The present invention relation to a method and system for determining a chronic mastitis index (CMi) for a specific cow being a member of a specific herd. The method typically comprises the steps of:

providing a Current Healthy Baseline (CHB) for the specific herd, the Current Healthy Baseline is a fitted curve representing relationship between absLDH and Day From Calving (DFC) specific for the farm,

- providing a baseline for the specific cow considered (VeryStableM, VSM), the baseline for the specific cow is a smoothed curve representing the relationship between absLDH and Days From Calving for the specific cow, determining the Chronic Mastitis index (CMI) at a selected time from calving (DFC) based on at least the baseline for the specific cow (VeryStableM) and the current healthy baseline (CHB).

BACKGROUND OF THE INVENTION

Mastitis in dairy cows causes problems in connection with milk productions and is it therefore advantageous if cows that have or even could develop mastitis are identifiable.

The article "A Model for Detection of Individual Cow Mastitis Based on an Indicator measured in Milk" (Chagunda et. al; J. Dairy Sci. 89: 2380-2998, 2006) has presented that a baseline may be established for the content of LDH in milk and has furthermore presented an index for the chronicity of a cow. The proposed index is a proposed degree of chronic mastitis and is calculated as

ChronDeg=fl(LDH, milk yield)/f2(Parity, DFC) where fl and f2 are two algebraic functions, LDH is Lactate Dehydrogenase and DFC is day from calving. According to Chargunda et. al. ChronDeg is expected to vary with breed and parity, which has led to the conclusion that a universal baseline for f2 is valid for a breed across herds. While this may be justified and intuitively seems correct by accepting the assumption that the mastitis risk for a member of a breed is a result of the breed's inherited resistance to mastitis, it has been found in connection with the present invention that classification of cow based on the model presented in Chargunda et. al. gives rise to number of false mastitis cases detected. While it may be reasonable easy to detect false mastitis cases by surveying how a particular cow evolves over time and compare the result with the mastitis prognosis determined, the concept of a universal baseline for a breed prevails as this is a fundamental theme within the field of mastitis detection. Thus, if and when false mastitis cases are detected, the fault is suspected to be in the universal baseline used and not in the concept of applying a universal baseline. Thus, the solution to the fault related problem would be to select better - if available - data which at the same time are statistical significant. As this is a general accepted problem, the generally accepted trend within data analysis in the field of the present invention is that false results are handled by removing

"outliers", that is cows within the breed which shows abnormal behaviour, and by increasing the amount of data to make the universal baseline increasingly more statistical significant.

OBJECT OF THE INVENTION

Following the above, an improved method for determining a chronic mastitis index would be advantageous, and in particular a more efficient and/or reliable method of evaluation lactating cows would be advantageous.

It is a further object of the present invention to provide a method of classifying cows into different class identified by their state of health with respect to suffering from mastitis. It is a further object of the present invention to provide an alternative to the prior art.

SUMMARY OF THE INVENTION

Thus, the above described object and several other objects are intended to be obtained in a first aspect of the invention by providing method for determining a chronic mastitis index (CMi) for a specific cow being a member of a specific herd, the method comprising

providing a Current Healthy Baseline (CHB) for the specific herd (farm), the Current Healthy Baseline is a fitted curve representing relationship between absLDH and Day From Calving (DFC) specific for the herd, providing a baseline for the specific cow considered (VeryStableM), the baseline for the specific cow is a smoothed curve representing the relationship between absLDH and Days From Calving for the specific cow, determining the Chronic Mastitis index (CMI) at a selected time from calving (DFC) based on at least the baseline for the specific cow

(VeryStableM) and the current healthy baseline (CHB).

In connection with the present invention, it has become apparent that universal baseline for a breed to be applied across different farms could not be found, as not only the progression of the curve but also the level of absLDH differs between farms. This means that if a universal baseline is used for determining a chronic index, healthy cows could in one farm be seen as suffering from mastitis whereas sick cows could be seen as not suffering from mastitis in another farm. It is noted, that contrary to the work reported in "A Model for Detection of

Individual Cow Mastitis Based on an Indicator measured in Milk" (Chagunda et. al; J. Dairy Sci. 89: 2380-2998, 2006) the CurrentHealthyBaseline, CHB, is specific for a specific farm. This, means that the CHB is designed to match the absLDH level of a specific farm which is contrary to apply a universal baseline, derived from dataset from a plurality of farms and "assembled" into one single baseline.

This also means that even though a Current Healthy Baseline in some situations (will be discussed in details below) is based on data from a plurality of farm, the Current Healthy Baseline being tailored to one specific farm may deviate from a Current Healthy Baseline being specific for another farm even though such

Current Healthy Baselines may originate from the same data.

The crux of the present invention may be seen as using a Healthy Baseline for a specific herd of a farm thereby discarding the prevailing and general accepted understanding that a universal healthy baseline for a breed exist, which universal baseline is valid for different farms as long as the breed considered is the same.

Further, in the work presented by Chargunda et al a degree of chronic mastitis is calculated as ChronDeg=Stable/ChronRef, where Stable is calculated as a rolling average of the LDH level (Mmol/min) for each cow over an interval of 7 days and ChronRef a fitted curve to a pool of data assembled from a number of farms (and thereby from a number of distinct herds) having the same breed (as no variations are to be expected within a breed). By estimating Stable as a rolling average, no smoothed curve representing the relationship between absLDH and DFC for the specific cow is provided. While this may seem as less important, the inventors of the present invention has realized that by using such a rolling average, the clinical history of the specific cow considered is discarded thereby increasing the risk of determining a CMI which does not reflect the true clinical situation of the cow considered.

As presented herein, the Current Healthy Baseline may preferably be a calculated curve representing a relationship between absLDH, determined as the per cow product of the milk yield in litre (L) and lactate dehydrogenase activity value ( mol/min per L), and Days From Calving (DFC) for the specific herd.

Further, absLDH as used in the baseline for the specific cow considered may preferably be determined as the product of the milk yield in litre (L) and lactate dehydrogenase activity value ( mol/min per L) and Days From Calving for the specific cow. The baseline for the specific cow considered may preferably be provided by fitting a smooth curve to a time series of smoothed absLDH-values obtained by measurements from the milk produced by the specific cow

considered.

The determination of the Chronic Mastitis index (CMi) at a selected time from calving (DFC) is, preferably, determined at least as a relation of values of absLDH determined from the baseline for the specific cow (VeryStableM, VSM) and from the Current Healthy Baseline. "Relation" preferably includes, but is preferably not limited to, one of the relations: VSM-CHB; VSM/CHB and (VSM-CHB)/CHB.

Further, additional factors, as disclosed herein, may preferably be added to the relation.

A particular problem arises by deviating from the concept of a universal baseline, namely the problem of determining a healthy baseline when the data available for the specific herd is not sufficient to be statistical significant for the herd. To solve this issue, one may consider using a universal baseline until sufficient data is available. However, based on the inventors findings, this may result in less useable determinations for CMi.

