- Email: [email protected]
Culling of dairy cows. Part II. Effects of diseases and reproductive performance on culling in Finnish Ayrshire cows
Preventive Veterinary Medicine 41 (1999) 279±294
Culling of dairy cows. Part II. Effects of diseases and reproductive performance on culling in Finnish Ayrshire cows P.J. Rajala-Schultza,b,*, Y.T. GroÈhna a
Department of Population Medicine and Diagnostic Sciences, Section of Epidemiology, College of Veterinary Medicine, Cornell University, Ithaca, NY 14853, USA b Department of Clinical Veterinary Sciences, Faculty of Veterinary Medicine, PO Box 57, University of Helsinki, Helsinki, FIN-00014, Finland Accepted 9 March 1999
Abstract The effects of 15 diseases and reproductive performance on culling were studied in 39 727 Finnish Ayrshire cows that calved in 1993 and were followed until culling or next calving. Survival analysis, using the Cox proportional hazards model, was performed with diseases and pregnancy status as time-dependent covariates. Parity, calving season and herd were included as covariates in every model. The effect of the number of inseminations was also studied. The farmer's knowledge of the cow's pregnancy status had a significant effect on culling. It varied according to the stage of lactation a cow was in; the earlier the farmer knew a cow was pregnant, the smaller was the risk of culling. If a cow had not been inseminated at all, her risk of culling was 10 times higher than if she was inseminated once. If a cow was inseminated more than once, she had a slightly lower risk of being culled than a cow inseminated only once. The effect of parity decreased when pregnancy status and number of inseminations were added to the model, indicating that part of the parity effect was accounted for by reproductive performance. Including diseases in the model with pregnancy status and the number of inseminations did not change the effects of reproductive performance on culling. Mastitis, teat injuries and lameness had the greatest effect on culling (whether adjusted for reproductive performance or not), increasing the risk of culling, followed by anestrus, ovarian cysts and milk fever. In general, the effects of diseases decreased when reproductive performance was also accounted for in the model. When pregnancy status was included in the model, the effects of anestrus and ovarian cysts became slightly more protective, but when the number of inseminations * Corresponding author. Tel.: +1-607-253-3572; fax: +1-607-253-3083 E-mail address: [email protected] (P.J. Rajala-Schultz) 0167-5877/99/$ ± see front matter # 1999 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 7 - 5 8 7 7 ( 9 9 ) 0 0 0 4 5 - 8
280
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
was also considered, they became non-significant at the beginning of lactation and they increased the risk of culling at the end of lactation. Sensitivity analysis, which was run to evaluate the effects of our censoring mechanism on the results, indicated that the censoring times (i.e., the time of next calving) were not fully independent of the event (culling) times; the effects of the diseases and pregnancy status at the very end of the lactation changed slightly from the original model. # 1999 Elsevier Science B.V. All rights reserved. Keywords: Cattle-management; Reproductive performance; Culling; Survival analysis; Time-dependent covariates
1. Introduction Milk is the major source of income in a dairy farm. Because cows produce milk only after calving and cows' daily milk yield decreases as time passes after parturition, the farmer wants his/her cows to conceive within a restricted period of time after calving and to calve at regular intervals. If conception is delayed, this reproductive inefficiency can also lead to milk production inefficiency. If a cow is not pregnant (is open) for an extended period after calving, at the end of the lactation the farmer can either dry her off and allow a long dry period or he/she can continue to milk, albeit with reduced daily milk production. Neither of the above mentioned options are economically most desirable. Beaudeau et al. (1995) showed that reproductive performance had a high impact on the length of productive life of dairy cows, i.e. the longer the days open the higher the risk of culling. GroÈhn et al. (1998) showed that once a cow had conceived, her risk of culling dropped sharply. Harman et al. (1996), on the other hand, showed that several diseases have an effect on fertility of dairy cows. Concurrent diseases can affect fertility of dairy cows and can cause delayed conception and they could therefore be an indirect reason for culling. Earlier we studied the effects of 15 diseases on culling per se, without adjusting for the pregnancy status of a cow (Part I in this series, Rajala-Schultz and GroÈhn, 1999). The purpose of this study was to evaluate the effects of diseases and reproductive performance on culling decisions and to see how the effects of diseases on culling change when adjusted for pregnancy status and the number of times a cow is inseminated. 2. Materials and methods 2.1. Data The data consisted of 39 727 Finnish Ayrshire dairy cows that calved during 1993 and were followed until the next calving or culling. The cows were in 2338 herds that belonged to the milk registry and the national health recording system. All herds used artificial insemination. The data have been described earlier (Rajala and GroÈhn, 1998; Part I in this series, Rajala-Schultz and GroÈhn, 1999). Data on insemination dates were
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
281
also available. In this analysis, insemination data were divided into five levels: cows that had not been inseminated at all, those inseminated once, twice, thrice, and more than three times. The stages of lactation for disease occurrence and culling were the same as in Part I, (Table 1, Rajala-Schultz and GroÈhn, 1999). The farmer's knowledge of the pregnancy status of the cow may have a different effect on his/her culling decisions depending on the stage of lactation the cow is in. The lactation was divided into four stages (0±150 d, 151±240 d, 241±305 d and >305 d) with respect to when pregnancy status could affect culling decisions. The pregnancy status was considered to change its value at the time when the farmer could have reasonably known whether a cow had conceived or not after the last insemination and when this information, therefore, might have influenced the farmer's culling decision. This was assumed to be 63 d after last insemination, because in Finland artificial insemination (AI) technicians do most of the pregnancy checks and they often do not do them until 9 weeks after the insemination. The time when the pregnancy status changed its value was calculated in six different ways: (1) For cows that had a subsequent calving (26 162 cows), the time when their Table 1 Comparison of partial likelihoods (ÿ2log L) for different diseases in the different original modelsa Covariate
Basic model Mastitis Lameness Teat injury Milk fever Cysts Anestrus DAc Hardware NP paresisd Rumene Hypomgf Dystocia Metritis Ketosis Ret. plac.g
Model 1
Model 2
Model 3
Change in ÿ2log L
Change in ÿ2log L
Change in ÿ2log L
b
240 715 500*** 264*** 181*** 108*** 95*** 90*** 89*** 71*** 69*** 59*** 57*** 38*** 36*** 25** 7
b
231 281 432*** 226*** 192*** 115*** 164*** 171*** 89*** 73*** 74*** 54*** 50*** 27*** 22*** 20* 2
221 754b 340*** 135*** 153*** 122*** 91*** 100*** 68*** 70*** 90*** 54*** 49*** 26*** 27*** 21* 2
Change in dfh
15 15 15 5 11 11 9 14 14 15 15 5 8 9 5
a Model 1: parity, season and herd; Model 2: parity, season, herd and pregnancy status; and Model 3: parity, season, herd, pregnancy status and number of inseminations. b ÿ2log partial likelihood for the basic model against which the change in ÿ2log L was calculated. c Displaced abomasum. d Non-parturient paresis. e Rumen disorders. f Hypomagnesemia. g Retained placenta. h Degrees of freedom. * P-value<0.05; ** P-value<0.01; *** P-value<0.001.
282
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
pregnancy status might have influenced the farmer's culling decision was calculated by subtracting 279 d (i.e. average gestation length in these data) from the subsequent calving date and adding 63 d. (2) If the cow was culled within 150 d after calving (3996 cows), she was considered to have been open all that time (or more precisely, it was assumed that the farmer considered her to be open and that pregnancy status was not yet an issue in the culling decision at that time). (3) If she was culled later than 150 d after calving and the farmer-stated reason for culling was ``non-pregnant'' (2540 cows), she was also considered to have been open all the time. (4) If she was culled later than 150 d after calving without the farmer-stated reason ``non-pregnant'', and she did not have any insemination dates (2198 cows), she was considered open all that time. (5) If she was culled later than 150 d after calving without the farmer-stated reason ``non-pregnant'' and she had been inseminated and the interval between the last insemination and culling was longer than 279 d, then she was considered to have been open all that time (163 cows). (6) If she was culled later than 150 d after calving without the farmer-stated reason ``non-pregnant'', she had been inseminated, and the interval between the last insemination and culling was less than 279 d, the time of change in her pregnancy status (or more precisely, the time of change in farmer's knowledge concerning her pregnancy status) was calculated in the following way: last insemination date plus 63 d (3169 cows) (i.e. these cows were assumed to have conceived at the last insemination). Because the last category (6) was based on an assumption, not verified pregnancy status, the effect on culling assuming that all these cows were open was also investigated. 2.2. Statistical analysis Survival analysis, using the Cox proportional hazards model (Cox, 1972), was performed to study the effect of disease and reproductive performance on culling. The statistical method has been described in Part I of this series (Rajala-Schultz and GroÈhn, 1999). The analyses were conducted using the ``Survival Kit'', a set of FORTRAN programs (Ducrocq and SoÈlkner, 1994). Diseases and pregnancy status were modeled as timedependent covariates. 2.3. Models Three basic models were compared; parity, calving season and herd were covariates in all models. Therefore, when referring to any model later on in the paper, it should be assumed that those three covariates are in the model, too. Model 1 included parity, calving season and herd and one disease at a time (15 models, one for each disease) (see Part I of the series). Model 2 also included pregnancy status and Model 3 included both pregnancy status and the number of inseminations. Fifteen models (one for each disease) were also fitted for the basic Models 2 and 3. It was of interest to see how the effects of diseases changed when also adjusted for pregnancy status and for both pregnancy status and the number of inseminations as compared to when they were only adjusted for parity, calving season and herd.
