Section 12 – Milk Analysis

Milk Analysis View

Field of application

These guidelines concern methods for fat, protein, lactose, urea and somatic cell count determinations in individual cow, goat and ewe milk. Milk samples are in most cases preserved with chemical substances. This will be taken into account in the procedures. These guidelines define:

1. Authorised reference methods.
2. Operation of validated routine methods.
3. Recommendations for controlling sample quality.
4. Recommendations for quality control of analyses.

Reference methods

The wording "reference methods" designates the methods used to calibrate the routine methods.

The reference methods should be internationally standardised methods (i.e. ISO, IDF, AOAC methods), although practical arrangements are permitted (see note below). The reference methods are listed in clause 8 (Appendix 2. Methods) below.

Note: Reference transfers

1. Rapid chemical methods can be used instead of a more time consuming reference method as far as results have shown to be equivalent to those from reference methods (i.e. Gerber method for fat, Amido Black method for protein, enzymatic method for lactose).
2. Master instruments may be used to produce "reference values" for other instruments and for other laboratories in case of a system with centralised calibration. Instrumental values may be considered equivalent to the values of the method used as reference for the calibration. Application of a centralised calibration concept must take into account sensitivity of routine methods to matrix effects (milk composition).

Routine (instrumental) methods

Routine methods should be methods which are fit for purpose on the basis of a performance evaluation by an expert laboratory and using a standardised protocol, or methods certified at the international level by ICAR. With this respect, conditions and procedure of evaluation, as well as requirements for ICAR certification, are defined in a standard protocol certified by ICAR (Procedure 1 of Section 12 of the ICAR Guidelines - as here) as relevant for the purpose of milk recording.

Specific recommendations for controlling the quality of DHI milk samples

Refer to Section 11 for guidelines on devices for collecting milk samples and sample sizes.

The quality of the sample is the first major requirement for a consistent analytical result. Good quality samples are a prerequisite to establish whether analytical quality requirements are met.

Bottles

In general terms, vials and stoppers must be suitable for their purpose (to bring milk without loss or damage to laboratories). For instance, a too large empty volume above the milk may facilitate churning during transport, especially with non-refrigerated milk. A too small empty volume above the milk may give rise to problems with mixing. Fat loss may occur with imperfectly tight stoppers.

Preservatives

Preservation of milk recording samples using chemical compounds should:

1. Maintain the physical and chemical properties of the milk during the period between sampling and analysis under the locally applicable temperature and transport conditions.
2. Not prevent from performing reference analysis, as the possibility of comparative analysis remains to the laboratories.
3. Have no effect on the results of analysis with the reference methods and no or only a limited but consistent effect on the reference method and routine method responses. A limited but consistent effect can be compensated for through calibration and/or applying a fixed correction.
4. Be innocuous to DHI and laboratory staff according to local health and safety regulations.
5. Be innocuous to environment according to local environmental regulations.

Notes

1. Sample preservation is promoted by working with clean milking and sampling equipment, by storage of samples at cool temperatures during limited time with a minimum of handling.
2. Appropriate preservatives are mentioned in relevant standards with guidance (ISO 9622 | IDF 141 and ISO 13366 | IDF 148). Nevertheless, in general care must be taken for:
   • the preservative excipient: depending on the excipient - generally salts - various effects can be observed for applied formulations where none exists in the pure form (case of potassium dichromate and bronopol in milk by mid infra red spectrometry);
   • some dyes which are used as colour tracer may interfere with the instrumental response (absorption of light or dye-binding with DNA). The accuracy or the sensitivity of a method may therefore be reduced. These dyes should be avoided.

Quality control in DHI laboratories

Quality control on reference methods

Any systematic error with the reference method leads to an overall systematic error on routine results. This type of error, which may exist between laboratories within a country (or organisation) and between countries co-operating within international frameworks such as ICAR, justifies performance evaluations at both levels, national and international.

External control

Every DHI routine laboratory should participate or otherwise be linked in with an interlaboratory proficiency testing (IPT) scheme. Proficiency testing should be organised preferably by a national reference or pivot laboratory appointed for that by the national DHI organisation. The reference laboratory will provide analytical precision traceability by its regular participation in international proficiency trials.

Note:

In situations where there are not sufficient laboratories to implement a national scheme, the laboratory can join PT schemes organised by a national or an international PT provider or the national DHI PT scheme of a neighbouring country.

The minimum frequency for participation in interlaboratory proficiency testing should be 2 times a year.

National reference laboratories should take part in international proficiency tests at a minimum frequency of twice per year. A more frequent participation is advised.

These trials are to be organised according to international standards, or failing that, international guidelines or agreements as indicated in this section.

Internal control

If available, reference materials (RMs) are advised for use to check the exactness and the stability of reference methods used by comparison with nominal values. They will be used preferably when reference analyses for calibration of routine methods are carried out.

These can be:

1. Certified reference materials (CRMs) produced by a recognised official organisation.
2. Secondary reference materials (SRMs) prepared by an external supplier.
3. In-house reference materials (IRMs) prepared by the laboratory itself, where traceability is established with CRMs, SRMs or interlaboratory proficiency tests.

Whatever the choice made by the laboratory, CRMs and SRMs are to be produced and provided in QA conditions and according to international standards, or failing that, international guidelines or agreements as indicated in clause 5.1.1 above.

Quality control on routine methods

Routine methods provide the results effectively used for DHI purposes and, therefore, their consistency has to be checked.

For this, reference is made to the standard ISO 8196-2 | IDF 128-2.

