in early 1997 when the industry was experiencing high mortality rates in some net pens of Atlantic salmon, Salmo salar L., in Bliss Harbour, Lime Kiln Bay, and Seal Cove areas of New Brunswick, Canada. This area has a high concentration of salmon farms, some only 500 m apart, and in 1996–97 it produced a large proportion of the 20 000 metric tonnes produced in New Brunswick. Although a viral aetiology was suspected, all diagnostic testing had been negative for known viruses up to that time. The gross and histopathological lesions were considered unique and the term ‘haemorrhagic kidney syndrome (HKS)’ was used to describe fish mortalities (Byrne, MacPhee, Ostland, Johnson & Ferguson 1998). The virus responsible for ISA in Norway was subsequently isolated from fish experiencing HKS in New Brunswick and confirmed to be the orthomyxovirus, ISAV, in September 1997 (Lovely et al. 1999).
The mortality patterns and case definitions used for this study were reported previously (Hammell & Dohoo 2005) and will be summarized briefly here. A net pen was considered a case when it experienced an outbreak of unexplained mortalities if it had 7 or more days with daily mortality rates >100 per 100 000 fish (i.e. 0.1% per day), or a cumulative mortality of >5% of the initial number of fish in the net pen (excluding losses in the immediate 30 days after seawater transfer). Net pens were excluded from the analyses if mortality data existed for <3 months, total net pen population was <1000 fish, or net pens did not contain fish on 1 April 1997.
A total of 106 (49%) of 218 net pens on 14 sites fitted the case definition for outbreaks and eight sites (57% of 14 sites) had >50% of their net pens (1996 year class fish only) with defined outbreaks. The median of net pen level maximum mortality rates for the 106 outbreak net pens was 492 mortalities per 100 000 fish per day. The mean and median loss of fish during an outbreak were 12.2% and 6.6%, respectively. Mortality patterns were highly variable at the net pen level.
The current study was a retrospective analysis of management factors associated with mortality problems attributed to ISAV. The major advantage of this type of epidemiology study is the fact that actual occurrences of disease under conditions normally experienced by the study farms is the focus of the investigation. Exposure to the disease agent occurred under natural conditions, conditions which would be difficult or impossible to recreate in the laboratory. However, this type of study cannot make conclusions about causality, except for those factors that do not change over time (Noordhuizen, Frankena, Thrusfield & Graat 2001). An observed association between a factor and the disease occurrence may be due to a cause-and-effect, a relationship with another factor which was not recorded or could not be measured (i.e. confounding variable), or purely by chance. Statistical analyses reduce the risk of chance associations being identified but associations must be evaluated in the light of possible confounding from other factors.
The epidemiology of ISA has been studied in Norway using case–control methods (Vågsholm, Djupvik, Willumsen, Tveit & Tangen 1994; Jarp & Karlsen 1997). In each case the unit of concern was the site or holding unit, with 37 case sites compared with 37 control sites as the minimum sample size (Jarp & Karlsen 1997). Although these case–control studies are very useful in evaluating factors associated with a site being positive for ISAV, they lack an evaluation of factors associated with the net pen having increased mortalities caused by ISAV. The current study examined associations between individual net pens having an outbreak (attributed to ISAV) and physical characteristics and husbandry practices occurring for each net pen.
Norwegian ISA epidemiology studies defined sites as ISA-positive through diagnostic results, confirmed by pathological and haematological investigations for one study (Jarp & Karlsen 1997) and confirmed by the state veterinary diagnostic laboratory for the other (Vågsholm et al. 1994). In both cases, the fact that ISA was a notifiable disease allowed the classification of sites as positive or negative through routine diagnosis (or suspicions) by site veterinarians and did not require the investigators to perform independent diagnostic testing to verify the presence of ISA at positive sites or the absence of ISA at negative sites. These investigations were performed in 1988–90 (Vågsholm et al. 1994) and 1992–93 (Jarp & Karlsen 1997). Neither study had the benefit of virus isolation or other diagnostic tests to detect the presence of the agent responsible for ISA, as the first isolation of the causative virus (Dannevig et al. 1995) was not reported until after the completion of the epidemiology studies.