The Current Healthy Baseline (CHB) is preferably selected either from

- a group of Global Healthy Baselines (GHB's), the Global Healthy Baselines being determined, preferable by calculation, on the basis of data sets of absLDH and Days From Calving (DFC) for a plurality of farms, or

as a Local Healthy Baseline (LHB) for the specific farm being determined, preferably by calculation, on the basis of data sets of absLDH and Days From Calving (DFC) for the specific farm.

Thus, the Current Healthy Baseline is preferably provided by a selecting the Current Healthy Baseline from a group of Global Healthy Baselines or as a Local Healthy Baseline. It is noted that a difference in general exist between a Global Healthy Baseline and a universal healthy baseline (as disclosed above) as will become clear from the following. Preferably, a Local Healthy Baseline is selected when sufficient data as disclosed herein is available.

Thus, when sufficient data for the specific herd is not available yet for determining a Current Healthy Baseline, a GHB may, preferably, be selected from a group of GHB, which selection results in that a GHB which comes closest to the level of the specific herd may be selected until sufficient data is available for the herd. As disclosed in the following, such GHB is not considered as a universal baseline e.g. as in Chagunda et al.

Preferably, according to the present invention, no measurements are in general removed from the data sets on which the Current Healthy Baseline (CHB) and the baseline for the specific cow considered (VeryStableM, VSM) are based. By executing the method according to the present invention, a measure for the Enzyme Load (EL) is obtained. It is noted that the method basically determines CMi as EL, but additional indicators may be added in order to obtain a refined CMi. In farms with a low number of cows and newly started farms (i.e. farms for which rather limited historical data is available) it is a challenge to produce a sufficient amount of data to create a valid local baseline for the farm. Until then a global baseline is used as a basis until sufficient data is available.

It will now be described how a suite of so-called "Global Healthy Baselines" can be derived according to preferred embodiments of the present invention : It is possible to define a "best" farm baseline by using all available historical data from a farm. In fig. 1 "best" farm baselines from five different farms are shown.

Although exhibiting the same general decay after calving, the curves differ substantially from each other both with respect to the level of the horizontal part of the baseline and the relatively rapid decrease in absLDH from calving day. To illustrate the principle imagine that e.g. a low, medium and a high fractile is calculated, day-by-day, from a large collection of such "best" farm baselines, representing a large variation in udder health. One would then obtain three dense, semi-smooth curves which can be defined as baselines representing low, medium and high degree of mastitis stress respectively. This concept can be refined so all fractiles between e.g. 5% and 95% in steps of 5 are calculated to give a suite of Global Healthy Baselines: GHB.n, where n denotes the fractile calculated. These baselines can be then be stored in a database, and for a newly started farm with only few measurements one can select the GHB.n closest resembling the current farm level. The farm level can be estimated e.g. as the 10% fractile of at least 100 point in the range DFC = 50 to 150. The selected GHB.n is the used as the current healthy baseline (CHB). When enough data is collected a local healthy baseline can be estimated using only local data and this will now be the current healthy baseline (CHB).

As much variation is present in the fast decaying first part of a farm baseline, an estimated global baseline (GHB.n) - for a given "new" farm - is, preferably, only used in the part which is substantially horizontal, e.g. after 50 days from calving.

The Chronic Mastitis index, CMi, may be according to preferred embodiments be determined as the ratio between the two parameters referred to as VeryStableM and CurrentHealthyBaseline: CMi=VeryStableM/CurrentHealthyBaseline.

As indicated in the terminology, CurrentHealthyBaseline is either a baseline selected from a selection of global baselines or a baseline fitted to local data. VeryStableM is a smooth curve, which is fitted to a time series of the already smoothed absLDH-values for a specific cow. Thus, CMi can be viewed as being determined for a specific cow with reference to a baseline selected to representing the level of the farm to which the specific cow belongs. It is, however, noted that in order to obtain a better value of CMi additional factors may be added to the above equation.

In the present content, a number of terms are used in a manner being ordinary to a skilled person. However, some of these terms are discussed below. absLDH is typically used to mean the product of LDH and milk yield, where LDH is lactate dehydrogenase measured in enzyme units and milk yield is measured in kg, that is absLDH is lactate dehydrogenase activity value [Mmol/minutes per liter milk] and milk yield is per cow per milking in litre. Thus, absLDH is the absolute amount of LDH in the milked milk portion of the specific cow.

Farm is typically used in the sense that a herd denotes the cows on a farm. A herd might be divided into one or more sub-groups. Farm is also used in the present context to refer to the characteristics of a herd on a particular farm - thus "farm" and "herd" are often used interchangeably herein to reference a particular selection of cows.

Days From Calving (DFC) is typically the time calculated in days for a particular cow since last calving. Baseline is on an conceptual level typically considered to be a curve that shows the basic (low) level of measurements sometimes showing positive deviations from the base level. The base level can be constant in time (CHB) or change with time (VSM). CurrentHealthyBaseline, which can be either a Global Healthy Baseline or a Local Healthy Baseline are typically a fit of an exponential function, e.g. absLDH = ai *exp(a ₂ *DFC) + as, to a (DFC, absLDH)-dataset. A preferred manner used to extract the (DFC, absLDH) data pairs used to estimate the coefficients in the function is given by an algorithm tailored to the specific situation ("Farm" or "Local"); this is described in details below. Other functions, which may be fitted to (DFC, absLDH) datasets is considered within the reach of the present invention.

"Best" Farm Baseline is typically a baseline calculated using all available data on a given farm. Together with similar baselines for other farms it forms the basis for the group of Global Healthy Baselines (GHB's). The "Best" Farm Baselines are, preferably, only used in the algorithm development phase and when the group of GHB.n should be renewed.

Global Healthy Baseline (GHB) is based on "Best" Farm baselines:

- GHB.n is the n ^th fractile of all "Best" Farm baselines. For each DFC, the n ^th fractile of absLDH-values across all available farms is calculated. This leads to a dense, semi-smooth (DFC, absLDH)-dataset, to which a smooth exponential curve is fitted. Typical values of n is [5, 10, 15, ... , 85, 90, 95]. Local Healthy Baseline (LHB) is typically calculated from a selection of the current farm absLDH-data available, e.g. the past 300 days based on e.g. DFC 2-150.