External control

A periodical check of the accuracy must be applied by an national expert laboratory, either through individual external control (IEC), by comparison of routine methods to reference analysis on samples representative of the laboratory area, or through participation in interlaboratory proficiency testing when it has been clearly demonstrated that a single calibration can be used for all the laboratories. In the latter case, recommendations in clause 5.1.1 above are to be followed. The minimum frequency recommended is 2 times a year.

Repeatability and suitability of calibration are the main parameters to be checked. Depending on the experimental design, additional aspects can be evaluated such as sample preservation and instrumental parameters such as linearity, intercorrections (with MultiLinear Regression (MLR)-based calibration models) and intra-laboratory reproducibility.

Internal control

Irrespectively of the parameter, an internal quality control on routine methods has to be carried out in routine testing at the laboratory.

In general the standard ISO 8196 | IDF 128 does not define limits to fulfil for each method and/or milk component. Therefore specific standards have to be applied where they exist:

1. Fat, protein, lactose and urea (mid infra-red spectrometry): ISO 9622 | IDF 141.
2. Somatic cell count: ISO 13366-2 | IDF 148-2.

Preparation of control or pilot samples, used for monitoring instrument stability, should be made under quality assurance (i.e. quality control for homogeneity and stability), thereby referring to relevant indications of international standards/guides for reference materials.

According to ISO 8196 | IDF 128, the major checks in quality control are on:

1. Repeatability.
2. Daily and short-term stability of instrument.
3. Calibration.

In addition, checkings are recommended for:

1. Carry-over effect (all methods).
2. Linearity (all methods).
3. Zero-setting (all methods).
4. Intercorrections (with MLR-based calibration models).
5. Homogenisation (infra-red).

It is advised to fulfil requirements about frequencies and limits as in clause 7 below.

Requirements for analytical quality control and quality assurance tools

Interlaboratory proficiency tests

Interlaboratory proficiency trials are to be organised in quality assurance conditions, according to international standards, or failing that, international guidelines or agreements:

1. ISO 17043.
2. ILAC-G13.
3. International Harmonized Protocol for Proficiency Testing of (Chemical) Analytical Laboratories (IDF Bulletin 342:1999).
4. ISO Standard 13528.

Reference materials

Reference materials used for DHI analytical purposes are to be produced in quality assurance conditions, according to international standards, or failing that, international guidelines or agreements:

1. ISO 17034
2. ILAC-G9.
3. ILAC-G12.

Choice of AQA service suppliers

Choice of Analytical Quality Assurance (AQA) service suppliers - i.e. proficiency testing and reference material - by DHI laboratories is to be made in tight relation with the overall DHI AQA system.

Services suppliers should operate under quality assurance and be able to provide documented proof of that.

Service suppliers should submit themselves to a periodical independent audit, i.e. a third party, in order to have the conformity of its QA system judged. These audits can be carried out by accreditation assessors, commissions of user representatives, experts acting on behalf of the national DHI national organisation, provided that their competence and independence are guaranteed and that the audits are conducted in line with ISO and ILAC recommendations.

Appendix 1. Analytical quality control in milk testing laboratories

It is to be expected that meeting these requirements will provide a satisfactory minimum quality level for analytical measurements, as well as comparability between laboratories and countries. If the following scheme cannot be immediately applied, it should be considered as a target.

Components of quality control and recommended minimum frequencies

Table 1. Components of quality control and recommended minimum frequencies.
Control	Frequencies	Mode
Reference methods
- External control	Half-yearly	IPT
- Internal control	Weekly (check of mean bias)	CRMs, SRMs, IRMs
Routine methods
- External control	Half-yearly	IPT/IEC
- Internal control	(see 1.1)	IRMs

IPT: Interlaboratory Proficiency Testing.

CRMs: Certified Reference Materials.

IEC: Individual External Control.

SRMs: Secondary Reference Materials.

IRMs: In-house Reference Materials.

Frequencies and limits for routine methods

Frequencies and limits stated hereafter in Table 2 are for a part defined in existing ISO | IDF standards or are derived from contained recommendations. Other values are indicative as they are not defined in a standard. Experience will show whether or not the latter ones are suitable for all laboratories.

Limits stated below in Table 2 are proposed as "action limits" for internal instrument management. They should only be considered as targets to users and not be used for external evaluations for which other (larger) values can appear more suitable.

Table 2. Minimum frequencies and limits for checking routine methods.
Checks	Frequencies	F P L Limits		SCC Limits
Instrumental fittings
Homogenization	Monthly	£ 1.0 % relative	(a)	Not applicable
Carry-over	Monthly	£ 1 % relative	(a)	£ 2 % relative	(b)
Linearity (curving)	Quarterly	£ 2 % of range	(a)	£ 2 % of range	(b)
Intercorrection	Quarterly	±0.02 % units	(a)	Not applicable
Calibration
Mean bias	Weekly	±0.02 % units	(c)	±5 % relative	(c)
Slope	Monthly	1.00±0.03	(c)	1.00±0.05	(c)
Overall daily stability
Repeatability (sr)	Daily/every	0.014 % units	(a)	6% relative	(b)
	start-up			at 150 000 cells/ml
				5 % relative
				at 300 000 cells/ml
				4% relative
				at 450 000 cells/ml
				3% relative
				at >750 000 cells/ml
Daily/short-term stability	³ 3/hour	±0.05 % units	(c)	±10 % relative	(c)
Zero-setting	³ 4/day	±0.03 % units	(c)	£ 8 000 cells/ml	(b)

(a):     Limit stemming from ISO 9622 | IDF 141

(b):     Limit stemming from ISO 13366-2 | IDF 148-2

(c):     Indicative limit as there is no value specified in corresponding international standards

Note 1:

In case calculated values are out of limits but do not differ from a statistical point of view, adjustments in instrumental settings are not justified. Therefore, representative and/or adequate sample sets should be used in such a way that any outside value is significantly different. Relevant aspects in this are type and number of samples, number of replicates and level of concentration.