A major finding of the case–control studies performed in Norway was the association between sites becoming ISAV positive and the proximity to another ISAV-positive site or a salmonid slaughterhouse not properly disinfecting its wastewater (Jarp, Gjevre, Olsen & Bruheim 1994; Vågsholm et al. 1994). These studies also reported that the following factors were associated with decreased ISA outbreaks: separating smolt from older fish for the first three months and size-sorting (possibly by removing slow growing and early sexually maturing fish). Increased ISA outbreaks were associated with purchasing smolt from multiple hatcheries and transporting smolt over greater distances.
The primary objective of this investigation was to identify host, environment and management risk factors associated with mortalities attributed to ISAV at the net pen and site levels.
Materials and methods
Mortality data
Mortality data collection methods and the reference population were described in Hammell & Dohoo (2005) and are summarized briefly here. All study sites were located in one of three geographical regions, Lime Kiln Bay, Bliss Harbour or Seal Cove, in southern New Brunswick. These were the only regions in New Brunswick known to have ISA in 1997. All sites in the affected area were visited in the summer and autumn of 1997 and their mortality records for the 1996 year class were collected (retrospectively) for the period starting with the time fish were transferred to net pens at each site (i.e. April to June 1996) until the time of data collection. A prospective data collection system was then established to monitor mortalities in the study area through November 1997.
Farm managers were consulted to better understand their records and a standardized format was used to record mortalities for the purpose of comparisons between sites. The inventory records from each net pen were examined to determine the length of time that each group was maintained as a distinct group of fish. Any alteration in the group through additions from other net pens from grading or combinations with other net pen groups was considered as an endpoint of the previous group and the start of the next group for reconciliation with management factors of the group. Harvests, transfers out of the group because of partial harvests or splitting of net pens, and mortalities, were acceptable as these did not alter the factors which the group experienced.
Definition of case net pen and ISA problem site
The study population was categorized into ISA cases or non-cases for net pens and ISA problem or non-problem sites. A net pen was considered a case when it experienced an outbreak of unexplained mortalities (at any time between the time fish were put in the net pen and November 1997) as defined by one of the following criteria: (1) seven or more days with daily mortality rates >100 per 100 000 fish (i.e. 0.1% per day), or (2) a cumulative mortality of >5000 per 100 000 of the initial number of fish in that net pen (excluding losses in the immediate 30 days after seawater transfer) (Hammell & Dohoo 2005). Sites were classified as ‘problem sites’ if 50% or more of the net pens were classified as cases.
Unfortunately, the presence of ISAV at each site or net pen remained unconfirmed due to the absence of recorded diagnostic results and the differing levels of diagnostic screening intensity at the sites involved in the study. Other diseases may have been present in some cases and may have interfered with the detection of associations with management factors. For this reason, this study is considered an evaluation of factors associated with unexplained fish mortality for which ISAV was believed to be the predominant but not necessarily the only cause.
Risk factor data collection
All sites were visited in late 1997 and a detailed personal interview questionnaire was administered by two trained interviewers to obtain information about potential risk factors. Initial interviews lasted 2–3 h with subsequent follow-up visits or telephone/facsimile communications for additional data from site veterinarians and source hatchery managers. Whenever possible, net pen level information was collected. Net pen level and site-level factors were distinguished by their applicable differences within sites or between sites, where a site was defined as a collection of net pens owned and operated as a unit.
Several characteristics of management or fish health could not be obtained in a reliable format. For example, sea lice counts were provided by many sites. However, after examination of the data quality, counts could not be transformed into a usable measure of lice burdens because of differences between sites with respect to fish sampling methods, counting methods, count frequency, net pens sampled and record quality.
Risk factors examined
Net pen level information was collected for all 1996 year class fish that were present at a site on 1 July 1997. Due to the movement of net pens at a site which may have altered such variables as ‘proximity to seal scaring device’, the information was collected for net pens as they existed on 1 January and 1 July 1997. Questionnaires were tested using three farm managers prior to conducting the final survey which was made between October 1997 and February 1998.
