PovtschHt sbe iehte
The use of laboratory prepared reference materials in nutrient analysis J. R. Cooke Laboratory of the Government Chemist, Waterloo Road, London SEI 8XY, England
Summary. The need for laboratory prepared reference materials is discussed as are experiences gained during five years of their use in the analysis of foods for trace nutrients. Ways of monitoring results are described and the possibility of providing certified reference materials for some of the B group vitamins in foods is examined.
Introduction It is an analyst's responsibility to do his best to obtain a result of the desired degree of accuracy or in many cases to get the best result of which he is capable. The extent to which the analyst is successful will depend on many things but probably most of all on the way he ensures that the selected method gives reliable results when applied to a sample of the type to be analysed. The analyst's problems fall into several categories only some of which are relevant to this particular paper. A method has to be selected or developed and then tested to establish its range of application and the standard of results to be expected from its use. It is this last category which will be the subject of further discussion. Methods in analytical chemistry give rise to results which are affected to some degree by systematic and random errors. It is desirable that these are kept to a minimum or at least reduced to a level which is acceptable for the user of the results. In either case, the analyst should establish the level of these two types of error for the sort of samples most frequently analysed. Random errors can be assessed by carrying out multiple assays on a homogeneous sample or by carrying out duplicate determinations on a range of samples. Systematic errors on the other hand can be much more difficult to assess. Ideally, a certified reference material of a similar type and composition to the sample of interest is analysed and the difference between the result and the certified value is the systematic error. In practice the random error has to be considered as does the fact that the certified value is a best estimate and the actual value probably lies within a stated range. The position is better if several suitable certified reference materials can be analysed. Then, provided the results are not consistently one side or the other of the certified values, the systematic error is probably low. However, with one or two notable exceptions, e.g. trace metals in biological materials, certified reference materials are not yet available. In these circumstances the analyst has to use all his skill and
much time and effort to attempt to assess the systematic error. This task is not something an analyst will wish to repeat with any frequency but it is advisable that any method be checked from time to time to ensure that all is still well. Again the presence of a certified reference material can be of considerable assistance to the analyst but its availability is likely to be the exception rather than the rule. However, an actual certified value is not the essential need if the analyst merely wishes to check that a method is still giving rise to the same result for an analysis. The essential need is for a reference material which has previously been analysed, preferably several times, in order to establish the level of the analyte. Such a reference material can serve other useful purposes. In even the best run laboratory, where equipment is regularly serviced and tested and where everything possible is done to ensure the reliability of results, there is always the possibility of equipment malfunction or an error on the part of the analyst. This can lead to one or more sets of incorrect results. However if a reference material can be analysed with each set of samples an incorrect set of results will be immediately obvious and corrective action can be taken. A certified reference material could be used for this purpose, if it were available, but the rate of use might well be unacceptable either on the ground of cost or availability. No standards body is likely to be able to maintain a supply of such materials if most laboratories use them so routinely. Hence the requirement is for a set of laboratory prepared reference materials which have compositions similar to the main types of samples which are usually analysed. These reference materials should contain between them as many as possible of the analytes which are usually determined.
Experimental
1. Preparation of reference mater&& Samples of skimmed milk powder, wholemeal flour, and mixed flours were not prepared in any way but were mixed well before bottling. Liver and peas were freeze dried and comminuted to fine powders. The liver was also defatted by extraction with hexane in a soxhlet.
2. Storage of reference materials The above materials have been stored in several ways over the years. Freeze dried liver and peas have always been kept in a deep freeze. Initially, flour was kept at room temperature Fresenius Z Anal Chem (1987) 326:665- 668 9 Springer-Verlag 1987
Lectures + 3S0
+
+
++++++++++'+,
~n
-+-T-+--+--+ 5
c. E
l
#.J
1o
E u
%+
o o
,
,u
2SD
+
1SD
+++
~----- ---~---Tt-- ~ - ~ +~-
Analysis number
+
++
70
I I I
e
+
+
l
+
~-+
+
Mean
+
+
+
65
L-+-+--T_----%
++
+
++ +- - ~s0
-5
+
a
I---. . . .
