Chain event graphs for chronic diseases
Martine J. Barons, Jane L. Hutton & James Q. Smith
Martine.Barons@warwick.ac.uk
go.warwick.ac.uk/MJBarons
J.L.Hutton@warwick.ac.uk
go.warwick.ac.uk/JLHutton
J.Q.Smith@warwick.ac.uk
Chain event graphs for the modelling and management for chronic diseases
such as Diabetes, Dystonia and Epilepsy
Motivation & Aims
•Chain Event Graph (CEG) is a new flexible class of graphical models.
•Health professionals explain to statisticians decisions required to make.
•CEG retains the framework of a probability tree in a more compact graph. •Patients and pharmaceutical statistician perspective included.
•Can CEG aid health professionals in decision-making?
•Models to enhance understanding of treatment options & progression.
Diabetes
• Treatment decisions are needed at key points:
• Pre-diabetes prevention of progression to
diabetes or delay of progression to diabetes.
• Diagnosis of Type I / Type II diabetes.
• Progression stages of
comorbid conditions (diabetic
retinopathy, hyperlipidaemia,
cardiovascular disease,
nephropathy, diabetic foot,
depression).
Dystonia
• Rare disease characterised by sustained
involuntary muscle contractions.
• Widespread spectrum mainly in young people.
• Types: primary dystonia, non-primary dystonia,
dystonia plus, paroxysmal dystonia.
• Different body locations affected.
Epilepsy
• People with epilepsy of different types will
respond differently at different ages, and stages
of the condition to the range of treatments
available [5].
• Effectiveness of treatment depends on age and
type of seizures.
• Types of dystonia respond to different treatments. • 50% of patients do not
have a recurrence after a
• Diagnosis and classification of dystonia are highly
first seizure.
relevant for providing appropriate management,
prognostic information, genetic counselling and
treatment.
Insulin pump. Mbbradford: Wikimedia Commons
Epi-dog showing the alarm that is connected
directly to the Medical Services. Elisabeth
Magnusson Rune. Wikimedia Commons
Chain event graphs
• Graphical models are useful in developing statistical models for complex data, as they allow statisticians and subject area experts to discuss how many
variables can interact over time.
• Bayesian decision analysis uses the power of graphical models to assist decision making in complex situations, when balancing benefits and risks is needed [4].
• Chain event graphs are a recent extension of graphical models, which are able to address situations in which, after one decision is made, or one variable takes
a particular value, give rise to a different set of possible actions from those arising from other decisions. The theory of chain event graphs has been developed
very recently [2],[3], [6] including methods for models selection [1]. Practical application of these is under-way.
• Missing data can be explored as part of a process of collecting data, or decisions made by people to provide information or attend appointments. Duration
between events and decisions might depend on previous events, as well as demographic characteristics, so time-dependent models are required.
Illustrative Chain event graph for
the MESS trial [5]: patients who
had had one or more seizures and
who, with their clinicians, were
uncertain whether to proceed
with treatment were randomised
to immediate treatment (t=1)
with optimum epileptic drug or
deferred treatment (t=0) until
clinician and patient agreed
treatment was necessary.
Follow-up at 3, 6 and 12 months
and annually thereafter.
Occurrence, type of seizure,
epileptic drug treatments and
side effects recorded. (S=0, no
seizure; S=1, seizure occurred).
Other possible actions: change
treatment (t=2), cease treatment
(t=(-1)).
Professor Jane Hutton
has been awarded a
NIHR research
methods opportunity
fund grant to bring
together clinical and
statistical experts to
explore opportunities
for applying and
extending this recent
development in
statistical theory and
modelling to
understanding and
managing chronic
medical conditions.
References:
1: FREEMAN,G. and SMITH, J.Q. (2011) Bayesian MAP Selection of Chain Event Graphs. J. Multivariate Analysis, 102, 1152-1165
2: J.Q. SMITH AND P.E. ANDERSON. Conditional independence and chain event graphs. Artificial Intelligence, 172(1):42–68, 2008.
3: P. THWAITES, J.Q. SMITH, AND E. RICCOMAGNO.(2010) Causal analysis with chain event graphs. Artificial Intelligence,174 (12- 13) :
889–909.
4: SMITH, J.Q. (2010) Bayesian Decision Analysis: Principles and Practice, Cambridge University Press, Cambridge.
5: MARSON, A., JACOBY, A., JOHNSON, A., KIM, L., GAMBLE, C., CHADWICK, D. (2002), Immediate versus deferred antiepileptic drug
treatment for early epilepsy and single seizures: a randomised control trial. The Lancet 365 2007-2013
6: SMITH, J.Q., ANDERSON, P.E, and LIVERANI, S. (2008) Separation Measures and the Geometry of Bayes factor selection for
Classification J Roy. Statist. Soc. B, Vol. 70, Part 5, 957 - 980