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
283
2.4. Sensitivity analysis A typical assumption made in standard methods of survival analysis is that the censoring times and the event times are independent. In other words, censoring carries no prognostic information about the subsequent event experience, i.e., those who are censored are at no higher or lower risk of experiencing the event of interest than the rest of the individuals in the sample. So, censoring is assumed to be independent and noninformative (Marubini and Valseechi, 1995). In order to get an idea of the possible impact of potential informative censoring on the results, a sensitivity analysis can be performed. The basic idea is to re-do the analysis under two extreme assumptions about the censored cases. One assumption is that censored observations experience an event immediately after they are censored, and the other assumption is that censored cases have longer times to events than any other subjects in the sample (Allison, 1995). If the effect estimates from the different analyses are similar, the violations of the assumption of independent censoring are not important. We performed a sensitivity analysis for the six diseases (mastitis, teat injuries, lameness, anestrus, ovarian cysts and milk fever) that most affected culling decisions (based on the change in ÿ2log likelihood when disease was added to the base model with parity, herd and season). Thus, we re-did the analysis by 1) assuming all the censored cows to have been culled at the time of their censoring and 2) assuming them to have been censored at the longest time observed in these data. 3. Results 3.1. Sensitivity analysis The two extreme scenarios of the sensitivity analysis (Scenario 1: all censored cows assumed to have been culled at the time of censoring; Scenario 2: all of them censored much later) did not, in general, change the effects of diseases on culling when compared to the standard analysis. The only observed changes occurred within the very last stage of the lactation (>240 d): the effects of fertility disorders were significant and more protective in Scenario 1 and they increased the risk of culling in Scenario 2 (when adjusted only for parity, calving season and herd). In Scenario 1 the RRs (risk ratios) for mastitis, teat injuries and lameness dropped at the end of lactation. The effects of milk fever also changed slightly at the end of the lactation. The effects of parity decreased (RRs dropped) in Scenario 1, but they remained the same in Scenario 2 as in the starting situation. The effects of pregnancy status did not change meaningfully in either scenario until the very end of the lactation (the last stage for pregnancy status>305 d) (results not shown). 3.2. Effects of diseases and reproductive performance on culling The culling percentage in these data was 31.6. The median time of last insemination was 102 d after calving and out of all the cows that had been inseminated, 46.2% were inseminated only once, 27.5% twice, 13.9% thrice, and 12.4% more than three times.
284
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
Including pregnancy status in Model 1 with parity, calving season, and herd significantly improved the fit, indicating that pregnancy status had an important role in the culling decisions the farmers make. Also, the number of inseminations had a significant effect on culling, when added to Model 2. This was judged by the change in the ÿ2log likelihoods between the two models being compared (Model 1 vs. Model 2: 240 715ÿ231 2819434; Model 2 vs. Model 3: 231 281ÿ221 7549527; Table 1). Table 2 presents the effects of parity and calving season on culling from different models. Parity had a significant effect on culling; the older the cow the higher the risk of culling. The effects (RRs) of parity on culling were greatest when adjusted only for calving season and herd (Model 1). When pregnancy status was included in the model, the parity effects decreased. A cow in her sixth or higher parity was four times more likely to be culled than a first parity cow when not adjusted for pregnancy status, but three times more likely when pregnancy status was controlled for in the model. When the number of inseminations was also included in the model the effect of parity on culling decreased even more. However, parity still had a significant effect. For instance, cows in their sixth or higher parity were at 1.4 times higher risk of being culled than parity 1 cows. Calving season also had a significant effect on culling. The risk of culling was lowest for cows who calved during fall in Models 1 and 2 (Table 2). There was no difference between the cows calving during summer and fall in Model 3. The knowledge about a cow's pregnancy status had a different effect on culling at different stages of lactation. The results suggest that the later the cow conceived (i.e. the later the farmer knew that she was pregnant) the higher was the risk of her being culled Table 2 Effects of parity and calving season on culling of 39 727 Finnish Ayrshire cows calving in 1993a Covariate
Risk ratio (RR) Model 1
Model 2
Model 3
Parity Missing 1 2 3 4 5 6
2.5*** 1.0 1.4*** 2.0*** 2.6*** 3.1*** 4.0***
2.2*** 1.0 1.4*** 1.8*** 2.3*** 2.6*** 3.0***
1.4*** 1.0 1.2*** 1.3*** 1.4*** 1.3*** 1.4***
Calving season Winter Spring Summer Fall
1.2*** 1.2*** l.1** 1.0
1.2*** 1.3*** 1.1* 1.0
1.1*** 1.2*** 1.0 1.0
a
Model 1 includes also herd, Model 2 includes herd and pregnancy status, and Model 3 includes herd, pregnancy status and number of inseminations as covariates. * P-value<0.05; ** P-value<0.01; *** P-value<0.001.
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
285
Table 3 Effects of pregnancy status and number of inseminations on culling in 39 727 Finnish Ayrshire cows calving in 1993a Pregnancy status
Risk ratio (RR) b
Stage of culling
Stage of conception
Model 2
Model 3
0±150 d
0±150 d 0 0±150 d 151±240 d 0 0±150 d 151±240 d 241±305 d 0 0±150 d 151±240 d 241±305 d >305 d 0
1.0 43.4*** 1.0 1.2*** 3.2*** 1.0 2.0*** 2.5*** 12.0*** 1.0 1.7*** 3.1*** 4.8*** 55.8***
1.0 25.8*** 1.0 1.3*** 1.6*** 1.0 2.1*** 2.6*** 4.9*** 1.0 1.7*** 2.9*** 3.8*** 23.9***
151±240 d 241±305 d
>305 d
Number of inseminations 0 1 2 3 3
10.0*** 1.0 0.9*** 0.9* 0.9**
a Model 2 include parity, season, herd and pregnancy status as covariates and Model 3 included also the number of inseminations. b The stage of lactation when the farmer was assumed to have knowledge that a cow was pregnant; 0 refers to open cows (i.e. cows that did not conceive at all). * P-value<0.05; **P-value<0.01; ***P-value<0.001.
(Table 3). Cows that had conceived at the beginning of the lactation (within 150 d after calving) were considered to be the reference category. Because pregnancy status was defined as ``open'' for all cows that were culled within the first 150 d after calving, the risk ratios for pregnancy status during that period (RRs 43.4 and 25.8, in Models 2 and 3) are therefore artificial and cannot be considered valid. The risk of being culled at the end of lactation (between days 241 and 305 after calving) was 2.0 and 2.5 times higher for cows that had not conceived by day 150 or 240, respectively, than for cows that had conceived during the first 150 d after calving (Model 2). Cows whose pregnancy status was still unknown by day 305 after calving were 12 times more likely to be culled between 241 and 305 d than cows that had conceived within 150 d. Including the number of inseminations in the model reduced the magnitude of effect of pregnancy status on culling, but it still had a significant impact (Table 3). Cows that had not been inseminated at all were at ten times higher risk of being culled than cows that had been inseminated only once. If cows had been inseminated more than once, that had a slightly protective effect on culling (RR0.9). Judging by the change in the ÿ2log likelihood and regardless of the model used, out of all the diseases mastitis always explained the largest portion of culling, followed by
286
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
lameness and teat injuries (in varying order in different models) (Table 1). After that, milk fever, anestrus, and ovarian cysts (in varying order, depending on the model) had the greatest effect on culling. Retained placenta did not affect culling decisions. The effects of diseases varied between models (Tables 4±6). When adjusted for pregnancy status (Model 2), the changes in the disease effects were not very remarkable in comparison with Model 1 and the effects mainly tended to decrease. In Model 3 (containing both pregnancy status and the number of inseminations) the effects of diseases, in general, seemed to decrease if the disease had occurred at the beginning or in the middle of the lactation and they tended to increase if the disease had occurred later on during the lactation, when compared to Model 1. Milk fever, dystocia, and metritis increased the risk of culling at the beginning of the lactation in all the models (Table 4) and also at the end of the lactation when adjusted only for parity, calving season and herd (Model 1). When adjusted for pregnancy status Table 4 Effects of milk fever, dystocia and metritis on culling in 39 727 Finnish Ayrshire cowsa Stage of disease Milk fever 0 (no disease) >0 d
Dystocia 0 (no disease) >0 d
Metritis 0 (no disease) 0±30 d
>30 d
Stage of culling
RR Model 1
Model 2
Model 3
0±30 d 31±60 d 61±150 d 151±240 d >240 d
1.0 2.5*** 0.8 0.9 1.1 1.2**
1.0 2.7*** 0.9 0.9 1.1 0.9
1.0 2.4*** 0.8 0.8** 0.9 0.7***
0±30 d 31±60 d 61±150 d 151±240 d >240 d
1.0 2.4*** 0.8 0.8 1.0 l.2***
1.0 2.4*** 1.1 1.1 1.1 0.9
1.0 1.9*** 0.8 0.9 0.9 0.7***
0±30 d 31±60 d 61±150 d 151±240 d >240 d
1.0 2.2*** 1.1 1.2 1.2 1.4**
1.0 2.3*** 0.9 1.1 0.9 1.1
1.0 1.8** 0.8 0.9 0.9 0.9
31±60 d 61±150 d 151±240 d >240 d
No events 0.7 1.0 1.3*
No events 0.7 0.8 0.9
No events 0.9 1.2 1.4***
a Besides the disease in question, Model 1 included also parity, calving season and herd as covariates, Model 2 also included pregnancy status and Model 3 also included number of inseminations besides all the others mentioned above. * P-value<0.05; ** P-value<0.01; *** P-value<0.00l.
