Economic losses associated with mastitis due to bovine leukemia virus infection

Bovine leukemia virus (BLV), which causes enzootic bovine leukosis and immunosuppression, is widely prevalent on Japanese dairy farms. However, in the absence of a national eradication scheme with compensation programs, it is important to estimate BLV-associated economic losses to raise farmers’ awareness. Mastitis (includes both clinical and subclinical) is a common disease in the dairy industry and the most common reason for culling. We hypothesized that immunosuppression due to BLV predisposes subclinical mastitis. A retrospective cohort study was conducted to trace Holstein cows at 9 commercial dairy farms in the Nemuro and Kushiro regions of Hokkaido Prefecture, Japan, where monitoring of BLV proviral load is routine. Information regarding Dairy Herd Improvement data, parity number, and delivery day was collected at each farm. Cows with no confirmed infection with BLV during lactation were defined as non-infected. Low-proviral-load and high-proviral-load (H-PVL) cows were defined as those in which proviral load was below and over 2,465 copies/50 ng of DNA, respectively, or 56,765 copies/10 5 cells, respectively, throughout the lactation period. Survival analysis was performed using the frailty model to estimate the hazard ratio of subclinical mastitis for BLV infection status using data from 1,034 dairy cows after adjusting for parity number and delivery season as confounding factors. Kaplan–Meier survivor curves demonstrated that half of the H-PVL cows developed subclinical mastitis within 52 d after calving. The hazard ratio of subclinical mastitis for H-PVL cows was 2.61 times higher than that of non-infected cows. In 2017, there were 264,443 clinical mastitis cases in Hok-kaido. Using field and published data, annual economic losses were estimated using Monte Carlo simulation.


INTRODUCTION
Bovine leukemia virus (BLV) belongs to the genus Deltaretrovirus of the family Retroviridae and is related to several clinically important viruses, such as human T-cell lymphotropic virus type 1 and 2 and simian Tcell lymphotropic virus type 1 and 2 (El Hajj et al., 2012). Bovine leukemia virus causes enzootic bovine leukosis (EBL) in cattle, leading to significant economic losses (Rodríguez et al., 2011;Bartlett et al., 2020). Bovine leukemia virus is prevalent worldwide, except in Western Europe (OIE, 2018). Most infections are subclinical, but a proportion of cattle (<30%) over 3 yr old develop persistent lymphocytosis, and less than 5% of infected cattle develop lymphosarcomas in various visceral tissues (EFSA, 2015). Clinical signs and symptoms depend on tumor site and include digestive disturbances, inappetence, weight loss, weakness, general debility, and sometimes neurologic manifestations (OIE, 2018).
The BLV infection has a significant economic impact on the dairy industry due to trade restrictions, replacement costs, reduced milk production, and immunosuppression resulting in increased disease susceptibility (EFSA, 2015;OIE, 2018). Only a few studies have investigated economic losses associated with the effects of BLV infection on dairy production in Asia (Yang et al., 2016). We reported that carcass weight loss of dairy cattle due to infection with BLV causes annual economic loss of $1,391,649 in Hokkaido Prefecture, Japan (Nakada et al., 2022).
Mastitis (includes both clinical and subclinical) is the most common clinical disease in US dairy cows (NAHMS, 2014). In the United States, the annual economic loss due to mastitis was estimated at between $1.7 and $2.0 billion (Kvapilík et al., 2015), and that in Canada was 400 million Canadian dollars (Carson et al., 2017). Nutrition, host resistance, environmental conditions, milking equipment, milking technique, and hygiene are associated with the risk of mastitis (Lam et al., 2013). Bovine leukemia virus infection induces abnormal immune function (Frie and Coussens, 2015). A negative association between herd-level milk production and BLV positivity (Ott et al., 2003;Erskine et al., 2012) has been reported. At the individual animal level, 305-d mature equivalent yield (Norby et al., 2016) and milk production in the early and middle stages (Yang et al., 2016) in older cows were associated with BLV positivity. A higher SCS in the early and middle stages in BLV-positive cows than BLV-negative in older cows has been reported (Yang et al., 2016). However, the causality has not been proven.
