Modeling the Determinants of Health in Complex Policy Environments: A System Dynamics Perspective


Published on

This presentation for the Centre for Research on Inner City Health addresses the need to develop modeling tools to understand complex systems and the social determinants of health.

Bob Gardner, Director of Policy
Aziza Mahamoud, Research Associate, Systems Science and Population Health
Follow us on twitter @wellesleyWI

Published in: Technology, Health & Medicine
1 Like
  • Be the first to comment

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide
  • POWER data age-standardized % of adults 2005overall patterns – 3 X as many low income as high report health to be only fair or poor self-reported = good proxy for clinical outcomes but exactly the point here, capturing people’s experience of their health
  • In: SDoH lead to gradient of health in chronic conditionsplus affect how people can deal with the conditionsOut: complex and reinforcing nature of social determinants on health disparities
  • A way of forcing us to think about the interconnections, to demonstrate in our work, the SDOH we`ve chosen reflects where put the emphasisThe social determinants of health are inter-connected, interdependent and dynamicMultiple levels of determinants and pathwaysDisplay cumulative and reinforcing effects over timeIntergenerational influences and accumulation of social disadvantage
  • a famous quote by a statistician George E. P. BoxAll models are essentially simplification of reality, but some can make better depictions than others
  • Systems are compelx and we cannot afford to use simplistic models that assume linear connections
  • A problem solving methodology
  • dynamic complexities – co behaviour of system as a results of interactions of agents over timeCounterintuitive behaviour – unintended consequence, as a results of the distal feedback effects of our decisions and policies that we do not anticipateLeverage points – finding where in the system should we interveneThe focus is on system structure, rather than events and patterns – with emphasis on questions such as what’s causing the events we are seeing and why are patters occurring
  • It’s a reiterative process, a co-evolution process whereby our mental models are the centre, both tranforming the process of modeling as well as being transformed by it as we become explicit about our assumptionsOften, the greatest value is gained through the modeling process as opposed to the models built, the end result....this is sometimes not so obvious as stakeholders may put all the emphasis on the outcome of the simluation
  • For cchs variable, for some there was only 2001 data, and others, both data years
  • This is our dynamic hypothesis, or the hypothesized system structure with causal pathways and how interventions are affecting them.The model
  • 4 feedback loops and two delays – key concepts in system structureAll operating through income, and most through disability and some through chronic illness
  • We are testing our theory, or hypothesized causal relationships in the initial model to see if these are supported by our data, and how significant, strong, or weak the relationships are, and then we refine the dynamic hypothesis in a reiterative fashionLinear regression – some of the variables had two metrics, and both were tested
  • We are assuming interventions operate exogenously, i.e. they are unidirectional, which means we are not capturing any feedback effects from the changing health conditions and determinants on the interventions themselvesMany of the challenges due lack of trend data - inability to reproduce the historical epidemiologic profile
  • To remind people, that we now will be talking about simulation model results under different assumtions, and how structure we have discussed derives behaviour, we are looking at model results given assumptionsOur hypothesised causal relationships that underlie the bevaviour – structure determines determinesbevaviour
  • Modeling the Determinants of Health in Complex Policy Environments: A System Dynamics Perspective

