Number 3 : JULY 1998


This month's Newsletter takes on a new format to match the WWW version. This approach means that the Newsletter can be posted onto the homepage as soon as it is ready for going to the press. In the past there was about a month's delay while I revisited the text, took it out of its dual column format and put in the HTML tags by a combination of manual and automated methods. Another change which was implemented with the last issue was the change from coloured paper to plain white. This had to be done because our printer could no longer absorb the cost of the coloured paper into the price; each A3 page cost 7 cents.

So far we have had sponsors for all but one of the issues and I am pleased to be able to announce that this issue has a sponsor : Analytical Reference Laboratories in Melbourne. They have included a double sided information sheet outlining their services in the analytical biochemistry area. This company has a longstanding record in the provision of those specialist analyses that most routine private and public hospital pathology laboratories are unable to provide. By virtue of their instrumentation base and expert staff they are able to achieve some economies of scale and provide these services to clinicians within turn around times that make the investigations clinically worthwhile.

While on the subject of sponsorship - this is open to any Company or Organisation with an interest in nutrition. The arrangement is that for a relatively modest contribution they are able to insert one A4 sized advert / information sheet into the issue being sponsored. So if you are a potential sponsor or have contact with a potential sponsor please encourage them to phone or write to me; address etc. are at the end of this Newsletter.

This issue contains a lot of information on the 1998 Annual Scientific Meeting and please remember that your abstracts are now due ! The closing date for abstracts is the 28th August. Instructions for preparing abstracts have been mailed to all Members but if you have mislaid them - visit the WWW homepage and you can reprint a copy out for yourself. I have also included a lot of information on the ESPEN Conference. There have been a number of requests for details on this Conference and it was not easy to obtain; I eventually found a very good contact involved with the European journal 'Clinical Nutrition' who very kindly sent the details. So for those of you who are going to be fortunate enough to attend the details are here; for those of us who will be unable to attend I think you will find the programme very informative as it reveals they ways in which the Europeans are practicing and researching in our common areas of interest.

Finally if you feel that you would like to write an article for the Newsletter then please get in contact. Members are always interested to get reports from Conferences, updates on research projects, updates on TPN and EN practice, special interest groups, etc.

CANBERRA : 23rd and 24th OCTOBER 1998

Sabita Rajeshwar
Dietitian, New Children's Hospital Westmead NSW.

As Convenor for this year's Conference, I have great pleasure in announcing this Meeting. We on the Conference Organising Committee have been working very hard to ensure that this will be a successful meeeting and look forward to your attendance. In addition, as an exciting first for AuSPEN, we will be holding a joint meeting with the Gastroenterological Society of Australia to coincide with Australian Gastroenterolgy Week.

Registration brochures are currently available from the Conference Secretariat and Abstracts for Free Papers are being called for. We would greatly value and encourage your scientific contribution to the Society in the form of a paper or poster presentation. It is a wonderful opportunity to showcase your work to like minded participants. You could also end up winning the Prize for the Best Paper or Poster, or you may be eligible for a Travel Award.

Topics of Interest

The main programme has virtually been finalised and includes the following topics :

International Guest Speakers

We are priviledged to have the following World renowned experts as Key Speakers

National Speakers

We are fortunate to be represented by a number of excellent national speakers covering a wide range of topics to cater for a variety of interests. They include


As well as the main programme, two concurrent workshops will be running on the morning of Saturday 24th October. These are :


Being located in Australia's Parliamentary city, the National Convention Centre is a major conference facility. It is outstandingly equipped and ideally located. A range of accomodation has been earmarked for the conference including the Park Royal Hotel in close proximity to the Convention Centre. Social functions being organised include a welcome cocktail party and a conference dinner, which will be held at the majestic High Court building. We hope you will be able to attend these and join us in making them an unforgetable experience.

Conference Organising Committee

The Canberra 1998 Committee are

Please do not hesitate to contact any of the above for further information. Registration Brochures and Calls for Abstracts can be obtained from the Conference Secretariat.

Sabita Rajeshwar : July 1998



Members who attended the 1996 Annual Scientific Meeting will remember Prof Alan Shenkin and his presentations on Trace Elements etc in TPN. He has recently published an abstract in the Proceedings of the Association of Clinical Biochemists (UK) National Scientific Meeting 1998 on page 41. The text is reproduced here :

'Changes in Antioxidant Status of Patients Receiving Intravenous Nutrition, as Assessed by Two Methods'M Baines and A Shenkin, Departments of Clinical Chemistry at Royal Liverpool University Hospital and University Hospital, London.

