SYSTEMS THINKING AND DYNAMIC MODELING FOR STRATEGIC MEDICAL PROBLEMS

1
SYSTEMS THINKING AND DYNAMIC MODELING FOR
(STRATEGIC) MEDICAL PROBLEMS

• Yaman Barlas
• SESDYN Lab. www.ie.boun.edu.tr/sesdynDept. of
  Industrial EngineeringBogaziçi
  UniversityIstanbul, Türkiye

2

• SYSTEMS THINKING
• A way of seeing the world, the problems around
  us.
• As systems, indivisible collection of
  interconnected parts
• Origins go back to 1930s (Ludwig von
  Bertalanffy, a biologist)
• DYNAMIC Problems
• Things change over time
• A dynamic problem is one that necessitates
  continuous monitoring and action (management).
  Chronic problems.

3
(b) Retail price per gram of cocaine (in 1990
dollars). (Homer J. A Dynamic Model of Cocaine
Prevalence).

• World population growth. (Fey W. et al. The
  ECOCOSM Paradox)

(d) Boom and Bust Sales and Operating Income of
Atari, Inc. (Sterman J., Market growth, collapse
and failures to learn from interactive simulation
games).
(c) U.S.A. pulp prices - Deflated by CPI and all
trends removed (Sterman J., 2000).
Figure 1. Illustrations of some typical dynamic
patterns observed in real data
4
Figure 2. Categories of basic dynamic behavior
patterns typically seen in systemic feedback
problems.
5

• SYSTEMIC Problems
• The dynamics are caused by the internal structure
  of the system, not imposed by uncontrollable
  exogenous forces.
• We are interested in systemic dynamic problems.
• EXAMPLES
• Economic development/environmental degradation
• Inventory/Business cycles
• Public budget deficits
• Chronic high inflation
• Educational problems
• Full democracy/terrorism dilemma
• Fast business growth followed by collapse
• Systemic medical problems (unstable blood
  pressure)
• Intervention dilemma quick relief vs long-term
  effects
• Drug use/drug dependency

6
Figure 3. Illustration of model structure and its
dynamic behavior
7

• COMPLEXITY of SYSTEMS
• Dynamics Dynamic problems are harder than static
  problems. There are time delays involved between
  causes and effects between actions and
  reactions.
• Feedback The problem is further complicated when
  dynamics are created by operation of feedback
  loops. It means that which way the system will
  move is not easily predictable the evolution
  path unfolds gradually and continuously
  determines its own path into the future.
  (Path-dependent dynamics).
• Non-linearity Most system dynamics problems are
  non-linear. This means that the cause-effect
  relations between variables are not proportional.
  Non-linear effects are subtle, because a certain
  effect observed in a one range may not be valid
  at all in another range. Non-linearity
  furthermore often means that there are
  interaction effects between variables.
• Cause and effect separated in time and space In
  a non-linear dynamic feedback model with several
  variables, the cause-effect relations become
  detached in time and space.
• Scale As the number of variables increases, the
  complexity of the problem increases nonlinearly.
  Even small size policy problems involve tens of
  variables. At this scale, a non-linear feedback
  problem immediately becomes impossibly hard to
  track mathematically and intuitively.
• Human dimension Typical system dynamics problems
  involve human actors. So we must model not only
  the physics of the system (including information
  flows), but also how people react to situations,
  make decisions, set goals, make plans, etc.

8

• PRINCIPLES/LAWS of SYSTEMS
• Principle Meaningful macro behavior emerging
  from the interactions of micro components. The
  macro dynamics is not built into the behavior of
  individual components nor is it obviously
  predictable from the action rules of these
  agents.
• Dividing an elephant in two does not produce two
  small elephants
• Principle Counter-intuitive nature of systems.
  We human beings are naturally equipped only to
  deal with cause-effect relation close in time and
  space. The baby touches the stove with his index
  finger his index finger burns and it burns now
  and he learns. Our intuitive ability is further
  impeded by delays, errors, omissions and bias in
  data/information that we use in real life.
• Systems may exhibit better-before-worse dynamics
  (or vice versa)
• Principle Systemic misperceptions, biases and
  omissions are typical in decision making in a
  dynamic feedback environment. Experiments show
  that we are poor decision-makers in dynamic,
  non-linear feedback environments. Our intuitive
  time and space-constrained notion of causality
  cannot cope with systemic complexities. We
  ignore, distort or misperceive feedbacks, time
  delays and non-linearities in making decisions.
• Yesterdays solutions can be todays problems
• Principle Learning by experience is difficult
  and flawed in complex systems. Perhaps the most
  critical of all, learning is not
  natural/intuitive in complex dynamic
  environments. Experimental evidence shows that,
  with our reductionist intuition of causality, we
  make incomplete or plain wrong causal inferences
  about effectiveness of actions/decisions
• There is no enemy out there
• Faster is slower

