Anticoagulant Rodenticide Toxicity in Terrestrial Raptors: Tools to Estimate the Impact on Populations in North America and Globally

Anticoagulant rodenticides (ARs) have caused widespread contamination and poisoning of predators and scavengers. The diagnosis of toxicity proceeds from evidence of hemorrhage, and subsequent detection of residues in liver. Many factors confound the assessment of AR poisoning, particularly exposure dose, timing and frequency of exposure, and individual and taxon‐specific variables. There is a need, therefore, for better AR toxicity criteria. To respond, we compiled a database of second‐generation anticoagulant rodenticide (SGAR) residues in liver and postmortem evaluations of 951 terrestrial raptor carcasses from Canada and the United States, 1989 to 2021. We developed mixed‐effects logistic regression models to produce specific probability curves of the toxicity of ∑SGARs at the taxonomic level of the family, and separately for three SGARs registered in North America, brodifacoum, bromadiolone, and difethialone. The ∑SGAR threshold concentrations for diagnosis of coagulopathy at 0.20 probability of risk were highest for strigid owls (15 ng g−1) lower and relatively similar for accipitrid hawks and eagles (8.2 ng g−1) and falcons (7.9 ng g−1), and much lower for tytonid barn owls (0.32 ng g−1). These values are lower than those we found previously, due to compilation and use of a larger database with a mix of species and source locations, and also to refinements in the statistical methods. Our presentation of results on the family taxonomic level should aid in the global applicability of the numbers. We also collated a subset of 440 single‐compound exposure events and determined the probability of SGAR‐poisoning symptoms as a function of SGAR concentration, which we then used to estimate relative SGAR toxicity and toxic equivalence factors: difethialone, 1, brodifacoum, 0.8, and bromadiolone, 0.5. Environ Toxicol Chem 2024;43:988–998. © 2024 His Majesty the King in Right of Canada and The Authors. Environmental Toxicology and Chemistry published by Wiley Periodicals LLC on behalf of SETAC Reproduced with the permission of the Minister of Environment and Climate Change Canada.


INTRODUCTION
Anticoagulant rodenticides (ARs) have provided an effective and cost-efficient tool for containing pest rodent infestations for more than seven decades (Jacob & Buckle, 2018).Their use is worldwide and is growing, with some sales projections exceeding $US 2 billion globally by 2027 (Fortune Business Insights, 2020).Such growth in demand for rodenticides is forecasted based on increases in human populations, and the associated demand for food and space, along with loss of habitat and biodiversity, including natural controls of rodent populations (Crist et al., 2017).Because progress on the development of alternative pest control is proceeding slowly (Witmer, 2018), those projections in sales growth seem likely to occur.
Diagnosis of AR poisoning relies on evidence of symptoms such as hemorrhage, unclotted blood, and pallor of tissues and organs, in conjunction with subsequent confirmation by finding AR residues in the liver (Murray, 2018).Development of threshold liver residue concentrations indicative of poisoning by coagulopathy has been attempted a number of times (Newton et al., 1990(Newton et al., , 1999;;Thomas et al., 2011; reviewed by Lohr, 2018 andRattner &Harvey, 2020).Such threshold/ benchmark values are available for a wide range of other chemicals in wildlife, and are widely applied in risk assessments (Allard et al., 2010;Beyer & Meador, 2011).
To further resolve criteria for interpreting the significance of the increasing reports of liver residue concentrations, we compiled a database of hepatic SGAR residues and postmortem evaluations from 951 terrestrial raptor carcasses collected throughout Canada and the United States from 1989 to 2021, representing 26 species of mainly hawks and owls.We focused on the three SGARs commonly used in both Canada and the United States; bromadiolone, brodifacoum, and difethialone.Our objectives were to (1) quantify the probability of diagnosis of coagulopathy as the primary cause of death as a function of the SGAR concentration in the liver; (2) compare probability curves and error estimates and determine critical ∑SGAR concentrations by taxonomic groupings; and (3) quantify, as available data permitted, probabilities and error estimates for the three SGAR compounds in the data set, as a proxy for relative toxicity of each SGAR compound, and to suggest toxic equivalence factors.