The objectives of the present study were to estimate the association between BLV infection status and occurrence of subclinical mastitis and extrapolate losses due to mastitis caused by BLV infection in Hokkaido Prefecture, the main dairy production area of Japan.

Study Design and Field Investigation
A retrospective cohort study involving 9 commercial dairy farms with BLV-infected cows was conducted. These farms, located in the Nemuro and Kushiro regions of Hokkaido Prefecture, Japan, are routinely surveyed once or twice per year to determine the BLV infection of cows based on either detection of anti-BLV antibody or BLV provirus using blood samples under veterinary clinical services. The study was conducted between April 2015 and March 2018. No cows infected with BLV were reported with malignant lymphoma-EBL within the 9 herds during the study period. The study was conducted in accordance with Strengthening the Reporting of Observational Studies in Epidemiology-Veterinary (STROBE-Vet; Sargeant et al., 2016).

Quantification of BLV Proviral Load
Genomic DNA was isolated from whole blood samples using a Wizard Genomic DNA Purification kit (Prome-ga). The proviral load (PVL) was measured using one of 2 different real-time quantitative PCR (qPCR) assays, namely, the BLV-CoCoMo-qPCR (Riken Genesis) or Cycleave BLV qPCR (Takara) assay. Both qPCR methods have been reported to have high sensitivity (100% for 1.56 provirus per 10 5 peripheral blood mononuclear cells, 3 of 3 samples each; Jimba et al., 2012). The high specificity of BLV-CoCoMo-qPCR, which did not amplify long-terminal repeat of other various retroviruses in experiments, has been reported (Jimba et al., 2010). The field veterinarians in charge decided using either the BLV-CoCoMo-qPCR or Cycleave BLV qPCR assay. Good agreement between the results of these assays was described in our previous study (Nakada et al., 2022). Entire details on the quantification of BLV-PVL both BLV-CoCoMo-qPCR and Cycleave BLV qPCR assay have been published elsewhere (Nakada et al., 2022).
Quantification of BLV-PVL using the BLV-CoCoMo-qPCR assay was performed at the Agricultural Research Department of the Hokkaido Research Organization Animal Research Center, and that using the Cycleave BLV qPCR assay was performed at the Hokkaido Higashi Agriculture Mutual Aid Association clinical laboratory (Nakashibetsu, Japan) or the Research Institute for Animal Science in Biochemistry and Toxicology (Sagamihara, Japan).
The field veterinarians decided on the use of qPCR assays employing either BLV-CoCoMo-qPCR or the Cycleave BLV qPCR, depending on access to the corresponding laboratories.

Classification of BLV-PVL
Bovine leukemia virus-infected cows were classified into the following 2 groups according to PVL: low PVL (L-PVL) and high PVL (H-PVL). According to our previous research, the L-PVL and H-PVL cutoff thresholds as determined using the BLV-CoCoMo-qPCR and Cycleave BLV qPCR BLV assays were 2,465 copies/50 ng of DNA, and 56,765 copies/10 5 cells, respectively. Complete details on the classification of the level of BLV-PVL have been published elsewhere (Nakada et al., 2022). Low-PVL and H-PVL cows were defined as those in which the PVL was below and above the indicated cut-off thresholds throughout the lactation period, respectively. Based on our previous studies (Nakada et al., 2018(Nakada et al., , 2022, the cows continuously having H-PVL more than 2 times were considered to have persistent lymphocytosis. Cows in which the BLV-PVL fluctuated with values above and below the cut-off threshold during lactation were excluded from the study. The ELISA-positive cows with PVL below detection limit were categorized into L-PVL.