    1. 1. Modeling the Determinants ofHealth in Complex PolicyEnvironments: A System DynamicsPerspectiveAziza MahamoudBob GardnerFebruary 14, 2013Centre for Research on Inner City Health1
    2. 2. Objective• Background• Introduction to simulation models andsystem dynamics• Overview of urban health model and userinterface• Hands-on experience with using the urbanhealth model and interface• Discussion22/14/2014 |
    3. 3. The Problem to Solve:Systemic Health Inequities in Ontario•there is a clear gradient in healthin which people with lowerincome, education or otherindicators of social inequality andexclusion tend to have poorerhealth•+ major differences betweenwomen and men•the gap between the health ofthe best off and mostdisadvantaged can be huge – anddamaging•impact and severity of theseinequities can be concentrated inparticular populations andneighbourhoods35/28/2013 |
    4. 4. these health inequities are basedin structured social and economicinequality – social determinants ofhealth• income inequality and poverty• inequitable access to childcareand early development resources• precarious employment, unsafework• racism, social exclusion• inadequate and unaffordablehousing• decaying social safety nets45/28/2013 |
    5. 5. Canadians With Chronic ConditionsWho Also Report Food Insecurity5
    6. 6. We live in a world that is increasinglymore complex, dynamic &interconnected6Better tools are needed to help us understand andmanage complexity!5/28/2013 |
    7. 7. Health Inequities = ‘Wicked’ Problems• this means they are:• shaped by many inter-related and inter-dependent factors• in constantly changing social, economic, community and policy environments• action has to be taken at multiple levels -- by many levels ofgovernment, service providers, other stakeholders and communities• solutions are not always clear and policy agreement can be difficult to achieve• effects take years to show up• have to be able to understand and navigate this complexityto develop solutions• we need to be able to:• identify the connections between multiple factors → the key pathways tochange → the mechanisms or levers that drive change along these pathways• specify the outcomes expected and the preconditions for achieving them• understand how to deploy these levers in specific social, institutional andpolicy contexts75/28/2013 |
    8. 8. Systems Approach at WellesleyInstituteWI has been working with stakeholders to explore theuse of systems thinking and modeling to• inform our understanding of the complexities ofthe social determinants of health• identify, assess and develop effective policyalternatives to advance health equity• consider how new approaches like this can beinformed by and connected to communityperspectives and policy needs85/28/2013 |
    9. 9. 9“All models are wrong, but some are useful”5/28/2013 | www.wellesleyinstitute.comGeorge E. P. BoxRobustness in the Strategy of ScientificModel Building, 1979
    10. 10. Why Develop Simulation Models?• Systems are complex• Help us be explicit about our mental models• Theory building and testing• A virtual world to design and assessintervention strategies• Tool for stakeholder engagement• Identify gaps in our knowledge of how asystem works105/28/2013 |
    11. 11. Systems Dynamics: What is it?• Field developed by Jay. W. Forrester at MIT inthe 1950s• “The use of informal maps and formal modelswith computer simulation to uncover andunderstand endogenous sources of systembehavior” (Richardson, 2011)Richardson, G.P. (2011). Reflections on the foundations of systemdynamics. System Dynamics Review, 27(3), 219-243.115/28/2013 |
    12. 12. System Dynamics Foundations• Complexity science• Focus on the whole rather than individual parts• Interdependency• Emergent behaviour• Stock and flow• Emphasis on feedback and non-linear thinking approachto solving problems• Provides tools and techniques that can help us:• Study a system from various perspectives• Look for emerging patterns and trends over time• Examine causes of policy failures and unintendedconsequences• Identify effective ways of intervening (leverage points)125/28/2013 |
    13. 13. ProblemDefinitionIdentifyingProblemCausesFocus on PolicyLeversModelformulation, testing & evaluationKnowledgeTranslationApplying the System Dynamics PerspectiveMentalModel135/28/2013 |
    14. 14. Wellesley Urban Health Model• a computer-based systems dynamics simulationmodel• helps us learn and understand the complex, anddynamic interconnections between a select numberof health & social factors• allows us to test what impact our decisions(interventions) will likely have on population healthoutcomes under various assumptions• offers insight into how these effects could play out, andover what timeframes145/28/2013 |
    15. 15. Model FrameworkPopulation health outcomesDeath rate Disability Chronic illnessSocial determinants of health interventionsSocial cohesionHealth careaccessAffordablehousingIncome/jobs BehaviouralChanging health & social conditionsAdverseHousingLowIncomeSocialcohesionunhealthybehaviourPoor healthcare accessDisabilityChronicillnessdeath155/28/2013 |