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
287
Table 5 Effects of mastitis, teat injuries and lameness on culling in 39 727 Finnish Ayrshire cows calving during 1993a Stage of disease Mastitis 0 (no disease) 0±30 d
31±60 d
61±150 d 151±240 d >240 d Teat injuries 0 (no disease) 0±30 d
31±60 d
61±150 d 151±240 d >240 d Lameness 0 (no disease) 0±30 d
31±60 d
Stage of culling
Risk ratio (RR) Model 1
Model 2
Model 3
0±30 d 31±60 d 61±150 d 151±240 d >240 d 31±60 d 61±150 d 151±240 d >240 d 61±150 d 151±240 d >240 d 151±240 d >240 d >240 d
1.0 1.3* 2.1*** 1.9*** 1.7*** l.5*** 0.6 2.6*** 2.0*** 1.4*** 2.4*** 2.1*** 1.4*** 1.8*** 1.0 0.5***
1.0 1.3* 2.1*** l.9*** 1.7*** 1.4*** 0.6 2.6*** 1.9*** 1.3* 2.4*** 2.0*** 1.4*** l.8*** 1.3* 0.8
1.0 0.9 1.5*** 1.4*** 1.3*** 1.1* 0.4 2.0*** 1.7*** 1.5*** 2.2*** 2.3*** 1.7*** 2.4*** 1.7*** 1.0
0±30 d 31±60 d 61±150 d 151±240 d >240 d 31±60 d 61±150 d 151±240 d >240 d 61±150 d 151±240 d >240 d l51±240 d >240 d >240 d
1.0 3.0*** 2.8*** 2.1*** 2.0*** 1.5** 2.3 3.0*** 3.3*** 2.1** 2.8*** 1.8*** 1.6** 2.6*** 1.1 0.6*
1.0 3.0*** 2.8*** 2.0*** l.9*** l.9*** 2.2 3.0*** 3.3*** 1.8* 2.1*** 1.8** 1.8*** 2.7*** 1.9*** 1.0
1.0 1.9*** 1.8* 1.4* 1.4* 1.7*** 1.6 2.0** 2.7*** 2.1** 3.1*** 2.1*** 2.3*** 2.9*** 1.3** 1.4
0±30 d 31±60 d 61±150 d 151±240 d >240 d 31±60 d 61±150 d 151±240 d >240 d
1.0 6.0*** 4.2*** 2.5*** 1.9*** 2.2*** 12.0*** 3.5*** 1.2 2.0***
1.0 6.1*** 4.2*** 2.5*** l.6*** 1.5** 12.2*** 3.4*** 1.1 1.6*
1.0 3.3*** 2.4*** 1.5* 1.1 1.1 6.3*** 2.2** 0.7 1.0
288
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
Table 5 (Continued ) Stage of disease
Stage of culling
61±150 d
61±150 d 151±240 d >240 d l51±240 d >240 d >240 d
151±240 d >240 d
Risk ratio (RR) Model 1
Model 2
Model 3
3.3*** 2.3*** l.8*** 3.1*** 1.3 1.4
3.2*** 2.0*** 1.4* 3.2*** 1.6 2.2**
2.4*** 1.9** 1.9*** 3.7*** 1.7* 3.1***
a Besides the disease in question, Model 1 included also parity, calving season and herd as covariates, Model 2 also pregnancy status and Model 3 also number of inseminations besides all the others mentioned above. * P-value<0.05; ** P-value<0.01; *** P-value<0.001.
Table 6 Effects of anestrus and ovarian cysts on culling in 39 727 Finnish Ayrshire cows calving in 1993a Stage of disease Anestrus 0 (no disease) 0±90 d
91±150 d 151±240 d >240 d Ovarian cysts 0 (no disease) 0±90 d
91±150 d l51±240 d >240 d
Stage of culling
Risk ratio (RR) Model 1
Model 2
Model 3
0±90 d 91±150 d 151±240 d >240 d 91±150 d 151±240 d >240 d 151±240 d >240 d >240 d
1.0 0.3** 0.6*** 0.7*** 1.0 0.3*** 0.5*** 1.1 0.3*** 1.2* 0.7
1.0 0.3** 0.6*** 0.6*** 1.0 0.2*** 0.4*** 1.0 0.2*** 0.8* 0.5**
1.0 0.5 0.9 1.0 1.5*** 0.4* 0.7* 1.5*** 0.4* 1.4*** 1.1
0±90 d 91±150 d 151±240 d >240 d 91±150 d 151±240 d >240 d 151±240 d >240 d >240 d
1.0 0.4** 0.5*** 0.6*** 1.0 0.3*** 0.6*** 1.0 0.6 1.2* 1.1
1.0 0.4** 0.5** 0.6*** 1.0 0.2*** 0.4*** 0.9 0.3*** 0.8** 0.6**
1.0 0.5 0.8 0.9 1.6*** 0.4* 0.8 1.5*** 0.8 1.5*** 1.1
a Besides the disease in question, Model 1 included also parity, calving season and herd as covariates, Model 2 also included pregnancy status and Model 3 also included number of inseminations besides all the others mentioned above. * P-value<0.05; ** P-value<0.01; *** P-value<0.001.
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
289
(Model 2), they no longer had a significant effect (P>0.05) at the end of the lactation. When adjusted also for the number of inseminations (Model 3), the effect at the beginning was slightly less than in Model I and protective during late lactation for milk fever and dystocia. Having diseases in as time dependent covariates and considering interactions between the time of disease occurrence and stage of lactation enabled us to see the lag-effects of the diseases at the end of lactation. Mastitis, teat injuries, and lameness affected culling decisions significantly throughout the whole lactation in all the models, increasing the risk of being culled (Table 5). When adjusted for pregnancy status (Model 2), their effects slightly decreased in comparison to Model 1, with the exception of teat injuries, the effect of which on culling at the end of the lactation increased; also, the effect of lameness became significant at the very end of the lactation. In Model 3, their effects were lower at the beginning of the lactation, but if the disease occurred in the middle of or late in lactation (after 150 d), its effects were greater (RRs were higher) than when it was only adjusted for parity, calving season, and herd. Fertility disorders (anestrus and ovarian cysts) had a protective role against culling in Models 1 and 2 (Table 6). When adjusted for pregnancy status, cows that had been treated for anestrus and ovarian cysts within 90 d after calving had only 0.3 and 0.4, respectively, times the risk of being culled during that period as cows without these disorders. Both of these disorders seemed to be protective in Model 2 even if diagnosed and treated later on during the lactation (e.g. RRs 0.5 and 0.6, respectively, after 240 d). When adjusted both for pregnancy status and the number of inseminations (Model 3), neither anestrus nor ovarian cysts occurring within 90 d after calving had a significant effect on culling before 240 d. They increased the risk of culling after 240 d, regardless of when the cow had been treated for these disorders. The effects of digestive disorders (traumatic reticuloperitonitis, rumen disorders and displaced abomasum) did not vary meaningfully between Model 1 (containing parity, calving season and herd; Part I in this series, Rajala-Schultz and GroÈhn, 1999, Table 7) and Model 2 (containing also pregnancy status) (results not shown). When adjusted both for pregnancy status and the number of inseminations, the effects of these diseases, in general, decreased. However, the effects of traumatic reticuloperitonitis increased towards the end of the lactation. In Model 1 cows diagnosed with traumatic reticuloperitonitis between 150 and 240 d after calving were at 3.9 times higher risk of being culled during that same period than cows without the disease (Part I in this series, Rajala-Schultz and GroÈhn, 1999, Table 7), but in Model 3 the risk of being culled for a cow diagnosed with traumatic reticuloperitonitis was 5.3 times higher than for cows without it during that same period (results not shown). The effects of non-parturient paresis and hypomagnesemia late in lactation increased when adjusted for pregnancy status and for both pregnancy and insemination. Cows with non-parturient paresis after 240 d had a 2.2 times higher risk of being culled than cows without paresis in Model 1 (Part I in this service, Rajala-Schultz and GroÈhn, 1999, Table 7), but 7.1 times higher risk of being culled in Model 3. Hypomagnesemia did not have a significant effect on culling in Model 1 when occurring after 240 d, but in Model 3, hypomagnesemic cows were 8.5 times more likely to be culled than cows without the disorder (results not shown).
290
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
The pregnancy status of category 6 cows (see Section 2) was based on an assumption they had conceived. The effect on culling was also studied when these cows were assumed to have remained open after the last insemination. In this latter scenario the effect of pregnancy increased remarkably, the effects of parity decreased, but the effects of diseases did not change meaningfully when compared to the other extreme scenario in which all these cows were assumed to have conceived at the last insemination. 4. Discussion 4.1. Sensitivity analysis Standard methods of survival analysis require that random censoring be noninformative (Allison, 1995); informative censoring could lead to biased estimates. In other words, it is required that censoring and event mechanisms be independent. Thus, in our situation we needed to make the assumption that for a censored cow with a particular set of characteristics, had she continued to be at risk for culling, her chance of culling would not be any higher or lower than for any other culled cow with the same characteristics. The results from the sensitivity analysis did not vary meaningfully from the original scenario, until the very end of the lactation. All the censored cows in these data calved at the time of censoring. The cohorts of censored cows and of cows experiencing the event of interest (i.e. culling) were comparable (i.e. at the same risk for culling) until the time when the first cow calved, i.e. was censored, around 300±350 d after the previous calving. After this point, the non-censored cows still remained at risk for culling while the censored cows, of course, did not. The fact that the effects of parity decreased in Scenario 1 (all censored cows assumed to be culled) seems reasonable; if all cows that actually were pregnant and expected to calve were culled, it only seems to suggest that parity did not matter very much in the culling decision. Because our censoring time was related to reproductive performance in the sense that all cows that calved for the next time were censored, it also seems understandable that the effects of pregnancy status and the diseases that might affect conception changed at the end of the lactation. Also, in Scenario 1 in which all cows were culled at time of censoring, it seems reasonable to expect that the effects of diseases like mastitis, teat injuries and lameness would lose their effect, because all cows were culled anyway. Thus, the results indicated that the effects of pregnancy status and the diseases at the very end of the lactation might be biased due to our censoring mechanism. They should be interpreted with caution. 4.2. Effects of reproductive performance on culling Culling is, in general, classified into two categories, voluntary and involuntary. Voluntary culling often refers to those animals that leave the herd because of low production (in the absence of a known disease problem) or because the herd has an excess of animals. Culling potentially increases profits and/or reduces costs by replacing sick or non-pregnant animals which are expensive to keep. According to a literature review
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
291
(Fetrow, 1988), the major reason for involuntary culling is reproductive failure, averaging 23% of all culls and 38% of involuntary culls. Pregnancy status of a cow was found to be an extremely significant factor for culling also in this study. Our results indicated that the earlier the farmer knew that a cow was pregnant, the lower the risk of culling was. This is in agreement with several other studies, which have reported that a failure to conceive at first service and longer days open increase the risk of culling (Martin et al., 1982; Erb et al., 1985; Beaudeau et al., 1995), and that after a cow has conceived her risk of culling decreases (GroÈhn et al., 1998). Also, the number of times a cow had been inseminated had a significant effect on culling, even if already adjusted for pregnancy status. If the cow was not inseminated at all, the risk of culling was 10 times higher than if only inseminated once. Out of the cows that never conceived, some the farmer had probably deliberately not inseminated and some failed to conceive despite several inseminations. By including the number of inseminations in the model it was possible to distinguish between cows in these two groups. In the former case (not inseminated at all), the major reason for culling was not related to the current lactation (e.g. pregnancy status), but to the previous lactation (e.g. disease history or production). It is only natural, therefore, that the risk of culling was so high for cows that had not been inseminated at all, if the farmer had decided to cull them without even trying to get them pregnant. For the same reason it is also natural that the effects of pregnancy status decreased when the number of inseminations was included in the model. The effect of insemination was slightly protective if the cow had been inseminated more than once. This probably indicates that the farmer was willing to