Current Healthy Baseline (CHB) is preferably the baseline currently selected to be applied on the farm. The Current Healthy Baseline is either GHB.n or LHB. It is noted that within the reach of the invention is considered a situation where no Current Healthy Baseline, CHB, can be selected; in which case the method of determining CMi according to the invention typically not is executed until such CHB can be selected. Enzyme Load (EL). As presented herein, the CMi is defined to be primarily dependent on EnzymeLoad (EL). EnzymeLoad typically describes the relation of VeryStableM to CurrentHealthyBaseline and is preferably calculated for each individual cow preferably every time a new LDH measurement is available. It is used to calculate CMi and may also potentially be used to calculate udder Stress index (USi), which describes the udder health in early lactation. As presented herein, CMi may, preferably, be equal to EL in situations where additional factors (see below) is not taken into consideration. Enzyme load, e.g. the relation of VeryStableM (VSM) to CHB, may be calculated in number of different way:

• VSM-CHB: This has the advantage that it may be easy for the farmer to relate to because it is presented in well-known units (absLDH)

• VSM/CHB: This has the advantage that is it similar to a Mastitis Detection index (MDi) providing a intuitive interpretation to the user (intuitive in the sense that l = baseline, while 2=double the baseline)

• (VSM-CHB)/CHB: This has the advantage that the value is standardized, enabling comparison between farms. The present invention aims at placing a cow within a class identified by its state of health. When defining such classes in a classification model it is important to take into consideration that the invention aims at long-term trend in individual cow udder health compared with the health status on the farm. According to this, the invention preferably makes use of three classes: Healthy, Silent and Chronic, which are discussed below with reference to preferences according to the present invention.

Healthy: A healthy cow is in the literature defined as a cow with no subclinical or clinical signs of mastitis. This definition is not fully comprehensive when defined as a healthy cow in the classification model, where we look at a cow's long-term health status (otherwise healthy people can catch a cold sometimes!). The acute cases of mastitis can be handled by a model as described in (Chagunda et. al; J. Dairy Sci. 89: 2380-2998, 2006 - cited above). A healthy cow is in the

classification model defined as a cow with no recurrent mastitis incidents and with a low LDH level in the milk compared to the base level at the farm. As described herein the Spike frequency may be designed so that a cow has to have 3-4 mastitis cases before the spike frequency contribute fully to the CMi. This is due to the fact that a small number of mastitis cases or dispersed mastitis cases are not enough to make a cow chronic. Likewise the increase in EL (dEL) has to be serious, before it gives rise to a change in the udder health class. A cow is in the classification model defined as healthy if CMi< 1.5.

Silent: A cow in the Silent group is defined as a cow in between the Healthy and Chronic groups. There are two reasons for a cow to be placed in the Silent group. First, the cow can have a minor increase in Spike Frequency, dEL or a high EL, but not high enough to be sure that she is truly Chronic. This type of Silent cow has a CMi between 1.5 and 2. Second, the Silent group is also used for cows with a sudden increase in CMi> 2, or more preferably CMi>2.5, because the cows cannot change directly from Healthy to Chronic, as explained in next section.

Chronic: A Chronic cow is defined as a cow with chronic udder infection, where the glandular tissue remains damaged, and bacteria are persistent in the mammary gland. These cows exhibit flare-ups of clinical signs, such as clots and flakes, or elevated LDH and somatic cell counts. These flare-ups are often related to stress and heats. A chronic cow is in the classification model defined as a cow with a high CMi. For a cow to be considered as Chronic it has to have a CMI>2, or more preferably CMi>2.5, over a certain time period as described herein.

Smooth is typically used as ordinary for the skilled person and includes the result and process of smoothing a data set to create an approximating function that attempts to capture important patterns in the data, while leaving out noise or other fine-scale structures/ rapid phenomena.

Curve as used herein, for instance in relation to Currently Healthy Baseline (CHB) being a curve representing relationship between absLDH and Day From Calving, is used to mean an algebraic formula being fitted to data (e.g. absLDH, DFC).

Preferably, the Current Healthy Baseline may be selected to be the Days From Calving, DFC, and parity dependant reference to which the baseline for the specific cow, (VSM) of a given cow may be compared. Preferably, the Global Healthy Baselines may be determined as a number of fractiles (GHB.n), where the n ^th fractile of the Global Healthy Baselines is the n ^th fractile of a plurality Best Farm Baselines, where each Best Farm Baseline may be a baseline for a specific farm calculated using all available data for the specific farm.

A LDH measurement for the specific cow may preferably be, preferably, performed frequently, such as once a day, and the baseline for the specific cow (VSM) may preferably be calculated each time a new LDH value is available.

In preferred embodiments of the invention, the baseline for the specific cow (VSM) may preferably be provided on the basis of exponentially weighted products of measured milk yield and LDH (absLDH) with the greatest weight on the lower values of absLDH.

Preferably, the Current Healthy Baseline may preferably be selected to be a Local Healthy Baseline being based on most recent datasets of absLDH and DFC, if at the time instant at which Chronic Mastitis index is to be determined :

· the number of measurements available for the herd, in the range of days from calving, DFC, between m and n ₂ , where n i and n ₂ are preselected integers, is above a preselected limit,

• the Local Healthy Baseline passes a validation test. Further, the Current Healthy Baseline may preferably be selected to be a most recent valid Local Healthy Baseline if at the time instant at which Chronic Mastitis index is to be determined :

• the number of measurements available for the herd, in the range of days from calving, DFC, between m and n ₂ , where n i and n ₂ are preselected integers, is above a preselected limit,

• a Local Healthy Baseline based on most recent data does not passes a

validation test. Preferably, the Current Healthy Baseline may preferably be selected from the Global Healthy Baselines determined as a number of fractiles (GHB.n) if at the time instant at which Chronic Mastitis index is to be determined :

• the number of measurements available for the herd, in the range of days from calving, DFC, between m and n ₂ , where n i and n ₂ are preselected integers, is above preselected limit (see e.g. fig. 4),

• a Local Healthy Baseline based on most recent data does not passes a

validation test, as will be disclosed in greater details below. In a particular preferred embodiments, the number of measurements in the range of days from calving DFC may preferably be for

n i equal 2 and to n ₂ equal 15 greater the 20 measurements,

n i equal 16 and to n ₂ equal 25 greater the 20 measurements, and

n i equal 26 and to n ₂ equal 50 greater the 50 measurements

Preferably, the validation test may be that the difference in calculated absLDH at two different days from calving is less than 1. Preferably, the two different days from calving is selected as days from calving such equal 350 and 100, preferably 300 and 100.

The Current Healthy Baseline may preferably be selected from the Global Healthy Baselines determined as a number of fractiles (GHB.n) if at the time instant at which Chronic Mastitis index is to be determined :

• the number of measurements available for the herd, in the range of days from calving, DFC, between m and n ₂ , where n i and n ₂ are preselected integers, such as 50 and 150 DFC, is above a preselected limit, such as 100.

Preferably, specific global baseline (GHB.n) closest to the level of the specific farm is selected as the current healthy baseline.

The Chronic Mastitis index (CMi) may preferably be determined as, or on the basis of, the ratio between

the baseline for the specific cow considered (VSM) and

- the Current Healthy Baseline (CHB), preferably with a contribution from additional factors.

The Chronic Mastitis index (CMi) may preferably be determined as, or on the basis of:

- the baseline for the specific cow considered (VSM) minus the Current

Healthy Baseline (CHB) (VSM-CHB), preferably with a contribution from additional factors.