To make the data collection interview more rapid, site maps were generated and used to describe such information as ‘usual order of mortality diving’ and then these maps were converted to ‘first, middle, and last one-third of net pens’. Farm records were utilized whenever appropriate. However, some characteristics were derived from the manager's memory. Interviewers encouraged interviewees to answer ‘don't know’ if they were unsure of the information.
Factors examined were divided into eight categories: (1) net pen data, such as dimensions, construction material, depth under net, walkway presence, current flow, relative proximity to shore, proximity to other year class net pens; (2) fish host data, such as date of smolt transfer, number stocked, smolt age and size at transfer, hatchery source, smolt transfer method, and mortalities in the first 30 days post-transfer; (3) site-level husbandry practices, such as proportion of net pens containing 1996 year class salmon, weight sampling frequency and methods, size-sorting practices, order of diving net pens at the site, diver characteristics, and mortality collection hygiene; (4) sea lice management practices, such as lice counts and counting methods, frequency of treatments (bath and oral), occurrence of treatment toxicities, and net pen order for bath treatments; (5) general health histories, such as disease diagnoses, antibacterial treatment frequency, occurrence of seal attacks, and relative proximity of net pen to seal scaring devices; (6) feeding practices, such as number of months using moist feed after smolt transfer and during cold water periods, feed brand and manufacturer, methods of transportation of feed to site, relative proximity of net pen to feed delivery point at site, feeding methods and relative net pen order of feeding; (7) harvesting and processing data, such as location of processing plant used, wharf used to transport harvested fish, proximity to closest processing plant; and (8) environmental assessment rating, as reported as part of the government lease permits. A copy of the questionnaire is available from the corresponding author upon request.
Statistical analysis
Net pen level risk factors were analysed using a dataset with one record per net pen with the outcome variable being a case as defined previously. Whenever appropriate, continuous variables were categorized using biologically sensible determination points. In most cases, there was a natural distinction between groups at the selected cutoff. Descriptive statistics, unconditional associations (between risk factors and the dependent variable), correspondence analyses, and logistic regression analyses were used to evaluate net pen level risk factors. The logistic model used was logit(p) = βo + ΣβjXj in which the outcome is the logit of the probability of the net pen becoming a case, βo is the intercept, and βj is the logistic regression coefficient for the jth net pen level predictor variable.
A site-level data set was created with one record per site with the outcome variable being problem sites as defined previously. Descriptive statistics and unconditional associations between risk factors and the dependent variable were evaluated.
A combined data set (one record per net pen) containing both net pen and site-level risk factors was created for use in survival analyses. In addition to risk factors previously identified as potentially significant, a time-dependent covariate which represented whether or not any other ‘case’ net pens had been identified at the site was created. Cox proportional hazards models were fitted to identify important risk factors. The model was h(t) = h0(t)eβX in which h(t) is the probability of a net pen becoming an outbreak at time t given that it has not occurred by time t, βX = β1X1 + β2X2 +⋯+ βkXki (the coefficient and predictors for all k cage and site-level predictors), h0 is the baseline hazard, and t is a point in time. The combined data set was also used to build a two-level random effects logistic regression model which allowed for the simultaneous analysis of factors from two different levels (net pen and site-level factors) and which took the clustering of net pens within a site into account. The random effects model was logit(pi) = βo + β1X1i+⋯+ βkXki +usite(i) in which logit(pi) is the log odds [ln(p/1 − p)] of the net pen becoming a case, and usite(i) is the random effect of the site which contains net pen i. The Xis are the predictor values for the ith net pen. These multilevel data were also analysed using general estimating equations (GEE) and ordinary logistic regression with robust standard errors (OLR-RSE) in addition to a random effects logistic regression model.
Unconditional associations were evaluated at P = 0.1 while the logistic regression and proportional hazards models used a significance threshold of P = 0.05. All models were fitted by initially considering all factors which had significant unconditional associations with the outcome of interest and subsequently removing non-significant terms. All statistical analyses were carried out using STATA, Version 5 (Stata Corp., College Station, TX, USA) except for the random effects logistic regression which was carried out using MlwiN (Multilevel Models Project, Institute of Education, University of London, UK).