Patients started on intravenous nutrition (IVN) may be under oxidative stress and are dependent upon the nutrition to maintain their antioxidant defences. This study investigated initial antioxidant status, and response to IVN, by two methods.

25 patients had plasma samples taken daily for a range of 1 to 14 days (average 6 samples per patient). Antioxidant status was assessed by an established radical suppression method (TAS, Randox Laboratories) and by a newer method measuring the ferric reducing ability of plasma (FRAP, Benzie and Strain, 1996). Albumin and urate were also measured.

On beginning IVN, the median TAS level was 1204 umol/L, range 878 - 1748, with 10 out of 16 patients below our lower reference limit (LRL)(female 1117 umol/L, male 1305 umol/L). At the end of sampling, after 4 - 16 days IVN, the median TAS level was 1194 umol/L, range 915 - 1440, with12 out of the 25 patients still below the LRL. Of the 10 patients who were initially low only 4 reached the reference range.

Corresponding figures for FRAP were : initial median 331 umol/L, range 198 - 1155 (7 out of 16 low), end median 434 umol/L, range 202 - 895 (5 out of 25 low), 4 improving of the 7 initially low.

The correlation between TAS and FRAP was : r = 0.57 and both showed significant positive correlation with urate (TAS, r = 0.53; FRAP, r = 0.95). TAS had only a weak correlation with albumin (r = 0.15, N.S.), whereas, surprisingly, FRAP had a significant negative correlation ( r = 0.43 ).

This study has demonstrated that a substantial proportion (63%) of IVN patients are initially antioxidant depleted and many remain so, despite short term IVN. The TAS method is less influenced by urate and albumin than FRAP and may therefore be preferable.

Readers will recall that in the January 1998 Newsletter, page 5, mention was made of a paper from M Ryan et al which attributed much of the antioxidant capacity of serum to urate. This work of Baines and Shenkin also found a strong correlation with serum urate. Perhaps urate should be tracked in all TPN patients from time to time. Being such a readily available biochemistry test it could be regarded as the next best alternative to these other more esoteric - and hence less available - assays


Readers will be interested in this article by Ambados and Brealey, (Pharmacy Dept., The Queen Elizabeth Hospital, Woodville S.A.), which appeared in the Australian Journal of Hospital Pharmacy, Vol. 28, #2, pp 112 - 114, 1998. They used a technique involving drawing a TPN mixture of Synthamin, glucose and a three component admixture into syrnges, storing them under room temperature and refrigerated conditions and then examining them for crystalline precipitates. The final mixture consisted of 500 mls of Synthamin 17 without electrolytes, 500 mls of 50% glucose solution, 10 mls of an ascorbic acid solution (1g/10mls), 20 mls of calcium chloride (14 mmol Ca2+/20mls) and 5 mls of Trace Element Solution from David Bull Laboratories. Visual examination and scanning electron microscopy revealed the presence of crystals after varying lengths of storage, anything from 3 days at room temperature to 12 days under refrigerated conditions. The appearance of these crystals was consistent with crystals of calcium oxalate. Their hypothesis is that the high concentration of ascorbic acid leads to the in vitro production of oxalate ions as the ascorbate ions decompose. Ca oxalate has very poor solubility in aqueous solutions and readily precipitates out. In view of this phenomenon the authors make the very practical suggestion that perhaps ascorbic acid additives should be alternated with the trace elements additive.

The paper also makes reference to a number of other turbidity / precipitation phenomena that have ben observed in TPN mixtures. Aminophylline solution when mixed with the trace element solution results in the formation of a 'white colloidal mass and discolours the surrounding solution mauve'. Sodium folate solution when mixed with the trace element solution results in the precipitating out of folic acid crystals which will redisolve on vigorous mixing. Mixtures of trace element solution and Intravite B Group solution have been associated with a colour change from yellow brown to dark orange brown.

It is not surprising that interactions occur in our TPN mixtures given that they have such a complex recipe. The opportunities for a host of reactions between the components is a multidimensional problem and we should not be too surprised when they are reported. They highlight the fact that we rely heavily upon the vigilance of the manufacturing pharmacists to ensure that any obvious interactions are investigated and wherever possible the mixture compunding protocol is modified in such a way as to minimise any such reactions. The multi component TPN mixture is here to stay but that must not make us complacent. Manufacturers and pharmacists must continue to gather information on the 'in vial' and the 'in bag' chemistries and the 'in vivo' biochemistries of these preparations. The results can only lead to benefits for all concerned - the manufacturers, the pharmacists, the prescribers and most importantly the TPN patients.