Note 2:

1. Milk with high fat and protein concentrations (milk of buffaloes, ewes, and specific cow and goat species). Because of variable high fat and protein contents, reliable limits for repeatability and short-term stability can be determined by multiplying limits for cows by the ratio of buffaloes (or ewes) average level versus cows average level.
2. Goats milk: Limits can be the same as for cows milk in case of similar fat and protein content. In case of high fat and protein contents, one will operate according to a).

Note 3:

The criteria are calculated according the ICAR protocol on milk analyser evaluation and ISO 8196-3 IDF 128-3, available here.

Checking

1. Check on homogenisation (only applicable with IR instruments): In infra-red analysis, the natural size of fat globules strongly affects the measurement of fat, therefore a fat size reduction is applied through an homogenisation before the measurement. Inefficient homogenisation results in poor repeatability and drifts of the signal.
2. Check on carry-over: In case of successive samples with strong differences of component concentrations, the result for a sample may have been affected by the former milk sample, e.g. by the residual volume of milk in the flow system or by the contamination by the stirrer and the pipette. The error is a proportion of the difference of concentration with the previous sample. The overall carry-over effect should be minimised, should not exceed limits stated and can be corrected for in routine operation by applying carry-over compensation factors.
3. Check on linearity: Specific sets of samples are prepared in order to cover the whole range of concentration and check that the instrumental measurement is proportional to the concentration of the component measured. The percentage of the bending can be estimated by the ratio (range of the residuals observed) x 100 /(range of the levels).
4. Check on intercorrections (with MLR-based calibration models): Specific sets of samples are prepared in order to create independent modification in respective components and verify that changes in one particular component do not affect significantly the measurement of the other components. Intercorrections are set in order to compensate the natural interactions due to a incomplete specificity of methods. The larger the range of concentrations of the correcting channel, the bigger the potential error due to an inadequate intercorrection adjustment for the corrected channel.
5. Check on the mean bias: Representative milk samples are used to check the validity of the calibration at a medium level and to indicate whether any drift has occurred due to changes in milk composition or progressive wear of instruments.
6. Check on the slope: Specific sets of (calibration) samples are prepared in order to cover the whole range of levels and check that the slope is within the stated limits. The larger the range of concentrations, the bigger the error for extreme values in case of an inadequate slope adjustment.
7. Repeatability: A repeatability check is indicating whether or not the instrument is working stable. Repeatability is evaluated at the start-up of each instrument on the basis of 10 times replicate analysis of one (control) milk sample. During routine testing a regular test can be made by replicate analysis of the control sample. The estimate of the standard deviation of repeatability should meet stated limits.
8. Daily and short-term stability: Every day and regularly along a working day, the so-called control samples (or pilot samples) are used to check instruments functioning at one or more concentration levels. Differences observed against assigned values should not exceed the stated limits +/-L. It is advised to complete the control using the calculation of the cumulative mean of the n successive differences which should not exceed the limits +/-L/√n, see ISO 8196|IDF 128.
9. Zero-setting: Rinsing the flow system and checking the "zero value" are periodically required to check for fouling on the walls of the measurement cells and/or (depending on instruments) to detect any drift of the basic signal.

Appendix 2. Methods

International reference methods

Table 3. International reference methods.
Fat
Gravimetry (Röse-Gottlieb)	ISO 1211 | IDF 1
Gravimetry (modified Mojonnier)	AOAC 989.05
Crude (or total) Protein
Titrimetry (Kjeldahl)	ISO 8968 | IDF 20, parts 1 and 3
AOAC 991:20
AOAC 991:21
AOAC 991:22
AOAC 991:23
Casein
Titrimetry (Kjeldahl)	ISO 17997 | IDF 29, parts 1 and 2
AOAC 998.05
AOAC 998.06
AOAC 998.07
Lactose
HPLC	ISO 22662 | IDF 198
Urea
Differential pH-method	ISO 14637 | IDF 195
Somatic cell count
Direct microscopic somatic cell count	ISO 13366-1 | IDF 148-1

Secondary international reference methods

Table 4. Secondary international reference methods.
Fat
Butyrometric method (Gerber)	ISO 19662 | IDF 238
	AOAC 2000.18
Babcock	AOAC 989.04
Protein
Dye-binding (Amido Black)	AOAC 975.17
Lactose
Enzymatic	ISO 5765 | IDF 79
	AOAC 984.15
Differential pH-method	ISO 26462 | IDF 214

Standardized routine methods

Table 5. Standardized routine methods.
Fat
Automated turbidimetry I	AOAC 969.16
Automated turbidimetry II	AOAC 973.22
Protein
Automated dye-binding (Amido Black)	AOAC 975.17
Fat-protein-lactose-urea
Mid infrared (MIR) spectrometry	ISO 9622 | IDF 141
	AOAC 972.16
Somatic cell count
Fluoro-opto-electronic methods	ISO 13366-2 | IDF 148-2
	AOAC 978.26

Procedure 1: Protocol for evaluation of milk analysers for granting ICAR certification

Foreword

The present protocol has been produced by the Working Group on Milk Testing Laboratories.

Though various standards or normative documents already treat the subject of the evaluation of instrumental or indirect or alternative methods, there are as yet no documents with sufficient practical indications on the way to execute, and on the specific technical requirements to fulfil, in the evaluation of analytical routine methods for the particular aspect of the Certification for milk recording by an (official) international body such as ICAR.