Results
Description of cases
Fourteen sites had sufficient 1996 year class data to permit examination of mortality rates at a net pen level (for details of mortality patterns, see Hammell & Dohoo 2005). Eight were ‘problem sites’ and six were non-problem sites. There were 218 net pens included in the study areas from 14 sites, of which 106 were considered cases and 112 non-cases.
Unconditional associations at the net pen level
There were 38 net pen level variables from the questionnaire. Table 1 presents the results from selected net pen level risk factors which had unconditional associations with the net pen being classified as a case (P < 0.10). Not all significant risk factors have been listed as many were highly correlated with factors listed in Table 1. For example, virtually all net pens with >12 000 fish were large, circular and made of polyvinylchloride (PVC), so the variables for net pen shape (circular) and material (PVC) were also significant. Correspondence analyses (Dohoo, Ducrot, Fourichon, Donald & Hurnik 1996) to identify correlated variables and investigator judgement to identify which of the correlated variables were most readily interpretable were used to select factors for further analysis.
| Variable | Levels | No. of cases (total no. of net pens) | Relative risk | Lower 95% CI | Upper 95% CI | P-value |
|---|---|---|---|---|---|---|
| Initial population at risk | <5000 | 15 (72) | 0.000 | |||
| 5000–12 000 | 28 (57) | 2.36 | 1.40 | 3.97 | ||
| >12 000 | 40 (49) | 3.92 | 2.45 | 6.27 | ||
| Fish density (fish m3) | <2.5 | 11 (55) | 0.000 | |||
| 2.5–5.0 | 43 (60) | 3.58 | 2.06 | 6.22 | ||
| >5.0 | 26 (49) | 2.65 | 1.47 | 4.81 | ||
| Cumulative mortality rate of all fish in net pen during 1996 | <0.003 | 12 (45) | 0.000 | |||
| 0.003–0.007 | 24 (68) | 1.32 | 0.74 | 2.37 | ||
| >0.007a | 34 (46) | 2.77 | 1.66 | 4.63 | ||
| Fish were weight sampled during 1997 | No | 40 (123) | 0.000 | |||
| Yes | 43 (55) | 2.40 | 1.80 | 3.21 | ||
| Bath treatments for lice during January to July 1997 | 0 | 37 (78) | 0.009 | |||
| 1 | 16 (36) | 0.94 | 0.61 | 1.45 | ||
| 2+ | 7 (38) | 0.39 | 0.19 | 0.79 |
- a Data from net pens which experienced outbreaks in 1996 were excluded as they would contribute to 1996 cumulative mortality.
Unconditional associations at the site level
There were 44 site-level variables from the questionnaire. Unconditional associations between a number of site-level risk factors and whether or not a site had more than 50% of the net pens affected by outbreaks are presented in Table 2. The level of 50% of net pens affected was selected as a cut-off in consultation with industry as the point above which sites were considered problem sites.
| Variable | Levels | Cases (sites) | Relative risk | Lower 95% CI | Upper 95% CI | P-value (exacta) |
|---|---|---|---|---|---|---|
| Mortality dives per week | 1 | 4 (11) | 0.051 | |||
| 2 | 3 (3) | 2.75 | 1.26 | 6.01 | (0.19) | |
| Diver visits multiple sites | No | 1 (5) | 0.094 | |||
| Yes | 6 (9) | 3.33 | 0.54 | 20.43 | (0.27) | |
| Multiple site company | No | 1 (6) | 0.001 | |||
| Yes | 7 (7) | 7.0 | 1.14 | 42.97 | (0.002) | |
| Moist feed fed after transfer | No | 6 (8) | 0.031 | |||
| Yes | 1 (6) | 0.22 | 0.004 | 1.39 | (0.10) | |
| Months moist feed fed between December 1996 and July 1997 | 0 | 6 (6) | 0.004 | |||
| 1–4 | 1 (4) | 0.25 | 0.05 | 1.36 | (0.03) | |
| 5–7 | 0 (4) | 0 | – | – | (0.005) | |
| Feed delivered by feed company | No | 2 (8) | 0.031 | |||
| Yes | 5 (6) | 3.33 | 0.95 | 11.66 | (0.10) | |
| Proportion of net pens with 1996 year class | <0.9 | 6 (7) | 0.008 | |||
| ≥0.9 | 1 (7) | 0.17 | 0.03 | 1.05 | (0.03) | |
| At least one other net pen at site had experienced a HKS outbreak | No | |||||
| Yes | NAb | 3.13 | 1.71 | 5.72 | 0.000 |
- HSK, haemorrhagic kidney syndrome; ISA, infectious salmon anaemia.
