The practical implications of using z-scores: Concern's experience in Angola
by Beth Matthews, Maureen Billiet (Concern field staff Angola) Annalies Borrell (Concern chief nutritionist Dublin).
In children the three most commonly used anthropometric indices are weight for height, height for age and weight for age. These indices can be used to assess wasting, stunting and underweight respectively. For the purpose of this paper we are discussing wasting (weight / height), and where the term malnutrition is used it refers to this index.
When a child's "weight for height" is assessed, the child's weight is being compared to a reference weight for a child of the same height. The reference weights for each height are known as the NCHS/WHO reference values and are derived from the combination of two distinct data sets, which cover two age groups:
- 2 to 18 years, based on nation-wide data from the US National Centre for Health Statistics (NCHS), and
- birth to 3 years from a more geographically and socio-economically restricted sample from the Fels Research Institute.
Though NCHS/WHO reference values are currently deemed the most appropriate for assessing the nutritional status of infants and young children, it is also recognised that they are not ideal. The WHO expert committee has recently recommended the development of a new reference conceming weight and height of infants and children.
- percentage of median: the ratio of a child's weight to the median weight of a child of the same height in the reference data, expressed as a percentage, e.g., if the median weight of the reference data for a particular height is I Okgs then to say that a child is 80% weight for height means that the child is 8kgs
- percentiles: the position of an individuals weight of a particular height in the reference values in terms of the percentage of values ex ceeded or equalled, e.g., if 10% of the refer ence population weigh less than the child being considered, then the child is in the 10th per centile.
- z-scores: by describing how far in (units called standard deviations) a child's weight is from the median weight of a child at the same height in the reference data.
While 'percentage of median' is the most commonly used index for selection criteria for feeding programmes it has not been deemed an appropriate index for use with the NCHS/WHO reference values. It has the advantage of being easy to use and, more influential perhaps, of being easy to understand. But this ease of use is paid for in loss of accuracy. The method cannot take into account the fact that the variation in weight found in normal children at one height will be different to the variation found at a different height.
Percentiles are considered appropriate descriptors for malnutrition, using the NCHS/WHO reference values, but do not distinguish well between children who are in the 3rd percentile. It is difficult therefore to classify further these children into moderately and severely malnourished. Z-scores and percentage of the median can be used to further distinguish between severely and moderately malnourished.
Z-scores use an application of statistical theory to describe how far a child's weight is from the average (the median for the NCHS/WHO values) weight of a child of the same height in the reference data. This "distance" is called a z-score. It is expressed in multiples of the standard deviation and is derived as follows:
|weight-for-height z-score =||observed weight - median weight|
where both the median weight and the standard deviation (how the different values are distributed about the mean) are those taken from the normalised growth curves derived from the NCHS /WHO reference values for the given height.
The z-score is a more sensitive descriptor than either percentiles' or 'percentage of the median'. The use of z scores has the advantage that it recognises that the spread of(or variation in) weights at one height may be different than that at a different height. Since the percentage of median' does not do this, its use may mean that children who are actually malnourished are not identified.
The table below illustrates how, as a child grows taller (older), s/he will be more likely tobe classified as malnourished by the use of z-score than by the use of percentage of the median.
|Classifications of boy's weights (kgs) at different heights using percentage of median and z-scores:|
For example a boy of 115 cms and 16.Skgs would not be admitted to supplementary feeding using percentage of median whereas using - 2 z-scores he would be selected.
Z-scores are useful in practice because they can:
- be applied to the individual or population;
- pinpoint any given weight and height, noting improvement or deterioration over time in relation to the reference values; and
- classify children of all ages and sizes equally.
Although there is substantial recognition that the z score is the most appropriate descriptor of malnutrition, health and nutrition facilities have been, in practice, reluctant to adopt its use as a means for individual assessment.
In programme use the practical requirements that a descriptor should fulfil are several. The descriptor should ensure the correct identification of malnourished children. It should allow estimation of absolute numbers of malnourished children through nutrition surveys. This would then lead to informed estimation of programme size and eventual programme coverage. Finally, the use of a particular descriptor should provide information on the relative seriousness of the nutritional situation, thus helping to identify the most appropriate type of intervention needed.
