Standardized analysis of German cattle mortality using national register data

Standardized analysis of German cattle mortality using national register data

Preventive Veterinary Medicine 118 (2015) 260–270 Contents lists available at ScienceDirect Preventive Veterinary Medicine journal homepage: www.els...

944KB Sizes 2 Downloads 256 Views

Preventive Veterinary Medicine 118 (2015) 260–270

Contents lists available at ScienceDirect

Preventive Veterinary Medicine journal homepage: www.elsevier.com/locate/prevetmed

Standardized analysis of German cattle mortality using national register data Gunter Pannwitz ∗ District Veterinary Office Vorpommern-Greifswald, Germany

a r t i c l e

i n f o

Article history: Received 30 April 2014 Received in revised form 17 November 2014 Accepted 22 November 2014 Keywords: Cattle Mortality Standardization Surveillance Health Welfare

a b s t r a c t In a retrospective cohort study of national register data, 1946 randomly selected holdings, with 286,912 individual cattle accumulating 170,416 animal-years were analyzed. The sample was considered to represent the national herd in Germany 2012. Within each holding, individual cattle records were stratified by current age (≤21 days, 3–6 weeks, 6–12 weeks, 3–6 months, 6–12 months, 1–2, 2–4, 4–8, and >8 years), sex, breed (intensive milk, less intensive milk, and beef), and mean monthly air temperature (<10 ◦ C and ≥10 ◦ C). Holdings were categorized by size (<100 and ≥100 animal-years), calving rate, slaughter rate, and federal state. 8027 on-site deaths (excluding slaughter for human consumption) were recorded, with cattle aged <6 months, 6–24 months, and >2 years contributing 50.0%, 15.4%, and 34.6% of deaths, respectively. Poisson regression and generalized estimating equations (gee) accounting for intra-herd clustering were used to model the number of deaths. In both models, most age bands differed significantly, with highest rates in calves ≤21 days, falling to lowest rates in 1–2 year olds, and rising again thereafter in females. Males exhibited higher mortality than females from birth to 2 years. All breed categories differed significantly with lowest rates in beef and highest in intensive milk breeds. Larger holdings, temperatures ≤10 ◦ C, calving rates >0–0.5 per animal year were all associated with higher mortality. Via interaction, intensive and less intensive milk breed cattle aging 6 weeks to 6 months and intensive milk breed females >4 years were associated with higher mortality. There were no significant differences between federal states and slaughter rates. The standardized deviations of modeled dead cattle numbers from occurred deaths per calendar year per holding were calculated and a 95% reference range of deviations constructed. This approach makes a standardized active monitoring and surveillance system regardless of herd size possible, offering a useful, inexpensive and easy implementable aid in the detection of holdings deviant from mortality levels of the national herd. © 2014 Elsevier B.V. All rights reserved.

1. Introduction Health-related monitoring and surveillance systems (MOSS) rely on valid, simple, timely, flexible and

∗ Correspondence to: District Veterinary Office VorpommernGreifswald, Bluthsluster Str. 5b, D-17389 Anklam, Germany. Tel.: +49 3834 8760 3817; fax: +49 3834 8760 9019. E-mail address: [email protected] http://dx.doi.org/10.1016/j.prevetmed.2014.11.020 0167-5877/© 2014 Elsevier B.V. All rights reserved.

cost-effective information (Zepeda and Salman, 2003). In human and animal medicine, the mortality level has been established as a useful parameter (European Commission, 2008; Broom, 1991, 1993; Hemsworth et al., 1995; Mellor and Stafford, 2004). Several studies in cattle also confirmed the high relevance of mortality measurements as a health and welfare indicator (Rushen et al., 2008; von Keyserlingk et al., 2009; Alvasen et al., 2012; Thomsen et al., 2004, 2006; Thomsen and Houe, 2006; Perrin et al., 2010; de

G. Pannwitz / Preventive Veterinary Medicine 118 (2015) 260–270

Vries et al., 2011, 2013) that may be used for targeted riskbased official veterinary inspections (Ortiz-Pelaez et al., 2008). In pig and poultry holdings, for example, monthly mortality risks exceeding certain thresholds may legally require veterinary investigations (Anon., 2013a,b). Last but not least, an increased mortality decreases the profitability of livestock production and is economically undesirable. In the EU, cattle identification, registration and documentation of movements have been harmonized. Each holding carries an individual ID (identity) number. Every keeper has to report births, deaths, and movement of cattle to and from the holding with the respective individual cattle ID-numbers and dates to a national computer database (Regulation (EC) No 1760/2000). Reports have to be done within 7 days of occurrence or less, depending on national law. Plausible data entry is motivated by database crosschecks comparing reports of consignor and consignee, notification of implausible records to official agencies, timely inspections, and coupled premium and direct payments to legal compliance (Regulation (EC) No 73/2009). Within the European Union, Germany is the biggest milk producer containing the second largest bovine population. This national herd is quite diverse both in respect to animal and holding characteristics (Destatis, 2013; Eurostat, 2013). As for other member states, basic descriptive mortality statistics are not easily available despite the outlined existence of comprehensive data and the usefulness of mortality measures. Therefore, the objectives of this study were threefold: first, to describe and analyze cattle mortality in Germany 2012, second, to outline a standardized approach for cattle mortality analyses using routinely available national register data, and third, to describe methods to detect holdings with higher than expected mortality. 2. Materials and methods 2.1. Study population and sampling The study covered a follow-up period of 1 calendar year (1.1.2012–31.12.2012) and targeted the German cattle population. The primary target units were all agricultural cattle holdings, sourced from the registered holdings in the national database HI-Tier (www.hi-tier.de). The initial sampling frame comprised a list of all registered agricultural holding ID-numbers at 17.9.2012 for every federal state. The states Hamburg, Berlin, Bremen, and Saarland were excluded from analysis due to low holding and cattle numbers. The final sampling frame was a list of agricultural holding ID-numbers for 12 of 16 federal states. According to the official statistic (Destatis, 2013), these states contained 99.5% of the 2012 German cattle population (12.45 of 12.51 million) and 99.4% of German agricultural holdings (160,444 of 161,453). The sampling frame itself did not contain any information regarding the risk factors to be examined. In a pilot study in March 2013, 60 holdings per state were randomly sampled and their 2012 cattle registries examined. The pilot sample revealed holdings that did or did not keep cattle in 2012 (i.e. active and non-active holdings) as well as an estimated general mortality rate of about