Chronic Mastitis index (CMi) may preferably be determined as, or on the basis the baseline for the specific cow considered (VSM) minus the Current Healthy Baseline (CHB) divided by the Current Healthy Baseline ((VSM- CHB)/CHB)), preferably with a contribution from additional factors. In particular preferred embodiments, the Chronic Mastitis index (CMi) may be determined as the ratio between

the baseline for the specific cow considered (VeryStableM) and

the current healthy baseline (CHB)

and/or

- an a contribution reflecting

the spike frequency (f(SpikeFrequency)) in, and

change in enzyme load (dEL) of

the timely variation of absLDH of the specific cow. The Local Healthy Baseline, and Global Healthy Baseline may each preferably be a fit of an exponential function :

absLDH= a i*exp(a ₂ *DFC)+a ₃

to a dataset comprising corresponding values of DFC and absLDH. In such embodiments, the validation test may preferably comprise or may preferably further comprise the restriction that a i< 100, a ₂ <0 and a ₃ >0 and a ₃ <50.

Preferably, measurement of the LDH and the milk yield for the specific cow considered may preferably be carried out at least once a day and a new baseline for the specific cow considered (VSM) may preferably be provided each time a new set of LDH value and milk yield are available. In a second aspect, the invention relates to assigning a specific cow to a class by its state of health. In this aspect, the invention preferably relates to a method for assigning a cow, which previously has been assigned to a class identified by its state of health with respect to suffering from mastitis, wherein the classes are Healthy, Silent, or Chronic, to a different one of these classes.

Preferably, the method comprising determining, for a specific cow, a Chronic Mastitis index (CMi) by the method according to the first aspect of the invention, at a plurality of consecutive instances of Days From Calving, and wherein the cow is assigned to a different class than the one it is currently assigned to in the following manner:

if the cow is assigned the class Healthy and a preselected first fraction of Chronic Mastitis indexes, preferably 0.8, determined over a preselected first time, preferably 14 days, are above a first threshold, preferably 1.5, then the cow is assigned to the class Silent;

if the cow is assigned to the class Silent and a preselected second fraction of Chronic Mastitis indexes, preferably 0.8, determined over a preselected second time, preferably 14 days, are below a second threshold, preferably 1.5, and when used in determining chronic mastitis index,

the spike frequency (f(SpikeFrequency)) in, and

change in enzyme load (dEL) of the timely variation of absLDH of the specific cow is less than a third threshold, preferably 0.25, then the cow is assigned to the class Healthy;

- if the cow is assigned to the class Silent and a preselected third fraction of Chronic Mastitis indexes, preferably 0.875, determined over a preselected third time, preferably 21 days, are above a third threshold, preferably 2.5, then the cow is assigned to the class Chronic;

if the cow assigned to the class Chronic and preselected fourth fraction of Chronic Mastitis indexes, preferably 0.9, determined over a preselected fourth time, preferably 28 days, are below a fourth threshold, preferably 2.5, then the cow is assigned to the class Silent.

Preferably, any assignments of cows over time from the class Silent to the class Chronic, or vice versa, includes an intermediate assignment to the class silent. Initially, that is preferably prior to assigning a cow to a different ones of the classes, a cow may preferably be assigned to the class healthy. This aspect of the invention is particularly, but not exclusively, advantageous in that the present invention may be accomplished by a computer program product enabling a computer system to carry out the operations of the apparatus/system of the first aspect of the invention when down- or uploaded into the computer system. Such a computer program product may be provided on any kind of computer readable medium, or through a network. Thus, in a further aspect the invention relates to a system configured to execute the method according to the first aspect of the invention.

Further embodiments and aspects of the invention are presented in the following..

The individual aspects of the present invention may each be combined with any of the other aspects. These and other aspects of the invention will be apparent from the following description with reference to the described embodiments. BRIEF DESCRIPTION OF THE FIGURES

The invention and in particular preferred embodiments therefore will now be described in more detail with regard to the accompanying figures. The figures show one way of implementing the present invention and is not to be construed as being limiting to other possible embodiments falling within the scope of the attached claim set.

Figure 1 disclose various "Best" Farm Baselines from different farms as function of Days from calving and absLDH,

Figure 2 is a graph showing : absLDH smoothed, CurrentHealthyBaseline (CHB) and VeryStableM,

Figure 3 is a graph showing examples of GlobalHealthyBaselines, GHB.n, Figure 4 is a flow chart pertaining to a preferred embodiments and which details steps typically involved in determining a Current Healthy Baseline, CHB, for a particular farm, Figures 5-14 details determination of SpikeFrequency according to a preferred embodiment of the present invention, and

Figures 15-17 details determination of dEL according to a preferred embodiment of the present invention,

Figure 18 illustrates rules imposed on transfer of a cow between the classes "Healthy", "Silent" and "Chronic".

Figures 19 and 20 illustrates results obtained by the method according to the present invention for two different cows; the figures illustrates time wise development in absLDHssm (ssm means smoothed), CHB, UdderHealthClass which includes the classes Healthy, Silent, Chronic, Acute Alarm based on the model as described in A model for Detection of Individual Cow Mastitis Based on an Indicator Measured in Milk (Chagunda et al. ; J. Dairy Sci. 89: 2380-2998, 2006; - cited above - and CMi determined as CMi=VSM/CHB+additional factors as disclosed herein,

Figure 21 illustrates time wise classification into the classes, Healthy, Silent, Chronic on one farm (herd) for parity 1, 2 and 3+ respectively (3+ indicates all parities greater than or equal 3 are included in this group).

DETAILED DESCRIPTION OF AN EMBODIMENT

As disclosed above, a Chronic Mastitis Index (CMi) is designed according to the invention to provide a measure of the level of predicted chronicity of an individual cow based on the absLDH level and additional risk factors, i.e. CMi = EL + addRisk. Accordingly, cows with a high CMi are expected to have compromised milk quality and to have either clinical or subclinical mastitis.

Within a farm, the level of EL is relative; i.e. cows with the greatest EL are also expected to have the poorest milk quality and to pose a potential threat to the other herd mates due to the risk of pathogen transmission. Similarly, cows with a low EL are expected to have a good milk quality. It is noted that EL, as implemented in the preferred embodiments disclosed, does not take acute infections into account, by definition. Therefore compromised milk quality due to acute mastitis is preferably not assessed by EL and hence CMi.

On an overall level, EL is calculated by

EL=fl(LDH, milk yield)/f2(Parity, DFC) where fl is a mathematical function (or mathematical operator) and is labelled VeryStableM (VSM) and f2 is a mathematical function (or mathematical operator) labelled CurrentHealthyBaseline (CHB). Parity is a cow's parity and is either 1, 2, 3 (3 includes all parities >= 3), etc. Hereby EL may be written as

EL=VSM/CHB.

Although this relationship may provide good results it is preferred to take into account further additional factors than what is directly expressed through VSM and CHB. According, CMi may be calculated as:

CMi=VSM/CHB +additional factors = EL+additional factors

VeryStableM (VSM) is a smooth curve, which is fitted to a time series of already smoothed absLDH-values.