September 16th - 19th : 20th ESPEN Conference at the Acropolis Convention Centre, 1 esplanade Kennedy, B.P. 4083 06302 Nice Cedex 4, France. Contact Development Department, 1 esplanade Kennedy, B.P. 4083 06302 Nice Cedex 4, France; Ph 33-4-9392-8156 Fax 33-4-9392-8338.

The Scientific Secretariat can be contacted at Prof Bernard Messing, Hopital Sait Lazare, 107 rue du Faubourg Saint-Denis, 75010 Paris, France; Ph 33-1-4800-5588 Fax 33-1-4800-5581. Unfortunately there are no w.w.w. sites or email addresses available.

Registration is FF 1900 for Members, FF 2250 for non-members

Programme :

There will be three debates

Symposia, Workshops and Round Tables :

Educational Programme

Case Presentations

General Sessions

September 18th-19th: 4th Annual Paediatric Nutrition Workshop: 'Infant Feeding' at Royal Children's Hospital, Brisbane, Queensland. Sponsored by Children's Nutrition Research Centre, Department of Paediatrics and Child Health, University of Queensland. Contact Charles Fleetwood, Consult Fleetwood Management Services, P.O. Box 79, Arana Hills, Qld 4054. Phone: 07 3264 5970, Fax: 07 3264 3520. Email:
Excerpts from the Programme : Growth of Infants by Ross Shepherd, Energy Needs by Peter Davies, Vitamin and Mineral Supplements by Geoff Cleghorn, The Latest in Assessing Infants with Dysphagia by Berenice Mathisen, Infants with Developmental Disabilities by Robyn Littlewood, Gastroeosophageal Reflux by Peter Lewindon, Infants of Vegan Mothers by Judy Wilcox, Infant Feeding on the Internet by Orrawin Trocki.

September 27th - 30th : 5th Conference of the International Society for Trace Element Research in Humans. Lyon, France. Contact Trace Element Institut for UNESCO, Immeuble Le Condorcet, 1 place de l'Ecole.

October 8th - 11th : The 23rd Australian & New Zealand Annual Scientific Meeting on Intensive Care, Adelaide Convention Centre. Prof Jonathan Cohen - Infection / Sepsis Dr Deborah Cook - Evidence Based Practice Prof Sergio Fanconi - Paediatric Intensive Care Prof Huda Huijer Abu-Saad - Nursing Research / Pain Assessment Prof Michael Matthay - Acute Lung Injury Ms Sandra Mattheson - Credentialling Prof Sander Van Deventer - Inflammation / Cytokines Ms Kathleen Vollman - Prone Positioning in Acute Lung Injury

Contact : Festival City Conventions, PO Box 949, KENT TOWN, South Australia 5071

Ph: 08-8363-1307 Fax: 08-8363-1604


October 19th - 22nd 1998 : The American Dietetic Association Annual Meeting and Exhibition, Kansas City Convention Center, Kansas City, MO. Contact The American Dietetic Association, 216 West Jackson Boulevard, Chicago, IL 60606-6995. Tel 312- 899-0040. Fax 312-899-0008.

email: Internet:

December 11th - 13th December : Clinical Controversies - Issues in Therapeutics , presented by the Society of Hospital Pharmacists of Australia, at the Wrest Point Convention Centre, Hobart, Tasmania.
Conference Secretariat : Clinical Controversies Conference, Mures Convention Management, Victoria Dock, Hobart, Tasmania, Ph 03-6234-1424 Fax 03-6234-4464 email:


Dr Tom Hartley Pathology Dept. Royal Hobart Hospital

With the approach of the Scientific Conference season I have written the following with a view to giving some pointers on the appropriate use of statistics in abstracts and conference presentations.

