Decision analysis

Decision Analysis
Prof. Carl Thompson
[email protected]
What is decision analysis?
• Based on a normative theory of decision
making: subjective expected utility theory
• Assumes
– Decision process logical and rational
– Works on basis that rational decision maker
will choose the option that maximises their
utility (the desirability or value attached to a
decision outcome)
What is decision analysis?
“Decision analysis is a systematic, explicit,
quantitative way of making decisions in
health care that can … lead to both
enhanced communication about clinical
controversies and better decisions.”
(Hunink, Glasziou et al, 2001, p.3.)
What is decision analysis?
• Assists in understanding of a decision task
• Divides decision task into its components
• Use of decision trees as a way of structuring the
decision task
– Uses evidence in the form of probabilities – so can
examine the risks associated with different options
– Examines the utility or cost associated with each
• Suggests the most appropriate decision option for
that particular situation
Stages in a decision analysis
• Structure the problem as a decision tree - identifying
choices, information (what is and is not known) and
• assess numerically the probability (chances) of every
choice branch
• assess numerically the utility (preferences) of every
outcome state
• identify option which maximises expected utility
• carry out a sensitivity analysis to explore effect of varying
• ‘toss -up’ if two options have same EU (Dowie, 1993)
Structure the problem
Elizabeth Harding: 78 year old lady with atrial
fibrillation (AF), currently not on any
anticoagulation therapy.
The decision problem: AF is associated with an
increased risk of thromboembolic stroke.
Treatment with anticoagulation therapy can
reduce the risk of stroke, however there is an
associated risk of bleeding. If prescribed
Warfarin, then regular monitoring of INR is
needed. If prescribed aspirin, there is a risk of
gastrointestinal side effects.
Structure the problem
• Illustrated using a decision tree
– structure of tree dictates values needed and
outcomes considered.
– should be structured with the aid of an expert
in the field
– need for compromise between simplifying
problem too much and including enough detail
for it to be relevant to the problem (Doubilet &
McNeil, 1988
The structure of a decision tree
• Square node
– Decision node
– Represents choice
between actions
• Circle node
– Chance node
– Represents uncertainty
– Potential outcomes of
each decision
Consequences or Chances
• Consequences of each decision option and
chance of event occurring
– Short term and long term
• Need best available evidence
– Includes risks and benefits of interventions
– Natural history of disease
– Accuracy and interpretation of diagnostic test
Clinical problem
• Alternatives for Mrs Harding include:
– No treatment
– Prescription of Warfarin
– Prescription of Aspirin
• Consequences or chances include:
– Stroke (either through thromboemolism or bleed)
• Which may or may not lead to severe disability
– Treatment side effects
• Use probability or chance of events occurring
• For each ‘branch’ in the decision tree, values have
to add up to 1 or 100%
• Specific measures of the uncertainty associated
with the decision
– Highlights the risks associated with each decision
• Probabilities should come from good quality
research evidence
Assessing Utility/Preference
• When there is more than one type of
consequence – valuation important
• Trade-offs between benefits and potential harms
of consequences
• Need clarification of the values involved
• Choice of intervention will often depend on the
values of the decision maker
• When considering values, need to consider
whether individual or societal
Measuring utility/values
• Need a strategy that weighs harms and benefits
explicitly in accordance with values of
• Utility measures
– Are a measure of the desirability of all the possible
outcomes in the decision tree
– Provide a numerical value attached to beliefs and
– Are measured on a numerical scale where 0 = worst
possible outcome and 100 = best possible outcome
Utility measures
• Quality Adjusted Life Years (QALY) commonly
used for population utility measures – 1 year in
perfect health = 1 QALY. Health states measured against
this (e.g. 2 years in health rated as 0.5 of perfect health =
1 QALY) Considers quantity and quality of life.
• Rating scales
• Standard Gamble
• Time trade off
Calculating expected utility
• Values are placed in decision tree by appropriate
• Expected value for each branch calculated by
multiplying utility with probability
• Expected values for each branch of tree added
together to give EU for each decision option
• Depending on nature of values, option with
highest/lowest value is the option that should be
Sensitivity Analysis
• Necessary if numbers used in analysis are
• Allows you to examine the effect different values
will have on outcome
• Known as sensitivity analysis
– vary uncertain variables over range that is considered
– Can calculate effect of uncertainty on decision
Sensitivity Analysis
Sensitivity Analysis on
Utility associated with having side effects of
Expected Value
No treatment
Utility associated with having side
effects of treatment
• If the value or utility of
having side effects
associated with the
treatment (with no
stroke) is less than 0.84,
then no treatment is the
preferred option. If the
utility is above 0.84, then
Warfarin is the preferred
Sensitivity analysis
• In this situation the decision is ‘preference
sensitive’ with regard to how the individual
feels about having side effects associated
with anticoagulation therapy.
• Decisions can also be ‘probability sensitive’
1 way sensitivity analysis HIV prophylaxis
• Makes all assumptions in a decision explicit
• Allows examination of the decision process used
• Way of integrating evidence into the decision
• Often insight gained during process more
important than the actual numbers used
• Can be used for individual decisions, population
level decisions and for cost-effectiveness analysis
• Probability estimates
– often data sets needed to estimate probability don’t
– Subjective probability estimates are open to bias:
overconfidence & heuristics
• Utility measures
– often ask individuals to rate a state of health that they
have no experience of
– Different techniques will result in different numbers
– Subject to framing effects
• Sources of Evidence
– Prodigy Guidance – Atrial Fibrillation
– Protheroe J, Fahey T, Montgomery AA, Peters TJ (2000) The impact of
patients’ preferences on the treatment of atrial fibrillation: observational
study of patient based decision analysis. BMJ 320: 1380-4
– Hankey GJ, Sudlow CLM, Dunbabin DW. Thienopyridine derivatives
(ticlopidine, clopidogrel) versus aspirin for preventing stroke and other
serious vascular events in high vascular risk patients. The Cochrane
Database of Systematic Reviews 1999, Issue 4. Art. No.: CD001246. DOI:
• Further Reading
– Llewelyn H, Hopkins A (1993) Analysing how we reach clinical decisions.
Royal College of Physicians of London.

similar documents