Therefore, it is the aim of the present document to define an overall procedure starting from the request for the certification, the procedure for the certification, the description of the technical evaluation needed, providing at the end the elements for a decision on certification.

The present document complies with ISO Standard 8196 (equivalent of IDF standard 128) and will concern milk of various species within the scope of ICAR (cows, goats, ewes, buffaloes) and the various components of interest for milk recording (fat, protein, lactose, somatic cell count, urea).

Introduction

Before being used for milk recording, a new analytical method or new equipment is to be submitted to an evaluation and must be approved for use by a competent body. At present, evaluations are carried out individually with, as a consequence, possible multiplication of evaluations in numerous countries. Moreover, the absence of a common protocol for such evaluations can result in incomplete and inaccurate technical information and numerous reports with non-comparable or partly comparable results.

The objective of this protocol is to define all relevant analytical parameters to be evaluated, providing respective limits to comply with in the relevant ranges for various animal species.

On the basis of this protocol, a limited number of evaluations should suffice to decide about an international certification on common ICAR rules for the application of analytical methods and/or equipment in milk recording.

Rules of the certification

Stages of the evaluation and general principles

1. Phase I: Every new instrument will be evaluated in specific conditions of test bed, within the period of time necessary to assess all the technical requirements prescribed in the present protocol. This part of the evaluation must be carried out by an expert laboratory specialised in analytical evaluations as well as experienced in (the) reference method(s) required. This laboratory should be accredited for this activity or be recognised as competent for this task by a competent body (national milk recording organisation and/or ICAR).
2. Phase II: The second phase of the evaluation starts after having succeeded with the first one. At least two new instruments will be used for a two-month period of observation in routine conditions in two different milk recording laboratories. They should fulfil the day-to-day quality control and satisfactorily respond to general convenience needs.
3. National certification: Request for an evaluation should be brought by manufacturers (or suppliers) to an official organisation (i.e. national milk recording, ministry, etc) who should appoint the laboratories to be involved in the evaluation and would give them an assignment for the work. Reports of both phases I and II will be examined by an official committee. Then, on the basis of technical reports produced by laboratories, a national certification can be pronounced.
4. International certification: For an international certification by ICAR, the total evaluation should be renewed successfully in three ICAR countries and on similar bases as defined in the protocol. Collation of reports and the request for ICAR certification should be made by manufacturers to ICAR. Milk analyser files will be submitted to the relevant ICAR Sub-Committee (Milk Analysis) for technical advice to the Board.

Then the ICAR board will pronounce itself about the request for certification.

Field of validity of the certification

An certification is given only:

1. For the field of application where the instruments has been evaluated (component, concentration range, animal species, etc.)
   • In case milks of different animal species are to be analysed, specific evaluations for every species concerned have to be carried out to assess that the instrument is appropriate for the expected use. Refer to Table 6 for species specific component ranges.
   • In case of breed with unusual milk fat and protein contents (i.e. Jersey breed with high fat and protein contents), the evaluation should be carried out within the same component range with milk of the specific breed.
2. For the specific instrument configuration used during the evaluation.
   • In case of configuration changes, the proof should be brought that it does not affect the precision and the accuracy beyond acceptable limits.

Animal species and particularities of configuration(s) assessed should be carefully noted in the evaluation report.

Table 6. Indicative milk component ranges at least to be covered by an evaluation.
	Cows	Goats	Ewes	Buffaloes	Units
Fat	2.0 – 6.0	2.0 – 5.5	5.0 – 10.0	5.0 – 14.0	g/ 100 g
Protein	2.5 – 4.5	2.5 – 5.0	4.0 – 7.0	4.0 – 7.0	g/ 100 g
Lactose	4.0 – 5.5	4.0 – 5.5	4.0 – 5.5	4.0 – 5.5	g/ 100 g
Urea	10.0 – 70.0	10.0 – 70.0	10.0 – 70.0	10.0 – 70.0	mg / 100 g
Cells	0 – 2000	0 – 2000	0 – 2000	0 – 2000	103 cells/ml

Course of operations of a technical evaluation:

Introduction to the principle of the evaluation (explanatory note)

Whatever the indirect method is, a standard measurement processing can be presented by the scheme in Figure 1. Each step does not necessarily exist in every instrument. This depends on manufacturers choices in relation to the principle of the measurement and the component measured – for example little or negligible effect (for instance step 3 in somatic cell count in cow’s milk) - or in some case can be merged (for instance, steps 2, 3 and 4 in particular infra-red devices). Nevertheless, in theory the different steps of the signal process can be set up in the instrument and remain available to be activated or not, through active or neutral mathematical matrices. On the other hand, interactions of major components or carry over effect can be eliminated by the method or the physical device (physical treatment, chemical reagents, tube length) and therefore no longer need numerical corrections

Figure 1. Example of a theoretical measurement process in conventional analysers. Every step of the measurement process corresponds to an element of the breakdown of overall accuracy of the method. Minimising the overall error is achieved through minimising every component thereby optimising every step of the measurement process. Then the experimental design for the evaluation of a milk analyser is defined in order to assess that every measurement step is correctly adjusted.

1 Measurement : Zero/blank, repeatability, stability, reproducibility.

2 Amplification : Sensitivity, measurement lower limit ; repeatability.

3 Linearisation : Linearity range ; upper limit; accuracy.

4 Interactions : Effect of other milk components ; accuracy.

5 Calibration : Suitability of manufacturer calibration system ; accuracy.

6 Carry over effect : Effect of previous milk intake ; repeatability, accuracy.

Every step of the evaluation described in the following paragraphs can be required to fulfil appropriate limits for each analytical criteria (component) before starting up the next step.