invest time and effort to get the cow pregnant because he/she wanted to keep her, e.g. for high production. We used 150 d as a cut-off time before which the pregnancy status was not assumed to have any effect on culling decisions. This might have been a conservative assumption and some cows probably had already conceived by that time; however, the median days in milk for the last insemination in these data was 102 d and often Finnish farmers do not check pregnancy of their cows until 9 weeks after insemination. Had an earlier cut-off time been used (e.g. 120 d), the risk ratios for culling for cows conceiving later in lactation would probably have been even higher than they were with the 150 d cut-off time. The pregnancy status of category 6 cows (see Section 2) was based on an assumption that they all conceived at the last insemination. Therefore, the opposite scenario, in which all were assumed to have remained open, was investigated to see how it affected the results. As expected, the effect of pregnancy status was even greater (higher RRs), because in this scenario, all culled cows were open and all censored cows calved another time. This did not, however, affect remarkably the estimates of the disease effects, indicating that they were not very sensitive to this change in the pregnancy status of some culled cows. In the scenario where these cows were assumed to have remained open, the effects of parity decreased. If all the culled cows are assumed to be open, the importance of pregnancy status is emphasized and the effect of parity loses significance. The last stage of lactation for pregnancy status was defined to be >305 d. The number of cows that had not conceived and were still alive by that time was, however, very small (699 cows out of the 38 228 cows, i.e. 1.8%). These cows can, in some sense, be
292
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
considered as outliers and that was the reason why we wanted to keep them in their own category (instead of merging them into the second to last category (241±305 d)). This could also be a reason why the results from the sensitivity analysis differed at the end of the lactation from the original scenario. Parity had a large impact on culling, the risk of culling increasing with increasing parity, as has been found in earlier studies (Dohoo and Martin, 1984; Beaudeau et al., 1995). However, when adjusted for pregnancy status or for pregnancy status and number of inseminations, the magnitude of its effect decreased, but still remained significant. This implies that part of the effect of parity is accounted for through reproductive performance of the cow. Also, in the study of GroÈhn et al. (1998), the effect of parity decreased when adjusted for conception status. 4.3. Effects of diseases on culling Milk fever, dystocia and metritis increased the risk of culling both at the beginning and at the end of the lactation, when the pregnancy status of the cow was not considered (Part I in the series, Rajala-Schultz and GroÈhn, 1999). They did not have a significant effect on culling at the end of lactation when pregnancy status was included in the model. This suggests that the pregnancy status of cows which had also experienced these diseases plays a more important role in culling decisions at the end of the lactation than the diseases per se. It also suggests that the disease effect is indirectly accounted for by including the pregnancy status in the model and that cows with these diseases might be more likely to stay open than cows without them. When the number of inseminations was also added to the model the effects of these diseases became protective at the end of the lactation. It simply suggests that the farmer still wanted to inseminate and to keep the cow. In general, Dohoo and Martin (1984), Erb et al. (1985), Milian-Suazo et al. (1989) and Beaudeau et al. (1994) also found that milk fever and dystocia increased the risk of culling. In Part I of this study (Rajala-Schultz and GroÈhn, 1999), without adjusting for pregnancy status, we found that mastitis, teat injuries and lameness had a significant effect on culling throughout the lactation. Also, Dohoo and Martin (1984), Erb et al. (1985), Milian-Suazo et al. (1989) and GroÈhn et al. (1997) found mastitis to increase the risk of culling. When adjusted for pregnancy status and insemination the magnitude of the effects of these diseases decreased at the beginning and in the middle of the lactation, but were greater at the end of lactation. This would suggest that the effects of these diseases were confounded by reproductive performance and that at the end of the lactation these diseases might have been the triggering factor for culling. Also, the digestive disorders and non-parturient paresis and hypomagnesemia seemed to be key factors in culling decisions if they occurred at the end of the lactation and when reproductive performance had been considered. Anestrus and ovarian cysts were protective against culling after the diagnosis and treatment of the disorder when adjusted for parity, calving season and herd. When pregnancy status was also in the model, their effects remained significant, and became more protective. The cows that were treated for fertility disorders at the end of the lactation, e.g. after 240 d, must have been very valuable to the farmer. The protective
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
293
effect of the treatment of fertility disorders at the end of the lactation is mainly an indication of the farmer's willingness and decision to keep the cow despite the fact that she was still open at that point in the lactation. When the number of inseminations was added to the model, the effects of anestrus and ovarian cysts became non-significant at the beginning of the lactation and they increased the risk of culling at the end of lactation. Cows with ovarian cysts and silent heat (anestrus) require more inseminations to conceive than cows without these problems. The results suggest that the effects of these disorders became more apparent when the number of inseminations was accounted for in the model; the actual disorder increased the risk of culling. It is important to keep in mind that these disease data are based on diseases diagnosed and treated by veterinarians; not only the occurrences of diseases. So by the time the disease diagnosis is recorded the farmer has already made the first, preliminary decision to keep the cow and have her treated. For the same reason, we were not able to study the effect of abortion on culling, because farmers do not usually call a veterinarian to treat a cow after an abortion. We did not take into account the milk production of the cows in the analysis. Farmers are usually more willing to treat and invest money in high producers than low producers and that could be one reason why for instance, cows with ovarian cysts were less likely to be culled than cows without this disorder. After all, it was shown earlier that high milk yield and ovarian cysts were associated in these same data (Rajala and GroÈhn, 1998). 5. Conclusions The results of this study indicate that reproductive performance of a cow, in addition to diseases, plays an important role in farmers' culling decisions. The knowledge about a cow's pregnancy status had a different effect on culling depending on the stage of lactation; the earlier the farmer knew that a cow was pregnant, the lower was the risk of culling. When the number of inseminations was accounted for, the effect of pregnancy status decreased, distinguishing between cows that failed to conceive and cows that were never even inseminated. If a cow was inseminated more than once, that had a protective role against culling, in general indicating that farmers are willing to invest time and effort to get certain cows pregnant in order to keep them. Mastitis, teat injuries, lameness, ovarian cysts, anestrus and milk fever were the most important diseases affecting culling, regardless of whether reproductive performance was accounted for or not. The effects of diseases varied slightly between different models: when adjusted for pregnancy status the changes in the disease effects were not very remarkable and the effects mainly tended to decrease, indicating that some of the disease effects were indirectly accounted for by reproductive performance. The results of the sensitivity analysis suggested that our censoring times were not fully independent of the event times at the end of the observation period (i.e. lactation). Therefore, the effects of pregnancy status and the diseases on culling at the very end of the lactation might be biased due to our censoring mechanism and should be interpreted with caution.
294
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
Acknowledgements This study was supported by the Finnish Academy, Walther EhrstroÈm Foundation and Foundation of Veterinary Medicine in Finland (Suomen elaÈinlaÈaÈketieteen saÈaÈtioÈ). The research was conducted using the resources of the Cornell Theory Center (Cornell University, Ithaca, NY), which receives major funding from the National Science Foundation (Arlington, VA) and New York State with additional support from the Advanced Research Projects (Arlington, VA), the National Center for Research Resources at the National Institute of Health (Bethesda, MID), IBM Corporation (Armonk, NY), and other members of the center's Corporate Partnership Program. References Allison, P.D., 1995. Survival Analysis Using the SAS System. A Practical Guide. Cary, NC, USA, SAS Institute Inc. Beaudeau, F., Ducrocq, V., Fourichon, C., Seegers, H., 1995. Effect of disease on length of productive life of French Holstein dairy cows assessed by survival analysis. J. Dairy Sci. 78, 103±117. Beaudeau, F., Frankema, K., Fourichon, C., Seegers, H., Faye, B., Noordhuisen, J.P.T.M., 1994. Associations between health disorders of French dairy cows and early and late culling within lactation. Prev. Vet. Med. 19, 213±231. Cox, D., 1972. Regression models and life tables. Journal of the Royal Statistical Society Series B: 187±202. Dohoo, I.R., Martin, S.W., 1984. Disease, production and culling in Holstein-Friesian cows. V. Survivorship. Prev. Vet. Med. 2, 771±784. Ducrocq, V., SoÈlkner, J., 1994. The Survival Kit, a FORTRAN package for the analysis of survival data, vol. 22. Guelph. Dep. Anim. Poultry Sci., Univ. of Guelph, pp. 51 and 52. Erb, H.N., Smith, R.D., Oltenacu, P.A., Guard, C.L., HilIman, R.B., Powers, P.A., Smith, M.C., White, M.E., 1985. Path model of reproductive disorders and performance, milk fever, mastitis, milk yield, and culling in Holstein cows. J. Dairy Sci. 68, 3337±3349. Fetrow, J., 1988. Culling of Dairy Cows. 20th Annual Meeting of AABP, Phoenix, Arizona, AABP. GroÈhn, Y.T., Ducrocq, V., Hertl, J.A., 1997. Modeling the effect of a disease on risk of culling: an illustration of the use of time-dependent covariates for survival analysis. J. Dairy Sci. 80, 1755±1766. GroÈhn, Y.T., Eicker, S.W., Ducrocq, V., Hertl, J.A., 1998. Effect of diseases on culling in New York State Holstein Dairy Cows. J. Dairy Sci. 81, 966±978. Harman, J.L., Grohn, Y.T., Erb, H.N., Casella, G., 1996. Event-time analysis of the effect of season of parturition, parity, and concurrent disease on parturition-to-conception interval in dairy cows. Am. J. Vet. Res., AJVR 57, 640±645. Martin, S.W., Aziz, S.A., Sandals, W.C.D., Curtis, R.A., 1982. The association between clinical disease, production and culling in Holstein-Friesian cows. Can. J. Anim. Sci. 62, 633±640. Marubini, E., Valsecchi, M.G., 1995. Analysing Survival Data from Clinical Trials and Observational Studies. Wiley, Chishester. Milian-Suazo, F., Erb, H.N., Smith, R.D., 1989. Risk factors for reason-specific culling of dairy cows. Prev. Vet. Med. 7, 19±29. Rajala, P.J., GroÈhn, Y.T., 1998. Disease occurrence and risk factors analysis in Finnish Ayrshire cows. Acta Vet. Scand. 39, 1±13. Rajala-Schultz, P.J., GroÈhn, Y.T., 1999. Culling of dairy cows. Part I. Effects of diseases on culling in Finnish Ayrshire cows. Prev. Vet. Med. 41, 195±208.