261

0.05 per animal year. The final sample per state was aimed to estimate a mortality rate of about 0.1 per animal year with desired 95% confidence, 80% power, and an absolute 2–5% margin of error. Regarding each holding as a cluster of individual cattle, stratified cluster samples of about 150 active holdings per state were calculated to be necessary (Scheaffer et al., 2006). In May 2013, 2820 holdings were sampled. Excluding non-active holdings, 1946 were examined (total holding sampling fraction 1.2%). The state-specific cattle census per 3.11.2012 was obtained (Destatis, 2013). The national cattle death total for the calendar years 2011–2013 was received from the national database HI-Tier in March 2014. 2.2. Data categorization All cattle were individually categorized by the fixed covariates sex and breed. Out of 85 possible breeds and crossbreeds, 68 were present in the sample and categorized as (1) intensive milk breeds: Holstein-Friesian and Red Holstein, (2) less intensive milk breeds: Fleckvieh, Braunvieh, Angler, Red Holstein Dual purpose, German black and white lowland cattle, Pinzgauer, Vorderwälder, Hinterwälder, Gelbvieh, Jersey, milk type crossbreeds, and (3) beef breeds: all others, mainly: mixed meat type, Limousin, Fleckvieh meat type, Charolais, Angus, Uckermärker, Galloway, Highland, Hereford, Blonde d‘Aquitaine, Dexter, Salers, Aubrac, Gelbvieh, etc. An individual’s animal-time in a holding was stratified by the time-dependent covariate current age. Using Lexis expansion, nine age strata were created (maximum possible animal-days per animal in brackets): ≤21 days (22), 3 to ≤6 weeks (23), 6 to ≤12 weeks (46), 3 to ≤6 months (91), 6 to ≤12 months (184), 1 to ≤2, 2 to ≤4, 4 to ≤8, and >8 years (366 each). After entering a holding (via birth or entry), an individual accumulates time in its age stratum until being right-censored, i.e. until to (a) exit holding alive for another holding or slaughter, (b) change to next age stratum, (c) be slaughtered for human consumption within holding, (d) die (regardless of cause, including euthanasia), (e) report as missing or (f) end of study period. Entering and leaving a holding on the same day counted as one animal-day. Cases were defined as on-site deaths (event d). Their number served as numerator for the calculation of mortality rates, expressed as deaths per animal-year (ay). For every state and calendar month, mean air temperatures were obtained from the German Weather Service (DWD 2013) and categorized as <10 ◦ C and ≥10 ◦ C. Calendar months were treated as a time-dependent covariate and categorized dependent on mean air temperature. For example, a calve born on 22nd April and leaving the holding on 7th May accumulated at total of 16 animal-days in the first age stratum, thereof 9 days at <10 ◦ C and 7 days at ≥10◦ (mean air temperatures in April and May: <10 ◦ C and ≥10 ◦ C, respectively). Herd size was categorized into <100 or ≥100 animalyears in total per calendar year. Births (excluding stillbirths/abortions that are not routinely recorded) were regarded both as a beginning of an individual’s contribution of animal-time (calves) and as exposure possibly affecting the dam’s mortality rate. Giving

262

G. Pannwitz / Preventive Veterinary Medicine 118 (2015) 260–270

birth counted as one calving event for the dam, regardless whether to singles, twins or triplets. For dams contributing two calving events, e.g. due to a calving in February and December, only the second event counted. The calving rate of females >2 years in every holding was obtained via division of calving events by animal-years of these females and categorized as: no calving, >0–0.5, 0.51–0.8, 0.81–1, and >1. Herd-level slaughter rates were categorized as: no slaughter, >0–0.03, >0.03–0.12, >0.12–0.3, and >0.3. All individual-level animal-times of the same age, sex, breed, and temperature strata were aggregated per holding. Federal state, holding size, and slaughter rate were treated as holding-level covariates, and calving rate as holding-level covariate applicable to females >2 years. 2.3. Data analysis and model building Individual cattle data within the sampled holdings were obtained from the database in comma separated files. Files were examined for plausibility with Microsoft ExcelTM and analyzed with R (R Development Core Team, 2008) using the packages Epi, epicalc, epiR, geepack, MASS, and stats. After initial univariate analysis, a correlation matrix of variables was examined for multicollinearity. Poisson regression models were fitted predicting the number of deaths with a logarithmic link function and animaldays as offset, initially assuming independent observations (formula (1)). Covariates were introduced manually and stepwise. At first the significant predictors were established. After that, a model was produced to evaluate all possible interactions. Significant interactions were examined manually in a stepwise fashion, guided by AIC values. When significant interactions were detected, interaction terms were introduced as separate variables to examine their significance in further modeling. To test the contribution of covariates and coefficients, Wald’s test was used. Models were compared using likelihood ratio tests and Akaike’s information criterion. Stratum-specific mortality rates were determined according to formula (2) log(Ei ) = ˛ + ˇ1 x1 + ˇ2 x2 + · · · + ˇi xi + log(ti ) log(i ) = log

E  i

ti

= ˛ + ˇ1 x1 + ˇ2 x2 + · · · + ˇi xi

(1) (2)

where Ei = expected number of deaths, ˛ = intercept (baseline rate), ˇi = coefficients, ti = animal-days (offset term), i = mortality rate; all relating to the ith covariate category (xi ). Model evaluation was performed as described by Dohoo et al. (2009). Investigation of Poisson model fit was carried out by Deviance and Pearson residual analysis. The covariates of the final Poisson model with the best fit were used to fit a Poisson generalized estimating equations (gee) model with a log link function. Initially, an independent working correlation matrix was defined to check identity of gee and final Poisson model. Then, an exchangeable working correlation matrix was applied regarding each holding as a cluster. The Huber–White sandwich estimator was used for the variance-covariance structure to obtain robust standard errors. The predictions of both models

were compared to the total national cattle mortality figures of 2011–2013. 2.4. Determination of deviant holdings For each holding, the deviation from the modeled deaths was characterized with the standardized mortality ratio (SMR) and the standardized mortality difference (SMD) as described by Stevenson (2003). Holdings with both SMR >2 and SMD >10 were marked as deviant as suggested by Pannwitz (2013). In addition, the difference of occurred and modeled deaths per holding was expressed in multiples of standard deviations (SD) of a Poisson distribution, where SD equals the square root of the modeled deaths total for that holding. These multiples were ranked and a reference range was defined comprising the lower 95 percentiles of ranks. Holdings in the 95th–100th percentile were regarded deviant. 3. Results 3.1. Descriptive sample statistics Following the removal of 17 implausible records (date or kind of cattle entry unknown), the dataset contained 290,424 records of 286,912 cattle (103,737 male, 183,175 female, cattle sampling fraction 2.3%), 170,416 animalyears (128,415 female, 42,371 male), and 8027 recorded deaths (5411 female, 2616 male). Sampled holding sizes ranged from 0.003 to 14,561 animal-years. The overall mean of animal-years per sampled holding was 87.6. 77% holdings were smallholdings (<100 ay), and 23% large holdings (≥100 ay). Smallholdings contained 19% and large holdings 81% of animal-years. In the official statistic (Destatis, 2013), 76% of holdings kept <100 and the remaining 24% holdings ≥100 cattle. The mean cattle number per German holding was 77.6, ranging per state from 47.9 (Hesse) to 174.1 (Mecklenburg-Western Pomerania). The official statistical herd size figures of all states were within the confidence limits of the stratified samples (data not shown). The official statistical and the sample data are summarized in Table 1. Female Holstein cattle predominated in all states except in Baden-Württemberg and Bavaria with mainly female Fleckvieh. The breed distribution by state was more heterogeneous in males. 46%, 32%, and 22% of all holdings contained one, two, or all three breed categories, respectively. 79% of all sampled individuals belonged to four breeds: Holstein-Friesian (52.6%), Red Holstein (5.0%, both intensive milk breeds), Fleckvieh (13.7%, less intensive milk breed), and mixed meat type (7.9%, beef breed). In total, 62,813 calves were registered born, thereof 27.4% beef and 72.6% milk breeds. Seven to nine percent of milk breed calves were born per month. In contrast, 42% of beef breed calves were born in March, April and May (15%, 14% and 13%, respectively). The crude rates of entry, exit, birth, on-site slaughter for human consumption, and on-site deaths are summarized in Table 2. 1.3% calves of early 2012 were excluded from further analysis as their dams calved again later in 2012. The