Study area and sample curation
Our North American database includes liver samples and postmortem evaluations from 951 terrestrial raptor carcasses collected throughout Canada and the United States from 1989 to 2021, representing 26 species of mainly owls and hawks (Table 1).For the British Columbia (Canada) data subset, curation of raptor carcasses and residue analyses are described in detail in the companion paper (Elliott et al., 2022).Birds were submitted to rehabilitation or veterinary centers either dead or debilitated.Sample submission across data sets was not consistent, with one source, California (USA), including only birds diagnosed as poisoned by rodenticides.The more recent data set from New York State (USA; Okoniewski et al., 2021) included all birds (n = 30) diagnosed at necropsy as having AR-facilitated hemorrhage, but another 41 birds with other diagnoses were also analyzed for rodenticide residues.In British Columbia, in addition to investigating the incidence of AR poisoning, study objectives included (1) tracking temporal exposure trends in relation to regulatory changes; and (2) assessing spatial patterns, to discriminate sources, whether agricultural or commercial/urban (Elliott et al., 2022;Hindmarch et al., 2017Hindmarch et al., , 2018)).
To pursue those objectives, we chemically analyzed livers from as many birds as the laboratory had capacity for, effectively selecting at random from a subset of the larger group of birds received.We defined that subset by the species we considered to be at most risk of AR exposure, such as the larger owls, hawks, and falcons that are capable of taking prey such as rats, and that will inhabit human-influenced agricultural and suburban habitats with extensive and intensive AR use.As a process to randomize selection, when we had collected a larger number of a targeted species for a given year and region, for example, the Lower Mainland of British Columbia, we selected for chemical analysis every second or third individual bird from the compiled list.However, we also selected for analysis any individual diagnosed at rehabilitation intake or necropsy with symptoms of AR poisoning.Thus, with the partial exception of the British Columbia data set, samples do not reflect the incidence of exposure to and toxicity of ARs in the wild.Instead, the intent of the current North American data set was to obtain an expanded sample of carcasses with necropsy results and with SGAR residues over a range of concentrations, with the goal of facilitating the identification of residue-and speciesspecific toxicity thresholds for terrestrial raptors.
Samples were retained in our North American assessment if they met a set of predetermined criteria (as per Thomas et al., 2011): (1) postmortem evaluations were conducted prior to  liver extraction and analysis and pathophysiological signs of rodenticide poisoning were described; (2) postmortem evaluations were conducted by a veterinary pathologist or other qualified individual; (3) at least one SGAR residue (second-generation hydroxycoumarins: brodifacoum, bromadiolone, or difethialone) was eventually detected at any concentration above the data set's methods reporting limit (MRL); and (4) no FGAR residues (first-generation hydroxycoumarins; indandiones) were detected at any concentration above the data set's MRL.Diagnosis of AR poisoning followed the principles and methods described by Murray (2018).Anticoagulant rodenticide poisoning was determined as the final cause of death when necropsy indicated hemorrhage, bleed-out, or pallor in the absence of other potentially lethal traumatic injury, disease, emaciation, or pesticide or lead poisoning, and there was no evidence of blood clotting.For our purposes, the final diagnosis also factored in the later detection of anticoagulant residues in the liver (Elliott et al., 2022;Hindmarch et al., 2019;Thomas et al., 2011), but cause of death was not diagnosed based on AR residue concentration alone, even if concentrations were elevated, without the other symptoms just listed.

Chemical analysis
In brief, livers were extracted from carcasses and analyzed for anticoagulant rodenticides (brodifaoum, bromadiolone, and difethialone) using liquid chromatography with tandem mass spectrometry.Chemical analysis for the data sets from British Columbia and Quebec followed methods described in detail in our preceding papers (Albert et al., 2010;Elliott et al., 2022;Thomas et al., 2011), with further details in the Supporting Information S1.Methods for the New York data set are further described in Stone et al. (1999) and Okoniewski et al. (2021); and for California in Hosea (2000) and Serieys et al. (2015).