Determination of VeryStableM

As presented above, VeryStableM is a smooth curve, which is fitted to a time series of already smoothed absLDH-values. Where e.g. CurrentHealthyBaseline, CHB, (see below) is a general measure for the absLDH-values and theirs progression for a herd (group of cows), VeryStableM reflects a specific cow's progression (in time) of absLDH-values as function of DFC.

As disclosed herein, a LDH measurement for a cow is performed frequently and VSM is preferably calculated each time a new LDH value is available. The measurements of LDH times the milk yield (absLDH) are exponentially weighted with the greatest weight on the lower values of absLDH (as disclosed below).

The specific absLDH-values for a specific cow could be smoothed, e.g. by a applying the basic smoothing principle of a Kalman filter.

As VSM is used as a cow's specific baseline, it should preferably ignore flare ups of absLDH due to acute mastitis, but instead capture the long-term trend in the enzyme load (absLDH). For this reason EWMA (Exponential Weighted Moving Average) is modified to put the highest weights on the lowest absLDH-values

(irrespective of the measurement time) and lowest weights are put on the highest data values. For VSM the measurements of the past 14 days are preferably used, and with the parameters currently used, only the lowest points among these are - in practise - used with the weights: [0.3679; 0.1617; 0.0361; 0.0024]. An example of VSM is shown in Fig. 2 along with the corresponding absLDH profile and CHB.

Determination of a Current Healthy Baseline, CHB

According to preferred embodiments of the present invention, the Chronic Mastitis index is determined as CMi=VSM/CHB (+ additional factors), where the

CurrentHealthyBaseline is either a locally fitted or a selected global baseline aiming to reflect the level of absLDH expected in a healthy cow in that particular herd. As presented above, a challenge arises when attempting to fit a local baseline, namely the challenge of sufficient data material. In newly started farms it takes a while before a sufficient amount of data are measured to produce a Local Healthy Baseline, and until then another valid alternative must - or at least should - be available. A solution to this is to use a GlobalHealthyBaseline, GHB, until sufficient data is available.

The question as to when sufficient data is available may in many case be decided on the basis on experience and test and criteria exist which may be used to determine when a significant amount of data is available. The task of deciding on sufficiency of data is considered within the reach of skilled person. In connection with the present invention, it has been found that one single Global Healthy Baseline is not sufficient; however several global baselines can be generated from the Best Farm Baselines (cf. above introduction of Best Farm Baselines) defined by different fractiles. These global healthy baselines are determined on the basis of absLDH data obtained from a plurality of farms.

In accordance with the present invention, global baselines (GHB.n) have been defined in the range of f5%-f95% as shown in fig. 3. The global baselines (GHB.n) are based on "Best" Farm Baselines from a large selection of farms (in fig. 3: 82 farms in ca. 6 countries) representing expected future variability. The GHB.n-suite are preferably renewed once a year.

The specific global baseline (GHB.n) closest to the level of the farm considered can be picked based on relatively few measurements of absLDH in a farm and create a valid reference point quickly. The GHB.n which is closest to the 10% fractile of the at least 100 points between 50 and 150 DFC in the last max. 300 days is typically selected.

Reference is made to fig. 4 which is a flow chart pertaining to a preferred embodiment of the invention and which details steps typically involved in determining a Current Healthy Baseline, CHB, for a particular farm. In preferred embodiments, the Current Healthy Baseline is selected to be the Days From Calving, DFC, and parity dependant reference to which the VeryStableM, VSM, of a given cow is compared and used in the determination of CMi=VSM/CHB

+additional factors (the addition of additional factors is preferably optionally and may therefore be left out in the determination of CMi).

The Current Healthy Baseline, CHB, is based on local farm date from the past e.g. 300 days, if sufficient LDH data is available. That means that all the LDH measurements available for all the cows of the particular farm are used in the determination of a CHB. If sufficient data is available, a local Healthy Baseline, LHB, defined to be valid from DFC=2 is calculated. If sufficient data is not available, a fitted global reference (Global Healthy Baseline, GHB.n) is defined to be valid from DFC=50. This fitted global reference is applied until enough local farm data is available. Typically data from 2 to 150 DFC is included in the estimation of the local baseline, and data from 50 to 150 DFC is included in the estimation of the global baseline (GHB.n). Again, the question of deciding on sufficiency of data is considered within the reach of a skilled person.

As disclosed in the flow chart of fig. 4, the determination of a CHB is initiated with determining which data is available. This involves evaluation of the rule:

• is the number of measurement above predefined limit(s) in the period DFC: [m=2, n ₂ = 150]? If the result of this is "Yes", then a Local Healthy Baseline, LHB, is estimated based on a regression model : absLDH = ai*exp(a ₂ *DFC) + as based on fractiles in a moving window of the available (DFC, absLDH)-dataset for the specific farm considered. However, the thereby determined LHB must pass a validation test in order to determine whether it can be assigned to be a Current Healthy Baseline, CHB. The validation test aims at ensuring the validity of the LHB and is implemented as a criteria to be fulfilled. In a preferred embodiment, the criteria which must be fulfilled is:

1) Flat curve: absLDH(DFC=300) - absLDH(DFC=100) < 1

In fig. 4, the criteria is indicated by (1) in the box "Is estimated LHB valid? (ΐ '

It is noted that the numbers DFC= 100, DFC=300 as well as the limits for a i and a ₃ may vary between different embodiments. It is further noted that for determining the values in 1) above, the functional relationship absLDH = al*exp(a2*DFC) + a3 is evaluated as it is a test for the validity of the curve which is imposed. While the above sub criteria 1) is formulated as ⁿ absLDH(DFC=300) - absLDH(DFC=100) < 1 " other formulations may be used such as

absLDH(DFC=300) - absLDH(DFC=100) ~ 0". Further, it may be of relevance to consider the numerical value of absLDH(DFC=300) - absLDH(DFC=100) in the criteria, whereby the criteria may be written as abs(absLDH(DFC=300)- absLDH(DFC= 100))< l, where the operator "abs" refers to the numerical value.

If this further validation is positive, then the Current Healthy Baseline is assigned to be equal to the Local Healthy Baseline:

CHB=LHB

If, on the other hand, the estimated local healthy baseline does not pass the further validation, then a previous valid Local Healthy Baseline, if available, is assigned to be equal to the Current Healthy Baseline:

CHB=most recent valid LHB.

If no such valid Local Healthy Baseline is available, then the Current Healthy Baseline is assigned to be equal to a Global Healthy Baseline:

CHB=GHB.n where "n" is selected as described previously.

If none of the rules results in a Current Healthy Baseline, then it is decided that no Current Healthy Baseline is available, i.e.

CHB=NaN and no determination of a Chronic Mastitis index, CMi, can be established.

Thus, to summarise, the method determines a Current Healthy Baseline (if one fulfilling the various restrictions is available) and this Current Healthy Baseline is applicable for the farm in the evaluation of the enzyme load for each individual cow in the herd. The CHB is, preferably, updated once a day.