G. Pannwitz / Preventive Veterinary Medicine 118 (2015) 260–270

263

Table 1 Cattle and holding numbers in German federal states; and number of holdings, number of animal-years and crude death rates in 1946 sampled cattle holdings in Germany 2012. Federal stateb

Official statistica Number of holdings

SH NI NW HE RP BW BY BB MV SN ST TH DEc

8407 22,935 19,016 9405 5709 18,949 53,787 4442 3128 7370 3109 4187 160,444

Sample Number of cattle 1,127,567 2,562,827 1,421,399 450,407 355,364 995,839 3,251,606 551,293 544,558 498,728 342,421 338,492 12,450,591

Mean cattle number per holding

Number of holdings

Number of animal-years

Mean animal-years per holding

Crude death rate

134.1 111.7 74.7 47.9 62.2 52.6 60.5 124.1 174.1 67.7 110.1 80.8 77.6

156 155 163 166 184 147 156 144 174 173 155 173 1946

18,779 16,184 11,686 7688 11,222 7257 9852 13,297 38,087 8177 13,628 14,560 170,417

120.4 104.4 71.7 46.3 61.0 49.4 63.2 92.3 218.9 47.3 87.9 84.2 87.6

0.047 0.053 0.037 0.037 0.044 0.044 0.041 0.059 0.043 0.052 0.046 0.059 0.047

a

Destatis, 2013. SH = Schleswig Holstein, NI = Lower Saxony, NW = Northrhine Westphalia, HE = Hesse, RP = Rhineland Palatinate, BW = Baden – Wurttemberg, BY = Bavaria, BB = Brandenburg, MV = Mecklenburg – Western Pomerania, SN = Saxony, ST = Saxony Anhalt, TH = Thuringia. c DE = Germany (sampled states). b

remaining 61,973 calves (58,309 singles, 3646 twins, 18 triplets) were born in 60,133 calving events. The majority of calving events (97.3%) occurred in >2-year-old females. 2752 deaths were recorded in these females, 24.4% thereof had calved ≤21 days before. This proportion was not significantly different for dams of different breeds and herd sizes (2 -test, 17 df, p = 0.22). The overall death risk for a dam ≤21 days post-calving was 1.15%. Females >2 years were present in 84.7% of holdings. The calving rates were 0 (i.e. no calving) in 20.6% of holdings, ≤0.5 in 7.6%, 0.51 to ≤0.8 in 27.1%, 0.81 to ≤1 in 33.7%, and >1 in 11.0%. These holdings contained 1.9%, 4.0%, 29.3%, 53.4%, and 11.4% of >2-year-old female animal-years, and 1.3%, 4.3%, 30.9%, 53.6%, and 10.0% of >2-year-old female deaths, respectively. 286 (14.7%) holdings reported a total of 2225 slaughters for human consumption. 1709 thereof occurred in three smallholdings that were regarded as slaughter facilities (Table 2). The mean overall mortality rate was 0.047 per animalyear. Crude sex and age-stratified rates are depicted in Fig. 1. With increasing age, crude mortality rates showed a reverse J-shaped pattern. Cattle aged ≤6 months, 6–24 months, and >2 years contributed 50.0%, 15.4%, and 34.6% of deaths, respectively. Males >2 years (0.83% of animal-years,

0.37% of deaths) showed no significant difference between their age-stratified mortality rates (p = 0.07, Fisher’s test), and as a group also not to males 1–2 years (p = 0.07, Fisher’s test). 54.3% of holdings, representing 8.6% animal-years, reported no deaths. 45.7% holdings with 91.4% of animalyears reported deaths (mean = 9, median = 3, 1–262 deaths per holding). These holdings consisted of 449 small and 441 large holdings, representing 30% and 96% of their holding size class total, respectively. In holdings reporting deaths, crude mortality rates ranged from 3.7 × 10−4 to 2.17 per animal-year (mean = 0.07, median = 0.045). Mean monthly air temperatures showed similar values among states, with a maximum of 4.5 ◦ C difference per month. Records were ≥10 ◦ C May–September and <10 ◦ C otherwise in all states, ranging from −4.52 ◦ C (February, Bavaria) to +18.92 ◦ C (August, RhinelandPalatinate). Temperatures ≥10 ◦ C compared to <10 ◦ C were associated with a lower crude mortality rate ratio (MRR) in calves (≤6 months old, crude MRR 0.74, CI 0.69–0.79), whereas older animals (>6 months to 2 years and >2 years) seemed unaffected (crude MRRs 0.99, CI 0.88–1.1 and 1.03, CI 0.95–1.11, respectively).

Table 2 Crude rates of reported entry and exit events per animal-year, by holding size, based on 1946 sampled cattle holdings in Germany 2012.

Entry Birth Live exit Slaughter Death Missingb a b

<100 ay

≥100 ay

Rate ratio (95% CI)

0.41 0.34 0.67 0.012a 0.03 7.4 × 10−4

0.33 0.37 0.64 0.001 0.05 2.4 × 10−4

0.82 (0.80–0.83) 1.09 (1.07–1.12) 0.96 (0.94–0.97) 0.09 (0.07–0.11) 1.55 (1.45–1.65) 0.32 (0.19–0.57)

377 slaughters, excluding 1709 slaughters recorded in three smallholdings, see text. Unclear, undefined exit.

264

G. Pannwitz / Preventive Veterinary Medicine 118 (2015) 260–270

Fig. 1. Relative distribution of sampled animal-years (ay) by age and sex (bars, left y-axis) with respective crude mortality rates per animal-year (MR, lines, right y-axis) based on 1946 sampled cattle holdings in Germany 2012.

Among federal states, crude age-stratified mortality rates in ≤6 months cattle appeared higher in the North-East compared to the South-West. There was no clear geographical pattern in older cattle.

