rss
J Epidemiol Community Health 64:16-21 doi:10.1136/jech.2008.085985
  • Theory and methods

Estimating the potential impacts of intervention from observational data: methods for estimating causal attributable risk in a cross-sectional analysis of depressive symptoms in Latin America

  1. N L Fleischer1,
  2. L C H Fernald2,
  3. A E Hubbard3
  1. 1
    Department of Epidemiology, University of Michigan School of Public Health, Ann Arbor, MI, USA
  2. 2
    Division of Community Health & Human Development, UC Berkeley School of Public Health, Berkeley, CA, USA
  3. 3
    Division of Biostatistics, UC Berkeley School of Public Health, Berkeley, CA, USA
  1. Correspondence to Ms. N L Fleischer, University of Michigan, Department of Epidemiology, Center for Social Epidemiology and Population Health, 109 South Observatory, 3rd Floor SPH Tower, Ann Arbor, MI 48109-2029; nancyfl{at}umich.edu
  • Accepted 16 May 2009
  • Published Online First 29 July 2009

Abstract

Background: The field of epidemiology struggles both with enhancing causal inference in observational studies and providing useful information for policy makers and public health workers focusing on interventions. Population intervention models, analogous to population attributable fractions, estimate the causal impact of interventions in a population, and are one option for understanding the relative importance of various risk factors. With population intervention parameters, risk factors are effectively standardised, allowing one to compare their values directly and determine which potential intervention may have the greatest impact on the outcome.

Methods: The difference between total effects and population intervention parameters was examined using naïve, G-computation and inverse probability of treatment weighting approaches. The differences between these parameters and the intuitions they provide were explored using data from a 2003 cross-sectional study in rural Mexico.

Results: The assumptions, specific analytic steps, limitations and interpretations of the total effects and population intervention parameters are discussed, and code is provided in Stata.

Conclusion: Population intervention parameters are a valuable and straightforward approach in epidemiological studies for making causal inference from the data while also supplying information that is relevant for researchers, public health practitioners and policy makers.

Footnotes

  • ▸ Additional data are published online only at http://jech.bmj.com/content/vol64/issue1

  • Funding This work was supported by the Oportunidades programme, the John D. and Catherine T. MacArthur Foundation Network on SES and Health; the Fogarty International Center of the National Institutes of Health [K01 TW06077 to L.F.]; and the National Institute of Child Health and Human Development [R01 HD40864 to Dr Paul Gertler].

  • Competing interests None.

  • Ethics approval Mexico; University of California, Berkeley.

  • Provenance and Peer review Not commissioned; externally peer reviewed.