Handling Uncertainty in a Medical Study Dietary Intake During Pregnancy

1
Handling Uncertainty in a Medical Study -
Dietary Intake During Pregnancy
Adele Marshall, Queens University Belfast David
Bell, Roy Sterritt, University of Ulster
2
Background
Previous research have shown

• an association with low birth weight and
  increased incidence of
• neonatal mortality and higher neonatal morbidity

• Impairment of foetal growth low birth weight
  due to inadequate maternal and foetal nutrition
  - may lead to increased risk of the infant to
  chronic diseases in adulthood

• Maternal diet may also effect the longer-term
  health of the offspring

Alternatively, excessive foetal growth
(macrosomia) leading to high birth weight also
has associated risks - difficulties in childbirth
and intensive postnatal care
3
Introduction
Research project concerned

• with the accurate recording of details of
  pregnant women at
• 28 weeks gestation

• analyse the potential connection between
  dietary intake and
• birth weight of the baby

The study is an extension of a major new research
project underway at the Royal Victoria Hospital,
supported by MRC as part of the HAPO
(Hyperglycaemia and Adverse Pregnancy Outcome)
Study.
4
HAPO Study
The USA National Institute of Health (NIH) has
recently approved an extensive international HAPO
(Hyperglycaemia and Adverse Pregnancy Outcome)
study in 16 key centres around the world
Each centre will recruit 1500 pregnant women over
a 2 year period
Information gathered on anthropometric
measures, socio-economics, family history,
metabolic status, measures of pregnancy outcomes.
Hypothesis Hyperglycaemia in pregnancy is
associated with an increased risk of adverse
maternal, foetal and neonatal outcomes
5
Extending HAPO Study
In addition to the HAPO variables, the RVH have
started a collection of variables on dietary
intake of the pregnant women
Interviewed at 28 weeks gestation for completion
of a food frequency questionnaire
Food frequency questionnaire designed to gather
as much information as possible regarding the
dietary intake of the pregnant mothers
Research the association between dietary intake
during pregnancy and birth outcome
6
Food Frequency Questionnaire
7
Study Objectives
The main objectives of this study will be the
evaluation of
(i) dietary intake in pregnancy compared to
non-pregnant women of childbearing age
(ii) links between diet and lifestyle/socio-econo
mic factors
(iii) links between diet and maternal/foetal
glucose
(iv) links between dietary intake and pregnancy
outcome
(v) possible follow-up of links between dietary
intake in pregnancy and both maternal and
outcome at two years
8
The System
The key aspects of this work are - to identify
potential causal relationships in the study data,
while - engineering an intelligent system.
Such a system is required to handle uncertainty
on several levels

• Uncertain implicit nature of the domain

• Uncertain nature of the research project - no
  hypotheses to test?

9
Bayesian Belief Networks (BBNs)

• Statistical graphical models which model
  potentially causal information

• BBN is a network/graphical structure of
  inter-related variables which has conditional
  probabilities attached

• The networks are based on Bayes Theorem which
  provides the conditional probability that an
  event occurs given that some other prior event
  has already happened

10
Bayesian Belief Networks (BBNs)

• Nodes - represent the variables
• Arrows - represent relationships between
  variables

11
Potential Causal Relationship
Sudden and severe restriction of energy and
protein intake during pregnancy reduces birth
weight up to 300g (Stein Susser 1975) - Siege
of Leningrad (1941-43) Dutch famine (winter
1944-45)
12
Hypothetical BBN
13
Preliminary Results
Currently the study has recruited and interviewed
294 women
108 babies have been born (so far) with birth
weights ranging from 1.9kg - 4.93kg
(4.18lb-10.85lb) with an average of 3.51kg (7.7
lb)
Variables identified as having a direct influence
on birth weight include the frequencies at which
pasta, bread and potatoes are consumed
Other potential influences are the amount of
cigarettes smoked, the number of children that
the mother already has and whether there are any
relatives with diabetes
14
Resulting BBN
15
Reduced BBN
16
Summary
A food frequency questionnaire is designed and
used to record dietary information. There is a
high level of complexity and uncertainty involved
in determining the nutritional value and content
of various different foods. In addition, the
project itself is a research project with
undiscovered knowledge and unproven
hypotheses. BBNs are used to capture significant
relationships on baby birth weight. As the data
set grows, it is expected that the number of
edges and complexity of the system will also
increase. With increased complexity, the need
for further models will increase - development
of an expert system.
17
Further Work

• Currently with 6 months data, continue for the
  next 6 months to identify
• potential causal relationships as the data set
  grows
• From 1 year, begin the trial predictions of
  birth outcome utilising learnt BBN
• Group/cluster variables within the BBN.
• As the study progresses, BBNs will be utilised
  , not only as a tool for
• representation of a final model, but also as a
  research development
• technique used for discussion purposes
  uncovering potential.

18
Acknowledgements
P.R. Flatt, M.J. Eaton-Evans, School of
Biomedical Sciences, Faculty of
Science, University of Ulster at Coleraine.
A. Hill, Nutrition Dietetics Royal Hospital
Trust Belfast
D. McCance, D.R.Hadden Metabolic Unit Royal
Victoria Hospital, Belfast
19
Developing the BBN
HUMAN Probabilities and Structure defined by
consultation with experts and literature
SYSTEM Learning of structure and probabilities
from data
COMBINE Expert structure amended with
discoveries and probabilities obtained from
data or the structure induced from data is
adjusted to include expert reasoning