60
0.6
+++++++
g 0.2
+
++
5
E
71 +l
+
._>_
10
15 Analysis I
~3 -o.2
1.3
f/I
2; 3T0 Analysis number
4'0
-
3513
5;
Fig. 2. Plot of results for magnesium in defatted freeze dried liver against analysis number, with lines indicating the mean and _+1, 2 and 3 standard deviations
number
-0.4
1//
-0.6
Fig. la, b. Plots of cumulative sum against analysis number for: a Magnesium in flour with the mask on point 16, indicating a satisfactory situation, b Nitrogen in defatted freeze dried liver with the mask on point 15, indicating a problem
in the dark while skimmed milk powder was kept in a refrigerator. Nowadays, all samples are bottled in small screw capped containers holding sufficient material for a few analyses. A plastic film (Nescofilm) is stretched over the top of the bottle before the lid is replaced. These bottles are stored in a deep freeze. The bottles of flour and skimmed milk powder currently in use are kept in a refrigerator. Moisture levels are monitored but the changes observed are not normally significant.
Results and discussion
Our experience with laboratory prepared reference materials now goes back five years. Initially, we used samples of flour and defatted freeze dried liver and these were analysed for nitrogen content by a Kjeldahl method and for certain nutritionally important trace metals by dry ashing followed by inductively coupled plasma optical emission spectroscopy. Considerable effort was put into ways of monitoring the results we obtained and for a long time we experimented with cusum charts as a paper describing the technique had recently been published [1]. Basically the technique involves a plot of the cumulative sum of the differences between each individual value and the expected value, as previously determined from a series of analyses, against the number of the determination. The 666
--
1'0
0.4
2SD
scaling of the axes is such that a cumulative sum equivalent to about twice the standard deviation for the analysis is equal to one unit on the number of determination axis. A V shaped mask, whose shape again depends on the standard deviation of the analysis, is placed on each point as it is plotted. A problem is indicated if the line of points cuts the mask. Figure 1 illustrates the use of the chart in two cases a) where all is satisfactory and b) where a problem is apparent. This is indicated by points lying outside the angle of the mask when it is placed on points 14 and 15. The cusum chart is an excellent way of showing up tendencies towards high or low results occurring for a limited period. In such instances, it is far superior to a straight forward plot of result against number of analysis. In practice we have found problems in the use of cusum charts. It is essential to obtain good results for the mean value and for the standard deviation of the analysis. This can take a great deal of time for lengthy analyses and for analyses which are not made on an almost daily basis. If an incorrect mean is used the cumulative sum will tend to rise or fall steadily, while an incorrect measure of the standard deviation will affect the scaling of the chart and lead to either too tight or too slack control by the V shaped mask. It is a good idea to start the cusum chart after say 10 results have been amassed. If, in time, it becomes apparent that either the mean or standard deviation on which the chart is based are not satisfactory, then the values should be recalculated using all the results and a fresh chart should be started. Results which can be classed as outliers should not be used in the calculations. Likewise, a rogue result, which by itself leads to the mask indicating a problem, should be deleted from future calculations but the reason for it should of course be investigated. If for any reason a cusum chart is considered to be unsuitable for the monitoring of results then at the very least a simple chart should be employed for the purpose. Initially this need be nothing more than a plot of result against number of analysis or some sort of timescale. A mean value
Fortschrit sber eh e 3.0
+
§ + ++§
+
2.0
§ +
++ ~++ ++ +++ +
=_ >
+
4- +
+
++
+
+ +
1.0
4+ +
+ +
4++
+ §
0 19~83
I
19'8/+
I
19'85
r
Date
Fig. 3. Plot of results for riboflavin in skimmed milk powder against date indicating a loss of vitamin with time
and the standard deviation can be calculated when sufficient results are available. Lines for the mean and for 1, 2, and 3 standard deviations in either direction can be drawn. A typical chart is illustrated in Fig. 2. Over the course of say 50 analyses it is easy to see that about 33 results lie within _+ 1 standard deviation of the mean and that only 2 or 3 results lie between _+ 2 and 3 standard deviations. If this is not so the mean and standard deviations are not correct or there might be a problem with the method leading to occasional high or low results. In any case a chart will help to draw attention to any problems as will a comparison of charts for closely related methods say for two or more trace elements in a food sample. As mentioned earlier, problems for the analyst are much greater when no certified reference materials are available and it is in these instances that laboratory prepared reference materials can be particularly valuable. The analysis of foods for the water soluble vitamins of the B group is just such a case. The lack of certified reference materials for vitamins in foods is almost certainly because such compounds are not indefinitely stable. However, the problems in a single laboratory are not as great as for an organisation