Data Collection and Management
Herd-level information such as the total number of cows and animal-level information such as breed, parity, and BLV test results and Dairy Herd Improvement (DHI) records was collected by field veterinarians in charge of the study farms via interviews with farm owners and checking farm records. Within the lactation period, event (subclinical mastitis) time was defined as the number of days from delivery until the first occurrence of subclinical mastitis. Dates of the first occurrence of subclinical mastitis were collected from DHI records of the 9 herds. Subclinical mastitis was defined as a linear SCC score of ≥5 (283,000 SCC/mL) in the DHI records. The cut-off value, 283,000 SCC/mL, was adapted based on a report of positive predictive value of 80% for dairy herds with 30% prevalence of IMI (Reneau, 1986). A high positive predictive value, 94.4%, has been reported from Belgium recently, by using 250,000 SCC/mL cut-off for IMI (note that positive predictive value is high when a prevalence is low) (Jashari et al., 2016). Cows with missing DHI records during the lactation period were excluded from the study.
Parity was categorized as 1st, 2nd and 3rd, 4th and 5th, or 6th and over. Delivery season was defined as follows: from January to March as winter, April to June as spring, July to September as summer, and October to December as fall. All data were digitized and handled using commercially available spreadsheet software (Excel 2013;Microsoft Corp.).

Statistical Analysis
According to the DHI data across Hokkaido Prefecture in 2017, 15% of milking cows recorded a liner score of ≥5 (Hokkaido Dairy Milk Recording and Testing Association, 2017). The sample size was calculated to detect the difference in time to event endpoint, for the 1.5 times increased rate of subclinical mastitis due to BLV infection, using the Freedman formula (Abel et al., 2015).
The frailty model is a random effect survival model that allows for unobserved heterogeneity or statistical dependence between observed survival data, and random effects are treated as continuous variables that describe excess risk or frailty (Rondeau et al., 2006). Frailties are useful in modeling correlations in multivariate survival and event history data, including recurrent events such as mastitis or lameness, in which an individual cow's frailty affects the occurrence of events, and community trials, in which different events within a community involve a common frailty (or shared frailty) shared by each individual within the community (Hanagal, 2011). Based on our hypothesis, a 2-level hierarchical causal web was constructed to illustrate the relationships between the explanatory and outcome variables ( Figure 1). Therefore, Cox regression model nested frailty was applied.
To test our hypothesis based on the data, a Cox regression model with 2 nested frailties (herd and cow levels) was considered. Frailty is typically defined as a clustering effect in survival analyses (Dohoo et al., 2009). The 2 nested frailties Cox model (Elghafghuf et al., 2014a, b) can be written as follows: where h ij represents for hazard of cow j in herd i, h 0 (t) represents for baseline hazard in the regression model, Z denotes the covariate vector, β(t) represents the corresponding vector of the regression parameter, and w i and w ij represent unobserved random effects common to all observations from cow j in herd i, conditional on the 2 nested frailties. Hazards must be positive in the Cox model, so w follows a log-normal distribution. Equation 1 can be transformed into random effects context as follows: where u i = log(w i ) and u ij = log(w ij ) indicate nested random effects with zero means and variances σ i 2 and σ ij 2 for the herd and cow levels, respectively. As shown in Figure 1, the explanatory variable of interest was BLV infection status (X). We had 2 potential confounders, including greater parity number (C 1 ), which can be associated with higher probabilities of BLV infection (and progression) and subclinical mastitis occurrence, whereas season of delivery (C 2 ) affects only subclinical mastitis occurrence.
Descriptive analyses were carried out for explanatory variables aggregated in the data set, and distributions and collinearities among variables were assessed. Univariable analyses of explanatory variables for the hazard of subclinical mastitis were performed using a standard Cox regression model with the Efron method for ties. The proportional hazards assumption was appraised for every predictor using Schoenfeld residuals (Dohoo et al., 2009).
Potentially important predictors based on the unconditional analysis results were then included in a multivariable model. Statistical associations were then inferred based on the causal web.