The use of z-scores in nutrition surveys
Concern, like many intemational agencies, analyses and reports nutrition survey results using z-scores. A comparison of the reported prevalence of malnutrition according to z-scores and 'percentage of the median' show that the two descriptors provide quite different results. See figure 1 below.
The difference in results has a significant impact in terms of programme planning and estimating food needs. In Malange (Angola), based on a population of 190,000 with a 17% under five population, the estimated programme sizes are described in the following table.
|Numbers based on z-scores||Numbers based on percentage of the median|
On average the use of the z-score selected an undemourished group which was 1.6 times greater than that selected by the use of the '80% of the median' method (range: 1.3 - 1.8). Similar differences were obtained from nutrition surveys carried out in other locations in Angola and in Sierra Leone.
|Survey results with % of median: z-score ratios|
|May '96||(ratio)||Nov. 96'||(ratio)|
|<80% of median||5.7%||(1.6)||4.5%||(1.7)|
|<80% of median||7.2%||(1.4)||3.6%||(1.6)|
|<80% of median||2.7%||(1.8)||-|
Again, the use of z-scores selected a group which was between 1.4 and 1.8 times greater then that selected by the use of the 'percentage of the median' method.
The difference in the results obtained from z-scores and percentage of the median may become larger when proportionally more older stunted children are included in the survey; for example, when the selection criteria for measurement is 'less than 110cm' as opposed to 'less than 5 years' and the former includes more older stunted children.
Z-scores as selection criteria
Z-scores were introduced as admission criteria in May and June 1996 in five Concem nutrition centres in the provinces of Huambo and Bie in Angola. The programmes were all dry supplementary feeding programmes.
The rationale for implementing z-scores was:
- its greater compatibility with the NCHS/WHO reference values and suitability for selecting malnourished children irrespective of height;
- the children being classified as malnourished in the nutrition surveys would be the same children being admitted into the nutrition centres thus ensuring compatibility between the programmes and the nutrition survey results;
- the Ministry of Heath in Angola together with WHO and UNICEF recognised and supported the use of z-scores;
- there had been no practical evaluation of the use of z-scores and its use in Angolan nutrition centres provided the opportunity for this sort of evaluation.
The impact of using z-scores to define admission criteria was evaluated through monitoring standard nutrition centre performance indicators like the number of admissions, length of stay etc. The study did not aim to evaluate differences in risk of mortality, illness or child development between the two indicators as there was no follow-up of children following discharge.
- Following the introduction of z-scores, a significant increase in admissions was reported. The increase was considered a direct consequence of the change in criteria since the methodology for case detection and the nutritional situation remained constant. These increases are described in Figure 2 and 3.
- There was an increase of 55 - 60% and 40-45% in the numbers of children being admitted to the nutrition programmes in Kuito and Huambo respectively.
- Once established, the nutrition centres would only have operated at 65% of their capacity had z-scores not been introduced.
- In November 1996, when programme closure was being considered, one fifth (20%) of those registered in the nutrition centres would have been discharged according to weight-for-height percentage of the median.
The impact on length of stay was evaluated following the introduction of z-scores. The criteria for discharge was set at - 1 z-score (approximately equivalent to 90% of the median) since no z-score cards with a calculated weight for -1.5 z-score (approximately 85% of the median) were available at the time of implementation. Average length of stay increased considerably in Kuito province but remained almost unchanged in Huambo province. These results are described in the table below. The average daily weight gain decreased correspondingly.
|Comparison of length of stay (in days) using percentage of median and z-score cut-offs|
|January-June (% of the median)||July-December (Z-Scores)|
Since the use of z-scores was a new concept, co ordination difficulties periodically arose between Concern and other agencies over issues arising from its use.
WFP co-operated with Concern during the initial period that z-scores were implemented and provided the necessary additional increase in food requirements. However, problems arose during food shortages later on during the year since numbers of children registered in the programme remained unusually high. The concept of z-scores was difficult for WFP field level food-monitors and staff at headquarters in Luanda.
At field level, staff experienced difficulties during the referral procedure to therapeutic centre with those children falling below -3 z-scores but not less than 70% weight-for-height. These children remained in the Concern supplementary feeding programme.
The implementation of z-scores was supported by the nutrition department at the Ministry of Health and also by UNICEF and WHO. However, MoH nurses at field level had difficulties with understanding z-scores.