Statistical methods
Method detection limits.Method reporting limits varied across regions and residues, and over time (0.0006-0.02 μg/g wet wt for bromadiolone; 0.002-0.05μg/g wet wt for brodifacoum; 0.001-0.02μg/g wet wt for difethialone).To retain as many samples as possible, we included in the statistical analyses individuals with SGARs detected at any concentration.However, we excluded individuals with FGARs detected at any concentration, including the British Columbia data set (Supporting Information, Table S1).The ∑SGAR represents the sum of all detected bromadiolone, brodifacoum, and difethialone residues.Detection frequencies are reported throughout as the percentage of raptors with a positive SGAR detection, and hepatic concentrations are reported as the geometric mean and range or geometric mean ± 95% confidence interval, because residue concentrations were inflated by low values.
Samples were assigned a binary code as positive (1) or negative (0) for pathophysiological signs of AR poisoning.As per Thomas et al. (2011), a positive coding (1) designated AR poisoning evident at necropsy.A negative coding (0) designated cases in which the primary cause of death was determined to be other, including trauma, disease, emaciation, other poisoning, or undetermined.
We compiled and analyzed necropsy records collected between 1989 and 2021 in Canada and the United States.Families represented in this data set include Accipitridae (n = 322), Ardeidae (n = 1), Cathartidae (n = 10), Corvidae (n = 2), Falconidae (n = 19), Strigidae (n = 446), and Tytonidae (n = 151;  Table 1).Among all 951 necropsy reports, 262 identified SGAR poisoning as the cause or major contributing factor in the mortality.Necropsy procedures and principles are outlined in Murray (2018).Other causes of mortality in this data set include disease, trauma, and starvation.Data originated from eight separately managed governmental, nongovernmental, and academic databases.The greater proportion were sourced from a British Columbia-based nongovernmental organization (n = 556) and an academic New York State-based repository (n = 242).The remaining six databases contributed between 2 and 55 (median of 22) records to the compiled United States and Canada-wide data set.Low sample numbers in the Corvidae and Ardeidae families precluded clade-level analysis of these individuals, and they were thus removed, for a final set of n = 948 for statistical analysis and modeling.
Within our full data set, a subset of 440 analytical results yielded a single compound, either bromadiolone, brodifacoum, or difethialone, as the primary SGAR residue.We considered a compound to be the primary residue when a single chemical accounted for at least 75% of the sum SGAR residues.Using that reduced single-exposure event data set, we modeled the probability of diagnosis of SGAR poisoning at necropsy for the single compounds brodifacoum, bromadiolone and difethialone.
We built generalized linear mixed-effects models (GLMMs) in Ver.4.2.3 of the statistical computing language R (R Core Team, 2020) with Ver.0.9-1 of the GLMM adaptive package (Rizopoulos, 2020) for mixed-effects modeling.Our GLMMs included random terms for records' data set of origin, as well as a year-effect to account for variation in chemical quantitation methods among laboratories and through time.Fixed-effect terms differed between the family-and compound-specific analyses, including either the main effects and interactions between ∑SGAR and family or the additive effect of SGAR type and concentration.We modeled probabilities of diagnoses of SGAR-induced mortality (poisoning symptoms) as a binomialdistributed response with a logit-link function.Estimates for probabilities of SGAR-induced mortality were obtained using population-averaged marginal coefficients of the random effects terms with the GLMM adaptive R package (Hedeker et al., 2018;Rizopoulos, 2020).Corresponding standard errors were calculated by Monte Carlo integration, also using GLMMadaptive (Rizopoulos, 2020).Model checks including tests for overdispersion and residual simulation were performed using Ver.0.4.6 of the DHARMa package (Hartig, 2022).Model terms were considered statistically significant when marginal coefficient p values were below a threshold of α = 0.05.Data are presented as marginal population-averaged estimates ± marginal standard errors (SEs) unless otherwise indicated.