Culling of dairy cows. Part II. Effects of diseases and reproductive performance on culling in Finnish Ayrshire cows P.J. Rajala-Schultza,b,*, Y.T. GroÈhna a
Department of Population Medicine and Diagnostic Sciences, Section of Epidemiology, College of Veterinary Medicine, Cornell University, Ithaca, NY 14853, USA b Department of Clinical Veterinary Sciences, Faculty of Veterinary Medicine, PO Box 57, University of Helsinki, Helsinki, FIN-00014, Finland Accepted 9 March 1999
Abstract The effects of 15 diseases and reproductive performance on culling were studied in 39 727 Finnish Ayrshire cows that calved in 1993 and were followed until culling or next calving. Survival analysis, using the Cox proportional hazards model, was performed with diseases and pregnancy status as time-dependent covariates. Parity, calving season and herd were included as covariates in every model. The effect of the number of inseminations was also studied. The farmer's knowledge of the cow's pregnancy status had a significant effect on culling. It varied according to the stage of lactation a cow was in; the earlier the farmer knew a cow was pregnant, the smaller was the risk of culling. If a cow had not been inseminated at all, her risk of culling was 10 times higher than if she was inseminated once. If a cow was inseminated more than once, she had a slightly lower risk of being culled than a cow inseminated only once. The effect of parity decreased when pregnancy status and number of inseminations were added to the model, indicating that part of the parity effect was accounted for by reproductive performance. Including diseases in the model with pregnancy status and the number of inseminations did not change the effects of reproductive performance on culling. Mastitis, teat injuries and lameness had the greatest effect on culling (whether adjusted for reproductive performance or not), increasing the risk of culling, followed by anestrus, ovarian cysts and milk fever. In general, the effects of diseases decreased when reproductive performance was also accounted for in the model. When pregnancy status was included in the model, the effects of anestrus and ovarian cysts became slightly more protective, but when the number of inseminations * Corresponding author. Tel.: +1-607-253-3572; fax: +1-607-253-3083 E-mail address: [email protected] (P.J. Rajala-Schultz) 0167-5877/99/$ ± see front matter # 1999 Elsevier Science B.V. All rights reserved. PII: S 0 1 6 7 - 5 8 7 7 ( 9 9 ) 0 0 0 4 5 - 8
280
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
was also considered, they became non-significant at the beginning of lactation and they increased the risk of culling at the end of lactation. Sensitivity analysis, which was run to evaluate the effects of our censoring mechanism on the results, indicated that the censoring times (i.e., the time of next calving) were not fully independent of the event (culling) times; the effects of the diseases and pregnancy status at the very end of the lactation changed slightly from the original model. # 1999 Elsevier Science B.V. All rights reserved. Keywords: Cattle-management; Reproductive performance; Culling; Survival analysis; Time-dependent covariates
1. Introduction Milk is the major source of income in a dairy farm. Because cows produce milk only after calving and cows' daily milk yield decreases as time passes after parturition, the farmer wants his/her cows to conceive within a restricted period of time after calving and to calve at regular intervals. If conception is delayed, this reproductive inefficiency can also lead to milk production inefficiency. If a cow is not pregnant (is open) for an extended period after calving, at the end of the lactation the farmer can either dry her off and allow a long dry period or he/she can continue to milk, albeit with reduced daily milk production. Neither of the above mentioned options are economically most desirable. Beaudeau et al. (1995) showed that reproductive performance had a high impact on the length of productive life of dairy cows, i.e. the longer the days open the higher the risk of culling. GroÈhn et al. (1998) showed that once a cow had conceived, her risk of culling dropped sharply. Harman et al. (1996), on the other hand, showed that several diseases have an effect on fertility of dairy cows. Concurrent diseases can affect fertility of dairy cows and can cause delayed conception and they could therefore be an indirect reason for culling. Earlier we studied the effects of 15 diseases on culling per se, without adjusting for the pregnancy status of a cow (Part I in this series, Rajala-Schultz and GroÈhn, 1999). The purpose of this study was to evaluate the effects of diseases and reproductive performance on culling decisions and to see how the effects of diseases on culling change when adjusted for pregnancy status and the number of times a cow is inseminated. 2. Materials and methods 2.1. Data The data consisted of 39 727 Finnish Ayrshire dairy cows that calved during 1993 and were followed until the next calving or culling. The cows were in 2338 herds that belonged to the milk registry and the national health recording system. All herds used artificial insemination. The data have been described earlier (Rajala and GroÈhn, 1998; Part I in this series, Rajala-Schultz and GroÈhn, 1999). Data on insemination dates were
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
281
also available. In this analysis, insemination data were divided into five levels: cows that had not been inseminated at all, those inseminated once, twice, thrice, and more than three times. The stages of lactation for disease occurrence and culling were the same as in Part I, (Table 1, Rajala-Schultz and GroÈhn, 1999). The farmer's knowledge of the pregnancy status of the cow may have a different effect on his/her culling decisions depending on the stage of lactation the cow is in. The lactation was divided into four stages (0±150 d, 151±240 d, 241±305 d and >305 d) with respect to when pregnancy status could affect culling decisions. The pregnancy status was considered to change its value at the time when the farmer could have reasonably known whether a cow had conceived or not after the last insemination and when this information, therefore, might have influenced the farmer's culling decision. This was assumed to be 63 d after last insemination, because in Finland artificial insemination (AI) technicians do most of the pregnancy checks and they often do not do them until 9 weeks after the insemination. The time when the pregnancy status changed its value was calculated in six different ways: (1) For cows that had a subsequent calving (26 162 cows), the time when their Table 1 Comparison of partial likelihoods (ÿ2log L) for different diseases in the different original modelsa Covariate
Basic model Mastitis Lameness Teat injury Milk fever Cysts Anestrus DAc Hardware NP paresisd Rumene Hypomgf Dystocia Metritis Ketosis Ret. plac.g
Model 1
Model 2
Model 3
Change in ÿ2log L
Change in ÿ2log L
Change in ÿ2log L
b
240 715 500*** 264*** 181*** 108*** 95*** 90*** 89*** 71*** 69*** 59*** 57*** 38*** 36*** 25** 7
b
231 281 432*** 226*** 192*** 115*** 164*** 171*** 89*** 73*** 74*** 54*** 50*** 27*** 22*** 20* 2
221 754b 340*** 135*** 153*** 122*** 91*** 100*** 68*** 70*** 90*** 54*** 49*** 26*** 27*** 21* 2
Change in dfh
15 15 15 5 11 11 9 14 14 15 15 5 8 9 5
a Model 1: parity, season and herd; Model 2: parity, season, herd and pregnancy status; and Model 3: parity, season, herd, pregnancy status and number of inseminations. b ÿ2log partial likelihood for the basic model against which the change in ÿ2log L was calculated. c Displaced abomasum. d Non-parturient paresis. e Rumen disorders. f Hypomagnesemia. g Retained placenta. h Degrees of freedom. * P-value<0.05; ** P-value<0.01; *** P-value<0.001.
282
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
pregnancy status might have influenced the farmer's culling decision was calculated by subtracting 279 d (i.e. average gestation length in these data) from the subsequent calving date and adding 63 d. (2) If the cow was culled within 150 d after calving (3996 cows), she was considered to have been open all that time (or more precisely, it was assumed that the farmer considered her to be open and that pregnancy status was not yet an issue in the culling decision at that time). (3) If she was culled later than 150 d after calving and the farmer-stated reason for culling was ``non-pregnant'' (2540 cows), she was also considered to have been open all the time. (4) If she was culled later than 150 d after calving without the farmer-stated reason ``non-pregnant'', and she did not have any insemination dates (2198 cows), she was considered open all that time. (5) If she was culled later than 150 d after calving without the farmer-stated reason ``non-pregnant'' and she had been inseminated and the interval between the last insemination and culling was longer than 279 d, then she was considered to have been open all that time (163 cows). (6) If she was culled later than 150 d after calving without the farmer-stated reason ``non-pregnant'', she had been inseminated, and the interval between the last insemination and culling was less than 279 d, the time of change in her pregnancy status (or more precisely, the time of change in farmer's knowledge concerning her pregnancy status) was calculated in the following way: last insemination date plus 63 d (3169 cows) (i.e. these cows were assumed to have conceived at the last insemination). Because the last category (6) was based on an assumption, not verified pregnancy status, the effect on culling assuming that all these cows were open was also investigated. 2.2. Statistical analysis Survival analysis, using the Cox proportional hazards model (Cox, 1972), was performed to study the effect of disease and reproductive performance on culling. The statistical method has been described in Part I of this series (Rajala-Schultz and GroÈhn, 1999). The analyses were conducted using the ``Survival Kit'', a set of FORTRAN programs (Ducrocq and SoÈlkner, 1994). Diseases and pregnancy status were modeled as timedependent covariates. 2.3. Models Three basic models were compared; parity, calving season and herd were covariates in all models. Therefore, when referring to any model later on in the paper, it should be assumed that those three covariates are in the model, too. Model 1 included parity, calving season and herd and one disease at a time (15 models, one for each disease) (see Part I of the series). Model 2 also included pregnancy status and Model 3 included both pregnancy status and the number of inseminations. Fifteen models (one for each disease) were also fitted for the basic Models 2 and 3. It was of interest to see how the effects of diseases changed when also adjusted for pregnancy status and for both pregnancy status and the number of inseminations as compared to when they were only adjusted for parity, calving season and herd.
