Mary Beth Landrum leads this project at HCP.

The CanCORS Statistical Coordinating Center (SCC) has assisted in the refinement of study hypotheses, in the design of data-collection instruments, in the building of data-management tools, and in the secure and timely transfer of data from source data sites. The coordinating center also provides statistical and administrative support for the primary data collections (PDCR) sites in CanCORS, provides guidance in the analyses of the data, and serves as a hub of communication for the CanCORS consortium.

In addition, investigators in the SCC are conducting research into new methods for the analysis of the longitudinal and cross-sectional data arising in studies of patterns of care, access to care for subpopulations, and outcomes in nonrandomized population-based studies. Two of these methodological projects are being conducted primarily in the Department of Health Care Policy:

Propensity score methods with hierarchically structured observational data
Population-based observational studies often are the best methodology for obtaining generalizable results on access to, patterns of, and outcomes from care when large-scale controlled experiments are infeasible. Comparisons between groups can be biased, however, when the groups are unbalanced with respect to measured and unmeasured confounders. Propensity score methods have been proposed as a robust alternative to regression adjustment for observed differences between treatment groups. However, propensity score methods were developed and have been applied in cross-sectional settings with unclustered data. Data collected in CanCORS and similar studies are typically clustered or hierarchically structured, in the sense that patients are grouped together in one or more ways that may be relevant to the analysis (for example, by treatment center or hospital). We are conducting research on use of propensity score methods with hierarchically structured observational data. The specific aims of this project are to:

• develop statistical methodology for causal inference in the context of hierarchically structured observational data;
• develop diagnostics, which make use of the clustered structure, for testing the sensitivity of causal inferences to important violations of assumptions;
• apply these methods and diagnostics to data from the CanCORS PDCR sites.

Combining data sources to improve measures of patterns and quality of care The CanCORS project has collected unusually consistent and well-validated information about treatment. Because of the expense of medical record reviews, however, the scale of data collection at each site was strictly limited. Collaborating institutions, including state cancer registries and health plans, also collect data on treatment of broader cancer populations using administrative data sources. However, administrative data systems typically contain data that are less complete than those obtained from medical record review. In this project, SCC investigators are developing and applying methods for using administrative data in research on processes and outcomes of care for cancer, combining them with data from medical records and survey. The specific aims of this project are to:

• develop models for imputation of corrected data using data recorded with error in administrative systems and a “validation sample” of data from matched administrative and medical records;
• develop algorithms and software for practical implementation of these models;
• apply these models and algorithms to data from the CanCORS PDCD sites and evaluate the inferences obtained under these models, focusing on their precision and validity relative to those based solely on medical record and survey data.