Including trauma cases in analyses
Trauma victims (i.e., by road kill or other accident) made up the majority of our database.This may include a substantial number of animals that otherwise could have been scored as 1 because they were hemorrhaging, but at the time of postmortem evaluation hemorrhaging would have been discounted based on a conservative assumption that such bleeding was caused by the trauma and not necessarily indicative of AR coagulopathy.Cases were classified as either "No Trauma" or "Trauma" victims, whereby a trauma case could not be distinguished from a potential AR poison victim (ambiguous pathophysiological signs).To address this question, we ran a set of preliminary analyses with the model.Probabilistic curves generated for the test runs indicated no statistical effect of including ambiguous trauma cases.Therefore, all such Trauma cases were included in the final analyses with the full data sets.

Family-level analyses
We built a GLMM that included fixed main effects for family and ∑SGAR as well as their interaction to test hypotheses concerning relative sensitivities of clades to ∑SGAR-poisoning (Figures 1 and 2 and Table 2).A nonparametric dispersion test of fitted residuals against simulated residuals indicated that data were not overdispersed with respect to the specified model (dispersion = 1.03, p = 0.43).In addition, residual plots indicated that the model specification was suitable for these data (Supporting Information, Figure S1).
Family-level intercepts were significantly different from the Accipitridae class (1.5 ± 0.4; z = 3.9; p < 0.001) for the Cathartidae (−1.0 ± 0.3; z = −3.1;p < 0.001) and the Tytonidae (−0.5 ± 0.2; z = −2.8;p = 0.002).Ratios were not significantly different for Falconidae (0.5 ± 0.4; z = 1.4; p = 0.2) or Strigidae (0.03 ± 0.06; z = 0.3; p = 0.6).In addition, we tested for differences in taxon sensitivities to ∑SGAR with the inclusion of an interaction effect between ∑SGAR and each family.We found that the Tytonidae responded to increases in ∑SGAR less steeply (−0.31 ± 0.08; z = −3.5;p < 0.001) than the Accipitridae, wheras the true owls, Strigidae, appeared marginally more rapidly affected by increases in ∑SGAR (0.09 ± 0.05;   As noted, concentrations are in ng g −1 .Because the method reporting limits varied across data sets and residues, and over time, SGARs detected at any concentration were included and FGARs detected at any concentration were excluded from the statistical analyses.Estimates are presented as missing (-) when the prediction occurred outside of the range of existing data.Uncertainty terms are reported as the standard deviation among SGAR concentrations producing the specified probability of ΣSGAR-poisoning, based on our accepted GLMM.FGAR = first-generation anticoagulant rodenticide; GLMM = generalized linear mixed-effects model.
z = 1.9; p = 0.072).The Cathartidae were also significantly less steeply affected by increases in ∑SGAR ( − 0.5 ± 0.1; z = −3.5;p < 0.001), and there was no emergent difference in susceptibility to ∑SGAR poisoning for the Falconidae (0.1 ± 0.2; z = 0.9; p = 0.48).Subject-conditional coefficients, including 95% confidence intervals, test statistics, and conditional p values are presented in the Supporting Information, Table S2.Toxicity threshold values for 0.05, 0.10, 0.15, 0.20, and 0.50 probability risks were lowest for barn owls, followed by true owls, falcons, and lastly accipiters (Table 2), although concentrations yielding probabilities of 0.05 and 0.10 were outside of the range of data for barn owls.Figure 3 presents the density of available data organized by diagnosis of AR or other cause of death, and by family grouping.Data density is well separated by family, with the exception of the barn owls, for which a second smaller peak of AR poisoning data appeared at comparatively low ∑SGAR concentrations, indicating an overlap in positive/negative AR mortality events.