Minimum necessary assessments for an evaluation

This part defines and describes the elements of the evaluation which are compulsory to evaluate.

Whatever the method and precision element assessed, an evaluation is to be carried out from test results displayed expressed in standardised units and no prior data transformation should be performed (e.g. log or square root for somatic cell counting). Evaluation results should comply with specifications stated in the following paragraphs.

Assessment of preliminary instrumental fittings

Before starting any further assessment, one has to verify basic criteria that indicate a proper functioning of the method or the instrument. These criteria are daily precision (including repeatability and short-term stability), carry-over and linearity.

Daily precision (repeatability and short-term stability)

Basically, a milk analyser should present a signal stability    which complies with the precision requirements. If not, the analyser is either in dysfunction (and should not be used) or its precision is not suitable for the objective of the analysis. Therefore, the instantaneous stability (repeatability) and the signal level stability have to be assessed prior to any other parameters.

Along a whole day period and every 15-20 minutes, analyse a same milk sample in triplicate by the instrument without any change in the adjustment of the calibration in order to obtain a minimum of 20 check test series. It should be preferably operated in as close as possible conditions as routine. Therefore, sufficient number of samples should be planned to keep the instrument running between the periodical checks.

The precision will be evaluated at three different concentrations of each component, low, medium and high. To achieve this three different milk samples can be split in as many identical sub-samples as necessary for the analyses.

Using a one-way ANOVA, calculate the estimate of the standard deviation of repeatability (Sr), the standard deviation between check series (Sc) and the standard deviation of daily reproducibility (SR), referring to Appendix 1 (section 6 on page 19):

${\text{SR}}=({\text{Sc}}^{2}+{\text{Sr}}^{2})^{1/2}$

The values Sr and SR obtained should comply with the limits stated for milk recording analysis (see Table 2 and Table 3).

One can check the significance of the non-stability using a F-test. Alternatively, a one-way analysis of variance can be carried out to confirm the non-stability of signal.

Carry-over effect

Strong differences in component contents between two successive milk samples analysed may influence the result of the latter one. It can happen because of an incomplete rinsing of the flow system and the measuring cell by milk circulation and/or a contamination of the former sample by the stirring device. The overall carry-over effect (including both sources of error) will be evaluated on the one hand and the rinsing efficiency of the flow system on the other hand.

Automated analysers often allow to apply on-line corrections to compensate the overall carry-over effect when necessary, therefore:

1. Rinsing efficiency of the flow system must be assessed by running tests without any correction (correction factor fit to zero) in manual mode (bypass the automated stirrer). Rinsing efficiency should not be less than 99 % or the internal carry-over should not exceed 1 %.
2. Overall carry-over effect will be assessed including the correction factors either set in the instrument or obtained using the method supplied by the manufacturer. It should not exceed the values stated for the component for milk recording purposes.

Method

1. Analyses Replicate as many times (n) as necessary the analytical sequence (LL,LL,LH,LH) where LL is a low component concentration sample and LH is a high component concentration sample.
2. Samples
   • Sufficient number of sub-samples of each sample L_L and L_H must be prepared prior to analysis in order to analyse each sub-sample only once.
   • L_L and L_H should preferably be milks or liquids of similar viscosity as milk.
   • Respective component concentrations must differ considerably. Depending on the component and the method, this can be achieved by using natural separation (creaming for fat), artificial separation (ultra-filtration for protein, micro-filtration for somatic cells) or addition (lactose and urea).
   • For biochemical component determinations, concentrations of L_L and L_H should better be extreme values in the measuring range. At the contrary, for somatic cell count, one will assess the carry-over for three different high cell contents (500, 1000, 1500 10³ cells/ml) and a single low cell content, preferably a zero-cell milk.
3. Calculation
   • Calculate the mean and the standard deviations of the differences d_Li = L_1i-L_2i and d_Hi = H_2i - H_1i, respectively , Sd_L,  _H, Sd_H.
   • calculate the mean difference of concentration

     

Then carry-over ratios C.O.R. and their standard deviations SC.O.R. are obtained using the following formulas:

As well, C.O.R. can be obtained by the equivalent formulas:

The two values obtained should not significantly differ from each other and should not exceed the limit (L_c.o.r.) stated for the component.

Note

1. Acceptable limit for conformity: At the worst, the carry-over effect should not produce in the extreme case of lowest and highest concentration of the measuring range (ΔC) an error higher than the repeatability admitted for the method r=2.√2.Sr. Therefore, the limit of c.o.r. can be defined as:

${\text{Lc.o.r.}}=\left({\frac {r}{\Delta C}}\right)\times 100$

A 1-2 % limit is generally recommended in standards.

1. Number (n) of analytical sequences: It can be defined in order to allow to estimate C.O.R. values with a ± 20 % maximum relative confidence interval (i.e. 1±0,2 %). Thus 2. SC.O.R. ≤ 0,20 . (C.O.R.)  

   

   
   Between 10 and 20 analytical sequences are generally recommended in standards.

Linearity

According to the classical definition of an indirect method, instrument signal should result from a characteristic of the component measured, thereby allowing to define a simple relationship with component concentration.

Nevertheless, newly developed indirect methods can be based on much less specific signal, still providing consistent results from multiple signals through multivariate statistical approaches. For these latter analysers linearity is no longer an absolute requirement in every case (though it must be in some specific utilisation of dairy industry, i.e. on processed milk with progressive contents stemming from concentration or dilution). Since then, for those methods and depending of analytical objectives, the step of linearity assessment can be discarded. The quality of the relationship with reference will be assessed in evaluating overall accuracy. In such a case, any routine measurement outside the calibration concentration range should be considered of doubtful quality and preferably not be used.