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
283
2.4. Sensitivity analysis A typical assumption made in standard methods of survival analysis is that the censoring times and the event times are independent. In other words, censoring carries no prognostic information about the subsequent event experience, i.e., those who are censored are at no higher or lower risk of experiencing the event of interest than the rest of the individuals in the sample. So, censoring is assumed to be independent and noninformative (Marubini and Valseechi, 1995). In order to get an idea of the possible impact of potential informative censoring on the results, a sensitivity analysis can be performed. The basic idea is to re-do the analysis under two extreme assumptions about the censored cases. One assumption is that censored observations experience an event immediately after they are censored, and the other assumption is that censored cases have longer times to events than any other subjects in the sample (Allison, 1995). If the effect estimates from the different analyses are similar, the violations of the assumption of independent censoring are not important. We performed a sensitivity analysis for the six diseases (mastitis, teat injuries, lameness, anestrus, ovarian cysts and milk fever) that most affected culling decisions (based on the change in ÿ2log likelihood when disease was added to the base model with parity, herd and season). Thus, we re-did the analysis by 1) assuming all the censored cows to have been culled at the time of their censoring and 2) assuming them to have been censored at the longest time observed in these data. 3. Results 3.1. Sensitivity analysis The two extreme scenarios of the sensitivity analysis (Scenario 1: all censored cows assumed to have been culled at the time of censoring; Scenario 2: all of them censored much later) did not, in general, change the effects of diseases on culling when compared to the standard analysis. The only observed changes occurred within the very last stage of the lactation (>240 d): the effects of fertility disorders were significant and more protective in Scenario 1 and they increased the risk of culling in Scenario 2 (when adjusted only for parity, calving season and herd). In Scenario 1 the RRs (risk ratios) for mastitis, teat injuries and lameness dropped at the end of lactation. The effects of milk fever also changed slightly at the end of the lactation. The effects of parity decreased (RRs dropped) in Scenario 1, but they remained the same in Scenario 2 as in the starting situation. The effects of pregnancy status did not change meaningfully in either scenario until the very end of the lactation (the last stage for pregnancy status>305 d) (results not shown). 3.2. Effects of diseases and reproductive performance on culling The culling percentage in these data was 31.6. The median time of last insemination was 102 d after calving and out of all the cows that had been inseminated, 46.2% were inseminated only once, 27.5% twice, 13.9% thrice, and 12.4% more than three times.
284
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
Including pregnancy status in Model 1 with parity, calving season, and herd significantly improved the fit, indicating that pregnancy status had an important role in the culling decisions the farmers make. Also, the number of inseminations had a significant effect on culling, when added to Model 2. This was judged by the change in the ÿ2log likelihoods between the two models being compared (Model 1 vs. Model 2: 240 715ÿ231 2819434; Model 2 vs. Model 3: 231 281ÿ221 7549527; Table 1). Table 2 presents the effects of parity and calving season on culling from different models. Parity had a significant effect on culling; the older the cow the higher the risk of culling. The effects (RRs) of parity on culling were greatest when adjusted only for calving season and herd (Model 1). When pregnancy status was included in the model, the parity effects decreased. A cow in her sixth or higher parity was four times more likely to be culled than a first parity cow when not adjusted for pregnancy status, but three times more likely when pregnancy status was controlled for in the model. When the number of inseminations was also included in the model the effect of parity on culling decreased even more. However, parity still had a significant effect. For instance, cows in their sixth or higher parity were at 1.4 times higher risk of being culled than parity 1 cows. Calving season also had a significant effect on culling. The risk of culling was lowest for cows who calved during fall in Models 1 and 2 (Table 2). There was no difference between the cows calving during summer and fall in Model 3. The knowledge about a cow's pregnancy status had a different effect on culling at different stages of lactation. The results suggest that the later the cow conceived (i.e. the later the farmer knew that she was pregnant) the higher was the risk of her being culled Table 2 Effects of parity and calving season on culling of 39 727 Finnish Ayrshire cows calving in 1993a Covariate
Risk ratio (RR) Model 1
Model 2
Model 3
Parity Missing 1 2 3 4 5 6
2.5*** 1.0 1.4*** 2.0*** 2.6*** 3.1*** 4.0***
2.2*** 1.0 1.4*** 1.8*** 2.3*** 2.6*** 3.0***
1.4*** 1.0 1.2*** 1.3*** 1.4*** 1.3*** 1.4***
Calving season Winter Spring Summer Fall
1.2*** 1.2*** l.1** 1.0
1.2*** 1.3*** 1.1* 1.0
1.1*** 1.2*** 1.0 1.0
a
Model 1 includes also herd, Model 2 includes herd and pregnancy status, and Model 3 includes herd, pregnancy status and number of inseminations as covariates. * P-value<0.05; ** P-value<0.01; *** P-value<0.001.
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
285
Table 3 Effects of pregnancy status and number of inseminations on culling in 39 727 Finnish Ayrshire cows calving in 1993a Pregnancy status
Risk ratio (RR) b
Stage of culling
Stage of conception
Model 2
Model 3
0±150 d
0±150 d 0 0±150 d 151±240 d 0 0±150 d 151±240 d 241±305 d 0 0±150 d 151±240 d 241±305 d >305 d 0
1.0 43.4*** 1.0 1.2*** 3.2*** 1.0 2.0*** 2.5*** 12.0*** 1.0 1.7*** 3.1*** 4.8*** 55.8***
1.0 25.8*** 1.0 1.3*** 1.6*** 1.0 2.1*** 2.6*** 4.9*** 1.0 1.7*** 2.9*** 3.8*** 23.9***
151±240 d 241±305 d
>305 d
Number of inseminations 0 1 2 3 3
10.0*** 1.0 0.9*** 0.9* 0.9**
a Model 2 include parity, season, herd and pregnancy status as covariates and Model 3 included also the number of inseminations. b The stage of lactation when the farmer was assumed to have knowledge that a cow was pregnant; 0 refers to open cows (i.e. cows that did not conceive at all). * P-value<0.05; **P-value<0.01; ***P-value<0.001.
(Table 3). Cows that had conceived at the beginning of the lactation (within 150 d after calving) were considered to be the reference category. Because pregnancy status was defined as ``open'' for all cows that were culled within the first 150 d after calving, the risk ratios for pregnancy status during that period (RRs 43.4 and 25.8, in Models 2 and 3) are therefore artificial and cannot be considered valid. The risk of being culled at the end of lactation (between days 241 and 305 after calving) was 2.0 and 2.5 times higher for cows that had not conceived by day 150 or 240, respectively, than for cows that had conceived during the first 150 d after calving (Model 2). Cows whose pregnancy status was still unknown by day 305 after calving were 12 times more likely to be culled between 241 and 305 d than cows that had conceived within 150 d. Including the number of inseminations in the model reduced the magnitude of effect of pregnancy status on culling, but it still had a significant impact (Table 3). Cows that had not been inseminated at all were at ten times higher risk of being culled than cows that had been inseminated only once. If cows had been inseminated more than once, that had a slightly protective effect on culling (RR0.9). Judging by the change in the ÿ2log likelihood and regardless of the model used, out of all the diseases mastitis always explained the largest portion of culling, followed by
286
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
lameness and teat injuries (in varying order in different models) (Table 1). After that, milk fever, anestrus, and ovarian cysts (in varying order, depending on the model) had the greatest effect on culling. Retained placenta did not affect culling decisions. The effects of diseases varied between models (Tables 4±6). When adjusted for pregnancy status (Model 2), the changes in the disease effects were not very remarkable in comparison with Model 1 and the effects mainly tended to decrease. In Model 3 (containing both pregnancy status and the number of inseminations) the effects of diseases, in general, seemed to decrease if the disease had occurred at the beginning or in the middle of the lactation and they tended to increase if the disease had occurred later on during the lactation, when compared to Model 1. Milk fever, dystocia, and metritis increased the risk of culling at the beginning of the lactation in all the models (Table 4) and also at the end of the lactation when adjusted only for parity, calving season and herd (Model 1). When adjusted for pregnancy status Table 4 Effects of milk fever, dystocia and metritis on culling in 39 727 Finnish Ayrshire cowsa Stage of disease Milk fever 0 (no disease) >0 d
Dystocia 0 (no disease) >0 d
Metritis 0 (no disease) 0±30 d
>30 d
Stage of culling
RR Model 1
Model 2
Model 3
0±30 d 31±60 d 61±150 d 151±240 d >240 d
1.0 2.5*** 0.8 0.9 1.1 1.2**
1.0 2.7*** 0.9 0.9 1.1 0.9
1.0 2.4*** 0.8 0.8** 0.9 0.7***
0±30 d 31±60 d 61±150 d 151±240 d >240 d
1.0 2.4*** 0.8 0.8 1.0 l.2***
1.0 2.4*** 1.1 1.1 1.1 0.9
1.0 1.9*** 0.8 0.9 0.9 0.7***
0±30 d 31±60 d 61±150 d 151±240 d >240 d
1.0 2.2*** 1.1 1.2 1.2 1.4**
1.0 2.3*** 0.9 1.1 0.9 1.1
1.0 1.8** 0.8 0.9 0.9 0.9
31±60 d 61±150 d 151±240 d >240 d
No events 0.7 1.0 1.3*
No events 0.7 0.8 0.9
No events 0.9 1.2 1.4***
a Besides the disease in question, Model 1 included also parity, calving season and herd as covariates, Model 2 also included pregnancy status and Model 3 also included number of inseminations besides all the others mentioned above. * P-value<0.05; ** P-value<0.01; *** P-value<0.00l.
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
287
Table 5 Effects of mastitis, teat injuries and lameness on culling in 39 727 Finnish Ayrshire cows calving during 1993a Stage of disease Mastitis 0 (no disease) 0±30 d
31±60 d
61±150 d 151±240 d >240 d Teat injuries 0 (no disease) 0±30 d
31±60 d
61±150 d 151±240 d >240 d Lameness 0 (no disease) 0±30 d
31±60 d
Stage of culling
Risk ratio (RR) Model 1
Model 2
Model 3
0±30 d 31±60 d 61±150 d 151±240 d >240 d 31±60 d 61±150 d 151±240 d >240 d 61±150 d 151±240 d >240 d 151±240 d >240 d >240 d
1.0 1.3* 2.1*** 1.9*** 1.7*** l.5*** 0.6 2.6*** 2.0*** 1.4*** 2.4*** 2.1*** 1.4*** 1.8*** 1.0 0.5***
1.0 1.3* 2.1*** l.9*** 1.7*** 1.4*** 0.6 2.6*** 1.9*** 1.3* 2.4*** 2.0*** 1.4*** l.8*** 1.3* 0.8
1.0 0.9 1.5*** 1.4*** 1.3*** 1.1* 0.4 2.0*** 1.7*** 1.5*** 2.2*** 2.3*** 1.7*** 2.4*** 1.7*** 1.0
0±30 d 31±60 d 61±150 d 151±240 d >240 d 31±60 d 61±150 d 151±240 d >240 d 61±150 d 151±240 d >240 d l51±240 d >240 d >240 d
1.0 3.0*** 2.8*** 2.1*** 2.0*** 1.5** 2.3 3.0*** 3.3*** 2.1** 2.8*** 1.8*** 1.6** 2.6*** 1.1 0.6*
1.0 3.0*** 2.8*** 2.0*** l.9*** l.9*** 2.2 3.0*** 3.3*** 1.8* 2.1*** 1.8** 1.8*** 2.7*** 1.9*** 1.0
1.0 1.9*** 1.8* 1.4* 1.4* 1.7*** 1.6 2.0** 2.7*** 2.1** 3.1*** 2.1*** 2.3*** 2.9*** 1.3** 1.4
0±30 d 31±60 d 61±150 d 151±240 d >240 d 31±60 d 61±150 d 151±240 d >240 d
1.0 6.0*** 4.2*** 2.5*** 1.9*** 2.2*** 12.0*** 3.5*** 1.2 2.0***
1.0 6.1*** 4.2*** 2.5*** l.6*** 1.5** 12.2*** 3.4*** 1.1 1.6*
1.0 3.3*** 2.4*** 1.5* 1.1 1.1 6.3*** 2.2** 0.7 1.0
288
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
Table 5 (Continued ) Stage of disease
Stage of culling
61±150 d
61±150 d 151±240 d >240 d l51±240 d >240 d >240 d
151±240 d >240 d
Risk ratio (RR) Model 1
Model 2
Model 3
3.3*** 2.3*** l.8*** 3.1*** 1.3 1.4
3.2*** 2.0*** 1.4* 3.2*** 1.6 2.2**
2.4*** 1.9** 1.9*** 3.7*** 1.7* 3.1***
a Besides the disease in question, Model 1 included also parity, calving season and herd as covariates, Model 2 also pregnancy status and Model 3 also number of inseminations besides all the others mentioned above. * P-value<0.05; ** P-value<0.01; *** P-value<0.001.