Using z-scores... what are the practical implications?
The availability of z-score tables is essential for enabling staff to use the index in the field. Z-score cards are available and the methodology for using these cards is similar to that of the percentage median.
- Concern has produced their own z-score cards with values for - 1.5 z-score and a lower standard deviation included (this is a value necessary for calculating exact z-scores).
- WHO has produced z score cards in target countries.
- Other agencies are producing z-scores tables e.g. MSF and Oxfam
The calculation of individual z-scores
Sometimes rather than saying a child is between -2 and -3 z-scores we may wish to calculate the exact z-score value. This calculation involves three figures (see formula above); an observed weight, an expected weight and an associated standard deviation. In the calculation of z-scores based on the NCHS/WHO reference values there are two relevant standard deviations, a lower standard deviation which is used in the calculation of z-scores for children less than the median weight at a specific height and an upper standard deviation used to calculate z-scores for children greater than the median weight at a given height. It follows, that for selection purposes to feeding centres only the lower standard deviation is necessary. However, in situations where all children (those greater and less then the median weight at particular heights) require classification e.g. in the analysis of nutrition surveys, both lower and upper standard deviations should be provided for each height. Generally, field staff can learn the mechanical calculation but experience difficulties with the concept of z-scores.
Measurement of Child Progress
Nutritional deterioration and improvement of individuals can be monitored using z-scores as long as the median and the standard deviation for a given height are known. In circumstances where this is difficult alternative means for monitoring progress can be used, for example expected weight for a 5kg boy. This would be 15g/kg/day i.e. a total of 15g x 5kg x 7days = 525 grams per week.
The admission criteria are similar to those used in defining moderate and severe malnutrition in nutrition surveys (<-3 z-score for severe malnutrition etc.). The discharge criteria from therapeutic feeding programmes is >-2 z-score and from supplementary feeding programmes would normally be > - 1.5 z score. Staff who have been used to working with percentage of median' cut-offs may consider discharge too early at -1.5 z-score since this is almost equivalent to less than 85% for younger children. However, discharge at - 1 z-score may be considered inappropriate as it may prolong the length of stay in the nutrition programme. Using z-scores, selective feeding programmes will tend to include proportionally more older (stunted) children, therefore targeting those who are suffering acute or chronic malnutrition.
Compatibility between Surveys and Feeding programmes
Children can be referred during nutrition surveys since z-scores can be calculated by hand using a pocket calculator and z-score tables during the nutrition surveys. Coverage of the target group, as identified by the survey assessment, is more feasible since there is compatibility between surveys and programrnes.
Accurate results from nutrition surveys can be obtained from analysis by field staff without computers provided that the lower and upper standard deviations are given and calculations are sex-specific.
The indicators for assessing the effectiveness of a nutrition centre, as established by MSF and other NGOs and based on 'percentage of the median' may be inappropriate for programmes using z-scores. Target weight gains and length of stay may be inappropriate, particularly where discharge is at -1 z-score.
Teaching z-scores is undoubtedly challenging. There is a need to address imaginative and interesting ways to teach z-scores using graphs, colours and scales.
A change in criteria to z-scores may have significant implications for food requirements for supplementary feeding programmes. Predicted estimates for food requirements based on survey findings may be one and a half times previous estimates. Furthermore, the actual food consumption could be more than double if length of stay increases (x 1.5) and the number of admissions increases (x 1.5). In many circumstances where food resources are limited, it may be necessary to alter the criteria for selection accordingly i.e. from -2 z-scoreto -1.8 or-l.5 etc.
- This study provides some evidence to suggest there is an increase in the size of the target population for supplementary programmes when using z-scores.
- A single descriptor of nutritional status compared to the NCHS/WHO reference values that can be used for individuals (selection) and populations (surveys) will be useful for field practitioners operating nutrition programmes.
- In circumstances where nutrition centres are taken over by or assimilated into MoH structures and supported by the WHO, the use of z-scores ensures compatibility in selection criteria.
Questions and Challenges
- More field studies to evaluate the use of z-scores in- practice are required. These should include the identification and follow up of children excluded by either method.
- Z-scores should be included into all nutrition in emergencies training courses for practitioners and management staff, using imaginative and simple tools.