3.2. Multivariable analyses The Lexis expansion of 1946 holdings resulted in a dataset of 83,816 observations (mean: 43 observations per holding, range 1–178). Males >1 year were summarized in one age stratum. In the Poisson model, moderate overdispersion was detected with the Pearson goodness of fit test but not with the deviance goodness of fit test (dispersion 3.91 and 0.38, respectively). Crude and adjusted mortality rate ratios (MRR) are reported in Table 3. The Poisson and gee models exhibited similar point estimates with wider confidence intervals in the latter. In the Poisson model, the variables age, breed, sex, herdsize, calving rate, air temperature and federal state contributed significantly to the model. In the gee model, no federal state had a significantly different mortality rate. In both models, calves <3 weeks had the highest mortality rates that declined in both sexes until the age of 1–2 years. In females, they rose again thereafter. Age-band mortality rates differed significantly from each other in the first 6 months of life. There was some overlap among older age strata. In both models, males exhibited higher mortality rates than females up to 2 years of age. All three breed categories differed significantly. Smaller herds had lower mortality than larger herds. Significant interactions were detected only between age and breed, and age and sex, i.e. intensive and less intensive milk breed cattle of 6 weeks to 6 months, and intensive milk breed females >4 years, were associated with higher mortality via interaction. Other interactions were either not significant or not defined because of singularities (data not shown). In the gee model, herd calving rates of >0 to <0.5 were associated with lower mortality, but there was no significant difference between no calving and calving rates >0.5. In both models, temperatures ≥10 ◦ C were associated with lower

mortality. Different slaughter rates were not associated with mortality (p ≥ 0.17, data not shown). All-year mortality rates obtained with the gee model excluding the covariates calving rate, temperature, and federal state, are shown in Table 4. Rates range from 0.006 (1–2-year-old female beef cattle in smallholdings) to 0.755 per animal year (male 0–3-week-old intensive milk breed calves in large holdings). In 2012, a total of 590,576 cattle deaths were recorded nationwide. The Poisson model predicted 588,406 and the gee model 596,142 deaths (based on total number of cattle and holdings, respectively). Using the same sample but with different monthly air temperature datasets for 2011 and 2013, the total national cattle deaths of these years were predicted with −4.2% and +0.9% accuracy in both models (Table 5).

3.3. Determination of deviant holdings Using the specified SMR and SMD limits on the sample, 37 holdings (3 small, 34 large) were deviant in the gee model. Applying the 95% reference range method on the sample, 98 holdings (49 small, 49 large) were deviant. These were holdings with recorded deaths >3.76 SD of expected deaths in the gee model and >2.93 SD in the Poisson model. The 95% reference range method used on both models differed in the evaluation of six holdings (kappa coefficient  = 0.97). The SMR and SMD limits used on both models led to a different evaluation of 14 holdings ( = 0.76). In the gee model, the reference range detected all but two holdings found deviant with the SMR and SMD limits ( = 0.50). In the Poisson model, the reference range included all deviant holdings recognized with the SMR and SMD limits ( = 0.37). In case deviant holdings had to be inspected on site, about 8130 (CI 6539–9722) German holdings would have to be visited using reference range method and gee model, thereof 4021 small (CI 2898–5145) and 4210 large holdings (CI 3079–5341). With the SMR & SMD limits, about 3070 (CI 2076–4064) German holdings would have to be

G. Pannwitz / Preventive Veterinary Medicine 118 (2015) 260–270

265

Table 3 Crude and adjusted mortality rate ratios (MRR) in a simple and a gee Poisson model based on 1946 sampled cattle holdings in Germany 2012. Crude MRR

Simple Poisson MRR

p

Gee Poisson MRR

p

Age, baseline 0–3 weeksa 3–6 weeks 6–12 weeks 3–6 months 6–12 months 1–2 years 2–4 yearsb 4–8 yearsb >8 yearsb

0.27 (0.21–0.34) 0.39 (0.35–0.42) 0.18 (0.17–0.20) 0.08 (0.07–0.09) 0.04 (0.04–0.04) 0.02 (0.02–0.02) 0.05 (0.04–0.05) 0.08 (0.07–0.08) 0.09 (0.08–0.10)

0.30 (0.28–0.33) 0.41 (0.38–0.45) 0.13 (0.11–0.16) 0.06 (0.05–0.07) 0.04 (0.04–0.04) 0.02 (0.02–0.02) 0.05 (0.05–0.06) 0.06 (0.06–0.07) 0.09 (0.08–0.11)

***

0.26 (0.22–0.31) 0.39 (0.33–0.47) 0.13 (0.09–0.17) 0.05 (0.04–0.08) 0.04 (0.03–0.05) 0.02 (0.02–0.03) 0.05 (0.04–0.06) 0.07 (0.05–0.08) 0.09 (0.07–0.11)

***

Sex, baseline = female Male

1.46 (1.39–1.53)

1.35 (1.27–1.42)

***

1.39 (1.25–1.55)

***

Breed, baseline = beef Less int. milk Int. milk

1.08 (0.99–1.16) 1.65 (1.55–1.75)

1.25 (1.16–1.37) 1.55 (1.44–1.67)

***

1.24 (1.07–1.44) 1.51 (1.31–1.75)

0.005

Herdsize, baseline ≤100 animal-years ≥100

1.55 (1.45–1.65)

1.31 (1.22–1.40)

***

1.37 (1.19–1.57)

***

Interactions Milkb and 6 weeks–6 months Int. milk and >4 yearsc

1.87 (1.73–2.01) 1.37 (1.29–1.46)

1.60 (1.35–1.89) 1.53 (1.36–1.73)

***

1.63 (1.20–2.22) 1.54 (1.24–1.92)

0.002

Calving rate of females >2 years, baseline = no calving >0–0.5 1.76 (1.52–2.05) 0.51–0.8 2.20 (2.03–2.39) 2.39 (2.21–2.58) 0.81–1 >1 2.12 (1.92–2.34)

1.65 (1.40–1.94) 1.56 (1.41–1.73) 1.33 (1.20–1.46) 0.97 (0.87–1.09)

***

1.57 (1.08–2.27) 1.38 (0.99–1.93) 1.29 (0.91–1.83) 0.85 (0.54–1.33)

0.02 0.06 0.16 0.48

Monthly air temperature, baseline ≤10 ◦ C 0.90 (0.86–0.94) ≥10 ◦ C

0.91 (0.87–0.95)

***

0.88 (0.80–0.96)

***

Federal state, baseline = SH NI NW HE RP BW BY BB MV SN ST TH

1.08 (0.98–1.19) 0.96 (0.86–1.08) 0.92 (0.80–1.05) 1.07 (0.96–1.19) 1.33 (1.16–1.52) 1.21 (1.06–1.38) 1.22 (1.10–1.35) 1.20 (1.10–1.32) 1.16 (1.03–1.31) 1.08 (0.97–1.21) 1.25 (1.13–1.38)

0.11 0.54 0.22 0.25

1.11 (0.84–1.45) 0.83 (0.60–1.16) 0.93 (0.72–1.19) 1.05 (0.81–1.36) 1.32 (0.97–1.80) 1.20 (0.89–1.62) 1.18 (0.79–1.75) 1.13 (0.86–1.49) 1.15 (0.85–1.56) 0.88 (0.58–1.34) 1.21 (0.91–1.63)

0.46 0.28 0.56 0.70 0.07 0.24 0.41 0.38 0.37 0.55 0.19

a b c ***

1.11 (1.01–1.22) 0.79 (0.70–0.88) 0.77 (0.68–0.88) 0.92 (0.83–1.03) 0.94 (0.82–1.06) 0.86 (0.77–0.97) 1.24 (1.12–1.36) 0.91 (0.84–0.99) 1.10 (0.98–1.23) 0.98 (0.88–1.08) 1.25 (1.13–1.37)

*** *** *** *** *** *** *** ***

***

***

*** ***

0.64

***

0.01 *** ***

0.02 0.15 ***

*** *** *** *** *** *** *** ***

***

***

Baseline rate, intercept. Intensive and less intensive. Females only. p < 0.001.