Compound-specific comparisons
We built a GLMM that included the additive fixed effects of compound type and concentration (Supporting Information, Table S3).Overall, for all families combined, and at all levels of risk of probability of exhibiting symptoms, threshold concentrations were approximately 1.5-to 2-fold greater for bromadiolone compared with brodifacoum.Probabilities of exhibiting symptoms of SGAR poisoning were not significantly different for difethialone compared with brodifacoum, and were approximately half of the estimated threshold for bromadiolone (Figure 4 and Table 3).Thus, toxicity to raptors was highest for difethialone, although it differed minimally from brodifacoum.Bromadiolone was somewhat less toxic, by a factor of 2 or less.A nonparametric dispersion test of fitted residuals against simulated residuals indicated that data were not overdispersed with respect to the specified model (dispersion = 1.02, p = 0.77).In addition, residual plots indicated that the model specification was suitable for these data (Supporting Information, Figure S2).
We used these results to determine SGAR toxic equivalence factors (TEFs) for birds of prey.Using the 0.20 level of probability of toxicity, and setting the most toxic compound, difethialone as (one) 1, then a TEF for brodifacoum would be 0.757,   rounded to 0.8.The TEF for bromadiolone be 0.449, rounded to 0.5.

Toxicity thresholds
Published threshold values for interpretation of ∑SGAR liver concentrations vary considerably (Lohr, 2018;Rattner & Harvey, 2020).Proposed numbers include 700 ng g −1 , using brodifacoum data as an example compound and dosing studies with barn owls, by the Rodenticide Registrants Task Force, described as an international organization of rodenticide active ingredient registrants organized in 1999 (Kaukeinen & Buckle, 1992).Following aerial application of brodifacoum and evaluation of data from 10 species of free-ranging birds in New Zealand, Dowding et al. (1999) determined a concentration of 500 ng g −1 .Using data from field and captive dosing studies with barn owls, Newton et al. (1990Newton et al. ( , 1999) ) suggested 100 to 200 ng g −1 .Members of the present authorship used logistic regression models to analyze data from three owl species, Tyto furcata, Strix varia, and Bubo virginianus.They proposed threshold values, depending on species, of 70 to 180 ng g −1 ∑SGARs at a probability level of 0.20 of exhibiting anticoagulant symptoms (Thomas et al., 2011).
Wide use has been made of numbers in the range of 100 to 200 ng g −1 with authors citing Newton et al. (1990) and Thomas et al. (2011), for example: Coeurdassier et al. (2019), Herring and Eagles-Smith (2017), Lohr (2018), andCooke et al. (2022).Our present results produced threshold numbers that are approximately 1 order of magnitude lower.At liver concentrations in the range of 100 ng g −1 , the probability of coagulopathy is 50%, putting one in two birds at risk.There may be increased bias in the present data set, given that a portion of the birds were definitively targeted for residue chemistry based on initial diagnosis of coagulopathy.However, that procedure would tend to produce a larger proportion of birds with greater liver concentrations in the analyzed samples, and, therefore, calculation of a higher threshold value than otherwise.The finding of lower threshold values in the present analysis is most likely due to differences in the statistical methods compared with Thomas et al. (2011).The present model was chosen from a series of candidates as being the most performant one, and it includes taxon-specific intercepts and interaction effects.That procedure allows for a difference in the placement and shape of the toxicity S-curve, which was not done in the previous analysis, and should mean increased confidence in the present results.
For three of the family groups, the Falconidae, Accipitridae, and Strigidae, there is consistency in the shape of the doseresponse curves, and thus the liver concentrations calculated at a given probability level.We propose, therefore, a common threshold from the mean of the concentrations determined for the three families: 10 ng g −1 at a probability of 0.20 and 80 ng g −1 at 0.50.The response curve is decidedly different, however, for the Tytonidae, the barn owls, suggesting that barn owls are more sensitive to ARs than the other raptors.Interspecific differences could be caused by factors such as the capacity to metabolize ARs or variation in structure of the vitamin K epoxide reductase (VKOR) enzymes, but there are limited pertinent data for birds on either factor (Horak et al., 2018;Huang et al., 2016;Rattner et al., 2014;Rattner & Harvey, 2021;Khidkhan et al., 2024).Dietary variation in vitamin K content is a possible factor (Abi Khalil et al., 2021), but does not seem likely in raptors, particularly owls, given the overlap in diet for many species (see Hindmarch & Elliott, 2014, 2015a, 2015b).