Determination of CMi - as VSM/CHB+additional factors

The above disclosure provides the basis for determining VSM and CHB. It is noted that while CHB is a farm baseline typically retrieved from a database, VSM is calculated, typically, when a new measurement is available for a specific cow. Thus, once a new measurement is available for a cow, CHB is selected according to the scheme above (if one is available), VSM is calculated and CMi is calculated as CMi=VSM/CHB. If additional factors are included in CMi, the magnitude of these are calculated and added to the ratio VSM/CHB to provide the CMi.

Additional factors

As presented herein, CMi may be determined from

CMi=VSM/CHB+ additional factors

Two such additional factors (or indicators) are: · Spike Frequency in absLDH (SpikeFrequency)

• changes in enzyme load, DeltaEnzymeLoad (dEL)

Cmi is thus calculated as

CMi=VSM/CHB+ SpikeFrequency + dEL.

These additional factors, SpikeFrequency and dEL, are based on the history of the specific cow.

SpikeFrequency

The purpose of the SpikeFrequency is to the describe the frequency of mastitis cases within a certain time period. Definition of a mastitis case

Before spike frequency (the frequency of cases) is defined, the definition of a mastitis case needs to be settled. In the present context, a mastitis case is preferably defined as

(1) one or more consecutive days, where the risk of acute mastitis

(OutRisk) is high, e.g. according to the mastitis model of Chagunda et al. ; wherein OutRisk = (absLDH(smoothed) - StableM)/35 where StableM is the sorted EWMA (Exponential Weighted Moving Average) of the last 7 points. The more "noisy" the risk profile (illustrated fig. 5), the more likely the cow is to be chronic; recurring cases throughout lactation is very typical for a chronically infected mastitis cow. In addition, a high level of noise in the middle band of the risk area (0.3 to 0.7) has proven to be another sign of a luring infection, even in cows that do not show a high level of full-blown alarms (Risk>0.7).

Based on this knowledge, the risk profile of a cow is very indicative of the cow's chronic status. To allow the model to automatically assess the cow's individual risk profile a rolling sum is calculated based on a set number of days back in time (NumDaysBack - selected by an administrator).

Because of the sampling pattern used, the number of measurements available per cow per day is not constant: healthy cows are measured less frequent than chronic cows or cows with an acute infection. This needs to be taken into account, when calculating the rolling sum based on a set number of days. If a cow is measured frequently, the sum of the risks in a given period of time will be higher than a cow that is measured less frequently, even if both cows are completely healthy. For this reason a fixed number of interpolated values is applied to ensure that the same number of measurements is available for every cow in that time period.

Fig. 6 shows an example of the interpolated values shown along with the actual measurement-based risk values. Bu using cubic interpolation it is possible to create a greater number of values to be used in the calculation of the normalized sum. For every DFC four interpolated values are calculated based on the actual measurement-based risk values. The Normalized sum is used to define the occurrence of a mastitis case and is defined as a rolling sum of the past 7 days (Num Days Back) interpolated OutRisks. To illustrate the method of calculating and using the normalized sum to define cases, a cow example is shown in fig. 7. This cow has a number of recurring mastitis incidences throughout lactation. The black triangles on the x-axis illustrate a Herd Navigator mastitis alarm.

First the OutRisk is used to calculate the interpolated OutRisks to fill out the gaps between measurements and to ensure that the number of measurements available for each DFC is constant. The OutRisk and interpolated OutRisk of the example cow is shown in fig. 8.

Next, a rolling normalized sum is calculated based on the interpolated OutRisks. The example cow's corresponding Normalized sum profile is shown along with normsum_limit in fig. 9. When the Normalized sum increases to above a set threshold (normsum_limit) that point in lactation is defined as a case (or a

"spike"). For the first 10 days, e.g., after a case has been registered, no new cases can occur, by definition. The reason for this is that acute alarms occurring within 10 days of the first registered case are expected to be a part of the same case. After 10 days a new case is allowed to occur if the Normalized sum (1) still exceeds the threshold or (2) increases to above the threshold yet again.

To summarize, cases are defined as:

· An increase of Normalized sum to above normsum_limit

• A new case cannot occur within 10 days of a previous case

• A Normalized sum which is still above normsumjimit 10 days after a

previous case Cases are utilized in the methods according to the present invention model with the basic concept that (1) the closer in time two cases occur, the greater the chance that they are related and (2) the more time has passed since the occurrence of a case, the less impact it has on the current status of the cow. For each case, the probability of relation to previous cases is evaluated using the sigmoid function shown in fig. 10a. By adjusting the parameters of the sigmoid function the output, which is the probability of relation in this case, can be adjusted. Note how adjusting just one parameter (flex), the curve is repositioned horizontally. Likewise, the steepness (the parameter "rate") of the curve can be adjusted to fit the desired output. The function applied in the model is tagged by an arrow in fig. 10a. Using these parameters, the probability of relation is very high for cases that occur 14-30 days apart, while the probability of relation for cases that occur with 60-70 days apart is rather low.

Second, the relative importance of the case is evaluated based on how much time has passed, since the occurrence of the given case. For example a case in the beginning of lactation is of little importance at the end of lactation, as it does not necessarily reflect the current udder health of the cow; on the other hand a case occurring within the past 30-40 days is of much greater relevance. For this reason, the value of each case is weighted relative to the days past since the given case using the sigmoid function shown in fig. 10b.

These two functions are now used to calculate the single contributions from each case at a given point in time. Meaning that the single contributions of each case is dependant on (1) days between cases and (2) the number of days passed since case.

SpikeFrequency is then defined as the sum of single contributions from each case.

Using the previous cow example, the principle of how the individual cases are utilized in the model for calculation of SpikeFrequency will be described in detail. In the upper plot of fig. 11, the Risk profile of the cow is shown (OutRisk) along with the automatically calculated cases ("spikes"). In the lower plot of fig. 11, the relative value of each case is shown. Each box corresponds to the relative contribution of a specific case.

In this example, the first case shown is the first case in lactation. First, the chance of relation to other previous cases is evaluated. As this is the first case in lactation, the only relation to be evaluated is the relation to itself. The chance that this case is related to itself is 1 by definition. Second, the number of days passed since the case is evaluated using the "Days past since last case" function (a sigmoid function). On the day of the case, full weight is applied and in the following days after the case, the weight will decrease according to days past since the case. The single contribution of this case is therefore the product of the probability of relation and the weight applied. This means, that if no other cases occurred after this case, then the SpikeFrequency would continue to decrease as time passes. When there is only one case to consider SpikeFrequency equals the single contribution from this case.

The upper plot of fig. 12 shows the progression in SpikeFrequency (the sum of single contributions) for this specific cow relative to DFC. In the lower plot of fig. 12 the slow decrease of SpikeFrequency in the days after the first case is shown to illustrate the effect of number of days passed since the case.

The second case occurs 22 days after the first case, which means that there is still a contribution from the first case. For the new case, the chance of relation to self is again 1. Because this is the second case in lactation, the probability of relation of this case to the previous is evaluated through the "probability of relation" function described previously. These cases occur very close to each other, therefore the chance of relation is still very close to 1 and the weight of the case is still high. The combined value of these two cases is close to 2.