Table 6 Effects of anestrus and ovarian cysts on culling in 39 727 Finnish Ayrshire cows calving in 1993a Stage of disease Anestrus 0 (no disease) 0±90 d
91±150 d 151±240 d >240 d Ovarian cysts 0 (no disease) 0±90 d
91±150 d l51±240 d >240 d
Stage of culling
Risk ratio (RR) Model 1
Model 2
Model 3
0±90 d 91±150 d 151±240 d >240 d 91±150 d 151±240 d >240 d 151±240 d >240 d >240 d
1.0 0.3** 0.6*** 0.7*** 1.0 0.3*** 0.5*** 1.1 0.3*** 1.2* 0.7
1.0 0.3** 0.6*** 0.6*** 1.0 0.2*** 0.4*** 1.0 0.2*** 0.8* 0.5**
1.0 0.5 0.9 1.0 1.5*** 0.4* 0.7* 1.5*** 0.4* 1.4*** 1.1
0±90 d 91±150 d 151±240 d >240 d 91±150 d 151±240 d >240 d 151±240 d >240 d >240 d
1.0 0.4** 0.5*** 0.6*** 1.0 0.3*** 0.6*** 1.0 0.6 1.2* 1.1
1.0 0.4** 0.5** 0.6*** 1.0 0.2*** 0.4*** 0.9 0.3*** 0.8** 0.6**
1.0 0.5 0.8 0.9 1.6*** 0.4* 0.8 1.5*** 0.8 1.5*** 1.1
a Besides the disease in question, Model 1 included also parity, calving season and herd as covariates, Model 2 also included pregnancy status and Model 3 also included number of inseminations besides all the others mentioned above. * P-value<0.05; ** P-value<0.01; *** P-value<0.001.
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
289
(Model 2), they no longer had a significant effect (P>0.05) at the end of the lactation. When adjusted also for the number of inseminations (Model 3), the effect at the beginning was slightly less than in Model I and protective during late lactation for milk fever and dystocia. Having diseases in as time dependent covariates and considering interactions between the time of disease occurrence and stage of lactation enabled us to see the lag-effects of the diseases at the end of lactation. Mastitis, teat injuries, and lameness affected culling decisions significantly throughout the whole lactation in all the models, increasing the risk of being culled (Table 5). When adjusted for pregnancy status (Model 2), their effects slightly decreased in comparison to Model 1, with the exception of teat injuries, the effect of which on culling at the end of the lactation increased; also, the effect of lameness became significant at the very end of the lactation. In Model 3, their effects were lower at the beginning of the lactation, but if the disease occurred in the middle of or late in lactation (after 150 d), its effects were greater (RRs were higher) than when it was only adjusted for parity, calving season, and herd. Fertility disorders (anestrus and ovarian cysts) had a protective role against culling in Models 1 and 2 (Table 6). When adjusted for pregnancy status, cows that had been treated for anestrus and ovarian cysts within 90 d after calving had only 0.3 and 0.4, respectively, times the risk of being culled during that period as cows without these disorders. Both of these disorders seemed to be protective in Model 2 even if diagnosed and treated later on during the lactation (e.g. RRs 0.5 and 0.6, respectively, after 240 d). When adjusted both for pregnancy status and the number of inseminations (Model 3), neither anestrus nor ovarian cysts occurring within 90 d after calving had a significant effect on culling before 240 d. They increased the risk of culling after 240 d, regardless of when the cow had been treated for these disorders. The effects of digestive disorders (traumatic reticuloperitonitis, rumen disorders and displaced abomasum) did not vary meaningfully between Model 1 (containing parity, calving season and herd; Part I in this series, Rajala-Schultz and GroÈhn, 1999, Table 7) and Model 2 (containing also pregnancy status) (results not shown). When adjusted both for pregnancy status and the number of inseminations, the effects of these diseases, in general, decreased. However, the effects of traumatic reticuloperitonitis increased towards the end of the lactation. In Model 1 cows diagnosed with traumatic reticuloperitonitis between 150 and 240 d after calving were at 3.9 times higher risk of being culled during that same period than cows without the disease (Part I in this series, Rajala-Schultz and GroÈhn, 1999, Table 7), but in Model 3 the risk of being culled for a cow diagnosed with traumatic reticuloperitonitis was 5.3 times higher than for cows without it during that same period (results not shown). The effects of non-parturient paresis and hypomagnesemia late in lactation increased when adjusted for pregnancy status and for both pregnancy and insemination. Cows with non-parturient paresis after 240 d had a 2.2 times higher risk of being culled than cows without paresis in Model 1 (Part I in this service, Rajala-Schultz and GroÈhn, 1999, Table 7), but 7.1 times higher risk of being culled in Model 3. Hypomagnesemia did not have a significant effect on culling in Model 1 when occurring after 240 d, but in Model 3, hypomagnesemic cows were 8.5 times more likely to be culled than cows without the disorder (results not shown).
290
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
The pregnancy status of category 6 cows (see Section 2) was based on an assumption they had conceived. The effect on culling was also studied when these cows were assumed to have remained open after the last insemination. In this latter scenario the effect of pregnancy increased remarkably, the effects of parity decreased, but the effects of diseases did not change meaningfully when compared to the other extreme scenario in which all these cows were assumed to have conceived at the last insemination. 4. Discussion 4.1. Sensitivity analysis Standard methods of survival analysis require that random censoring be noninformative (Allison, 1995); informative censoring could lead to biased estimates. In other words, it is required that censoring and event mechanisms be independent. Thus, in our situation we needed to make the assumption that for a censored cow with a particular set of characteristics, had she continued to be at risk for culling, her chance of culling would not be any higher or lower than for any other culled cow with the same characteristics. The results from the sensitivity analysis did not vary meaningfully from the original scenario, until the very end of the lactation. All the censored cows in these data calved at the time of censoring. The cohorts of censored cows and of cows experiencing the event of interest (i.e. culling) were comparable (i.e. at the same risk for culling) until the time when the first cow calved, i.e. was censored, around 300±350 d after the previous calving. After this point, the non-censored cows still remained at risk for culling while the censored cows, of course, did not. The fact that the effects of parity decreased in Scenario 1 (all censored cows assumed to be culled) seems reasonable; if all cows that actually were pregnant and expected to calve were culled, it only seems to suggest that parity did not matter very much in the culling decision. Because our censoring time was related to reproductive performance in the sense that all cows that calved for the next time were censored, it also seems understandable that the effects of pregnancy status and the diseases that might affect conception changed at the end of the lactation. Also, in Scenario 1 in which all cows were culled at time of censoring, it seems reasonable to expect that the effects of diseases like mastitis, teat injuries and lameness would lose their effect, because all cows were culled anyway. Thus, the results indicated that the effects of pregnancy status and the diseases at the very end of the lactation might be biased due to our censoring mechanism. They should be interpreted with caution. 4.2. Effects of reproductive performance on culling Culling is, in general, classified into two categories, voluntary and involuntary. Voluntary culling often refers to those animals that leave the herd because of low production (in the absence of a known disease problem) or because the herd has an excess of animals. Culling potentially increases profits and/or reduces costs by replacing sick or non-pregnant animals which are expensive to keep. According to a literature review
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
291
(Fetrow, 1988), the major reason for involuntary culling is reproductive failure, averaging 23% of all culls and 38% of involuntary culls. Pregnancy status of a cow was found to be an extremely significant factor for culling also in this study. Our results indicated that the earlier the farmer knew that a cow was pregnant, the lower the risk of culling was. This is in agreement with several other studies, which have reported that a failure to conceive at first service and longer days open increase the risk of culling (Martin et al., 1982; Erb et al., 1985; Beaudeau et al., 1995), and that after a cow has conceived her risk of culling decreases (GroÈhn et al., 1998). Also, the number of times a cow had been inseminated had a significant effect on culling, even if already adjusted for pregnancy status. If the cow was not inseminated at all, the risk of culling was 10 times higher than if only inseminated once. Out of the cows that never conceived, some the farmer had probably deliberately not inseminated and some failed to conceive despite several inseminations. By including the number of inseminations in the model it was possible to distinguish between cows in these two groups. In the former case (not inseminated at all), the major reason for culling was not related to the current lactation (e.g. pregnancy status), but to the previous lactation (e.g. disease history or production). It is only natural, therefore, that the risk of culling was so high for cows that had not been inseminated at all, if the farmer had decided to cull them without even trying to get them pregnant. For the same reason it is also natural that the effects of pregnancy status decreased when the number of inseminations was included in the model. The effect of insemination was slightly protective if the cow had been inseminated more than once. This probably indicates that the farmer was willing to invest time and effort to get the cow pregnant because he/she wanted to keep her, e.g. for high production. We used 150 d as a cut-off time before which the pregnancy status was not assumed to have any effect on culling decisions. This might have been a conservative assumption and some cows probably had already conceived by that time; however, the median days in milk for the last insemination in these data was 102 d and often Finnish farmers do not check pregnancy of their cows until 9 weeks after insemination. Had an earlier cut-off time been used (e.g. 120 d), the risk ratios for culling for cows conceiving later in lactation would probably have been even higher than they were with the 150 d cut-off time. The pregnancy status of category 6 cows (see Section 2) was based on an assumption that they all conceived at the last insemination. Therefore, the opposite scenario, in which all were assumed to have remained open, was investigated to see how it affected the results. As expected, the effect of pregnancy status was even greater (higher RRs), because in this scenario, all culled cows were open and all censored cows calved another time. This did not, however, affect remarkably the estimates of the disease effects, indicating that they were not very sensitive to this change in the pregnancy status of some culled cows. In the scenario where these cows were assumed to have remained open, the effects of parity decreased. If all the culled cows are assumed to be open, the importance of pregnancy status is emphasized and the effect of parity loses significance. The last stage of lactation for pregnancy status was defined to be >305 d. The number of cows that had not conceived and were still alive by that time was, however, very small (699 cows out of the 38 228 cows, i.e. 1.8%). These cows can, in some sense, be