Table 4 Age-stratified point estimate mortality rates per animal-year by sex, breed and herdsize of a Poisson gee model, based on 1946 sampled cattle holdings in Germany 2012.a Herdsize

<100 Animal-years

Breed

Beef

Age\sex

Female

0–3 weeks 0–6 weeks 6–12 weeks 3–6 months 6–12 months 1–2 years 2–4 years 4–8 years >8 years

0.263 0.104 0.033 0.014 0.011 0.006 0.014 0.017 0.024

a

≥100 Animal-years Less int. milk

Int. milk

Male

Female

Male

Female

Male

Female

0.366 0.144 0.046 0.020 0.015 0.008

0.325 0.128 0.067 0.029 0.013 0.007 0.017 0.021 0.029

0.453 0.179 0.093 0.041 0.018 0.010

0.397 0.157 0.081 0.036 0.016 0.008 0.021 0.040 0.056

0.553 0.218 0.114 0.049 0.022 0.012

0.359 0.142 0.045 0.020 0.014 0.008 0.019 0.024 0.032

Including both interaction terms but no other covariates.

Beef

Less int. milk

Int. milk

Male

Female

Male

Female

Male

0.500 0.197 0.063 0.027 0.020 0.011

0.444 0.175 0.091 0.039 0.018 0.009 0.023 0.029 0.040

0.618 0.244 0.127 0.055 0.025 0.013

0.542 0.214 0.111 0.049 0.022 0.011 0.028 0.056 0.076

0.755 0.298 0.155 0.067 0.030 0.016

266

G. Pannwitz / Preventive Veterinary Medicine 118 (2015) 260–270

Table 5 Number of occurred dead cattle, extracted from the database HI-Tier, in Germany 2011–2013, and predicted dead cattle based on a Poisson and a gee Poisson model based on 1946 sampled cattle holdings in Germany 2012. Year

Number of occurred deaths

2011 2012 2013

617,033 590,576 601,319

Table 6 Percentage of deviant holdings detected with the reference range method and with SMR (standardize mortality ratio) and SMD (standardized mortality difference) limits, by herdsize, using a gee model on 1946 sampled cattle holdings in Germany 2012. Herdsize

Reference range, in % (CI)

SMR and SMD limits,a in % (CI)

<100 ay ≥100 ay Total

3.3 (2.4–4.2) 10.8 (7.9–13.7) 5.0 (4.1–6.0)

2 × 10−3 (0–4 × 10−3 ) 7.5 (5.0–9.9) 1.9 (1.3–2.5)

a

SMR > 2 and SMD >10 concurrently.

regarded as deviant using the gee model, thereof 80 small (CI 0–168) and 2921 large holdings (CI 1962–3880, Table 6). Fig. 2 shows the SMR of all sampled holdings alongside the SMD of holdings outside the 95% reference range.

Number of predicted deaths Poisson model

Gee model

590,909 (95.8%) 588,406 (99.6%) 595,209 (99.0%)

621,464 (100.7%) 596,142 (100.9%) 580,042 (96.5%)

4. Discussion This study adds information to previous studies that have examined the mortality in cattle of different age, breed, sex and purpose of animal (for example, reviewed for dairy cattle by Thomsen and Houe, 2006). However, despite the existence of comprehensive cattle databases in many countries, representative overviews over mortality rates within a cattle population of a country or region are few. This is both in contrast to the reported need for cattle demographic data and to public concerns regarding farm animal welfare (Thomsen and Houe, 2006; Maher et al., 2008; von Keyserlingk et al., 2009; EFSA, 2009, 2011, 2012). The first aim of this study was to analyze the 2012 cattle mortality in Germany. To do this, a clear case definition of on-site death was necessary. The examined data did not

Fig. 2. Standardized mortality ratios (SMR), standardized mortality differences (SMD) by holding size, determined by a gee model based on 1946 sampled holdings in Germany in 2012.

G. Pannwitz / Preventive Veterinary Medicine 118 (2015) 260–270

allow a reliable differentiation of unattended death and euthanasia. In cattle slaughtered on site, it was impossible to distinguish normal from emergency slaughter, and if carcasses were condemned at meat inspection. Therefore, cattle experiencing death but not slaughter were regarded as the best measure of priori unwanted outcomes and considered the most reliable case definition. Cattle holdings are diverse, open and dynamic cohorts including dairy farms, heifer or veal calf production, intensive or extensive beef units, and any combination of these. Therefore, calculating rates is preferable over risks provided the animal-time of observed individuals can be accurately determined. This is no problem with national cattle register data. Certain management or reporting habits may have introduced some bias into the data, e.g. concerning stillborn calves, calves born and dead within the first week of life, (emergency) slaughter on site for human consumption, missing cattle, and cattle dying at transport or abattoir. Other sources of bias may have occurred with lacking access to emergency slaughter facilities or management practices prohibiting slaughter on site. However, registered database deaths are constantly cross-checked with rendering plant data, and implausible records are reported to keeper and competent authority. Thus, data quality was regarded sufficient to describe the status quo. The covariates current age, sex, breed, and holding size had the most influence on mortality. Other covariates (temperature, calving rate) were found of less significance when within-herd clustering was also considered. Initially, mortality levels were differentiated by federal state, as crude age-stratified analyses suggested regional differences. However, it was found that breed, sex, and holding size confounded crude age-stratified state mortality rates rendering differences among states statistically insignificant. It was therefore feasible to model cattle mortality for all German cattle holdings regardless of location. This simplified the statistical analysis for the second study objective, i.e. to outline a standardized approach for examining mortality of national register data. The regional indifference appears to contrast previous studies that have established divergent cattle mortality levels in various parts of a country. For instance, separate regional dairy cow mortality rates have been described for Sweden (Alvasen et al., 2012), Italy (Crescio et al., 2010), Denmark (Thomsen et al., 2004), France (Raboisson et al., 2011), US (McConnel et al., 2008) and for dairy calves in Norway (Gulliksen et al., 2009). However, these studies focused primarily on adult dairy cattle and did not concurrently adjust for age, breed, sex, holding size, and intra-herd clustering throughout. When considering herdlevel, regional, or national mortality, any divergence may best be tested with a standardized approach, as previously proposed by Thomsen and Houe (2006), in order to make results more comparable. The stratification by nine current age bands and sex as used here has previously been proposed in a small case series (Pannwitz, 2013). A comparable stratification by five current age bands and two breeds was used on a regional level by Perrin et al. (2010, 2012). In the present study, the stratification by age, breed, sex, and holding size was found