- a Pearson's chi-square P values (Fisher's Exact P values).
- b Value changed with time so unconditional association (hazard ratio) was determined from a survival analysis with a time-dependent covariate.
Logistic regression of net pen level risk factors
The results from the final logistic regression model of net pen level risk factors are presented in Table 3. These results must be interpreted with caution as net pens were clustered within site and hence were not independent. The Hosmer–Lemeshow goodness-of-fit test was marginally significant (P = 0.06) suggesting that the model did not adequately fit the data. No obvious explanations for this apparent poor fit were evident from regression diagnostics. Net pens with small initial populations or the lowest category of fish density were at reduced risk of experiencing an outbreak, as were net pens which generally had a higher health level during their first summer – autumn following transfer to sea water (i.e. during 1996).
| Variable | Level | Odds ratio | Lower 95% CI | Upper 95% CI | P-value |
|---|---|---|---|---|---|
| Initial population at risk | <5000 | ||||
| 5000–12000 | 4.42 | 1.44 | 13.49 | 0.009 | |
| >12000 | 15.60 | 3.86 | 63.08 | 0.000 | |
| Fish density (fish m3) | <2.5 | ||||
| 2.5–5.0 | 7.27 | 2.16 | 24.48 | 0.001 | |
| >5.0 | 1.71 | 0.48 | 6.07 | 0.404 | |
| Cumulative mortality rate of all fish in net pen during 1996 | <0.003 | ||||
| 0.003–0.007 | 1.44 | 0.49 | 4.28 | 0.508 | |
| >0.007 | 10.06 | 2.77 | 36.61 | 0.000 |
Analyses of both site and net pen level risk factors
Survival analysis
The results of the final Cox proportional hazards model are presented in Table 4. It was based on a model in which the start of the time period at risk was defined as the point in time at which the fish were put in the net pen (usually the spring of 1996) and net pens were either classified as cases if they experienced an outbreak or were censored at the end of the data collection period (November 1997). Inclusion of a time-dependent covariate representing whether or not another net pen at the site had been classified as a case resulted in a very unstable model with unrealistic estimates of coefficients so that particular variable could not be included in the final model.
| Variable | Level | Hazard ratio | Lower CI | Upper CI | P-value |
|---|---|---|---|---|---|
| Proportion of net pens with 1996 year class | <0.9 | ||||
| ≥0.9 | 0.38 | 0.22 | 0.67 | 0.001 | |
| Feed delivered by feed company | No | ||||
| Yes | 1.69 | 1.00 | 2.85 | 0.048 | |
| Months moist feed fed between January and July 1997 | 0 | ||||
| 1–4 | 0.45 | 0.23 | 0.90 | 0.024 | |
| 5–7 | 0.08 | 0.03 | 0.19 | 0.000 | |
| Cumulative mortality rate of all fish in net pen during 1996 | <0.003 | ||||
| 0.003–0.007 | 1.52 | 0.73 | 3.19 | 0.27 | |
| >0.007 | 3.61 | 1.63 | 8.01 | 0.002 |
Having a site that contained primarily one year class reduced the hazard of outbreaks, as did feeding moist feed. For the latter variable, the hazard was lowest when moist feed was fed for the longest period of time. Higher levels of mortality during 1996 were associated with an increased hazard of outbreaks. As with the logistic regression analyses, the results from this multivariable model must be interpreted with caution as net pens were not independent within sites, and this clustering was not accounted for in the proportional hazards model.