292
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
considered as outliers and that was the reason why we wanted to keep them in their own category (instead of merging them into the second to last category (241±305 d)). This could also be a reason why the results from the sensitivity analysis differed at the end of the lactation from the original scenario. Parity had a large impact on culling, the risk of culling increasing with increasing parity, as has been found in earlier studies (Dohoo and Martin, 1984; Beaudeau et al., 1995). However, when adjusted for pregnancy status or for pregnancy status and number of inseminations, the magnitude of its effect decreased, but still remained significant. This implies that part of the effect of parity is accounted for through reproductive performance of the cow. Also, in the study of GroÈhn et al. (1998), the effect of parity decreased when adjusted for conception status. 4.3. Effects of diseases on culling Milk fever, dystocia and metritis increased the risk of culling both at the beginning and at the end of the lactation, when the pregnancy status of the cow was not considered (Part I in the series, Rajala-Schultz and GroÈhn, 1999). They did not have a significant effect on culling at the end of lactation when pregnancy status was included in the model. This suggests that the pregnancy status of cows which had also experienced these diseases plays a more important role in culling decisions at the end of the lactation than the diseases per se. It also suggests that the disease effect is indirectly accounted for by including the pregnancy status in the model and that cows with these diseases might be more likely to stay open than cows without them. When the number of inseminations was also added to the model the effects of these diseases became protective at the end of the lactation. It simply suggests that the farmer still wanted to inseminate and to keep the cow. In general, Dohoo and Martin (1984), Erb et al. (1985), Milian-Suazo et al. (1989) and Beaudeau et al. (1994) also found that milk fever and dystocia increased the risk of culling. In Part I of this study (Rajala-Schultz and GroÈhn, 1999), without adjusting for pregnancy status, we found that mastitis, teat injuries and lameness had a significant effect on culling throughout the lactation. Also, Dohoo and Martin (1984), Erb et al. (1985), Milian-Suazo et al. (1989) and GroÈhn et al. (1997) found mastitis to increase the risk of culling. When adjusted for pregnancy status and insemination the magnitude of the effects of these diseases decreased at the beginning and in the middle of the lactation, but were greater at the end of lactation. This would suggest that the effects of these diseases were confounded by reproductive performance and that at the end of the lactation these diseases might have been the triggering factor for culling. Also, the digestive disorders and non-parturient paresis and hypomagnesemia seemed to be key factors in culling decisions if they occurred at the end of the lactation and when reproductive performance had been considered. Anestrus and ovarian cysts were protective against culling after the diagnosis and treatment of the disorder when adjusted for parity, calving season and herd. When pregnancy status was also in the model, their effects remained significant, and became more protective. The cows that were treated for fertility disorders at the end of the lactation, e.g. after 240 d, must have been very valuable to the farmer. The protective
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
293
effect of the treatment of fertility disorders at the end of the lactation is mainly an indication of the farmer's willingness and decision to keep the cow despite the fact that she was still open at that point in the lactation. When the number of inseminations was added to the model, the effects of anestrus and ovarian cysts became non-significant at the beginning of the lactation and they increased the risk of culling at the end of lactation. Cows with ovarian cysts and silent heat (anestrus) require more inseminations to conceive than cows without these problems. The results suggest that the effects of these disorders became more apparent when the number of inseminations was accounted for in the model; the actual disorder increased the risk of culling. It is important to keep in mind that these disease data are based on diseases diagnosed and treated by veterinarians; not only the occurrences of diseases. So by the time the disease diagnosis is recorded the farmer has already made the first, preliminary decision to keep the cow and have her treated. For the same reason, we were not able to study the effect of abortion on culling, because farmers do not usually call a veterinarian to treat a cow after an abortion. We did not take into account the milk production of the cows in the analysis. Farmers are usually more willing to treat and invest money in high producers than low producers and that could be one reason why for instance, cows with ovarian cysts were less likely to be culled than cows without this disorder. After all, it was shown earlier that high milk yield and ovarian cysts were associated in these same data (Rajala and GroÈhn, 1998). 5. Conclusions The results of this study indicate that reproductive performance of a cow, in addition to diseases, plays an important role in farmers' culling decisions. The knowledge about a cow's pregnancy status had a different effect on culling depending on the stage of lactation; the earlier the farmer knew that a cow was pregnant, the lower was the risk of culling. When the number of inseminations was accounted for, the effect of pregnancy status decreased, distinguishing between cows that failed to conceive and cows that were never even inseminated. If a cow was inseminated more than once, that had a protective role against culling, in general indicating that farmers are willing to invest time and effort to get certain cows pregnant in order to keep them. Mastitis, teat injuries, lameness, ovarian cysts, anestrus and milk fever were the most important diseases affecting culling, regardless of whether reproductive performance was accounted for or not. The effects of diseases varied slightly between different models: when adjusted for pregnancy status the changes in the disease effects were not very remarkable and the effects mainly tended to decrease, indicating that some of the disease effects were indirectly accounted for by reproductive performance. The results of the sensitivity analysis suggested that our censoring times were not fully independent of the event times at the end of the observation period (i.e. lactation). Therefore, the effects of pregnancy status and the diseases on culling at the very end of the lactation might be biased due to our censoring mechanism and should be interpreted with caution.
294
P.J. Rajala-Schultz, Y.T. GroÈhn / Preventive Veterinary Medicine 41 (1999) 279±294
Acknowledgements This study was supported by the Finnish Academy, Walther EhrstroÈm Foundation and Foundation of Veterinary Medicine in Finland (Suomen elaÈinlaÈaÈketieteen saÈaÈtioÈ). The research was conducted using the resources of the Cornell Theory Center (Cornell University, Ithaca, NY), which receives major funding from the National Science Foundation (Arlington, VA) and New York State with additional support from the Advanced Research Projects (Arlington, VA), the National Center for Research Resources at the National Institute of Health (Bethesda, MID), IBM Corporation (Armonk, NY), and other members of the center's Corporate Partnership Program. References Allison, P.D., 1995. Survival Analysis Using the SAS System. A Practical Guide. Cary, NC, USA, SAS Institute Inc. Beaudeau, F., Ducrocq, V., Fourichon, C., Seegers, H., 1995. Effect of disease on length of productive life of French Holstein dairy cows assessed by survival analysis. J. Dairy Sci. 78, 103±117. Beaudeau, F., Frankema, K., Fourichon, C., Seegers, H., Faye, B., Noordhuisen, J.P.T.M., 1994. Associations between health disorders of French dairy cows and early and late culling within lactation. Prev. Vet. Med. 19, 213±231. Cox, D., 1972. Regression models and life tables. Journal of the Royal Statistical Society Series B: 187±202. Dohoo, I.R., Martin, S.W., 1984. Disease, production and culling in Holstein-Friesian cows. V. Survivorship. Prev. Vet. Med. 2, 771±784. Ducrocq, V., SoÈlkner, J., 1994. The Survival Kit, a FORTRAN package for the analysis of survival data, vol. 22. Guelph. Dep. Anim. Poultry Sci., Univ. of Guelph, pp. 51 and 52. Erb, H.N., Smith, R.D., Oltenacu, P.A., Guard, C.L., HilIman, R.B., Powers, P.A., Smith, M.C., White, M.E., 1985. Path model of reproductive disorders and performance, milk fever, mastitis, milk yield, and culling in Holstein cows. J. Dairy Sci. 68, 3337±3349. Fetrow, J., 1988. Culling of Dairy Cows. 20th Annual Meeting of AABP, Phoenix, Arizona, AABP. GroÈhn, Y.T., Ducrocq, V., Hertl, J.A., 1997. Modeling the effect of a disease on risk of culling: an illustration of the use of time-dependent covariates for survival analysis. J. Dairy Sci. 80, 1755±1766. GroÈhn, Y.T., Eicker, S.W., Ducrocq, V., Hertl, J.A., 1998. Effect of diseases on culling in New York State Holstein Dairy Cows. J. Dairy Sci. 81, 966±978. Harman, J.L., Grohn, Y.T., Erb, H.N., Casella, G., 1996. Event-time analysis of the effect of season of parturition, parity, and concurrent disease on parturition-to-conception interval in dairy cows. Am. J. Vet. Res., AJVR 57, 640±645. Martin, S.W., Aziz, S.A., Sandals, W.C.D., Curtis, R.A., 1982. The association between clinical disease, production and culling in Holstein-Friesian cows. Can. J. Anim. Sci. 62, 633±640. Marubini, E., Valsecchi, M.G., 1995. Analysing Survival Data from Clinical Trials and Observational Studies. Wiley, Chishester. Milian-Suazo, F., Erb, H.N., Smith, R.D., 1989. Risk factors for reason-specific culling of dairy cows. Prev. Vet. Med. 7, 19±29. Rajala, P.J., GroÈhn, Y.T., 1998. Disease occurrence and risk factors analysis in Finnish Ayrshire cows. Acta Vet. Scand. 39, 1±13. Rajala-Schultz, P.J., GroÈhn, Y.T., 1999. Culling of dairy cows. Part I. Effects of diseases on culling in Finnish Ayrshire cows. Prev. Vet. Med. 41, 195±208.