Estimation of Economic Loss
Economic loss due to BLV infection was defined as an increase in economic loss due to the increment of clinical and subclinical mastitis occurrences caused by BLV infection in a year, compared with cows not infected with BLV. The definition applied to losses associated with both individual cows and Hokkaido Prefecture. Economic loss was categorized into 2 types as follows: (1) Costs of discarded milk and intramammary antimicrobial agents associated with the treatment of BLV-induced clinical (veterinarian-treated) mastitis.
In this analysis, secondary economic losses, such as increased work load, mental burden, and culling of mastitis cows, were not included. Economic loss due to BLV-associated mastitis (includes both clinical and subclinical) across Hokkaido Prefecture was estimated. These data were used to calculate the loss per individual cow for both H-PVL and non-H-PVL cows (including L-PVL cows and the cows not infected with BLV by ELISA test, as the hazards of subclinical mastitis do not differ). In the estimation process, Monte Carlo simulations were used to calculate some parameters, and an interval estimate approach was selected for the final estimations.
Increased economic loss due to BLV infection across Hokkaido Prefecture was estimated as follows. First, the economic loss due to clinical mastitis in H-PVL cows was estimated (Losscmas HPVL , i). Second, the loss due to reduced milk production caused by subclinical mastitis in H-PVL cows was estimated (Lossred HPVL , ii). The loss associated with L-PVL was not included in these analyses as the loss from BLV infection because the hazard of subclinical mastitis in L-PVL cows was not significantly different from that of non-infected cows. Third, these 2 types of losses were summed (Loss HPVL , iii = i + ii). Fourth, baseline losses due to clinical (Base cmas , iv) and subclinical mastitis (Base scmas , v), if these H-PVL cows were not infected with BLV, were calculated (Base, vi = iv + v). Finally, the baseline economic loss (vi) was subtracted from the economic loss for H-PVL cows (iii) to calculate the increased economic loss in Hokkaido Prefecture due to mastitis resulting from BLV infection (Eloss, vii = iii − vi). Theoretical causal web for the occurrence of subclinical mastitis due to bovine leukemia virus (BLV) infection in 9 Japanese dairy herds. Two-level hierarchy affects survivorship from mastitis. The factor of interest is X, and the outcome is Y. C 1 and C 2 are potential confounders.
To estimate the economic loss due to clinical mastitis in H-PVL cows in Hokkaido Prefecture (i), the number of H-PVL clinical mastitis cases (Ncmas HPVL ) was first estimated by multiplying the number of clinical mastitis cases reported in Hokkaido Prefecture in 2017 (Ncmas), 264,443 (Hokkaido-NOSAI, 2017), with the estimated proportion of H-PVL cows among clinical mastitis cases in the prefecture (PHPVL cmas ), which is as follows:

Ncmas
Ncmas PHPVL HPVL cmas = × . [3] The proportion of H-PVL cows among clinical mastitis cases, PHPVL cmas , was estimated using the proportion of H-PVL cows among subclinical mastitis infections (PHPVL scmas ), assuming their similarity. [4] The number of subclinical mastitis infections among H-PVL cows (Nscmas HPVL ) was estimated by multiplying the number of dairy cows in Hokkaido Prefecture (Ncows) as of 2017 (n = 496,400; MAFF, 2018a) with the following parameters: animal-level BLV prevalence (PrevBLV), 24.1%, as used in our previous study (Nakada et al., 2022); the proportion of H-PVL cows among BLV-infected cows in this study (PHPVL overall ), 19.7%; and the mean probability of subclinical mastitis occurrence among H-PVL cows (Pscmas HPVL ). The number of subclinical mastitis infections among H-PVL cows was calculated as follows: Similarly, the number of subclinical mastitis infections among non-H-PVL cows (L-PVL cows and cows not infected with BLV: Nscmas nonHPVL ) was estimated as follows using the number of non-H-PVL cows (Ncows nonHPVL ) and the mean probability of subclinical mastitis occurrence among non-H-PVL cows (Pscmas nonHPVL ): Nscmas N cows Pscmas nonHPVL n onHPVL nonHPVL where Ncows nonHPVL represents the value resulting from subtracting the number of H-PVL cows (Ncows × Pre-vBLV × PHPVL overall in Equation 5) from Ncows.