    16. 16. Model ScopePopulation: City of TorontoDistinguishes people by:• Ethnicity (Black, White, E Asian, SW Asian, Other)• Immigrant status (Recent, Established, Native-born)• GenderCaptures:• 5 areas of intervention: Healthcare access, Healthbehavior, Income, Housing (lower & non-lowerincome), Social cohesion• Outcomes: Changes in overall deaths and healthconditions, and disparity ratiosTimeframe: 2006 – 2046Age: 25-64165/28/2013 |
    17. 17. Outcome measures & definitionsUnhealthy behaviour & obese: the prevalence of peoplewho are smokers or obese (POWER 2009).Chronic illness: having two or more of 12 chronic conditionsas specified by the Association of Public HealthEpidemiologists in Ontario (POWER 2009)Access to health care: the ease of getting an appointment forprimary careDisability: limitation in activities of daily livingMortality: age-standardized death rateAdverse housing: overcrowding (insufficient bedrooms)Social cohesion: feeling “strong sense of communitybelonging "175/28/2013 |
    18. 18. Data Sources and Parameter EstimationAll data or estimates broken out by 30 subgroups:5 ethnicities x 3 immigrant statuses x 2 gendersCensus 2001 and 2006, Ages 25-64• Population sizes• Disabled % (“often or sometimes”)• Low income• Adverse housing for lower income and higher incomeDeaths per 1000 ages 25-64, City of Toronto combined 2000-05(ethnic differences estimated, not available)CCHS combined 2001-08 (4 cycles), Ages 25-64• Chronically illness• Healthcare access• Unhealthy behaviour• Social cohesion185/28/2013 |
    19. 19. Dynamic Hypothesis19The figure maps causal pathways in the model. The variables in red are the intervention options. The orange arrows indicatestabilizing effects, and blue arrows indicate reinforcing effects.Low income %Unhealthybehaviour %Poor access toprimary care %Disabled %Chronically ill %Death rateSocialcohesion %Adversehousing %Employment/incomeinterventionsHealth careinterventionsBehaviouralinterventionsSocial cohesioninterventionsHousinginterventions5/28/2013 |
    20. 20. Feedback loops in the model20- Blue arrows have reinforcing (+) effects- Red arrows have stabilizing (-) effects- Large + signs depict positive feedback loop% Low-incomePrevalence ofdisabilityPrevalence ofchronic illnessPrevalence ofunhealthy behaviour& obesityPoor health careaccess %AdversehousingSocial cohesioninterventions+Health care accessinterventionsUnhealthybehaviourinterventionsHousinginterventionsSocial cohesion--Employment/incomeinterventions----5/28/2013 |
    21. 21. Hypothesis Testing• Multivariate regression analysis was conducted totest causal connections and to produce effectestimates to parameterize the simulation model• Conducting analysis at the subgroup level (notindividual)• treat each subgroup as a single observation• Controlling for demographic variables215/28/2013 |
    22. 22. Limitations• Other important SDoH not included• Interventions are aggregate• Community support and care not captured• Lack of historical data to do trend analysis• Measurement issues associated with certain variables• Lack of projections for poverty and housing225/28/2013 |
    23. 23. Model Uses1. planning, strategizing and advocating for improvingpopulation health outcomes2. a learning tool to ground policy development & analysisfor dynamically interacting and complex SDoH• Introduce systems thinking3. allows decision-makers to ask "what if" questions andtest different courses of action4. building a shared understanding and consensus amongdiverse groups with differing views on issues5. eliciting stakeholder views and knowledge6. strengthening community dialogue235/28/2013 |
    24. 24. How do interventions work?• There are 5 intervention options to choose from• Interventions are ramped up over the period2011-15 and stay in force through 2046• Range from 0 to 100%• Broad-based• For example:• implementing 30% of the behavioural interventionreduces unhealthy behaviour by 30% in allpopulation segments245/28/2013 |
    25. 25. Interface & Scenario Demonstration255/28/2013 |
    26. 26. Discussion Questions• How could you imagine using the model?• Who would you use the model with?• What would need to be developed to facilitatethat use?265/28/2013 |
    27. 27. For more informationMahamoud A. Roche B, Homer J. Modeling theSocial Determinants of Health and SimulatingShort-Term and Long-Term InterventionImpacts for the City of Toronto, Canada. SocSci Med (in press).275/28/2013 |
    28. 28. © The Wellesley Institutewww.wellesleyinstitute.comAcknowledgementCollaborators1. Jack Homer, Homer ConsultingModeling2. Dianne Patychuck, Steps toEquityData collection3. Carey Levinton, Equity MagicStructural equation modelingAdvisors1. Nathaniel Osgood, University ofSaskatchewan2. Peter Hovmand, WashingtonUniversity3. Bobby Milstein, US CDC285/28/2013 |
    29. 29. THANK YOUPlease visit us atwww.wellesleyinstitute.com5/28/2013 |