9

• PRINCIPLES OF DYNAMIC SYSTEMS MODELING
• Operational causality rather than correlations
• Circular causality rather than static one-way
  causality
• Internal structure as main cause of dynamic
  behavior
• Focus on dynamic patterns rather than events
• Multi-disciplinary, rich system boundary
  (systems approach)

10
A negative feedback loop
11
(a)
Housing Pressure
Polution Pressure
(b)
Figure 4. Comparing (a) an exogenous, static
model of city population with (b) an endogenous,
dynamic model.
12
IN5050t
S
OUT20
(a)
OUT5050t
S
OUT200t
IN15050t
S
IN20
(b)
(c)
Figure 5. Dynamics of a stock (a) when the inflow
is increasing linearly (outflow constant), (b)
when the outflow is increasing linearly (inflow
constant) and (c) when both flows change linearly
and cross.
13
Figure 6. Inflow is oscillating and the outflow
is constant. Observe that the oscillations in the
stock are lagging behind the inflow oscillations,
with a phase lag of p/2 Integration creates
delays
14
(b)
(a)
(c)
Figure 7. Density-dependent growth (a) Causal
loop diagram, (b) stock-flow structure and (c)
possible dynamic behaviors
15
Ot Et ?(It-It) ?(SLt- SLt) WhereIt
kEt SLt ?Et
(a)
Figure 8. The general stock adjustment problem
applied to inventory management. (a) Causal loop
diagram (b) Stock-flow structure
(b)
16
(a)
(b)
Figure 9. The behavior of the stock adjustment
structure (a) Goods in supply line are ignored
in the order decision (b) supply line information
is taken into account. All other parameters are
the same.
17

• APPLICATION TOOLS/PLATFORMS
• Causal Loop Diagrams (maps)
• Stock-flow diagrams (models by maps)
• Simulation experimentation
• Interactive simulation games
• Micro-worlds (Virtual worlds)
• Learning laboratories
• Group model building
• (Organizational) learning environments

18

• SOME RESEARCH QUESTIONS
• Meaningful macro behavior emerging from the
  interactions of micro components.
• Modeling/methodology research
• Systemic misperceptions, biases and omissions in
  decision making in a dynamic feedback
  environment.
• Experimental interactive gaming research
• Difficulties/flaws in learning by experience in
  complex systems.
• Experimental interactive gaming research
• Applications
• University Management (UNIGAME), Total Quality
  Management (TQM GAME), Insurance Management
  (INSGAME), Shallow Lake Ecology, Dynamics of
  genetically modified agriculture, Endocrine
  disorders (diabetes, hyper/hypo-tyroid), body
  temperature regulation, dynamics of long-term
  blood pressure progression, BWATER Game
  (Management of Body Water Homeostasis disorders)

19
BWATER Game (Interactive simulator for
Management of Body Water Homeostasis disorders)

• To develop a system dynamics model that
  represents the structure of the body water and
  sodium balance for an individual normal adult
  subject
• To study body water regulation and its disorders
  by focusing on the fundamental feedback
  mechanisms in the normal and disease physiology
• To develop an interactive simulation model for
  treatment of a particular body water disorder,
  i.e. Water intoxication/ hyponatremia

20
Disorders of Water Metabolism
Dysnatremias Blood sodium concentration is
maintained between 135-145 mEq/L Hypernatremia gt
than normal Hyponatremia (water
intoxication/water poisoning) lt than
normal Hyponatremia is the most common
electrolyte abnormality in hospitalized patients
(Shafiee et al., 2003). Management of
hyponatremia Hyponatremia is clinically
important, its diagnosis management
constitutes a challenging problem. Severe
hyponatremia associated with substantial
mortality, rapid correction can lead to death To
date, all present therapies have limitations
(Verbalis, 2003). Treatment should weight
risks of hyponatremia against risks of
correction.
21

• Fundamentals of Body Water Homeostasis
• Major division of body water Extracellular (EC)
  Intracellular (IC) compartments
• Main electrolyte of EC is sodium (Na), main
  electrolyte of IC is potassium (K)
• EC sodium concentration mEq/L Amount of sodium
  contained in 1 liter of EC water
• Control of ECNa concentration is almost the same
  as controlling the EC osmolality, the number of
  osmoles per liter of water
• Concentration and content of Na regulated by
  two different systems

22
Control of EC Osmolality Body Water

• Hypothalamus controls TBW via a negative feedback
  mechanism thirst-ADH system.
• What is the advantage of maintaining a constant
• EC osmolality in water balance?
• Control of EC osmolality controls IC volume.
• The constancy of the IC volume is important for
  maintaining optimum function of most cells, and
  particularly important for the brain.