We think the most likely cause for this difference is interspecific variability in the distribution of available hepatic concentration data for birds diagnosed with coagulopathy, and therefore effectively an artifact of the available data.Barn owls have reduced data availability over the whole distribution, leading to a difference in the curve for barn owls.Additionally, the detection limits for some of the earlier data sets are too high to allow coverage of data over the full range of relevant concentrations for barn owls, something that could be resolved only by collecting more data, which is not an option in the present study.Data close to the analytical detection limits also present challenges in calculating sums for the analysis.The second minor peak evident in the density data distribution of AR-poisoned cases at relatively low ∑SGAR for the Tytonidae, where there is overlap in positive/negative cases, is likely the result of data imputation lumping cases together when there is a gradient in concentrations that is being misrepresented.Thus, unfortunately, the analytical methods employed in some of the earlier studies were less sensitive and did not adequately cover the range of exposure values of Tytonidae.Alternatively, we would have needed to collect more data for barn owls.Given these factors, we have less confidence in the threshold for the Tytonidae, and recommend that users employ it with added caution.
In comparing their results with available thresholds, a number of authors have questioned why investigators regularly encounter raptors with liver concentrations exceeding the published threshold values of 100 to 200 ng/g, but without obvious coagulopathy symptoms (Murray, 2011(Murray, , 2018;;Quinn, 2019;Rattner & Harvey, 2020).As discussed in several reports (Khidkhan et al., 2024;Lohr, 2018;Murray, 2018;Thomas et al., 2011), and considered at length by Rattner and Harvey (2020), many variables influence measured liver residue concentrations, and whether the measured degree of exposure results in toxicity.Proposed factors include taxonomic differences in sensitivity; however, our results suggest that is relatively low.With the exception of the Tytonidae (as just discussed), the other families of raptors exhibited markedly similar exposure-response curves.We suggest that the greatest sources of variability in these studies of fortuitously collected wildlife result from the timing of exposure, that exposure is rarely to a single compound, and that there is the potential for multiple exposures and varying intervals between exposure events.As a mechanism, Rattner & Harvey (2020) refer to Huckle et al. (1989), who proposed that hepatic AR sequestration would prevent AR toxicity until accumulation sites were saturated, at which point VKOR binding sites would become blocked by free AR.Thus, healthy free-ranging animals could experience repeated AR exposures and accumulate a burden of multiple AR compounds in liver other tissues without developing and exhibiting overt toxicosis, until further AR exposure event(s) exceeded the hepatic AR sequestration threshold.In reviewing the available literature pertinent to interpreting hepatic AR residues in nontarget species and referring specifically to Thomas et al. (2011) and Walker et al. (2019), Rattner & Harvey (2020) concluded: Residues in liver provide evidence of exposure and some indication of the likelihood of effects.Intra-and interspecific differences in sensitivity, and the timing of exposure, albeit recent and repeated, are also important determinants of the onset of AR toxicosis.
As just discussed, the last point seems the most important factor influencing exposure and response.Virtually all cases reported in the literature came from opportunistic sampling efforts, and thus we know virtually nothing about the timing and incidence of AR exposure.Similar to Rattner & Harvey (2020), we suggest that the reader use these threshold values as guidelines for the probability of AR toxicity.In Table 2, we provide values for the different raptor families and a range of probabilities.The end user can select a liver concentration that best suits their own situation and risk tolerance.At the 0.5 probability level, the mean of 80 ng g −1 could be rounded to 100 ng g −1 , thus keeping the number that several investigators have used, based on Newton et al. (1990) and Thomas et al. (2011).Given the higher level of probability, one in every two birds exhibiting symptoms at that concentration, some users may have more confidence in using that number.Applied to our own data set, that is accepting a high level of poisoning risk.For example, in the case of the published data for Western Canada, the median ∑SGAR concentration for barred owls was 130 ng/g, based on a sample of 208 birds (Elliott et al., 2022).
For endangered and threatened species, loss of individuals can be critical.Thus, a lower guideline is likely to be preferred, particularly for populations experiencing a high incidence of SGAR exposure (Berny & Gaillet, 2008;Coeurdassier et al., 2019;Cooke et al., 2022;Herring et al., 2022;Hindmarch et al., 2017;Hong et al., 2019;Huang et al., 2016;Oliva-Vidal et al., 2022;Pay et al., 2021;Roos et al., 2021).At the 50% probability of toxicosis in the Tytonidea, the concentration is 40 ng g −1 , and could be considered as a guideline for listed species.An even lower threshold may be preferred for endemic island species.For the Tytonidae, the threshold at the 20% probability level is 0.32 ng g −1 ; however, that concentration may be too low to be widely endorsed, particularly given the concerns we have raised about the reliability of the barn owl data at the lower end of the exposure-response curve.