To estimate the probabilities of subclinical mastitis for H-PVL (Pscmas HPVL ) and non-H-PVL (Pscmas nonHPVL ) cows, respectively, the proportion of subclinical mastitis cows was randomly sampled based on Kaplan-Meier survival curves for subclinical mastitis in H-PVL and non-H-PVL cows at 310 d postcalving; therefore, these proportions represent the prevalence of subclinical mastitis at any time point. The Monte Carlo simulations were iterated 5,000 times for both H-PVL and non-H-PVL cows. We selected 310 d postcalving because in 2017, the calving interval mode for dairy cows in Hokkaido Prefecture was 357 d, and the dry period was approximately 60 d (LIAJ, 2017), resulting in 297 d in milking. Considering cows for which the calving interval was not calculated due to replacement, the milking period was set longer.
To estimate the cost of clinical mastitis per cow (Losscmas cow ), it was assumed that milk was discarded (i.e., wasted) for 7 d (Day waste ) for a single clinical mastitis treatment. Based on an average amount of 305-d milk of 9,626 kg in Hokkaido Prefecture in November 2017 (LIAJ, 2017), the average daily milk production per head was 31.6 kg (Vol day ). The raw milk unit was assumed to be ¥100/kg (Price milk ), and the cost of intramammary antimicrobials used to treat mastitis was ¥100 (equivalent to $0.95 based on the November 24, 2021, exchange rate of ¥105.0, Price AM ) per day. Clinical mastitis treatment was assumed to be intramammary antibiotic therapy for 3 d (Day treat ). The cost of clinical mastitis per cow was calculated as follows: [7] Finally, the economic loss due to clinical mastitis in H-PVL-cows in Hokkaido Prefecture (Losscmas HPVL , i) was calculated, as follows, from the product of the number of clinical mastitis cases in H-PVL cows (Ncmas HPVL ) and the unit cost (Losscmas cow ): To estimate the loss due to reduced milk production caused by subclinical mastitis associated with BLV infection (Lossred HPVL , ii), the mean decline in milk production of a cow with subclinical mastitis was multiplied with the total number of milking days under subclinical mastitis conditions of affected cows in Hok- The baseline loss due to clinical mastitis (Base cmas , iv) among H-PVL cows (the loss if these cows did not exhibit H-PVL status) was estimated by applying the incidence rate of clinical mastitis in non-H-PVL-cows to the estimated H-PVL bovine population (Ncmas HPVL ). The baseline loss due to subclinical mastitis (Base scmas , v) among H-PVL cows (again, the loss if these cows did not exhibit H-PVL status) was estimated by applying the proportion of subclinical mastitis cows among non-H-PVL cows described above (Pscmas nonHPVL ) and the median number of days under subclinical mastitis conditions among non-H-PVL cows to the H-PVL bovine population (Ncmas HPVL ; Equation 10). The number of days under subclinical mastitis conditions among non-H-PVL cows was estimated using the same approach with H-PVL cows. Non-H-PVL cows had subclinical mastitis or were in the post-treatment state for 64.2% of the lactation period (Pdayscmas nonHPVL ). The baseline loss due to subclinical mastitis was calculated as follows:  (Therneau, 2020).