Perhaps many of us have been guilty of carrying out a study, tabulating the data and then handing them over to a statistician in the hope that he or she will be able to detect the message in the data. What this situation reveals is that we, the investigators, failed to write down before we started the hypothesis(es) that we wished to test. In fact what we are probably hoping is that the statistician will find some 'statistically significant' features in our data set around which we will write a hypothesis. A sort of reverse engineering. Actually I would regard this as being an overly harsh account of what actually happens. From my experience most people who come to me for statistical advice have in fact had very good reasons for designing the experiment / project the way they have. They are the experts in their field and often what they have done is respond to their professional sixth sense and gone and done the instinctively necessary investigations. Their usual mistake has been to omit to measure a particular parameter which would have given a great deal more statistical power to their study. So their statistical report, which is good, often lacks that all important finishing touch which would make the report both excellent and thoroughly convincining. So with those thoughts in mind I would urge you to write down some hypothesis(es) now relating to the work you are planning to present, before launching into your statistical analysis.


A hypothesis is best kept simple eg.- changing 'x' will improve 'y'. Anything more complicated will make it difficult to present in a 10 minute paper or describe on a poster that most viewers will look at for less than 5 minutes.

Avoid drawing up Hypotheses that contain conditional clauses eg. - changing 'x' in patients without significant symptoms of 'y' will show an improvement in 'z'. It is better to split the hypotheses : First we will prove that our patients had little or no evidence of symptoms 'y'. Then we will prove that in these same patients changing 'x' actually improved 'z'.


Statisticians prefer to write papers and articles in the language of algebra. Most of do not think in algebraic terms - so if I continue to use algebraic language in this article many readers will read no further. Instead I will give the examples fictitious titles - and equally fictitious data - that should sound a bit more inviting to the reader.

Supplements of Parenteral Glutamine Preserve Villous Height in TPN Fed Rats.

This title immediately reflects a simple hypothesis - if we give more glutamine in the TPN mix then villous atrophy is reduced. The fictional experiment simply involved giving ten rats 3 days of glutamine free TPN - so that gut atrophy would become established - and then putting glutamine supplements into their TPN mix and measuring intestinal villous height after a further 3 days.

The glutamine supplements were calculated as mg/kg body weight/day and the villous heights were measured in mms. The results were as shown in Table 1.

TABLE 1 : Glutamine Doses and Villous Heights
Glutamine Height Glutamine Height
16 0.41 35 0.50
20 0.38 40 0.54
20 0.44 40 0.60
25 0.50 48 0.58
30 0.50 55 0.66

To reveal whether or not glutamine supplements have affected villous height it is appropriate to draw a graph as shown in Figure 1. The rule that must be observed in plotting such a graph is that the parameter which we have manipulated must be put on the 'x' axis and the parameter we have measured must be put on the 'y' axis. Then we can calculate the equation to the line of best fit through the points and the correlation coefficent which gives a measure of the goodness of fit. The safest way to calculate these is to use a statistical calculator or a 'tool' in a computer spreadsheet program such as Excel or Cricket Graph. However, if you do not have access to these then it is not too difficult to do it with a simple calculator taking care to record the results on a piece of paper as you go :

Now work out three intermediate results :

Now the Slope of the Line of best fit is easily calculated by dividing A by B = 0.00634.

The Intercept of the line on the Height, (y), axis is easily calculated :

d - Slope x b = 0.511 - 0.00634 x 32.9 = 0.302

The Correlation Coefficient is also easy to calculate. First multiply B by C and take the square root of the answer before dividing it into A.

A / squareroot(B x C) = 9.581 / 10.173 = 0.942

We can now express the results of our experiment in an equation :

Villous Height = 0.00634 x Glutamine Supplement + 0.302

So from this experiment we can describe the outcome as follows : There was a progressive improvement in villous height with glutamine supplementation of the TPN mixture such that over the range of supplements given, 16 - 55 mg/kg/day, there was a statistically significant increase in villous height of 0.25 mm.

Now that statement can only be made because of our knowledge of the equation to the line of best fit. One of the most common ommissions in abstracts is not to make use of this equation which is odd because so much effort goes into calculating it! The majority focus on the correlation coefficient alone in which case all they can say is that there was a statistically significant effect observed. The equation to the line gives you the functional relationship between the supplement and the villous height. It enables you to describe the 'dose response' of your experiment in very practical terms : over the range of supplements given, 16 - 55 mg/kg/day, there was a statistically significant increase in villous height of 0.25 mm. How did I derive this 0.25 mm, well I substituted 16 mg/kg/day into the equation and calculated the corresponding villous height, 0.40 mm. Then I substituted 55 mg/kg/day into the equation and calculated that villous height, 0.65 mm. The difference between these is 0.25 mm which I now claim to have occurred as a consequence of supplementing with glutamine.