Multilevel modelling of risk factors
Due to the fact that net pens are clustered within sites which have similar management factors and HKS (ISA) cases also tend to be clustered within sites, it was difficult to separate the effects of the different clusters. Through the use of a multilevel statistical model (random effects logistic regression), both net pen and site-level risk factors were analysed simultaneously, while accounting for the clustering of net pens within sites. The results of this model are presented in Table 5. All of the variables found to be significant in either the net pen level logistic regression or the combined site and net pen level survival analysis were included in the analysis. However, density (fish per cubic meter) was dropped due to the failure of the model to converge whenever that variable was included.
| Variable | Level | Odds ratio | Lower CI | Upper CI | P-value |
|---|---|---|---|---|---|
| Proportion of net pens with 1996 year class | <0.9 | ||||
| >0.9 | 0.138 | 0.033 | 0.582 | 0.01 | |
| Months moist feed fed between December 1996 and July 1997 | 0 | ||||
| 1–4 | 0.020 | 0.002 | 0.268 | 0.00 | |
| 5–7 | 0.002 | 0.000 | 0.036 | ||
| Cumulative mortality rate of all fish in net pen during 1996 | <0.003 | ||||
| 0.003–0.007 | 3.274 | 0.916 | 11.705 | 0.05 | |
| >0.007 | 9.631 | 1.485 | 62.480 |
- The above table was based on random effects models which investigated all factors except fish density (fish m3). Inclusion of this variable caused the model to fail to converge. Models were fitted using the computer program MlwiN with parameter estimates being based on the second-order PQL (Penalised Quasi-likelihood) estimation process. The final model P < 0.001.
For comparison purposes, the multilevel data (both site and net pen factors) were also analysed using OLR-RSE and using a GEE procedure and then compared with the random effects model discussed above. OLR-RSE and GEE were used because these types of models also adjust for clustering of net pens within the farms. OLR-RSE does not adjust the estimate of the factor's coefficient, but will estimate more conservative standard errors. GEE uses a population-averaged approach to estimate the variation at the site and net pen levels and results in adjustments to both the estimate of the coefficient and standard error of each factor. Similarly, the random effects model results in adjustments to both coefficients and standard errors, but does so using a subject-specific approach (Dohoo, Martin & Stryhn 2003). Each of the three methods uses a different statistical algorithm to arrive at the results. The results are presented in Table 6. Although the coefficients of each factor vary slightly amongst the different modelling techniques, the relative values remained unchanged. The random effects term (variance at the site level) was also provided to estimate if most of the variation of the model was between sites (site level) or within sites (net pen level). The site-level variance is small and in fact the standard error for the site-level variance was larger than the variance estimate, thus, there is not much evidence for site-level variation, leaving most of the variation within the net pen level. The three risk factors which remained in all of the multiple level models were: single year class sites, months fed moist feed between December 1996 and July 1997, and cumulative mortality during 1996. While the application of multiple statistical modelling techniques and the finding of consistent effects estimates across techniques increases the level of confidence in the findings, it does not eliminate the potential problem of confounding by unmeasured variables.
| Variable | Level | Logistic | GEE | Random effects | |||
|---|---|---|---|---|---|---|---|
| β | SE | β | SE | β | SE | ||
| Proportion of net pens with 1996 year class | <0.9 | ||||||
| >0.9 | −1.90 | 0.53 | −2.05 | 0.48 | −1.98 | 0.74 | |
| Months moist feed fed between December 1996 and July 1997 | 0 | ||||||
| 1–4 | −3.70 | 0.40 | −3.66 | 0.35 | −3.90 | 1.32 | |
| 5–7 | −5.91 | 0.56 | −5.96 | 0.56 | −6.11 | 1.42 | |
| Cumulative mortality rate of all fish in net pen during 1996 | <0.003 | ||||||
| 0.003–0.007 | 1.23 | 0.62 | 1.56 | 0.69 | 1.19 | 0.65 | |
| >0.007 | 2.34 | 0.86 | 2.92 | 0.78 | 2.27 | 0.95 | |
| Constant | N/A | 3.67 | 0.45 | 3.49 | 0.45 | 3.89 | 1.37 |
| Variance (site) | 0.28 | 0.41 | |||||
Discussion
ISA diagnosis
This cross-sectional epidemiology study examined factors associated with ISA cases as defined by elevated unexplained mortality rates at sites that were subsequently known to be exposed to the ISAV. However, the actual presence of the causative agent, ISAV, in mortalities could not be verified due to lack of diagnostic investigations in many instances. Nevertheless, the mortality patterns were most consistent with the clinical diagnosis of ISA (Hammell & Dohoo 2005). Misclassification of net pens was potentially possible because mortality rates may have changed as some net pens were split in an effort to reduce losses once mortality rates started to rise. These newly formed net pens were excluded in this risk factor study because they had limited mortality data and so pen level factors were not considered twice for the same original net pen.