Contribution of SpikeFrequency to CMi

The calculated Spikefrequency contributes to Chronic Mastitis index by use of a sigmoid function (see fig. 13). "Translating" the calculated SpikeFrequency to a relative contribution enables a greater level of control of how much a high number of cases is allowed to influence CMi and also how many cases is "a lot".

First, note the upper level of contribution. This upper level defines, that regardless of how great a SpikeFrequency is calculated, the contribution is never greater than 0.5.

Also note that single cases (SpikeFrequency= l) do not contribute at all, while a spike frequency of 3 gives almost full contribution. The reason for this is that a single case does not equal a chronic mastitis; it may simply be an acute case from which the cow recovers. However, within a time span of 70-80 days 3 cases is a high prevalence and this indicates a chronic mastitis.

To illustrate how SpikeFrequency influences CMi, the absLDH-, EnzymeLoad-, CMi- and SpikeFrequency profile of the chronic cow from the previous example is shown in fig. 14. Note how the contribution of SpikeFrequency (SpFr in fig. 14) is relatively low at the second spike and then increases to almost full load at the third spike. The CMi curve is shown without the contribution from dEL (to be explained below), so for the purpose of illustration CMi is the sum of only

EnzymeLoad and contribution from SpikeFrequency.

DeltaEnzymeLoad (dEL) - (changes in enzyme load)

The purpose of DeltaEnzymeLoad (dEL) is to capture long-term positive changes in EnzymeLoad of the individual cow as lactation progresses. This is done by using the lowest EnzymeLoad in lactation as an "anchor point" to which the steepness of the EnzymeLoad curve is evaluated. In the online calculation of dEL, the current lowest EL in lactation is applied. If a new lowest point is registered, then this will be applied in future calculations; meaning that a new reference point does not affect previous calculations of dEL.

It may be debated whether using the single lowest point as a reference in calculations is risky, because of the risk of bias from measurement outliers.

However, the values used in the shown calculations have been smoothed numerous times before entering this model; both in the calculation of smoothed values from raw values using the Kalman filter - the smoothed values that the mastitis model is also acting on - and in the calculation of VeryStableM. For this reason, it is considered safe enough to allow the reference point in these calculations to depend on a single value. DeltaEnzymeLoad is defined as: dEL= (ELcurrent-ELmin)/(DFCcurrent-DFCmin) where EL is calculated as VSM/CHB. Below, the principle of calculating dEL is shown based on a cow's EL profile. With reference to fig. 15 the reference point is marked with a triangle, while two random days in lactation (200DFC and 320DFC) are marked with a diamond and square, respectively.

For 200DFC the calculation of dEL is: dEL200 = (1.582-0.543)/(200-87.1) = 0.0092

The calculated dEL contributes to Chronic Mastitis index by use of a sigmoid function, see fig. 16, similar to the contribution of SpikeFrequency. "Translating" the calculated dEL to a relative contribution enables a greater level of control of how much a given increase of EL is allowed to influence CMi. Fig. 16 shows A sigmoid function is defined by the four parameters bottom, top, rate and flex.

By adjusting the parameters of the contribution function, the relative contribution of a given increase of EL can be adjusted. Four different sets of parameters with varying flex were evaluated; these are shown on in fig. 16. The final contribution function chosen is function number 1 (flex = 0.008) (fig. 16).

Similar to the contribution function of SpikeFrequency, the upper limit of the dEL contribution function is 0.5 to limit the maximum contribution.

To illustrate how dEL influences CMi a cow with a slow increase of EL is shown in fig. 17. This cow was infected with a Staph Aureus that never caused an acute alarm. Nevertheless, when evaluating the progression in absLDH and hence EnzymeLoad it is evident that the cow is not healthy.

Comments on Global Healthy Baseline

As discussed above, one of the advantageous features of the present invention is its capability of providing a measure for the enzyme load for individual cows eventhough sufficient data is not available for the farm to provide Local Healthy Baseline based on farm data as such. The advantage is provided by including absLDH measurements for a plurality of farms, determine Global Healthy

Baselines - not as a universal baseline - but as a plurality of Global Healthy Baselines (GHB.n) where that of these GHB.n's which fits the level of a specific farm (not necessarily included in the dataset used to defined GHB.n) is selected to be the Current Healthy Baseline for the specific farm until sufficient data is available to determine a Local Healthy Baseline based on local data only.

This has been found by inventors having access to a large number of datasets (absLHD, DFC) which have been acquired (for other purposes) through a number of years and stored in a database.

Until now, it has been found that only a universal baseline should or even could be established and any determination of enzyme load should be based on a comparison between such a single universal GHB and an actual measurement. However, extensive research done by the inventors has surprisingly resulted in the finding that use of a single GHB has at best a tendency to provide false indications as to the health state of a cow and that a Local Healthy Baseline advantageously can be used in combination with a plurality of Global Healthy Baselines, GHB.n

The invention can be implemented by means of hardware, software, firmware or any combination of these. The invention or some of the features thereof can also be implemented as software running on one or more data processors and/or digital signal processors.

Calibration of methods according to the invention

As presented herein, a method according to the invention is calibrated e.g. by specifying the intervals for DFC of fig. 4, specifying the number of data being sufficient for a curve fit etc. Such calibrations are typically carried out through empirical calibration in the sense that an administrator makes a number of educated guesses for the various parameters to be set and evaluates the operation of the method based on the selected parameters. If the administrator is satisfied with the performance of the method, these selected parameters are "locked" in the software and the software released for use.

This also implies, typically, that once the software is released for use, the user of the software has not access to change of the parameters. Implementation of the invention in a distributed farm structure

In a typical implementation, a central server stores data used for determining GHB.n and other parameters of the model which uses global data (contrary to local data which is data from a specific farm).

Each farm, applying the method according to the present invention, are connected via, e.g. an internet connection, to the server to access global data needed in order to determine CMi and to upload local data obtained from the herd of the farm. Typically, all calculations are performed locally at the farm, however, the calculations may be carried out at the server.

The individual elements of an embodiment of the invention may be physically, functionally and logically implemented in any suitable way such as in a single unit, in a plurality of units or as part of separate functional units. The invention may be implemented in a single unit, or be both physically and functionally distributed between different units and processors.

On classification of cows according to the present invention

As presented herein, an aim of the invention is to assign a specific cow to one of the classes Healthy, Silent or Chronic (see above). Transfer of a cow between classes is defined based on the CMi and, if used, the additional indicators of the cow. In the preferred embodiment, the transfer is governed by rules: i) a cow cannot be move directly from "Healthy" to "Chronic", she must always pass though the "Silent" class. This is shown in fig. 19 ii) a transfer of a cow from one class to another is not immediate: cows may not change classes frequently. It must be taken into account what the label "chronic" means to the management of the individual cow, and cows should not drop in and out of this class unless it is certain at least to a high degree that the cow is either Chronic, Silent or Healthy.

