عنوان

Using longitudinal measurements to identify undernutrition :

Number

TLets705559

Title Proper

Using longitudinal measurements to identify undernutrition :

General Material Designation

[Thesis]

First Statement of Responsibility

Tough, Fraser

Title Proper by Another Author

a statistical investigation

Name of Publisher, Distributor, etc.

University of Glasgow

Date of Publication, Distribution, etc.

2016

Dissertation or thesis details and type of degree

Thesis (Ph.D.)

Text preceding or following the note

2016

Text of Note

Understanding the ways in which practitioners can identify and manage undernutrition is important within developing world countries. There is still much uncertainty when it comes to understanding which measures of undernutrition are the most effective predictors of adverse outcomes. This thesis explores how children grow and applies statistical methodology to three longitudinal growth datasets with frequent measurements in the first two years, seeking new insights into how measures of undernutrition can be used to predict future adverse outcomes. The three datasets are diverse - from Malawi, South Africa and Pakistan, the latter of which contains 4 subsets of different socioeconomic groups. The large number of children within the sets made it possible to test several different hypotheses. Growth charts (or reference charts) are charts which allow practitioners to compare a given infant's anthropometric measurements with a reference population. We developed growth charts from the available datasets using Generalised Additive Models for Location Scale and Shape (GAMLSS), a method which allows users to flexibly model distributions of measurements over time. The reference charts we developed describe the growth of samples of children, many of whom will not have grown at a healthy rate. It is preferable to compare children with healthy infants from a composite external standard. The World Health Organization (WHO) growth standard was developed from a variety of populations from across the globe which describes the growth of a 'healthy' population. This suggests an aspirational model, as opposed to a reference, which describes how a sample of children actually grow. In this thesis GAMLSS was used to determine whether real populations of pre-school children from the developing world fit this international standard. We found that relatively affluent populations fit the standard well, or even outperform it, while more deprived populations fall away to varying degrees, then mainly track parallel to the WHO mean beyond 6 months. This suggests that after the first 6 months children from the developing world have rates of weight gain roughly on par with the standard, although the children are much lighter. Plotting measurements on growth charts identifies those whose weight Z score or centile is falling relative to the reference. However, children initially at the extremes tend to regress toward the mean. Conditional weight gain (CWG) takes this expected movement into account, but can only be used within the population in which the child originates, due to certain statistical assumptions. We developed a generalised measure of CWG for use with the WHO external standard. This measure requires the correlation between pairs of groups of measurements at different time points, as the amount of regression to the mean is synonymous with this correlation. If data are not available at these time points, they can be interpolated by firstly computing correlations between all available data, then modelling the resulting matrix. We found that these correlation matrices are heterogeneous within the developing world. Therefore, constructing a generalised correlation model was not possible. This makes the use of the new generalised measure of CWG impractical without access to correlation models computed from local data. However, the measure may be useful within the developed world, where correlation matrices may be less variable. The analysis then explored the ways that children move between different nutritional states, defined as healthy, thin (wasted) and/or short (stunted), over 3-6 month (m), 6-9m and 9-12m timeframes, and the probability these states will lead to death. We used stochastic models to explore the probability of moving state conditional on previous state, exploring the pathways children take through different states over time. Within all timeframes, children who were wasted as well as stunted were more likely to die than wasted children, who were in turn more likely to die than stunted children. Furthermore, as children age, the conditional risk of death in the next time period decreases. However, relative to healthy children, all children were less vulnerable within the middle period (6-9m) regardless of state. Children who were wasted were at significantly higher risk than healthy children of later wasting, or becoming stunted as well as wasted, over all timeframes. However, wasting alone significantly increased the risk of later stunting only in the 3-6m timeframe. Across the 3-6-9m timeframes children were much more likely to move from either healthy to healthy to stunted, or healthy to stunted to stunted, than from healthy to wasted to stunted. This indicates children are more likely to move directly into a stunted state than from healthy to stunted via wasted. Change in weight (growth) has been shown to be a predictor of mortality in populations of children, but it is not clear if this measure is more predictive than the latest weight (size). Using weighted Cox proportional hazards models, we determined which of these measures is the most valuable predictor of mortality for the majority of children within each individual dataset, conducting analyses using variable levels of weightings for children at the extremes. We included weight-for-age and height-for-age as predictors within our models to determine what combination of predictors best predict mortality. In all unweighted analyses, size was the best predictor of time until death. However, as the weighting increased, growth entered as the best predictor in populations with low rates of undernutrition. In contrast, size always remained the strongest predictor within populations with high rates of undernutrition, since in these populations, such a high proportion of children fall away from within the centre of the normal range, making growth pattern non-discriminating. This programme of work applied statistical techniques to three diverse longitudinal datasets, gaining insights into how children grow between different socio-economic backgrounds. We investigated measures of size and measures of growth, utilising methods that control for the inevitable fact that healthy children at the centre of the population distribution tend to dominate analyses. Furthermore, these methods were both multidimensional and time dependant, providing us with a useful framework to assess child growth while controlling for influential factors. The results should improve understanding of both the aetiology of undernutrition and its clinical management.

HA Statistics ; Q Science (General) ; RJ Pediatrics

Tough, Fraser

University of Glasgow

Electronic name

p

[Thesis]

276903

a

Y