I have described the response as having been statistically significant without giving any justification. My justification lies in an assessment of the value of the correlation coefficient of 0.942. Suffice it to say that statisticians have drawn up tables in which the statistical significance of any value of correlation coefficient derived from any number of data pairs have been drawn up. It it difficult to describe succintly how to look up these tables but the secret is to take 2 away from the number of data pairs in your experiment, run your finger down the left hand column until you reach that number, in our case 8, and then move your finger across until you reach a number in the column marked '0.05'. If your correlation coefficient is greater than this number then relax - your correlation coefficient is significant at the p =< 0.05 level. A very truncated table for assessing your correlation coefficient is shown in Table 2.

TABLE 2 : Significance of the Correlation Coefficient Abbrev.
Number of Observations minus 2
p=0.05 Significance Level
Number of Observations minus 2
p=0.05 Significance Level

Now let us move on to another type of paper :

Improved liver function tests following treatment with the parenteral amino acid solution Metamine.

In this study two groups of patients were investigated. The first group of twenty one patients had their liver function assessed in terms of their GGT levels after five days of TPN with their routine TPN mix. The second group of twenty one patients received the new amino acid solution Metamine and their GGT levels were measured after five days. The results were as shown in Table 3 and as Histograms in Figure 2.

The histograms show that the data in the two groups are not normally distributed ie they do not have the characteristic bell shape of the Normal Distribution Curve. This immediately alerts us to beware of using means, standard deviations etc. because these statistical tools all assume that the data that you are providing for analysis are 'normally distributed'. Instead we should use non-parametric measures. The non- parametric measure of central tendency is the median. Because we have an odd number of data points in each data set it easy to locate the median value. It is the data point that is the eleventh from the top or bottom of the data sorted in increasing order. In the Routine Formula data the median is unequivocal - it is 500. But in the Metamine data the eleventh data point is alongside two other points of the same value ie 240. This means that our median is between 240 and 250; in fact it is 2/3rds of an interval up above 240. This is 246.6 but because we have only measured GGT values to the nearest 10 the best relevant estimate of the median of these data is the rounded up value of 250. You can calculate confidence limits to the median; this is akin to calculating the Standard Error of the Mean but as you will see the limits are assymetric because of the assymetry of the data distributions. The formula you have to use returns the position numbers of the lower and upper limits in your list of sorted data.

TABLE 3 : Serum GGT Levels
Routine Formula TPN
160 240 240 300 350 400
400 460 460 460 500 540
540 540 540 540 600 600
600 650 700

Metamine Formula TPN
80 100 130 160 160 200
200 200 240 240 240 270
300 300 300 300 350 400
400 450 600

The procedures are as follows:

So for our data sets the findings are shown in Table 4.

TABLE 4 : Medians and Their Confidence limits

Lower Confidence limit Median Upper Confidence Limit 95% Tolerances
Routine Formula
-100, +150
Metamine Formula
- 50, + 60

Clearly the difference of 250 between the medians at a worst case is more tahn 2.5 times the size of the largest tolerance, -100. So the Metamine formula appears to be a lot less hepato irritant than the Routine formula. If we were using parametric statistics we would use a 'Studet's t Test' to assess the difference between the two means and on getting a 't' value in the vicinity of 2.5 we would conclude that the Metamine had a very significant beneficial effect. The non-parametric equivalent of the Student's t Test is the Mann-Whitney Test. This involves pooling the two data sets and organising them into a single table sorted in increasing order taking care to tag each item as to which data set it was originally drawn from - the Metamine set or the Routine set in our case. Our pooled data set is shown in Table 5 with the Routine data set items tagged with the letter 'r' and the Metamine data set items tagged with the letter 'm'

Table 5 : Pooled Data and their Rankings

080m 100m 130m 160m 160m 160r 200m
Item # 1 2 3 4 5 6 7
Ranking 1 2 3 5 5 5 8

200m 200m 240m 240m 240m 240r 240r
Item # 8 9 10 11 12 13 14
Ranking 8 8 12 12 12 12 12

270m 300m 300m 300m 300m 300r 350m
Item # 15 16 17 18 19 20 21
Ranking 15 18 18 18 18 18 21.5

350r 400m 400m 400r 400r 450m 460r
Item # 22 23 24 25 26 27 28
Ranking 21.5 24.5 24.5 24.5 24.5 27 29