Causal-loop diagram for body water/ osmolality
control
23
Control of EC Volume Body Sodium

• Na is the principal determinant of ECV.
• Maintenance of normal ECV necessitates a balance
  between Na intake and excretion
• The fundamental control is on the output (not on
  the intake)
• Kidneys adjust Na excretion
• Na excretion mainly involves three factors
• Filtered load
• Aldosterone Hormone
• Atrial Natriuretic Hormone

Simplified causal-loop diagram for sodium and
ECFV regulation
24
OVERVIEW OF THE MODEL

• 9 sectors under 5 sector groups
• Body Water Sector
• Sodium (Na) Sector
• Endocrine Sector Group (3 sectors)
• Antidiuretic Hormone (ADH)
• Aldosterone (ALD)
• Atrial Natriuretic Hormone (ANH)
• Urinary Na (UNa) concentration sector
• Treatment sector group (3 sectors)
• Diuretic
• Aquaretic (ADH-Antagonists)
• Saline Infusion

25
OVERVIEW OF THE MODEL
Simplified causal loop diagram of the overall
model
26
BODY WATER SECTOR

• Structure simulates body water and its
  distribution between the EC and IC compartments,
  drinking and urine flow dynamics ...
• Drinking, insensible loss urine
• flow are routes of water intake
• and excretion.

Blood volume as a function of EC volume
27
BASE BEHAVIOR
key variables in the equilibrium run..
Equilibria of key variables with discontinuous
drinking...
28
BASE BEHAVIOR Water Loading
Base dynamics of urinary excretion following
ingestion of 1 L of water...

• data from Baldes and Smirk, (1934),
• data for eight subjects
• data for one subject
• (taken from Uttamsingh, 1985)

29
Sensitivity of Blood Volume to Different Levels
of Daily Water Intake
Approximate and simulated effects of changes in
daily water intake on blood volume (from Guyton,
2000)
Under normal conditions, blood pressure is not
affected by changes in water intake
30
Sensitivity of ECNa concentration to Different
Daily Sodium Intakes
Sodium intake varied between 0.2 of normal salt
intake and 5 times normal intake, a range of 25-
fold
ECNa concentration is kept within 1 control
limits when all feedbacks are intact
Simulated levels of ECNa concentration with
different daily sodium intakes
ECNa concentration is controlled with reasonable
effectiveness even with large changes in sodium
intake, as long as water intake is enough to
balance the losses
31
Effect of ADH-thirst feedback system on ECNa
concentration

• Effect of changes in sodium intake
• on ECNa conc - from (Guyton, 2000)
• under normal conditions
• after the ADH-thirst feedback has been blocked

each one of ADH thirst systems can control the
ECNa conc. with reasonable effectiveness
if both of them are blocked simultaneously,
ECNa conc. changes tremendously
32
Effect of ALD feedback system on ECNa
concentration

• Effect of changes in sodium intake
• on ECNa conc - from Guyton (2000).
• under normal conditions
• after the ALD feedback has been blocked

ECNa concentration almost equally well
controlled with or without ALD feedback control
33
EXTENDING THE MODEL

• Some sectors/structures are added to the original
  model (Diuretic, Aquaretic and Saline Infusion)
  variables for game related measurements

34
Development of hyponatremia
Dynamics of key indicators when only ADH or
thirst is dysregulated..
Appearance of hyponatremia when both ADH
thirst are dysregulated
35
THE INTERACTIVE DYNAMIC SIMULATOR (BWATERGAME)
36
Results of the Game Tests by Players
ECNa concentration
Total body water
Saline infusion decisions
Blood pressure
37
Results of the Game Tests by Players
Dynamics of ECNa concentration for five
players...
Dynamics of total body water...
38
Results of the Game Tests by Players
Dynamics of mean arterial pressure for five
players..
Dynamics of hourly correction rate..
39
Results of the Game Tests by Players
Dynamics of Na intake resulting from
decisions. for five players..
Dynamics of total water intake..