Compound-specific toxicity
The three SGARs we considered are very similar in chemical structure and in toxicity to test rodents.Data for birds are more limited, particularly for bromadiolone, with the available information indicating lower avian toxicity than brodifacoum (Rattner & Mastrota, 2018).We found that bromadiolone was less toxic than brodifacoum, but the difference was relatively small.Our results show bromadiolone to be very toxic to raptors, and to be poisoning birds across North America.We would suggest, therefore, reconsideration of the decision by US Environmental Protection Agency and the Canadian Pesticide Management and Review Agency to continue to allow outdoor application of bromadiolone around buildings and along fence lines in virtually all commercial, residential, and agricultural settings.An option is to limit SGARs to essential services, as implemented recently by the Province of British Columbia (British Columbia Ministry of Environment and Climate Change Canada, 2022), or restrictions imposed by the State of California (University of California Agriculture and Natural Resources, 2020).Of further consideration, if wide-scale and effectively prophylactic outdoor use of SGARs continues, it seems inevitable that resistance will spread and reduce the effectiveness of these compounds (Berny et al., 2018).
Monitoring data have documented significant increases in bromadiolone and decreases in brodifacoum liver concentrations in birds of prey collected in Western Canada following federal changes to regulations implemented in 2013 (Elliott et al., 2022).Thus, at least in British Columbia, the AR risk mitigation measures do appear to be having a measurable impact on the relative amounts of SGARs deployed, and the mean hepatic concentrations in raptors.At the same time, given that our present data show that bromadiolone is more toxic than previously reported (see Rattner & Mastrota, 2018), even though exposure patterns have changed, the degree of poisonings has not declined.In theory, this outcome should be extendable across North America, given that the changes in regulations were similar in both countries (Pesticide Management Regulatory Agency (PMRA), 2010; US Environmental Protection Agency [USEPA], 2014).However, AR temporal trend data from other jurisdictions are limited.The one report from the US northeast found no decrease in SGAR exposure of birds of prey subsequent to implementation of regulations (Murray, 2017).
Finally, summation of hepatic SGAR concentrations for diagnostic purposes is the standard approach used by all investigators, and that approach has been questioned (Rattner & Harvey, 2020;Rattner & Mastrota, 2018).Determination of TEFs provides a possible solution, and we think that our calculated TEFs are likely to be reasonably robust and an alternative for researchers to calculate ∑SGARs and assess the toxicity of exposure to multiple SGARs.