To make sure that cows do not readily move in and out of classes a number of limitations has been defined. These limitations is designed to make sure that the decision to transfer a cow from one class to another is based on a certain number of measurements within a certain number of days, and that based on these measurements the model insists that the cow should change group.

Transfer from "Healthy" to "Silent"

When considering the transfer from healthy to silent it is very important to take into account the sampling pattern of the mastitis model. Silent cows may be for example "climbers" i.e. cows with a slow increase in absLDH and these cows do not trigger an acute risk in the mastitis model. A cow without an acute risk is considered healthy by the mastitis model, which means that these cows may be sampled as rarely as every 2 ¹ /_-3 days.

In addition, when transferring a cow from the healthy to the silent class it is preferred to be reasonably sure that the cow is really no longer healthy. For this reason a certain amount of time must be allowed to pass, to ensure that the cow is truly no longer healthy.

Based on these considerations it is defined that before a cow is allowed to switch from the healthy to the silent class at least 4 measurements should be available within 14 days. In addition, at least 80% of the measurements should trigger class transfer.

Class transfer from Healthy to Silent is initiated when CMi> 1.5. This means that before a cow is allowed to switch class, her CMi must have been above 1.5 for at least 3 of 4 measurements the past 14 days. Similar to the transfer from Healthy to Silent it is preferred to take into account the LDH sampling pattern for cows both with and without an acute alarm. For this reason, at least the same requirement as for transfer from Healthy to Silent should preferably apply. In addition, the additional indicators alone should preferably be considered. For example a cow may have a CMi right below the threshold (1.5) allowing her to transfer from Silent to Healthy, but she might still have a rather high contribution from e.g. SpikeFrequency which indicates that she should not be moved. For this reason a requirement has been put on the additional indicators for transfer back from Silent to Healthy; the Additional indicators (f(SpikeFrequency)+f(dEL)) must be less than 0.25. Class transfer from Silent to Healthy is initiated when CMi< 1.5 and Additional indicators<0.25. The cow is not allowed to switch class until these two conditions have been fulfilled at least 80% of the measurements for the past 14 days. Chronic is a very serious label to put on a cow, and the possible impact of putting a cow into a chronic class must be considered carefully. Chronic cows are treated differently in daily management: these cows are less likely to be inseminated (hence more likely to be culled) and less likely to be treated with antibiotics in case of an acute mastitis case. In some farms, chronic cows are put into separate physical groups to prevent them from infecting other cows, which means that if they were not chronic to begin with, they will be after being moved into a group with other chronic cows because of the increased risk of infection.

Based on these considerations it is defined that before a cow is allowed to switch from Silent to Chronic at least 8 measurements should be available within the past 21 days. In addition, 90% of the measurements should trigger class transfer. Class transfer from Silent to Chronic is initiated when CMi>2.5. So in practice, before a cow is allowed to switch class from Silent to Chronic, her CMi must have been above 2.5 for at least 7 of 8 measurements the past 21 days.

The considerations to be made when moving a cow from Chronic to the Silent group are similar to the considerations made for cows transferring from the Silent to the Chronic group. Because of the significance of the Chronic "label", it is preferred that cows do not move in and out of this class readily. In plain terms: It must be difficult to move into the Chronic class, but it should be even more difficult to move out again.

Based on this, a cow should preferably not be allowed to switch from Chronic to Silent unless at least 10 measurements are available for the past 28 days and at least 90% of the measurements trigger a class transfer.

In practice, a cow is preferably only allowed to switch class from Chronic to Silent if her CMi has been below 2.5 for at least 9 of 10 measurements the past 28 days. Cow examples of class transfer

Based on the conditions listed above, the individual cows are allowed to transfer from class to class. In the following a number of examples are shown to illustrate the class transfer of individual cows (figures 19-20).

In the examples, UdderHealthClass=0 is Healthy, UdderHealthClass= l is Silent and UdderHealthClass=2 is Chronic.

In the first example, see figure 19, a previously healthy cow does not recover from a mastitis incidence at 135DFC, indicated by the fact that her LDH level does not return to normal level after the incidence. In addition, 23 days later, she gets another flare-up.

The next example, see figure 20, illustrate a cow that transfers back to the Healthy class. At the end of lactation (around day 285) the infected quarter is dried off prematurely and the cow moves from Chronic, through Silent and to the Healthy class.

On an application

An aim of the present invention is to provide a farmer with a tool that enables him to know the long-term udder health of individual cows based on daily LDH recordings. Knowledge on the long-term udder health of the cow first of all allows a more precise treatment of acute cases; for example a strong antibiotic treatment to previously healthy cows and a less aggressive homeopathic treatment to ease the symptoms of chronic cows. In addition, knowledge on the long-term udder health situation of the cow enables more strict management strategies; for example application of additional cleaning procedures after milking of chronic cows in herds or even physical isolation of chronic cows in one group to reduce the risk of transmission of pathogens to other cows. The dynamics of the present invention allows a very practical approach to evaluation of the cure of a previously healthy cow after an acute case on a more short-term basis in addition to the evaluation of the long-term udder health status of the cow. After an acute mastitis case the invention will move the cow to the Silent class if the LDH level of the cow indicates that she has not recovered. After an additional 21 days (e.g.) the cow may be moved to the chronic class, if her LDH level is so elevated that she qualifies for this. On the other hand, the cow will move back to Healthy from the Silent class if her LDH level decreases again.

The classification on herd level

In addition to using the invention to classify individual cows, the result obtained by the invention can be summarized to describe the current udder health situation on herd level.

By evaluating the distribution of cows within the three classes over time or relative to days from calving, the development in udder health over time on farm level can be optimized. Knowing how management changes affect the udder health at farm level may be difficult to assess, because management changes have long-term effect. Having access to the relative distribution of healthy, silent and chronic cows enables the farmer and his advisor to assess the effect of these management changes. Figure 21 illustrates an example fora farm (that is in this case a single herd) of the distribution of cows within udder health classes over time. In figure 21, the cows are classified into the classes Healthy, Silent and Chronic for parity 1 (upper part of fig. 21,) parity 2 (middle part of fig.21) and parity 3+ (lower part of fig. 21). Parity 3+ means cows with parity 3 or higher. The plot spans the period August 19 ^th to September 1 ^th . Thus, as points in the plot is made with reference to a particular date all the lactations stages (DFC) for the herd is represented at each date.

Although the present invention has been described in connection with the specified embodiments, it should not be construed as being in any way limited to the presented examples. The scope of the present invention is to be interpreted in the light of the accompanying claim set. In the context of the claims, the terms "comprising" or "comprises" do not exclude other possible elements or steps. Also, the mentioning of references such as "a" or "an" etc. should not be construed as excluding a plurality. The use of reference signs in the claims with respect to elements indicated in the figures shall also not be construed as limiting the scope of the invention. Furthermore, individual features mentioned in different claims, may possibly be advantageously combined, and the mentioning of these features in different claims does not exclude that a combination of features is not possible and advantageous.