Connecting population health researchers with complex systems modelers: Why? How? Carl Simon Professor of Mathematics, Economics, Complex Systems and Public Policy The University of Michigan
Why? Key insight about problem-solving: Everything is a system, composed of interdependent and interacting components. Ignoring these interconnections often leads to unanticipated consequences.
Failures for lack of Systems Thinking Natural Systems DDT Bring some wild rabbits to Australia so you can still hunt Health Systems Hospital director Use only the strongest antibiotics Use antibacterial soaps Personal physician Indiscriminate prescription of medications
Systems thinking: first steps Identify key variables and main actors Especially those in the problem under consideration Draw (causal) diagrams to illustrate connections and feedbacks If possible, quantify some of those connections Result: a model heuristic, computer, mathematical
Start with Simple Models K.I.S.S. Principle Capture the bare elements of the system under study and its main components Make many simplifying assumptions
How Swine Flu Got Started
BENEFITS OF MODELING THE SPREAD OF DISEASE (C. Simon) Models can be used to identify trends, make general forecasts, or estimate uncertainty in forecasts. Models can shed light on large-scale dynamics that occurs as a result of complex biological and sociological interactions. Models can provide conceptual framework for data collection so that the most data are collected and are properly integrated. Models can help researchers more accurately estimate biological and sociological parameters associated with the spread of disease. In epidemiology, models are the basis of theoretical and simulation experiments in a realm where field experiments are unethical and/or impractical. Models allow for comparison and contrast of spreads of different diseases, at different times, or in different populations. Models can lead to a better understanding and more careful formulation of the assumptions used to study the mechanisms that influence the spread of disease. Models provide structures for organizing, coalescing and cross-checking diverse pieces of information.
Simple system of disease spread c New infection b c Recovery Susceptibles Infected Removed Loss of immunity Background death
Characteristics of Simple Systems (economics, ecology, biology, business, ) Homogeneity ( representative agent ) Equilibrium (no or simple dynamics) Random mixing (no structure or organization) No feedback; no learning/adaptation Deterministic No connection between micro and macro phenomena
What do we gain and what do we miss with these simple models? Gain: insights into connections e.g., R 0 (tipping points) ecological cycles, consumer demand decreasing with price Gain: Simple solutions, buzzwords, panancea
What do we gain and what do we miss with these simple models? How much can we trust these insights? The real world may be much more complex than these simple models suggest. What do we miss? Keep it simple, but not too simple. (Einstein) Natural question: what happens when we relax these simplifying assumptions or add more realism?
Characteristics of Simple Systems (economics, ecology, biology, business, ) 1. Homogeneity ( representative agent ) 2. Equilibrium (no or simple dynamics) 3. Random mixing (no structure or organization) 4. No feedback; no learning/adaptation 5. No connection between micro and macro phenomena
Complex System Approach Heterogeneous agents/ diversity Nonlinear dynamics Contact structure; networks; organization Feedback, adaptation, learning, evolution Stochasticity Emergence, multi-level
Some Complex Systems Techniques Dynamical systems Agent-based modeling Cellular automata Discrete event systems Computational social and decision science Game theory Thresholds, Tipping Points Networks Genetic algorithms****
Hooking up with Complex System Modelers Spoiled at University of Michigan Thin walls, with strong top-down encouragement of cross disciplinary research projects UM Center for the Study of Complex Systems Multiple courses in ABMs, networks, GAs, system dynamics CSCS Computer Lab (Rick Riolo) NSF IGERT (CS and social science) Dynamic modeling courses in Epid, Biol, Bus,
Model Development (UM-CSCS Approach) Analytically solvable dynamical systems model Math analysis, phase portraits, calculus, Computer solvable dynamical systems model = Systems Dynamics Math-Lab, Stella, Madonna, Sterman: Bus. dynamics Agent-based models (when heterogeneity kicks in) Netlogo, Repast, Python,.
Pluses and Minuses Analytically solvable dynamical systems model +: Solid general results +: Straightforward listing of assumptions and parameters -: Strong assumptions required (e.g., homogeneity) Systems Dynamics -: general results require MANY runs +: Straightforward listing of assumptions and parameters +/-: Weaker homogeneity assumptions Agent-based models -: Convincing general results very difficult to obtain -: Rare and tedious to list underlying model assumptions (Netlogo; Tom Boyce) +: Allows for realistic heterogeneity, feedback, networks, -: difficult to get published -: not enough to show that one run matches real world data -
Publishing ABM models PLoS One Nature Complexity Some tightly focused field journal Conference Proceedings New Journal of Policy and Complex Systems Dr. Mirsad Hadzikadic, editor (UNC Charlotte) Liz Johnson, managing editor
Learning about Complex Systems Short courses Patty Mabry (OBSSR) Institute on Systems Science Santa Fe Institute summer courses New England Complex Systems Institute (NECSI) Nate Osgood Agent-Based Modeling Bootcamp & Incubator August 18-23, 2014 http://tinyurl.com/healthmodelingbootcamp2014
Learning about Complex Systems On-line resources Nate Osgood s videos http://tinyurl.com/abmforhealth Scott Page s MOOC and Learning Company DVD Netlogo on-line tutorial
Final thoughts on modeling population health issues There is a full tradition of publishing mathematical models of infectious disease transmission. More biology and less social structure than population health issues. of social spread of tobacco use. National Cancer Institute Cannot imagine how to do a systems approach to population health concerns without complex systems modeling.
Final thoughts on modeling population health issues There is a full tradition of publishing mathematical models of infectious disease transmission. More biology and less social structure than population health issues. of social spread of tobacco use. National Cancer Institute Cannot imagine how to do a systems approach to population health concerns without complex systems modeling. Thank you, OBSSR, for realizing this too!