EB Research: Causal Inference and Biomarker Methods Development

The use of biomarkers to assess exposure and investigate biomedical questions has become an essential component of epidemiological research. Principled methods are needed to overcome the challenges inherent to using biomarker data in this research, specifically in the measurement, study design, and analysis phases. Of particular interest are issues of intra-individual variability and instrument sensitivity (i.e., limits of detection) on the performance of naive methods, in addition to the economic inefficiency of averaging costly biomarker measurement replicates or inserting arbitrary values for values below the assay limits of detection.

One area that may improve statistical and fiscal efficiency when using biomarkers is novel study design, such as pooling and outcome-dependent sampling, to reduce the number of expensive samples required. However, the literature on pooling study designs had previously been primarily limited to binary exposures and outcomes, and the outcome-dependent sampling design is only effective if analytical tools can make proper use of the data.

EB is focused on improving biomarker-centric methodology so that it is cost-efficient and incorporates understanding of biomarker biology and measurement. Within this area of research, EB is especially interested in the following problems.

In the context of high-cost assays, novel study designs that reduce cost and leverage statistical efficiency are a major focus, including one or more of the following:

  • Pooling samples
  • Collapsing calibration information prior to analysis
  • Utilizing efficient modeling strategies
  • Handling data appropriately below the limit of detection
  • Measuring biomarkers in a principally designed subset of the full cohort, guided by the outcome of interest

Branch researchers continue to adapt these designs and develop analytic methods to reduce various forms of bias and measurement error, and to adapt outcome-dependent sampling to a broad spectrum of study designs.

In studies of biomarkers, measurement error can occur in a variety of measurement-specific or more general ways, including intra-individual variability and instrument sensitivity. EB research informs the epidemiologic community of sources and impacts of measurement error, while developing and implementing methodological tools that maximize statistical efficiency while properly accounting for biomarker measurement error.

EB researchers also have extended the methodological framework for causal inference to develop methods for studying reproductive and perinatal epidemiology. The objective of this research is to develop methods using causal inference tools, specifically to improve researchers' understanding of confounding and bias.

Biomarker Methods Publications (PDF 196 KB)

Principal Investigator

Sunni L. Mumford, Ph.D.

EB Collaborator

Elizabeth A. DeVilbiss, M.S., M.P.H., Ph.D.

Selected Publications

Rudolph, J. E., Naimi, A. I., Westreich, D. J., Kennedy, E. H., & Schisterman, E. F. (2020). Defining and identifying per-protocol effects in randomized trials. Epidemiology (Cambridge, Mass.), 31(5), 692–694. PMID: 32740471. PMCID: PMC7400733 (available on September 1, 2021)

Schisterman, E. F., DeVilbiss, E. A., & Perkins, N. J. (2020). A method to visualize a complete sensitivity analysis for loss to follow-up in clinical trials. Contemporary Clinical Trials Communications, 19, 100586. PMID: 32577583. PMCID: PMC7300145

Perkins, N. J., Weck, J., Mumford, S. L., Sjaarda, L. A., Mitchell, E. M., Pollack, A. Z., & Schisterman, E. F. (2019). Combining biomarker calibration data to reduce measurement error. Epidemiology (Cambridge, Mass.), 30 Suppl 2, S3–S9. PMID: 31569147

Perkins, N. J., Cole, S. R., Harel, O., Tchetgen Tchetgen, E. J., Sun, B., Mitchell, E. M., & Schisterman, E. F. (2018). Principled approaches to missing data in epidemiologic studies. American Journal of Epidemiology, 187(3), 568–575.  PMID: 29165572. PMCID: PMC5860376

Schildcrout, J. S., Schisterman, E. F., Aldrich, M. C., & Rathouz, P. J. (2018). Outcome-related, auxiliary variable sampling designs for longitudinal binary data. Epidemiology (Cambridge, Mass.), 29(1), 58–66. PMID: 29068841. PMCID: PMC5718926

Malinovsky, Y., Albert, P. S., & Schisterman, E. F. (2012). Pooling designs for outcomes under a Gaussian random effects model. Biometrics, 68(1), 45–52. PMID: 21981372. PMCID: PMC4159259

top of pageBACK TO TOP