Descriptive Statistics
The farms included in the study kept between 57 and 284 cows, with a mean and median of 133 and 87 cows, respectively. Between April 2015 and March 2018, the proportion of cows within a herd diagnosed with sub-clinical mastitis ranged between 18.8 and 56.5%, with a mean and median of 41.1 and 48.2%, respectively. The total number of cows studied was 1,034, which included 868 cows not infected with BLV, 135 L-PVL cows, and 31 H-PVL cows. The total number of cows, 1,034, does not include the 35 cows excluded (35/1,069, 3.3%) due to the change of infection status between L-PVL and H-PVL (including H-PVL to L-PVL at a border of the cut-off value) during the study period. The overall proportion of cows diagnosed with subclinical mastitis during the study period in the 9 herds examined was 42.6% (440/1,034), and the proportions of cows diagnosed with subclinical mastitis within 50, 110, and 210 postpartum days were 15.0, 25.0, and 36.7%, respectively. The median day of subclinical mastitis diagnosis and censoring day were 92 (25 and 75 percentiles: 33 and 197.75) and 263.5, respectively.
The descriptive statistics for cow-level predictor variables included in the analysis are shown in Table 1. Kaplan-Meier survival curves for subclinical mastitis events by BLV infection status (non-infected, L-PVL, and H-PVL), parity (1st, 2nd and 3rd, 4th and 5th, and 6th and over), and delivery season (spring, summer, fall, or winter) are shown in Figures 2, 3, and 4, respectively. The survival curve of L-PVL remained lower than that of non-infected cows, but significant difference was not observed between the 2 lines. In contrast, the survival curve of H-PVL decreased earlier than the other 2 lines  Low-proviral-load and high-proviral-load cows were defined as those in which the proviral load was below and above the cut-off threshold throughout the lactation period, respectively. ( Figure 2). The survival probability decreased as parity increased (Figure 3), but was not different between delivery seasons (Figure 4).
In log-rank tests for multi-collinearity, the P-values for 3 predictors (BLV infection status, parity, and delivery season) were < 0.2, suggesting no collinearity.

Multivariable Analysis
The sample size required for a standard Cox regression analysis was 448 cows, and the number of cows studied exceeded the requirement. A multivariable analysis was performed based on BLV infection status (non-infected, L-PVL, and H-PVL), parity (1st, 2nd and 3rd, 4th and 5th, and 6th and over), and delivery season (spring, summer, fall, or winter). Figure 1 suggests the possibility of an interaction between X and C 1 . However, Kaplan-Meier plots did not suggest an existing interaction, and the multivariable model did not include the interaction term. Results of the multivariable model are tabulated in Table 2. The hazard ratio (HR) for subclinical mastitis for H-PVL cows was 2.61 times higher than that for cows not infected with BLV. The HR of subclinical mastitis increased with parity number. Delivery season was not associated with the HR of subclinical mastitis.     Table 3 summarizes the economic loss due to mastitis (includes both clinical and subclinical) associated with BLV infection. The annual economic loss due to mastitis in H-PVL cows in Hokkaido Prefecture (iii in Table  3) was almost $9.7 million. By subtracting the baseline loss-the loss, which would have occurred even in the absence of BLV infection-in these cows (vi), the estimated increased loss due to BLV-associated mastitis in Hokkaido Prefecture (vii) was $6.1 million.

Estimation of Annual Economic Losses
At the individual cow level, the annual loss due to mastitis per H-PVL cow ($418.59) was 2.73 times greater than that per non-H-PVL cow ($152.96). The increased cost of BLV-associated mastitis per cow ($265.63) demonstrates the magnitude of economic impact; for example, it was even greater than the baseline loss due to mastitis per non-H-PVL cow ($265.63).

Mastitis (including both clinical and subclinical) is
an inflammation of the mammary gland caused primarily by bacterial infection. Typically, 72.8% of cows with clinical mastitis recover and remain in the herd, whereas 24% are removed or sold (USDA 2014). Risk factors for mastitis have been widely investigated (Green et al., 2007;Lam et al., 2013). However, the relationship between BLV infection and the occurrence of mastitis remains unclear. To our knowledge, this study is the first to report that BLV infection in H-PVL cows is associated with the occurrence of mastitis.
