Health Forecasting at UCLA


Population Health Forecasting

Population health forecasting can produce quantifiable information on the changes in health outcomes at different times in the future. It can estimate health benefits of action and harms of inaction both for the overall population and subpopulations of interest, taking into account their differences in demographic, socio-economic and behavior characteristics.

Although it is acknowledged that many of the quantitative and qualitative tools underlying current forecasts are imprecise, there have been dramatic improvements in forecasting methodologies over the past two decades (Armstrong, 2001). New techniques enable researchers to reduce forecasting error, but importantly also to estimate and assess uncertainty in the forecasting estimates. These advancements are driven partly by a better understanding of uncertainty in forecasting and the development of statistical tools to reduce forecasting uncertainty, as well as dramatic improvements in computing power which support extensive sensitivity analysis and statistical sampling methods to estimate uncertainty in the estimates. Forecasting provides a useful framework to anticipate the future impact of current decisions and actions or inaction. Policy makers, businesses, and individuals use them regularly to evaluate alternatives and make decisions.  A number of efforts to forecast disease- and risk-specific aspects of human health have been undertaken in the last two decades.

Micro simulation Modeling

Micro simulation models have been used to predict the impact of policy and behavioral changes on population health for more than two decades.  There are two major types of microsimulation models – static and dynamic – that are widely used to model potential impacts of social policies. Static models operate using cross-sectional databases that provide a snapshot of the population at a point in time.  In contrast, dynamic models operate using longitudinal databases that contain individual histories (Citro, 1991).

In the late 1980's, Milton Weinstein at Harvard developed the Coronary Heart Disease Policy Model, a static micro simulation model of policy and technological advances on the incidence, prevalence, and mortality from coronary heart disease, and changes in the cost of health care. Lee Goldman at UCSF, Department of General Internal Medicine continues to use this discrete-time, state-transition model to project future trends and assess the impact of interventions. He used the CHD Policy Model to estimate the effects of investments designed to change coronary risk factors between 1981 and 1990 on the incidence, prevalence, mortality and costs of CHD during this period and projected through 2015.

Population Health Modeling

More recently, researchers at Statistics Canada have developed a dynamic and continuous-time Population Health Model (POHEM), to assess the impact of different policy interventions and technologies on the health of the Canadian population (Will et al., 2001a). By utilizing continuous-time modeling the researchers simplified the modeling of multiple processes with many events that in a discrete-time model would results in an explosion of the number state transitions. This also simplifies the inclusion of additional factors and processes, whether dependent or independent of processes already in the model. Thus the model is easily expandable, and can grow over time as new evidence and a better understanding of each of the processes becomes available. Lastly, the modeling of individual level information over time enables the researchers to incorporate the joint distributions of variables that determine health outcomes, thus reducing the complexity of modeling covarying behaviors, comorbidity, and competing risks.

California Health Forecasting Prototype

We have adopted the population health model from Statistics Canada to reflect the California population using earlier work by Wolfson (1994) proposing that a comprehensive population framework provides an understanding and thinking about human health and can serve as a practical implementation foundation integrating many concepts and ideas. A continuous-time microsimulation framework provides sufficient flexibility to include a large number of factors and outcomes without exponentially increasing the complexity of the model. Inclusion of calendar time supports forecasting relating to the current and future populations in the state.

The existing model is a prototype built for the California population. It incorporates detailed demographic information on births, deaths, migration by age for five race/ethnic group starting early in the 20th century and extending through 2050. The demographic data is augmented with behavior information on exercise and obesity. Outcomes are calculated conditional on demographic and behavior data, as well as changes in base rates which approximate a combination of omitted variables and technological progress. Since the model simulates lifetime histories of individuals living in the state, it is possible to model delayed impacts of behaviors and outcomes through a variety of feedback mechanisms. For example physical activity is predictive of exercise levels later in life, but also lowers an individual’s weight. Similarly, incidence of disease can reversely impact physical activity levels; this can also be incorporated into the model. This approach to integrating information into a single framework is flexible and enables the incorporation of evidence from a range of sources. Furthermore assumptions on each of the relations are necessarily made explicit and thus transparent to users of the model.

The outcomes that are currently included in the prototype model are coronary heart disease incidence and mortality, all cause mortality, and medical expenditures associated with physical activity and obesity. Thus the model can already be used to make statements about the impact of interventions that target physical activity and obesity on disease incidence, mortality and direct personal medical expenditures into future decades. In particular the model provides additional insights into how the burden of disease is changing as migrant populations age, and trends in physical activity and obesity levels underscore the increasing costs to society of poor health behaviors.


Data Sources

Population and Demographics

1.      Natality
  •   Birth statistics, fertility rates, population numbers (Census), California Department of Finance (CA-DOF)
2.      Mortality
  •   Canada statistics (1870-1900), U.S. statistics (1900-1994), CA-DOF (1994 onward)
3.      Migration
  • Census and CA-DOF provide population numbers, net migration is residual of population + births – mortality -> out migration is estimated and in migration is the residual
4.      Education
  •   California Department of Education (CA-DOE)
5.      Marital Status
  •   Survey of Income and Program Participation (SIPP)
6.      Fertility
  •   California Department of Public Health (CA-DPH)

Risk Factors & Health Outcomes

1.     Physical Activity
  • National Health and Nutrition Examination Survey I (NHANES)
  • NHANES Epidemiologic Follow-up Study (NHEFS)
  • California Behavioral Risk Factor Surveillance System (CA-BRFSS)
2.     Overweight and Obesity
3.     Coronary Heart Disease
  • Literature review
  • National Center for Health Statistics – cause of death (NCHS)
  • American Heart Association- heart and stroke statistics (AHA)



  Evidence-based model to support advocacy of public health, research, and programs