marketing certified reference materials. The organisation has to have a suitable type of material prepared in bulk and then this has to be tested for homogeneity. The material has to be packed and finally analysed, using several different analytical techniques, some of which may require considerable periods of time in order to obtain results. Finally, the certified values have to be agreed upon and the necessary leaflets printed. The material has to be stable for long enough to enable an acceptable level of sales to be achieved. Hence, a times span of 3 - 5 years without significant deterioration of the analyte is required. In the laboratory the timescale is much less exacting. A kilogram or two of material can be prepared and used immediately. A useable life of only a few months may be acceptable for an analysis which is carried out on almost a daily basis, while a year or two may well suffice for an analysis carried out every week or so. A material
containing a vitamin which is slowly deteriorating may still be usable as the analyst will be alerted by an unusually high or low result. With the above in mind, we began late in 1982 to routinely analyse a sample of flour each time we carried out a microbiological assay for all except one of the eight B group vitamins we normally measure. The exception was vitamin B12 which does not occur in flour. It soon became obvious that the flour sample was very satisfactory as none of the vitamins was deteriorating at any appreciable speed. We also began to have a much better idea than we had in the past of the variation in analytical results from assay to assay. Microbiological assays are notorious for their variability from day to day and this was much in evidence. In the middle of 1983 we extended the experiments and started to analyse a sample of skimmed milk powder along with the flour. This gave us the opportunity to see if the results for the two samples behaved in a similar way and it also enabled us to include the assay for vitamin B12 as this vitamin is present in milk. Late in 1983 the mixed flour sample ran out and we continued the experiments with a wholemeal flour. Measurements on this new flour and on the milk powder were made continuously until the beginning of this year, i.e., for periods of 27 and 30 months, respectively. Small amounts of the two samples have been retained and it is our intention to continue monitoring the levels of the vitamins but with a much reduced frequency. New samples have been prepared for regular use. Unfortunately, the trends we have observed in the results over 2 - 21/2 years could have been affected by improvements made to the procedures and by other changes resulting in the substantial level of automation we now employ. However, there is little doubt about some of the conclusions that can be drawn. Odd exceptionally low or high results are obtained on occasions and these are not unexpected in microbiological assays. With two reference materials it is easy to see whether both are affected in the same way. If both produce high or low values it is almost certain the assay is faulty whereas if, as occasionally happens, only one is unsatisfactory then it is probably a rogue result. This advantage of using two reference materials will apply equally to other types of assay. However, it takes time and hence costs runny but it can be worthwhile in some cases where the cost of repeating a batch of samples can be high. The other interesting aspect of this work has been the stability of most of the B group vitamins in these two particular types of sample. Folaein, both free and total, has decreased by about 50% in 2 years in both samples as has biotin in skimmed milk powder. Riboflavin in skimmed milk powder has lost about 75% of its original amount in the same time (Fig. 3). However, apart from these the other vitamins have not altered appreciably. It seems possible that wholemeal flour and skimmed milk powder may be worthy of consideration as certified reference materials for 6 of the B group vitamins, i.e. thiamin, niacin, v i t a m i n B6, vitamin B12 (milk powder only), pantothenic acid and possibly biotin (flour only). Also it seem unlikely that these two materials, chosen because they were readily available as fine homogeneous powders, are the only foods which will be suitable for this purpose. However, one problem might remain even if these or other foods are considered to be suitable for use as certified reference materials for some of the B group vitamins. It is usual to obtain certified values by making measurements 667
Lectures using several different analytical procedures and in this instance the range of possible procedures is limited. Also, the different procedures might not measure exactly the same substances and this would, lead to variation in the results. Finally, if certified reference materials for compounds such as vitamins are to become available it will be necessary to find laboratories capable of carrying out the accurate analyses needed for certification. Laboratory prepared reference materials will greatly assist analysts to achieve the high standards required for this work. Acknowledgements. The author would like to acknowledge the part played by Irena Agater, Gillian Holeombe, and Brian Stockton in
668
the above work and also to thank many other analysts whose results have contributed to this publication. Crown copyright, reproduced by permission of the controller of H. M. Stationary Office.
Reference 1. Chamberlain UD (1980) Anal Proc 17:172-176 Received September 15, 1986