267

reasonably robust to model expected deaths per holding and also to estimate national herd-level mortality rates. A further subdivision of the first age band into 0–7 days and 8–21 days would be possible, as rates may differ. For example, beef calf mortality risks of 6.4% in the first week and 2.8% between 1 week and weaning have been reported (Azzam et al., 1993). However, in the author’s experience, holdings prefer to report births and movements once a week although some exclude calves being born and dead within 7 days of age. This is legally possible in Germany but introduces bias in the very first age stratum depending on reporting habits. The first stratum was set to end at 21 days because <14 day old calves may not legally be transported in Germany and (male) dairy calves are usually traded once a week. The other age strata were so divided to be easily related to certain subunits or management practices. Tarres et al. (2005) reported no significant mortality differences in beef calves between 33 and 180 days, which could not be confirmed in this dataset. Nevertheless, some age strata may be summarized if desired but, in the author’s experience, the used strata are easily related to any cattle farm and accepted by farmers. To enable valid analyses in various holding structures, this standardized stratification is suggested. The levels of age-stratified cattle mortality are comparable to previous reports (Roy, 1990; Gardner et al., 1990; Karuppanan et al., 1997; Gulliksen et al., 2009; Thomsen et al., 2004, 2006; Ortiz-Pelaez et al., 2008; McConnel et al., 2008; Maher et al., 2008; Perrin et al., 2010; Dechow and Goodling, 2008; Dechow et al., 2011; among others). For example, dairy cow mortality risks of 0.02–0.035 in Denmark 1990–1999 (Thomsen et al., 2004), 0.048 in North America 2002 (McConnel et al., 2008), mortality rates of 0.051–0.066 in Sweden 2002–2010 (Alvasen et al., 2012) have been reported. Despite Harris, 1989 reporting no age-dependence of mortality in dairy cows, the age-based approach used here accounts for parity, which is correlated with age. Increasing parity is associated with higher mortality in dairy cows, a pattern also found in this dataset. For example, Thomsen et al. (2004) report a roughly doubled mortality in ≥3 parous vs. younger Danish dairy cows. This is comparable to this study, where intensive milk breed (i.e. Holstein) cows aged 4–8 years and 2–4 years exhibit mortality rate ratios of about two in both large- and smallholdings. About 24% of all cow deaths are registered ≤21 days post-calving, regardless of breed. This is comparable to data from Denmark (30–40% cows deaths ≤30 days of lactation, Thomsen et al., 2004) and the US (Dechow and Goodling, 2008). The overall death risk for a dam ≤21 days postcalving is 1.15%, similar to 1–1.5% deducted from data of Thomsen et al. (2004). An increased mortality of males ≤2 years compared to females was established, as previously reported for calves by Roy (1990). The calculated MRR for males of 1.39 (CI 1.25–1.55) is similar to the male veal calf mortality odds ratio in Swiss holdings of 2.7 (CI 1.5–4.9, Bähler et al., 2012). Their higher figure may be due to cohort differences or that mortality included unwanted slaughter. In general, the higher male mortality seems to mirror risk factors of biology and market value.

268

G. Pannwitz / Preventive Veterinary Medicine 118 (2015) 260–270

The applied breed categories are similar to those used by Alvasen et al. (2012), although an individual rather than herd-level classification was applied. This avoids calculating the predominant breed per herd (based on individual numbers, animal-time, or otherwise), and seems to best represent holdings with various breeds. Still, Holsteins as intensive milk breed have repeatedly exhibited higher mortality rates (e.g. Thomsen et al., 2006; Raboisson et al., 2011; Alvasen et al., 2012). This was also found here, reflecting both the high metabolic and management demands of intensive dairy cattle. Herds ≥100 animal-years are associated with increased mortality (adj. MRR 1.37, CI 1.19–1.57) again confirming previous studies (e.g. Roy, 1990; McConnel et al., 2008; Alvasen et al., 2012). Gulliksen et al. (2009) found a similar odds ratio (1.81, CI 1.44–2.27) in herds >50 cow-years. The increased mortality in larger herds has been attributed to factors like less time available for individual animals and less pasture grazing, among others (Gulliksen et al., 2009; Thomsen and Houe, 2006). In addition, management decisions and slaughter facility access may also play a role. From the author’s experience, large holdings may prefer euthanasia to on site (emergency) slaughter for human consumption but may use approved emergency slaughter facilities if available. Nevertheless, although on-site slaughter was more common in smallholdings (crude slaughter rate ratio 11.1, CI 9.1–14.3), it was not significantly associated with herd-level mortality. This study examined the effect of mean monthly air temperatures within each federal state. Two categories, i.e. temperature ≥10 vs. <10 ◦ C (alias May–September vs. October–April) were examined, with former rates lower than the latter (adj. MRR 0.88, CI 0.80–0.96). This is comparable to Azzam et al. (1993) reporting an increased beef calve mortality at <10 ◦ C combined with precipitation. In contrast to this study, Vitali et al. (2009), Crescio et al. (2010), and Alvasen et al. (2012) describe increased dairy cow mortality in the summer, depending on air temperature, humidity and region. However, as the total herd-level mortality was modeled in this study, calves contributed 50% of deaths. Temperature effects on calves may have dominated and/or the summer temperatures of 2012 in Germany may have not been high enough for significant effects. Calving rates >0 and <0.5 per animal-year were associated with higher cow mortality in the gee model. This is in concordance with Alvasen et al. (2012) reporting calving intervals >392 and >492 days to be associated with an increased mortality odds ratio of 1.1 and 1.29, respectively. Multivariable methods are appropriate for cattle register analysis, as carried out by Alvasen et al. (2012), Thomsen et al. (2006, 2007), McConnel et al. (2008), among others. In this study, a Poisson regression model was used, similar to Perrin et al. (2010), but predicting the number of expected deaths per holding. Thereby, the determination of deviant mortality in any holding, but also at regional level is possible. A main assumption of Poisson models is a homogenous event rate for different subjects in a stratum. It was accounted for by the described risk factor stratification. Still, the assumption may not necessarily hold, as certain