460r 460r 500r 540r 540r 540r 540r
Item # 29 30 31 32 33 34 35
Ranking 29 29 31 34 34 34 34

540r 600m 600r 600r 600r 650r 700r
Item # 36 37 38 39 40 41 42
Ranking 34 38.5 38.5 38.5 38.5 41 42

The next step is to add up the position numbers of all the Routine data set items :

5 + 12 + 12 + 18 + 21.5 + 24.5 + 24.5 + 29 + 29 + 29 + 31 + 34 + 34 + 34 + 34 + 34 + 38.5 + 38.5 + 38.5 + 41 + 42 = 604

Do the same for the Metamine data set :

1 + 2 + 3 + 5 + 5 + 8 + 8 + 8 + 12 + 12 + 12 + 15 + 18 + 18 + 18 + 18 + 21.5 + 24.5 + 24.5 + 27 + 38.5 = 299

Note that because there are several sets of ties, the rankings are not identical to the Item number (#). Wherever there are ties you must add the all the respective Item Numbers involved together, divide by the number of items involved in that tie and then that value equals the ranking of each of the items in that tied group.

For example there are five items tied at a GGT value of 240; the sum of their item numbers is 10 + 11 + 12 + 13 + 14 = 60, which divided by 5 equals 12. Hence these 5 items all have the same ranking of 12.

Now if we had a situation where the data from each set was randomly intermingled with each other then when we would expect the sums of the rankings from each data set to be the same. In this example we have a total of 42 items, the sums of the rankings of all of them equals 903 and hence we would expect the sums of the rankings from the Routine Formula to equal the sums of the rankings of the Metamine Formula which in turn would equal 903/2 = 451.5

Our sums of the rankings are quite different from that. How do we assess the statistical significance of that difference? Well we do a Standard Normal Deviate style test for which we need a difference and a standard deviation. We have a difference - the difference between 451.5 and by convention the lower observed sum of rankings ie. the Metamine Formula data set with a rank sum of 299. That difference is 152.5. Statistians have determined that a standard deviation for this sum of ranks procedure can be calculated as follows :

So in our case this we are after the square root of 21 X 21 X 43 / 12 which is the square root of 1580.3 = 39.8

A Standard Normal Deviate, usually represented by a capital Z, is now easily calculated as the difference divided by the SD; in our case

152.5 / 39.8 = 3.8

The attractive feature of Z is that provided we have more than 10 data items each group then the interpretation criterion is the same ie. a Z value that is equal to or greater than 1.96 is statistically significant at the p < 0.05 level.

So in this example we can see that the sums of the rankings are both significantly different from the theoretical mean of the pooled data, and per se are significantly different from each other.

So in our abstract and presentation we can claim with considerable confidence that the Metamine Formula TPN was associated with a very significant amelioration in the LFT's as monitored by the serum GGT activities.

One final point; this discussion of the Wilcoxon Mann-Whitney Test has been illustrated with equal numbers of items in both treatment groups. Please contact the author if you want to apply the test to two groups of unequal size; there is a minor and easy modification step in the calculation of the theoretical rank sums.

The third type of paper I would like to consider is one in which when you plot the data there is clearly a profile to the results but it is non-linear. The question then is what tool should you use to assess the significance of this data profile. The title of the hypothetical paper is

Exercise levels produce similar responses in Body Cell Mass of both normal and immunosuppressed individuals.

In this study the activity levels of two groups were assessed, activity level 1 was assigned to individuals took no additional exercise through to activity level 12 which signified that those individuals were involved in a routine of daily body building exercise. Grades between 1 and 12 were levels of progressively increasing exercise. Table 6 shows the data for Group A, who were the immunosuppressed individuals, and Group B who were the normal individuals ie controls. When these data sets are plotted out then you can see in Figure 3 that progressive levels exercise were associated with increasing Body Cell Mass, (BCM), up to about Exercise Level 4, then tended to decline as the exercise level was increased to about Level 8, and thereafter BCM progressively increased again. So the graph immediately suggests that the responses to exercise in the two groups had very similar profiles and that the only real difference was that the immunosuppressed individuals, Group A, had lower BCMs and could only approach normal values if they embarked upon a conscious programme of body building exercise at Level 12.