We hypothesized that BLV suppresses immune function, leading to bacterial IMI, and consequently increasing incidence of subclinical mastitis. Numerous researchers have reported that BLV infection reduces milk productivity (Ott et al., 2003;Erskine et al., 2012;Bartlett et al., 2013;Nekouei et al., 2016;Norby et al., 2016), but the mechanism has not been elucidated. The Kaplan-Meier survivor curves constructed in the present study demonstrated that a half of H-PVL cows suffered from subclinical mastitis within 52 d after calving, and among the subclinical mastitis cases observed within 310 d after delivery, one-half occurred within 30 d. Milk fever and ketosis, which frequently occur during the postpartum period until peak lactation, are associated with nutritional management (Goff, 2006). Negative energy balance from the postpartum period to peak lactation inhibits immune function (Sordillo, 2016). In addition to these known factors, our survival analysis results identified H-PVL status as a significant risk factor for subclinical mastitis, which reduces milk productivity. Several previous studies reported the mechanism of immunosuppression in BLV infection. Bovine leukemia virus is harbored in the mammary glands of BLV-infected cows with subclinical mastitis (Yoshikawa et al., 1997), where it can cause immunosuppression. A higher percentage of CD5 − /CD11b − B cells in the milk of BLV-infected cows with persistent lymphocytosis compared with non-infected or BLVinfected aleukemic cows has been reported, indicating dysfunction of milk neutrophils (Della Libera et al., 2015). Moreover, the concentration in milk of lingual antimicrobial peptide, a natural immunity factor that is indicative of immune function in the mammary gland, is lower in H-PVL cows than L-PVL cows (Watanabe et al., 2019). This study demonstrated that subclinical mastitis associated with BLV infection may be induced by immunosuppression when H-PVL status (European Community's key for lymphocytic status) occurs, rather than infection with BLV.
The survival analysis of the present study identified another significant factor associated with subclinical mastitis, which is parity number. The incidence of mastitis reportedly increases with the increment parity number (Hiitiö et al., 2017;Zeryehun and Abera, 2017). In contrast, unlike previous studies (Olde Riekerink et al., 2007;Kurjogi and Kaliwal, 2014), delivery season was not associated with the occurrence of subclinical mastitis in our study. The study area was located near 43° north latitude, which is characterized by a cool climate. Cows on the farms studied are therefore less likely to undergo heat stress during the summer, thus reducing the likelihood of seasonal subclinical mastitis.
Several farm-level factors, such as farm dairy hygiene (Fujimoto et al., 2020;Miyama et al., 2020), and individual animal-level factors, such as sensitivity to mastitis (Martin et al., 2018), are known to affect the occurrence of mastitis. However, our study employed a frailty model to adequately control for such clustering effects (Hanagal, 2011), in showing the pure effect of BLV infection on subclinical mastitis. This method can be applied to other countries and be recommended in quantifying economic loss from mastitis due to BLV infection.
This was the first economic study of BLV-associated mastitis in Japan as far as we know. The estimated annual economic loss in a BLV-infected cow was $206.8. Nakada et al.: ECONOMIC LOSS ASSOCIATED WITH BOVINE LEUKEMIA VIRUS According to a study in the United States, BLV-seropositive cows produced $59 less annually than non-infected cows (Ott et al., 2003). The methods for calculating economic losses were different in these studies, but considering the difference in raw milk unit prices between the United States (raw milk unit price per kilogram in the United States was $0.41 as of 2021 September) and Japan ($1.00 per kilogram), our estimate of the annual economic loss in BLV-infected cows appears reasonable.