exits from a holding may concern more fragile animals (e.g. culled dairy cows, Pannwitz, 2013), or, in other circumstances, fitter individuals may be sold and others kept longer. Further Poisson process assumptions are rare and independent outcomes lacking overdispersion. These may be regarded as sufficiently met with dispersion values between one and three (Anderson et al., 1994). Still, the moderate overdispersion detected here may indicate that relevant explanatory variables were missing in the examined data, including factors associated with of feeding, herd management, disease, and others. Further, a moderate overdispersion is consistent with the intra-herd mortality clustering as detected via the widened confidence intervals in the gee-Poisson model. Mortality clustering has been described in calf and dairy cattle holdings (Moore et al., 2002; Bähler et al., 2012; Shahid, 2013). This corresponds to the view that mortality levels in holdings are multifactorial and often management-dependent (Welfare Quality, 2009; de Vries et al., 2011, 2013). In consequence, a substantially increased mortality in a holding makes violated model assumptions likely. These may include welfare problems, insufficient health care or feeding practices, and infectious diseases like Bluetongue, FMD, and BSE (de Vries et al., 2013; Stevenson, 2003; Perrin et al., 2010). These and any other possibilities remain to be analyzed when examining mortality in a concrete deviant holding. To detect deviant holdings, Pannwitz (2013) has suggested the concurrent use of SMR >2 and SMD >10. This approach reflects the fact that both an increased mortality rate and a minimum number of excess deaths need to be present to be considered relevant. In this dataset, these limits were not sensitive enough, although SMR and SMD limits may be adjusted relative to holding size. However, the high agreement of the reference range method in both models makes this method preferable for the detection of deviant holdings. The difference of occurred deaths to modeled deaths was expressed in terms of multiples of standard deviations (SD) of a Poisson distribution. Other distributions, like the Gamma distribution, may also be used (Dohoo et al., 2009). In any case, both models appeared to predict the total German deaths in 2011 and 2013, indicating that both can now be validated in the field. The mortality level is a useful health indicator in community medicine, animal health, and animal welfare (Broom, 1991, 1993; Hemsworth et al., 1995; Mellor and Stafford, 2004; European Commission, 2008). In the EU, standardized human mortality levels in member state regions are consistently analyzed. Results are freely available (European Commission, 2008). Similar cattle mortality analyses are currently not performed, despite their importance as welfare and health indicator has repeatedly been stressed (Laster and Gregory, 1973; Mellor and Stafford, 2004; Rushen et al., 2008; von Keyserlingk et al., 2009; Welfare Quality, 2009; de Vries et al., 2011, 2013; Kelly et al., 2013). This is in contrast to the usefulness of this measure to relevant stakeholders and the existence of available data. Still, a current proposal on European animal health law envisions the role of computerized databases to be widened in the future (Anon., 2013c). The predictive value of increased mortality to detect welfare problems has been examined to decide which

G. Pannwitz / Preventive Veterinary Medicine 118 (2015) 260–270

holding should be inspected (Nyman et al., 2011; OrtizPelaez et al., 2008; EFSA, 2011). Ortiz-Pelaez et al. (2008) find positive and negative predictive values of 26.9% and 65.4% when holdings were (blindly) inspected 18–24 months after mortality measurement. Nyman et al. (2011) reported a specificity and sensitivity of 56% and 96%, respectively. This indicates that increased mortality is a more sensitive than specific welfare measure, but aiding in shortlisting holdings for on-site inspection. An increased mortality, as compared to the national herd, may initially be reported to and managed by the holder, who may decide to change routines, to increase manpower, to seek veterinary support, etc. Unduly high death counts may then be officially followed up as appropriate. In consultation with stakeholders, practical model capabilities need to be further evaluated after their implementation into the national database. In conclusion, the standardized mortality analysis offers a useful, inexpensive and easy implementable tool in the detection of holdings deviant from mortality levels of the national herd. Acknowledgements The author wishes to thank Mario Ziller, Federal Research Institute for Animal Health, and two anonymous reviewers for helpful questions and comments leading to an improved paper. References Alvasen, K., Jansson Mörk, M., Sandgren, H.C., Thomsen, P.T., Emanuelson, U., 2012. Herd-level risk factors associated with cow mortality in Swedish dairy herds. J. Dairy Sci. 95, 4352–4362. Anderson, D.R., Burnham, K.P., White, G.C., 1994. AIC model selection in overdispersed capture–recapture data. Ecology 75, 1780–1793. Anon, 2013a. Verordnung über hygienische Anforderungen beim Halten von Schweinen, www.gesetze-im-internet.de Anon, 2013b. Verordnung zum Schutz gegen die Geflügelpest, www. gesetze-im-internet.de Anon, 2013c. Proposal for a Regulation of the European Parliament and the Council on Animal Health, http://ec.europa.eu/prelex/detail dossier real.cfm?CL=en&DosId=202630 (accessed 26.09.13). Azzam, S.M., Kinder, J.E., Nielsen, M.K., Werth, L.A., Gregory, K.E., Cundiff, L.V., Koch, R.M., 1993. Environmental effects on neonatal mortality of beef calves. J. Anim. Sci. 71, 282–290. Bähler, C., Steiner, A., Luginbühl, A., Ewy, A., Posthaus, H., Strabel, D., Kaufmann, T., Regula, G., 2012. Risk factors for death and unwanted early slaughter in Swiss veal calves kept at a specific animal welfare standard. Res. Vet. Sci. 92, 162–168. Broom, D.M., 1991. Animal welfare: concepts and measurement. J. Anim. Sci. 69, 4167–4175. Broom, D.M., 1993. Assessing the welfare of modified or treated animals. Livest. Prod. Sci. 36, 39–54. Crescio, M.I., Forastiere, F., Maurella, C., Ingravalle, F., Ru, G., 2010. Heatrelated mortality in dairy cattle: a case crossover study. Prev. Vet. Med. 97, 191–197. de Vries, M., Bokkers, E.A.M., Dijkstra, T., van Schaik, G., de Boer, I.J.M., 2011. Invited review: associations between variables of routine herd data and dairy cattle welfare indicators. J. Dairy Sci. 94, 3213–3228. de Vries, M., Bokkers, E.A.M., van Schaik, G., Botreau, R., Engel, B., Dijkstra, T., de Boer, I.J.M., 2013. Evaluating results of the welfare quality multicriteria evaluation model for classification of dairy cattle welfare at the herd level. J. Dairy Sci. 96, 6264–6273. Dechow, C.D., Goodling, R.C., 2008. Mortality, culling by sixty days in milk, and production profiles in high- and low survival pennsylvania herds. J. Dairy Sci. 91, 4630–4639. Dechow, C.D., Smith, E.A., Goodling, R.C., 2011. The effect of management system on mortality and other welfare indicators in Pennsylvania dairy herds. Anim. Welf. 20, 145–158.