Linearity expresses the constancy of the ratio between the increase of milk component measured and the corresponding increase of the instrument measurement. Therefore linearity of the instrument signal is in most cases essential to maintain a constant sensitivity along the measuring range and to allow easy handling of calibration and fittings. Moreover, it allows in routine (to some extent) measurements beyond the concentration range of calibration through a linear extrapolation of calibration within the assessment range. Since then it can help to cope with possible particular limitations of reference methods (e.g. somatic cell count for goat’s milk).

It can be assessed using sets of (n=8 to 15) samples with component concentrations regularly distributed all over the measuring range:

1. Samples should preferably be milks or liquids of similar physical characteristics (i.e. density, viscosity) as milk obtained by accurate dilution (weighing) of a high content sample by a low content one.
2. Concentrations should vary in regular intervals. Depending on the component, this can be obtained using various ways such as natural separation (creaming for fat), artificial separation (ultra-filtration for protein, micro-filtration for somatic cells)and pure solutions (lactose and urea).
3. Assessment concentration range should be at least the ones stated in Table 1, §2.2. Nevertheless, it is up to the evaluator to extend linearity assessment range in order to determine the upper limit for acceptable measurements.
4. Reference for linearity will be either the volume mixing ratio (volume/volume or mass/volume) or theoretical concentrations calculated from the concentrations of the initial samples (one can refer to Annex A of IDF Standard 141).

Note

Independently of expression units, reference for linearity should be according to the intake measurement principle, that is volumetric in all milk analysers developed till today, at the opposite of milk weighting quite impracticable. Since then the theory would require volume/volume or mass/volume ratio.

Nevertheless, using mass/mass ratios provides identical figures when mixing liquids with the same density.

Analyse each sample in triplicate, first in the order of increasing concentrations, second in the order of decreasing concentrations and calculate the linear regression equation y=bx+a (y=instrument, x=dilution ratio) and the residuals ei (ei=yi-(bxi+a)) from the means of replicates and dilution rations. Plot the residuals ei (y axis) versus the dilution ratio (x axis) on a graph. A visual inspection of the data points will usually yield sufficient information about the linearity of the signal.

Calculate the ratio of the residual range to the signal values range:

${\frac {De}{DC}}={\frac {e_{max}-e_{min}}{C_{max}-C_{min}}}$

where:

e_max and e_min = the upper and lower residuals, respectively

C_max and C_min = the upper and lower signal values, respectively

DE/DC should not exceed the limit stated for the component (generally 1-2 %):

Criteria	F	P	L	Urea	SCC
Limits for De/DC	0.01	0.01	0.02	0.02	0.02

Alternatively, a one-way analysis of variance can be carried out to confirm the statistical significance of non-linearity and statistical tests of comparison of variances can be applied to confirm the significance of difference between residual variances (see Annex).

One way is to calculate polynomial regressions with a progressive increase of the degree to determine the most appropriate adjustment of the signal that is, providing minimum standard deviation Sy,x^k (the degree of the polynomial should better not exceed 3 with significant coefficient) and to compare the estimate Sy,x^k with Sy,x of linear regression on the basis of significant ratio or F-test.

The final judgement on linearity adjustment of instrument is:

• Good              if the value Sy,x ≤ Sy,x^k
• Correct          if Sy,x > Sy,x^k and DE/DC ≤ limit%
• Incorrect      if Sy,x > Sy,x^k and DE/DC > limit%

Using the statistical test for comparison of residual variances or standard deviations (see Appendix 1: Usual statistical formulas for method evaluations on page 19).

Measurement limits

Limits of an instrumental method measurement exist at both extremities of the analytical range, lower limit and upper limit.

It is not required to determine these limits in case where natural concentration ranges for the respective components and species are normally located far from zero (general case for biochemical components, i.e. fat, protein, lactose, urea) and within the range of linearity of the method. Determination and assessment of measurement limits are carried out with the evaluation of linearity.

Lower limits

Lower limits evaluation is not treated by ISO 8196 (equivalent IDF 128) therefore reference can be made to standard EN ISO 16140:2000, which is dedicated to alternative microbiological methods, for definition and general principles.

At the date of the redaction of this document, only somatic cell counting is concerned by a lower limit evaluation for milk recording.

Definition

Lower limits are defined in three ways depending on the risk of error accepted and a priori precision requirements:

•      Critical level (CL) or decision limit which is the smallest amount which can be detected (not null), but not quantified as an exact value (risk b =50 %). Below it cannot be assumed that the value is not null:

CL = u1_-a . s or CL = 1.645 . s      with a = 5 %                         (1)

•      Detection limit (DL) for which the second type of error is minimised up to a defined level, generally equal to the level of risk s (5 %). It consists in the lowest result, which differs significantly from zero (first type error a), that can be produced with a sufficiently low probability (second type error b) of including the blank value (zero) and with a sufficient confidence interval

DL = (u1-α + u1-β) . σ    or    DL = 3.29 . σ      with α = β = 5 %         (2)

•      Quantification limit (QL) or determination limit which is the smallest amount of analyte which can be measured and quantified with a defined relative standard deviation SD% (or coefficient of variation CV%):

QL = kq . s and SD% = s /QL => kq = 1 / SD%

                                       QL = DL => kq = 3.29 => SD% = 30 %                                         (3)

Limit values to fulfill

In somatic cell counting, DL of cell milk counters should not be higher than 5000 cells/ml and SD% (CV%) at the lower level (close to zero) should not exceed 30 %, with QL equal to DL.