TABLE 6 : Exercise Levels and Body Cell Mass Data from Patients and Controls
Exercise Level Group A Patients, BCM, kg/m Group B Normals, BCM, kg/m Difference between Rankings Differences Squared
1 9.8 (1) 14.6 (1) 0 0
2 10.3 (3.5) 15.4 (3) 0.5 0.25
3 10.9 (7.5) 15.5 (4.5) 3 9
4 11.4 (9) 16 (8.5) 0.5 0.25
5 10.9 (7.5) 16 (8.5) -1 1
6 10.8 (5.5) 15.7 (6) -0.5 0.25
7 10.3 (3.5) 15 (2) 1.5 2.25
8 10 (2) 15.5 (4.5) -2.5 6.25
9 10.8 (5.5) 15.9 (7) -1.5 2.25
10 11.7 (10) 16.6 (10) 0 0
11 11.9 (11) 17.2 (11) 0 0
12 13.2 (12) 17.8 (12) 0 0

Plotting the two data sets, Group A versus Group B, would give you a fairly good straight line but you could not perform a linear regression analysis and calculate a correlation coefficient on this and expect a statistician to agree with your approach. For a start the data in each set are not random samples etc. etc. So what you need to do is perform a non-parametric correlation analysis called the Spearman Rank Order Correlation Analysis. Notice that there is a number in brackets alongside each data item in the Table 6. This is the ranking of that data item relative to the other data items in that column.

Working these rankings out is the hardest part of this test. You must rank each item without moving out of its row. So first of all write down the numbers 1 to 12 on a piece of paper. Then find the smallest value in the column and write (1) next to it, cross off 1 on your piece of paper, and then go hunting for the next highest value in the column, write (2) next to it and cross off 2 on your piece of paper, and so on ... Without that piece of paper with 1 to 12 written on it, you can easily get mixed up with your sorting ! Notice that in this data set there are tied values and we used the rule of applying the mean of the tied rankings to the members of the ties just as for the Wicoxon Mann-Whitney Test.

The next column in Table 6 shows the differences between the rankings of the items in Group A and Group B, and the final column is the square of these differences.

The last step is to solve a simple equation :

The significance of this Correlation can be assessed using the abbreviated Table 7 below

TABLE 7 : Spearman Rank Correlation Signifcance Table Abbrev.
Number of Observations p=0.05 Significance Level Number of Observations p=0.05 Significance Level

So we can see from this table that our Spearman Rank Order Correlation Coefficient of 0.925 is highly significant and we can justifiably claim in our paper and abstract that the exercise response of individuals in the two groups are almost identical, at the p<0.05 level of significance, and that they are separated by an almost constant difference of about 5 kg/m at every level of exercise.

So in conclusion the messages are to define your hypothsis(es) clearly and then select the appropriate parametric or non-parametric statistics tool to corroborate your conclusions.

Bibliography :

Nonparametric Statistics for the Behavioural Sciences, S. Siegel and N. John Castellan, McGraw-Hill Book Co., 2nd Edition, 1988, ISBN 0-07-057357-3

Statistical Methods, G. W. Snedecor and W. G. Cochran, Iowa State University Press, 7th Edition, 1980, ISBN 8138-1560-6

An Introduction to Medical Statistics, M. Bland, Oxford University Press, 1987, ISBN 0-19-261502-5


Dr Julie Bines, President of AuSPEN, Dept. of Gastroenterology and Clinical Nutrition, Royal Children's Hospital, Melbourne.
Ph +61-3-9345-5060. Fax +61-3-9345-6240

Dr Margaret Allman, Honorary Secretary of AuSPEN , Human Nutrition Unit, Biochemistry Dept.,The University of Sydney, Sydney 2006.
Ph +61-2-9351-3758 Fax +61-2-9351-6022


The views and opinions expressed in this Newsletter are not necessarily the views and opinions of the Australian Society of Parenteral and Enteral Nutrition. Reports and articles on techniques, procedures and products are provided for the information of the Members of the Society and their inclusion does not imply any endorsements from the Australian Society of Parenteral and Enteral Nutrition. No liability can or will be accepted by AuSPEN or its agents for the third party use of information in this Newsletter.

Dr Tom Hartley, Editor, 19/07/98.

Newsletter of the Australian Society for Parenteral and Enteral Nutrition
Editor : Dr Tom Hartley, 36, Pregnells Road, Sandfly, Tasmania 7150, Australia

Ph 03-6239-6475 (AH) 03-6222-8780 (BH) Fax 03-6231-3145

AuSPEN WWW Homepage :