269

Dohoo, I., Martin, W., Stryhn, H., 2009. Modelling count and rate data. In: Veterinary Epidemiologic Research, 2nd ed. VER Inc., Charlottetown, Canada, pp. 445–466. Destatis, 2013. Viehbestand am 3. November 2012. Federal Bureau of statistics – Statistisches Bundesamt, Fachserie 3, Reihe 4.1, Land- und Forstwirtschaft, www.destatis.de EFSA, 2009. Scientific opinion of the panel on animal health and welfare on a request from the European Commission on the overall effects of farming systems on dairy cow welfare and disease. EFSA J. 1143, 1–38. EFSA, 2011. Review of methodologies applicable to the validation of animal based indicators of welfare. SCIENTIFIC REPORT submitted to EFSA by Presi, P., Reist, M. Project report CT/EFSA/AHAW/2010/04. EFSA, 2012. Guidance on risk assessment for animal welfare. EFSA Scientific opinion. EFSA J. 2513, 1–29. European Commission, 2008. Health Statistics-Atlas on Mortality in the European Union. Office for Official Publications of the European Communities, pp. 212, ISBN 978-92-79-08763-9. Eurostat, 2013. Agriculture, Forestry and Fishery Statistics. Chapter 2: Eurostat Pocketbooks, 2013 ed., pp. 249, ISBN 978-92-79-33005-6. Gardner, I.A., Hird, D.W., Utterback, W.W., Danaye-Elmi, C., Heron, B.R., Christiansen, K.H., Sischo, W.M., 1990. Mortality, morbidity, casefatality, and culling rates for California dairy cattle as evaluated by the National Animal Health Monitoring System, 1986–87. Prev. Vet. Med. 8, 157–170. Gulliksen, S.M., Lie, K.I., Loken, T., Osteras, O., 2009. Calf mortality in Norwegian dairy herds. J. Dairy Sci. 92, 2782–2795. Harris, B.L., 1989. New Zealand dairy cow removal reasons and survival rate. N. Z. J. Agric. Res. 32, 355–358. Hemsworth, P.H., Barnett, J.L., Beveridge, L., Matthews, L.R., 1995. The welfare of extensively managed dairy cattle: a review. Appl. Anim. Behav. Sci. 42, 161–182. Hi-Tier, 2013. Herkunftssicherungs- und Informationssystem für Tiere – HI-Tier, www.hi-tier.de Karuppanan, P., Thurmond, M.C., Gardner, I.A., 1997. Survivorship approaches to measuring and comparing cull rates for dairies. Prev. Vet. Med. 30, 171–179. Kelly, P.C., More, S.J., Blake, M., Higgins, I., Clegg, T.A., Hanlon, A.J., 2013. Validation of key indicators in cattle farms at high risk of animal welfare problems: a qualitative case-control study. Vet. Rec. 172, 314. Laster, D.B., Gregory, K.E., 1973. Factors influencing peri- and early postnatal calf mortality. J. Anim. Sci. 37, 1092–1097. Maher, P., Good, M., More, S.J., 2008. Trends in cow numbers and culling rate in the Irish cattle population, 2003 to 2006. Irish Vet. J. 61, 455–463. McConnel, C.S., Lombard, J.E., Wagner, B.A., Garry, F.B., 2008. Evaluation of factors associated with increased dairy cow mortality on United States dairy operations. J. Dairy Sci. 91, 1423–1432. Mellor, D.J., Stafford, K.J., 2004. Animal welfare implications of neonatal mortality and morbidity in farm animals. Vet. J. 168, 118–133. Moore, D.A., Sischo, W.M., Festa, D.M., Reynolds, J.P., Atwill, E.R., Holmberg, C.A., 2002. Influence of arrival weight, season, and calf supplier on survival in Holstein beef calves on a calf ranch in California, USA. Prev. Vet. Med. 53, 103–115. Nyman, A.K., Lindberg, A., Sandgren, C.H., 2011. Can pre-collected register data be used to identify herds with good cattle welfare? Acta Vet. Scand. 53 (Suppl. 1 S8), 1–6. Ortiz-Pelaez, A., Pritchard, D.G., Pfeiffer, D.U., Jones, E., Honeyman, P., Mawdsley, J.J., 2008. Calf mortality as a welfare indicator on British cattle farms. Vet. J. 176, 177–181. Pannwitz, G., 2013. Standardisierte Mortalitätsraten und andere Parameter zur Überwachung von Rinderbeständen. Amtstierärztlicher Dienst und Lebensmittelkontrolle 20, 1–8. Perrin, J.P., Ducrot, C., Vinard, J.L., Morignat, E., Gauffier, A., Calavas, D., Hendrikx, P., 2010. Using the National Cattle Register to estimate the excess mortality during an epidemic: application to an outbreak of Bluetongue serotype 8 epidemics. Epidemics 2, 207–214. Perrin, J.P., Ducrot, C., Vinard, J.L., Morignat, E., Calavas, D., Hendrikx, P., 2012. Assessment of the utility of routinely collected cattle census and disposal data for syndromic surveillance. Prev. Vet Med. 105, 244–252. R Development Core Team, 2008. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria, ISBN 3-900051-07-0. http://www.R-project.org Raboisson, D., Cahuzac, E., Sans, P., Allaire, G., 2011. Herd-level and contextual factors influencing dairy cow mortality in France in 2005 and 2006. J. Dairy Sci. 94, 1790–1803. Regulation (EC) No. 1760/2000, 2000. EC Off. J. L 204, 1–10 (as amended). Regulation (EC) No. 73/2009, 2009. EC Off. J. L 30, 16–99 (as amended). Roy, J.H.B., 1990. The Calf. Management of Health, vol. 1. Butterworths, Basingstoke, Kent, UK, pp. 1–16.

270

G. Pannwitz / Preventive Veterinary Medicine 118 (2015) 260–270

Rushen, J., de Passillé, A.M., Keyserlingk, v.M.A.G., Weary, D.M., 2008. Health, disease and productivity. Chapter 2. In: The Welfare of Cattle. Animal Welfare, vol. 5. Springer, Berlin, ISBN 978-1-4020-6557-6. Scheaffer, R.L., Mendenhall, W.M., Ott, R.L., 2006. Elementary survey sampling. In: Cluster Sampling, 6th ed. Thomson Brooks, Cole, pp. 265–302 (Chapter 8). Shahid, M.Q., Dissertation 2013. Cow Mortality in Midwest Dairy Herds. University of Minnesota. Stevenson, M.A., (Ph.D. thesis) 2003. The Spatio-Temporal Epidemiology of Bovine Spongiform Encelopathy and Foot-and-Mouth Disease in Great Britain. Massey University, Institute of Veterinary, Animal and Biomedical Sciences, Palmerston North, New Zealand. Tarres, J., Casellas, J., Piedrafita, J., 2005. Genetic and environmental factors influencing mortality up to weaning of Bruna dels Pirineus beef calves in mountain areas. A survival analysis. J. Anim. Sci. 83, 543–551. Thomsen, P.T., Houe, H., 2006. Dairy cow mortality. A review. Vet. Quart. 28, 122–129. Thomsen, P.T., Kjeldsen, A.M., Sorensen, J.T., Houe, H., 2004. Mortality (including euthanasia) among Danish dairy cows (1990–2001). Prev. Vet. Med. 62, 19–33.

Thomsen, P.T., Kjeldsen, A.M., Sorensen, J.T., Houe, H., Ersboll, A.K., 2006. Herd-level risk factors for the mortality of cows in Danish dairy herds. Vet. Rec. 158, 622–626. Thomsen, P.T., Ostergaard, S., Houe, H., Sorensen, J.T., 2007. Loser cows in Danish dairy herds: risk factors. Prev. Vet. Med. 79, 136–154. Vitali, A., Segnalini, M., Bertocchi, L., Bernabucci, U., Nardone, A., Lacetera, N., 2009. Seasonal pattern of mortality and relationships between mortality and temperature–humidity index in dairy cows. J. Dairy Sci. 92, 3781–3790. von Keyserlingk, M.A.G., Rushen, J., de Passille, A.M., Weary, D.M., 2009. Invited review: the welfare of dairy cattle—key concepts and the role of science. J. Dairy Sci. 92, 4101–4111. Welfare Quality® 2009. Welfare quality® assessment protocol for cattle. Welfare Quality® consortium, Lelystad, the Netherlands. Zepeda, C., Salman, M.D., 2003. Planning survey, surveillance, and monitoring systems – roles and requirements. In: Salmon, M.D. (Ed.), Animal Disease Surveillance and Survey Systems, Methods and Applications. , 1st ed. Iowa State Press (Chapter 3).