Standard deviations

In milk recording analysis, where only single determinations are carried out in routine, s is the standard deviation of random error of the measurement that is, in the best case, the repeatability standard deviation at the proximity of zero content.

Standard deviation s can be estimated in different ways:

• Repeatability is dependent on concentration levels: standard deviation of repeatability (Sr) of the blank (zero) or estimated standard deviation at concentration values close to zero;
• Repeatability is not dependent on concentration levels: standard deviation of repeat- ability (Sr) estimated by taking benefit of replications at different levels in linearity assessment,
• Repeatability and sample variance are not dependent on concentration levels: standard deviation Sy₍₀₎ of the single estimate y₍₀₎ for x=0 using linear regression equation calculated in a linearity assessment in a linear part close to zero:

Sy₍₀₎ = S_y,x. (1 + 1/q +  / SCE_X)^1/2

Note

In that case, Sy₍₀₎ slightly overestimates σ as it takes into account sample errors and line estimation error in addition to repeatability.

Upper limit

Upper limit corresponds to the threshold where the signal or the measurement deviates significantly from linearity (cf. linearity assessment).

An upper limit met on the range of concentration concerned by the evaluation will produce a ratio De/DC exceeding accepted limits (see linearity). Plotting linearity assessment results on a graph will provide necessary information on the shape of the curve response.

One can check if measured upper values deviating from linearity yU differ significantly from y(xU) which should be obtained with the linear equation (prediction) calculated on the linear range without taking into account that result:

tobs = | y_U – y_(xU) | / Sy(x_U)

with S y(x_U) = Sy,x . (1 + 1/q + (x_U -)2 / SCE_X)^1/2 and q-2 d.f. and α = 0,05

·      if t_obs ≤ t_1-α/2   =>   no deviation from linearity at that point

if t_obs > t_1-α/2   =>  significant deviation from linearity at that point

Evaluation of the overall accuracy

One can refer to IDF standard 128 for a general information of this part of the evaluation.

The overall accuracy is composed by the sum of repeatability error, accuracy (or error of estimates versus reference) and error of calibration which occur in routine analytical conditions.

Each part of the overall accuracy is measured through the analysis of individual milk samples and herd milk samples of the specified animal species. Herd milk samples are to be collected in addition to individual milk samples in order to measure more accurately the part of variance related to herd effects.

The evaluation is to be performed on the instrument in the same state (working parameters, speed, calibration) the manufacturer intend to provide customers (users) with.

In case different analytical speed are available, parts of the overall accuracy will be assessed for the higher and the lower ones.

a.      Calibration

A preliminary calibration (or pre-calibration) is required and should be set in the instrument (or supplied with it) by the manufacturer with a detailed calibration procedure appropriate to the instrument.

In case the instrument is to be used directly without any local calibration (set-in), instrumental analyses of the evaluation will be directly performed on appropriate (representative) milk samples.

In case local calibration is necessary, prior calibration will be performed according to manufacturer recommendations and using instrument facilities, before starting up the evaluation.

b.     Samples.

Milks have to be sampled and collected in optimum conditions such as no damages should occur and could produce erroneous repeatability estimate. Individual milks should cover the maximum concentration range of the component according to Table 1.

-       Calibration samples. They will be samples prepared according to recommendations of relevant standards for the criteria or, if no standardised procedure exits, in a similar way as prediction samples (half part for calibration and the other part for prediction).

-       Prediction samples. Minimum numbers of 100 individual milk samples collected in 4-6 different herds and 50 herd milk samples should be used.

c.      Reference methods

Reference methods should be standardised methods and, in all cases, the method used should be in a close agreement with one or more of the international reference methods (ISO, IDF, AOAC).

Assessment of repeatability

Repeatability is the main criteria which indicates whether an instrument allows suitable results according user requirements or not and it is a major element of internal quality control. Therefore every new instrument has to fulfil a maximum limit for repeatability value stated in the relevant international standard in order to satisfy to certification criteria.

Milk samples are to be analysed on the instrument calibrated according the manufacturer recommendations, preferably in duplicate. Indeed this minimum replicate number keeps closer to true conditions of repeatability and prevents from possible damage on fat. Series of 15-20 milk samples are successively analysed twice after recovering initial analytical condition (i.e. temperature by heating) when necessary.

Then standard deviation of repeatability will be calculated from duplicate results obtained from the whole set of data and, for criteria covering a wide range of concentration –that is more than 1 log scale- (case of somatic cell count), part-by-part after splitting of the whole concentration range in different parts, three parts for the minimum (i.e. low, medium and high).

The standard deviation of repeatability will be calculated with the formula of IDF Standard 128 (see Appendix 1: Usual statistical formulas for method evaluations on page 19):

Sr = ( ∑ w_i² / 2_q )^1/2

where w_j is the difference between duplicates of sample i (w_i = x_1i – x_2i) and q the sample number.

Compare the values obtained (Sr) with the standardised repeatability values (σr) defined for the criteria and the application in Tables 2 and 3. It is expected that

Sr ≤ σr . (Χ²_1-α /q)^1/2.

Assessment of accuracy of the mean

According to IDF Standard 128 , the error of accuracy of the mean is broken down in the error of exactness of calibration and the error of accuracy (accuracy of estimates).

Statistical parameters to be used are those indicated in IDF Standard 128 and summed up in Appendix 1 (starting on page 19):  ; Sd ; Sy,x ; slope (b) ; student t test for  and b.

