Page 101 - Contributed Paper Session (CPS) - Volume 3

CPS1954 Vincent C. et al.
1. Introduction
According to the latest joint malnutrition estimates by United Nations
Children’s Fund, World Health Organization (WHO), and World Bank Group
(2018), it is estimated that stunted growth is prevalent in 22.2% of the children
population under the age of 5 in 2017 or over 150 million children worldwide.
This is particularly serious in low to medium income countries where the rate
of stunting is 35.0%. A major contributor to stunted growth is prolonged
faltering, which comes with adverse consequences such as increased
susceptibility to diarrhoea and respiratory infections (Kossmann et al., 2000),
abnormal neurointegrative development (Benitez-Bribiesca et al., 1999) and
capital loss to the labour market (Hoddinott et al., 2013). Therefore, it is
imperative to take early preventive measures so that these impacts can be
minimised. In order to implement the preventive measures, we need to first
identify faltered children in the population of interest. In addition, it is also
important to distinguish between the different types of growth patterns as
each type represents particular growth behaviour. For example, children who
caught up after faltering may have taken nutritional supplements. The strategy
can then be replicated to other children in the cohort to improve their growth.

2. Methodology
A popular method for modelling longitudinal growth data in the
epidemiology literature is the broken stick model defined as follows:

for = 1, . . . , , = 1, . . . , where ∈ ℝ denotes the height-for-age z-

score (HAZ) for child on the j-th measurement occasion at age , gives
+

the positive part of x and = ( , . . . , ) is an ordered vector of
1

predetermined internal knots or change points such that < ··· < , . The
1

random intercept and error are both assumed to be normally distributed

with parameter vectors given by ( , ) and such that < ··· < , . The
2

1

random intercept and error are both assumed to be normally distributed

with parameter vectors given by (0, ) respectively. The child specific and
2

time invariant controls for the heterogeneity in the HAZ at birth around the

population mean , and it is assumed to be uncorrelated with the error term

. The broken stick model fits + 1 piecewise linear segments with breaks

at to model the growth trajectory calibrated in terms of the HAZ. The
formulation in (2.1) enables an individual child’s growth velocity to be
90 | I S I W S C 2 0 1 9