In considering the magnitude of economic burden in dairy production in Hokkaido Prefecture, the elevated loss associated with mastitis due to BLV infection accounted for 0.17% of total raw milk production ($6.1 million/$3.5 billion in 2017; MAFF, 2018b (Rollin et al., 2015), approximately the same as the loss in H-PVL cows in the present study. An economic study of animal disease is a powerful tool to develop control measures for chronic animal disease with the consent from stakeholders (Isoda et al., 2019). The annual economic loss in H-PVL cows was 2.73 times higher than that in non-H-PVL cows ($418.59/$152.96), suggesting that there is economic merit in removing H-PVL cows from the herd. Isolation and culling of H-PVL cows have been recommended as effective methods for decreasing BLV infection because these cows are particularly infectious (Ruggiero et al., 2019). In Japan, however, in the absence of a national eradication scheme with compensation programs, dairy farmers are reluctant to cull H-PVL cows at their cost. The messages from our study should provide clear incentive for owners to cull H-PVL cows earlier and thus prevent the spread of BLV infection on their dairy farms. Moreover, the number of EBL notifications has tripled compared with a decade ago (MAFF, 2020), and an increase in infection BLV prevalence reduced milk production (Ott et al., 2003;Erskine et al., 2012). This suggests that the damage to the Hokkaido dairy industry will be even greater than the estimation in this study, if control measure is not taken now. From the viewpoint of efficient milk production and prevention of the spread of BLV infection, the government should consider some kind of compensation for the culling of H-PVL cows.
This research has 4 limitations. First, we only assumed that clinical mastitis was treated using intramammary antimicrobial agents; however, treatment of clinical mastitis included veterinary care. The cost of veterinary services was not calculated because the system in Japan involves publicly funded livestock insurance. Therefore, the economic losses estimated in this study may be underestimated. Second, mastitis was the second most common reason for dairy cow removal in Hokkaido Prefecture in 2017, as approximately 10% of mastitis-affected cows were culled (LIAJ 2017). However, this economic study did not assume the cost of replacing culled cows, which could have led to further underestimation of economic losses. Third, the definition of subclinical mastitis relied on the cut-off value of linear score, and the estimation was not adjusted for the sensitivity and specificity as they are not available. Fourth, the reduction in milk productivity was modeled to begin at the occurrence of subclinical mastitis and continue throughout the lactation period; thus, these losses could be slightly overestimated. Actually, the reduction of milk may be a function of SCC and type of responsible bacteria, but detailed modeling was avoided as this problem is too complex for a population-wise estimation. The economic analysis was conducted based on several assumptions that the sampled data can be extrapolated to entire Hokkaido Prefecture. Occurrence of EBL is observed in entire Hokkaido Prefecture, and the study site was one of the intensive dairy production areas. Therefore, the economic loss calculated in this study can represent Hokkaido Prefecture. However, there can be some areas where voluntary BLV control is weak. In such areas, the proportion of H-PVL cows may be higher than the present study. There are such data gaps in estimating the economic loss due to BLV infection, and uncertainties in the parameters were presented using probability distributions. The economic loss of mastitis may share a large part of the loss due to BLV infection. However, the entire picture of economic loss due to BLV infection has not been quantified. Integrative assessment of the economic loss is needed in future.
In conclusion, this study revealed that BLV-infected H-PVL cows exhibit a higher HR for subclinical mastitis after adjustment for parity number, delivery season, and clustering effects at the farm and individual cow levels. BLV-infected H-PVL cows are associated with significant economic losses in Hokkaido Prefecture. Priority removal of BLV-infected H-PVL cows is recommended in terms of both BLV infection control and economics.

ACKNOWLEDGMENTS
This study was carried out by the Animal Research Center of Hokkaido Research Organization (Shintoku, Japan) as research commissioned by the Hokkaido Higashi Agriculture Mutual Aid Association (Nakashibetsu, Japan). The authors thank the veterinarians of the Hokkaido Higashi Agriculture Mutual Aid Association for collecting blood samples. Many thanks go to the dairy farmers who participated in this study. Author contributions are as follows: SN, JK, and KM conceptualized the study. Methodology was developed by SN, YF, JK, and KM. Data were collected by SN and JK, diagnosed by SN and JK, and analyzed by SN, YF, and KM. SN and JK were funded for the study by Hokkaido Higashi Agriculture Mutual Aid Association (Nakashibetsu, Japan). Supervision was done by JK and KM. All authors approved the manuscript. The authors have not stated any conflicts of interest.