They are obtained from a simple linear regression calculated using means of duplicate instrumental results (x) and so-called reference results (y) obtained with a reference method recognised by ICAR (analyse in duplicates).

Assessment of accuracy

Accuracy is assessed for individual animal milks and herd milks separately.

It is measured through the residual standard deviation Sy,x of the simple linear regression of instrumental results (x) and reference results (y).

It is expected that the differences to the regression line are normally distributed, therefore any outlying result should be carefully scrutinised. In case of outlying results, an other split sample of the same milk should be reanalysed by reference and the analyser when possible. When not or if outlying figure remains, reporting should present Sy,x estimates and graphs including all data – with the outliers identified, their number and respective biases - and the same Sy,x calculation after discarding outliers. Statistical methods used to identify outliers should be specified in the evaluation report. The proportion of outliers should not exceed 5 %.

The estimate value of Sy,x should fulfil respective limits σy,x defined for individual milk samples and herd milk samples in Tables 2 and 3.

It is expected that

Sy,x ≤ σ_y,x .[X²_1-α / (q-2)]^1/2.

For criteria covering a wide range of concentration –that is more than 1 log scale- (case of somatic cell count), accuracy evaluation should be performed for the whole range and for successive parts of the range after splitting of the whole concentration range in different parts, three parts for the minimum (i.e. low, medium and high).

Table 7. Precision values for medium content milk samples (cows, goats).
	ICAR limits
Criteria (units)	F (g/ 100 g)	P (g/ 100 g)	L (g/ 100 g)	Urea (mg/ 100 g)	SCC (%)
Repeatability
Average Sr L / M / H	0.014 (1)	0.014 (1)	0.014 (1)	1.4 (2)	4 % (1) 8 % / 4 % / 2  %
Reproducibility
Average SR (SR%) L / M / H	0.028 (1)	0.028 (1)	0.028 (1)	2.8 (2)	5 % (1) 10 % / 5 % / 2.5  %
Accuracy
Animals Sy,x	0.10 (1)	0.10 (1)	0.15 (2)	6.0 (2)	10 % (2)
Herds Sy,x	0,07 (1)	0,07 (1)	0,07 (2)	4.0 (2)	10 % (2)

⁽¹⁾ Limits in accordance with IDF Standard 141C and 148A.

⁽²⁾ Limits derived from experimental results and IDF 141C (SR~2.Sr).

Note

For lactose IDF Standard 141C recommends the same limits for Sy,x as for fat and protein which are difficult to fulfil with regards to poor chemical method available as reference at that date.

Table 8. Precision values for high content milk samples (ewes, buffaloes, particular cow/goat species). Derived from medium levels limits by applying relevant level ratios: 2 for F and P; 1 for L and urea.
	ICAR limits
Criteria (units)	F (g/ 100 g)	P (g/ 100 g)	L (g/ 100 g)	Urea (mg/ 100 g)	SCC (%)
Repeatability
Average Sr (Sr%) L / M / H	0.028 (0.35 %)	0.028 (0.4 %)	0.014 (0.3 %)	1.4 (2 %)	4 % 8 % / 4 % / 2 %
Reproducibility
Average SR (SR%) L / M / H	0.056 (1) (0.7 %)	0.056 (1) (0.8 %)	0.028 (1) (0.6 %)	2.8 (2)	5 % 10 % / 5 % / 2.5 %
Accuracy
Animals Sy,x (Sy,x%)	0.20 (2.5 %)	0.20 (3.0 %)	0.15	6.0	10 %
Herds Sy,x (Sy,x%)	0,14 (1.75 %)	0,14 (2.0 %)	0,07	4.0	10 %

1.1.1.1.1         Assessment of exactness of calibration

Prior to analyses, the instrument is calibrated according to the procedure recommended by the manufacturer and expressed in the same units as reference method used for the evaluation. Since then raw signals are not concerned and further statistical comparisons can be made at a same scale for both instrumental and reference values, allowing classical tests of identity and assessments against standardised target values. For this purpose, individual animal and herd samples will be analysed to provide the relevant information on the quality of the adjustment.

Depending on the principle of the method, quality of calibration can be more or less influenced by the representativeness of calibration samples in addition to calibration technique applied (i.e. mathematical model, experimental design, process). Therefore, sources of error of representativeness shall be reduced at the maximum for instance by sampling calibration samples in close or identical condition as for prediction milk samples.

Exactness of calibration is to be assessed using the parameters of the regression y=b.x+a: the mean bias  and the slope b (see Appendix 1 - starting on page 19, and IDF Standard 128) taking care of eventual outlying results as in Assessment of accuracy on page 13. Estimates  and b should normally fulfil the limits in

Table 9. Failing that goal should normally imply further investigations or explanations.

Table 9. Tentative indicative ICAR limits for exactness of calibration assessment.
a.      Medium level (cows, goats)
Criteria	F	P	L	Urea	SCC
Mean bias	±0.05 (1)	±0.05 (1)	±0.05 (1)	±2.5 (2)	±5 % (2)
Slope b	1±0.05 (1)	1±0.05 (1)	1±0.05 (1)	1±0.05 (2)	1±0.05 (2)
b.     High level (ewes, buffaloes, goats)
Mean bias	±0.10 (1)	±0.10 (1)	±0.10 (1)	±2.5 (2)	±7 % (2)
Slope b	1±0.05 (1)	1±0.05 (1)	1±0.05 (1)	1±0.05 (1)	1±0.07 (2)

⁽¹⁾ Limits in accordance with IDF Standard 41C.

⁽²⁾ Limits derived from experimental results.