Net pen level risk factors for ISA mortalities
The final ordinary logistic regression model with only net pen variables identified three factors (initial population at risk, fish density, and cumulative mortality rate during 1996) associated with the risk of net pens experiencing an outbreak. Net pens with larger initial populations at risk (PAR > 12 000 fish) were more likely to experience an outbreak compared with net pens which had initial populations of <5000 fish. The greater number of fish within a net pen was highly correlated to the net pen type (larger PAR tended to be reared in 20(+) m diameter PVC circular net pens, whereas smaller PAR were more often reared in 15 m steel square net pens). The salmon farming industry has gradually converted to larger net pens with greater numbers of fish in each pen over the past decade to reduce capital and operating costs. The reason that larger net pen populations appear to increase the probability of experiencing elevated mortality rates in the presence of an infectious agent is unknown, but may be associated with specific management practices at farms with larger net pens.
The number of fish per cubic metre of water volume was calculated using fish, not biomass, per cubic metre because fish weight measurements were not comparable between sites. The effect of fish density within net pens was difficult to interpret because higher relative risk was associated with moderate density compared with high density. Lower numbers of fish per unit volume of water were at the lowest risk. That moderate density was a higher risk than high density may indicate that low fish densities may reduce the risk, but once a threshold of density is exceeded, there is less of an effect of increasing densities on the risk of experiencing elevated mortality rates attributable to HKS (ISA).
Reduced general health of salmon smolts, as measured by cumulative mortality rate during the first year of seawater growth, was associated with a greater risk of experiencing an outbreak of HKS (ISA) in the second seawater growth season. Better survival rates in the first warm water season after smolt transfer may be an indirect measure of many health management factors.
There were two factors (weight sampling and sea lice control) that were unconditionally associated with the risk of a net pen outbreak that were not in the final model. Net pens in which weight samples were taken during 1997 were more likely to experience an outbreak. Weight samples can induce additional stress on the group of fish being measured and this may account for the increased risk of HKS (ISA). However, weight sampling may also be an indirect measure of the intensity of management.
Sea lice have been reported to transmit various pathogens including ISAV (Nylund, Wallace & Hovland 1993; Nylund, Hovland, Hodneland, Nilsen & Løvik 1994). Farmers in this study suspected sea lice infestation levels as contributing to the risk of HKS (ISA). As an indirect measure of reduced level of sea lice infestations, the number of delousing bath treatments was used as the variable in which greater bath frequency was considered an indication of a more aggressive sea lice control policy. Greater than two delousing bath treatments was protective against HKS (ISA) outbreaks compared with one or no treatment.
Site-level risk factors for ISA mortalities
To evaluate site-level variables, a Cox proportional hazards and a random effects logistic regression model were fitted using both site and net pen level factors. Both models identified the proportion of net pens with 1996 year class, the amount of time fish ate moist feed, and the cumulative mortality rate during 1996 as associated with whether (logistic model) and when (Cox proportional hazards model) the net pens experienced an outbreak. The Cox proportional hazards model also identified feed delivery by feed company as an important association with when the net pen experienced an outbreak.
Net pens from sites which were least exposed to other year classes (i.e. <10% of the site's net pens contained another year class) had a much reduced risk of experiencing HKS (ISA) outbreaks, possibly due to reduction of transmission of virus from older to young salmon. Sites with multiple generations had greater frequency of primary ISA outbreaks in one Norwegian study (Vågsholm et al. 1994). The results from this study suggest that despite the close proximity of sites to each other (e.g. usually <1000 m), year class separation is an important consideration in reducing the risk of HKS (ISA).