SUMMARY AND CONCLUSIONS
Using specific probability curves generated for toxicity of ∑SGARs at the taxonomic level of the family, we determined threshold concentrations for falcons (Falconidae), hawks and eagles (Accipitridae), common owls (Stigidae), and barn owls (Tytonidae).Due to the commonality of the curves for all families (with the exception of barn owls), we suggest a common threshold from the mean of the concentrations determined for the three families: 10 ng g −1 at a probability of 0.20 80 ng g −1 at 0.50.The response curve is decidedly different, however, for the Tytonidae, the barn owls.We attributed that finding to reduced data availability over the whole concentration distribution, and possibly that the detection limits for some of the earlier data sets was too high to allow coverage of data over the full range of relevant concentrations.Because the barn owl is listed as threatened in British Columbia, we suggest the use of the lower guideline value of 40 ng g −1 at a probability level of 0.50.Investigators may also wish to cautiously consider the more conservative guideline of 0.32 ng g −1 (equating to a probability level of 0.20) for other listed species, especially island endemics.We also compiled data on single-compound exposure events for all families collectively, and generated probability curves for three SGARs registered for use in both Canada and the United States, that is, brodifacoum, bromadiolone, and difethialone.Our results found that brodifacoum and difethialone were similar in toxicity to raptors.Bromadiolone was less toxic, but not sufficiently so as to justify its continued widespread outdoor application.Using the compoundspecific data, we determined that the TEFs were 1 for difethialone, 0.8 for brodifacoum, and 0.5 for bromadiolone.
Supporting Information-The Supporting Information is available on the Wiley Online Library at https://doi.org/10.1002/etc.5829.
Acknowledgments-We dedicate this publication to our dear friend and colleague, the late Richard Shore, who was instrumental in the conceptualization of this research and will be dearly missed by the scientific, and especially the wildlife ecotoxicology community, worldwide.We thank the Orphaned Wildlife Rehabilitation Society (OWL), Monika's Wildlife Shelter, the Wildlife Rescue Association (WRA), the Mountainaire Avian Rescue Society (MARS), Fur and Feather Taxidermy, the South Okanagan Rehabilitation Center for Owls (SORCO), the North Island Wildlife Recovery Association (NIWRA), and the following individuals, L. Wilson (Environment and Climate Change Canada), K. Langelier, and M. McAdie, and the agencies, the BC Ministry of Environment, the Yukon Ministry of Environment, the Canadian Wildlife Service, and the general public for submitting raptor carcasses for our study.We thank the lab services staff at the National Wildlife Research Center for specimen archiving and rodenticide residue analysis.We thank the following organ- Disclaimer-The views expressed in this publication are those of the authors and do not necessarily reflect the official policy or position of their affiliated institutions.
limits varied across data sets, residues, and over time, second-generation anticoagulant rodenticides (SGARs) detected at any concentration were included and first-generation anticoagulant rodenticides (FGARs) detected at any concentration were excluded from the statistical analyses.

FIGURE 1 :
FIGURE 1: Probability of second-generation anticoagulant rodenticide (SGAR) poisoning, based on diagnosis of coagulopathy in raptors by family taxonomic level.Samples are from various sites across North America, 1989 to 2021.Shaded regions indicate 95% confidence intervals.

FIGURE 2 :
FIGURE 2: Comparison of the probability of second-generation anticoagulant rodenticide (SGAR) poisoning, based on diagnosis of coagulopathy in raptors by family taxonomic level.Samples are from various sites across North America, 1989 to 2021.Shaded regions indicate family-specific 95% confidence intervals.

FIGURE 3 :
FIGURE 3: Density of data available by second-generation anticoagulant rodenticide (SGAR) concentration and by raptor family and diagnosed cause of death.Samples are from various sites across North America, 1989 to 2021.

FIGURE 4 :
FIGURE 4: Probability of second-generation anticoagulant rodenticide (SGAR) poisoning symptoms in single-compound exposure events as a function of SGAR concentration.The compound with the most quantifiable detections was brodifacoum in 337 cases, bromadiolone in 80 cases, and difethialone in 23 cases.Samples are from various sites across North America, 1989 to 2021.Shaded regions indicate compound-specific 95% confidence intervals.
izations and researchers for submitting data from outside British Columbia: California samples (S.McMillin), US Geological Survey (C.Franson), Earlier New York samples (W.Stone), US Environmental Protection Agency 2004 samples, n = 2 only (W.Erickson and D.Urban).The funding for data analysis and manuscript writing was primarily from Environment and Climate Change Canada, Ecotoxicology and Wildlife Health Directorate.Funding for sample collection, curation, necropsy, and analytical services was from the respective co-author organizations.

TABLE 1 :
Summary of sample size by species and database, arranged in descending order of total samples, collected across NorthAmerica, 1989America,  -2018

TABLE 2 :
Expected toxicity thresholds for four families of raptors with exposure to second-generation anticoagulant rodenticide

TABLE 3 :
Expected Uncertainty terms are reported as the standard deviation among SGAR concentrations producing the specified probability of SGAR poisoning, based on our accepted GLMM.GLMM = generalized linear mixed-effects models; SGAR = second-generation anticoagulant rodenticide.