- There is a need for population based research focusing on the relationship between the various indices and outcomes such as mortality, illness, child development, etc.
1Gorstein J, Sullivan K, Yip R, de Onis Metal. Issues in assessment of nutritional status using anthropometry. Bulletin of the World Health Organisation 1994, 72:273 - 283
2de Onis M Habicht JP. Anthropometric reference data for international use. recommendations from a World Health Organisation expert Committee. Am. Journal of din. Nutrition. 1996, 64:650-658.
3WHO Technical Report No. 854. Physical Status. The use and interpretation ofAnthropometry pp.7-9 and pp. 219-224.
More like this
en-net: WHO Growth Standards
Why does the WHO Growth Standards have median in their z-score tables instead of mean? If they already have mean, why don't they have the percentage of median? It is easy to...
This new 2006 WHO Growth standards: What will they mean for emergency nutrition programmes? Dear Editor Whilst welcoming the principles which have driven the development of...
Summary of research1 Severely malnourished children managed in the MSF-run Centre de Récupération Nutritionnelle Intensive (CRENI) in Maradi, Niger A recent study by...
Summary of published research1 The need for a widely applicable growth reference for older children and adolescents has increasingly been recognised by countries attempting to...
Summary of Published Paper Mortality rates among children with severe malnutrition vary considerably between different treatment centres. This variation is due to differences...
en-net: Urgent: Weight for Height Z-scores: Unisex Z-score table used in West African countries vs. calculated Z score values
In the 2011 Guidelines for THE INTEGRATED MANAGEMENT OF SEVERE ACUTE MALNUTRITION published by ACF, Annex 4 depicts a Weight-for-Height table. This same table has been adopted...
Summary of published research1 More younger admissions to therapeutic feeding programmes are one of the implications of moving to the new WHO Growth Standards. A recent study...
Dear Field Exchange, I have the following comments to make about the article "The practical implications of using Z-scores: CONCERN's experience in Angola" (Field...
Experiences of ACF in Kabul city Claudine Prudhon is the head of the nutrition department in ACF HQ. Claudine is interested in improving the nutritional management of the under...
en-net: malnutrition rate
Dear All, Urgently want to know in any nutritional survey what is the difference we can expect in the percentage of SAM children we will find using WFH criteria and with MUAC...
FEX: Effect of nutrition survey ‘cleaning criteria’ on estimates of malnutrition prevalence and disease burden: secondary data analysis
Summary of research1 Location: Global What we know: Standardised methods for collection and reporting malnutrition prevalence data in nutrition surveys are used. What this...
Summary of paper presented at the Intenational Child Health Meeting on 'Managing Childhood Malnutrition' held in Birmingham At the recent Birmingham meeting reported on in...
Hello, I am passing on a question from a colleague. I am keeping his wording: "What is the lowest WAZ score that would be "compatible with human life"? I know that Epi-Info...
I am using Stata for analysis my data from Africa focusing on children aged from 6 to 24 months and need to calculate MUAC-for-age z-score and WAZ.
Circumstances leading to the setting up of SFPs (from WFP guidelines) Blanket SFP Targeted SFP A generalised SFP for prevention purposes maybe implemented in the absence of a...
Summary of published paper1 An infant having MUAC measured during the study in Kenya Current WHO guidelines for the management of severe malnutrition in children recommend...
I collected anthropometric data for my PhD, that I am analysing in STATA. I used the WHO macro for STATA to calculate Z-scores (WAZ, HAZ and WHZ) from my...
en-net: Analysis of survey data
Analysis of survey data in EPI INFO 6 and SMART (both against NCHS reference) seems to give different results. Has anyone else faced with the same problem? What is the reason?...
I tend to use the WHO "biological plausibility" flagging criteria with anthropometric data. These are very similar to the older CDC / NCHS flagging...
With the adoption and implementation of 2006 WHO Child growths standards for Emergency nutrition Programmes, It was also expected that - Treatment duration was to be shortened...
Reference this page
Beth Matthews, Maureen Billiet, Annalies Borrell (1997). The practical implications of using z-scores: Concern's experience in Angola. Field Exchange 1, May 1997. p5. www.ennonline.net/fex/1/practical