It is common practice for some sites in Atlantic Canada to use moist feed (i.e. feed with >30% moisture by volume) in the first few months following smolt transfer (the spring of 1996 for the study population) due to the perception that immediate post-transfer appetite and growth are enhanced with moist feed. Most farms will convert to dry, pelleted feed within 4 months of seawater transfer (i.e. late in the first summer). Another common practice is to feed moist feed during colder winter months when appetites are reduced and fish appear to be more stimulated by moist feed. Increasing the number of months feeding moist feed after smolt transfer reduced the risk of HKS (ISA) outbreaks in the study population, possibly due to improved nutrition and energy levels.
Feed delivery to the site by the feed company was more common with dry feeds and not with moist feeds. However, feed delivery was considered a potential method of viral transmission to a net pen due to the fact that the delivery boats travelled between sites. It is interesting to note that in the final Cox proportional hazards model, the effect of feed delivery remained a significant risk factor for when HKS (ISA) outbreaks occurred in a net pen.
There were three other factors that were unconditionally associated (at P ≤ 0.1) with HKS in >50% of the net pens of a site but were not in the final models: more than two mortality dives per week, divers who visited more than one site, and sites belonging to multiple-site companies.
The frequency of mortality dives were probably influenced by the presence, or the suspicion, of HKS (ISA) at the site and therefore were more likely a result rather than a cause of HKS (ISA) outbreaks. However, the divers often worked for more than one company and usually disinfected gear but rarely maintained distinct sets of gear for each site. As a result, it is possible that the diving equipment could travel from an infected net pen to an uninfected net pen, which might contribute to the spread of HKS (ISA). Mucus has been shown to be capable of transmitting ISAV under laboratory conditions (Totland, Hjeltnes & Flood 1996). Therefore, divers and other equipment could carry the ISAV to other sites and infect each individual net pen upon entry, thus increasing the risk of the site becoming a problem site.
Sites belonging to a multiple site company may have increased risk of becoming a problem site because of transfer of equipment or people between sites. Repeated sharing of equipment and personnel may result in multiple net pens on a site being exposed to the virus. These sites may also be quite distinct in their management styles and demands from company management may not always be directed solely towards the best interests of a single site.
Jarp & Karlsen (1997) identified 5 km as a safe distance between sites to reduce the risk of ISA transmission. The majority of sites in this study were all located within a 5-km radius and each site was undoubtedly influenced by the close proximity to other infected sites. Despite this close proximity, a previous HKS (ISA) outbreak in a net pen at a site was associated with an increased risk of the site having more than 50% of its net pens experiencing HKS (ISA) outbreaks. This indicates that managing risk factors at the site level may be able to reduce the risk of mortalities from infectious disease agents originating on nearby sites.
Conclusion
It was useful to categorize risk factors which might primarily affect the risk of introduction of ISAV into a net pen or site and risk factors which primarily affect host resistance. The distinction may be important when devising disease containment policies which reduce the probability of transmission to new sites or areas, compared with policies for managing health in situations of known exposure.
The factors identified in this study by modelling with logistic regression and survival analysis, as well as with unconditional associations which can be explained as having a role in transmission of ISAV to a new net pen include: (1) sea lice as vectors; (2) diver visiting multiple sites; (3) sites sharing equipment or personnel within a multiple-site company; (4) feed delivery methods; (5) exposure between year classes at the same site; and (6) proximity to other infected net pens.
The factors which can best be explained as having a role in modifying host resistance to clinical outbreaks when net pens are in the vicinity of affected sites include: (1) transmission between individuals within larger groupings of fish in a net pen; (2) overall health and productivity following smolt transfer; (3) stress of husbandry procedures during growout; and (4) health and productivity of the net pen population during the colder water periods, specifically feeding practices.
In conclusion, management factors appear to modify the relative risk of experiencing mortalities caused by HKS (ISA). Increasing the health and survival of smolt in the first summer in sea water and feeding moist feed during winter and spring were associated with reduced risk. Aggressive lice control and lower initial stocking numbers reduced the probability of HKS (ISA) outbreaks. Multiple year class sites and previous exposure to ISA at the site were also important risk factors.
Acknowledgements
The authors wish to acknowledge the support of the New Brunswick Salmon Growers Association, individual farms, and funding provided by the Canadian Department of Fisheries and Oceans for this study.
