Previous research has shown that beyond factors related to the individual, supraindividual factors such as where a person lives or where they receive medical care may be relevant to both their risk of developing a disease and their patterns of severity, treatment, and outcome following diagnosis. Research focused on documenting and explaining such supraindividual variability in health and health care use can be broadly divided into two methodologically disparate research traditions: the small area variation (SAV) tradition and the neighborhoods and health (NAH) tradition. Each tradition, however, has its own limitations: With the SAV tradition, the lack of salient details regarding individuals raises the possibility of substantial confounding in results related to omitted patient-level variables, thus making it difficult to tell a cogent story about the role of supraindividual factors in health care use. The NAH tradition has also been hindered by omitted information, including that describing the availability of local health care.

Given the broad clinical and policy relevance of understanding the mechanisms underlying supraindividual variation in cancer risk and cancer care, this study employs a new approach to overcome these limitations, integrating the two research traditions through application of multilevel analytic approaches to appropriately detailed data (i.e., containing information about individuals, their neighborhoods, and their health care providers). Such an approach may allow better isolation of the role of the individual and the role of place in determining health outcomes and better understanding of the causes of supraindividual variation in these outcomes. The research seeks to evaluate the importance of supraindividual-level features in individual cancer patients’ treatment outcomes through leveraging a novel, linked data source comprised of clinical trial data that has been appended with data from administrative sources. The specific aims of this study are to:

• create a new, two-level dataset where patients are nested within treatment centers using existing CALGB lung cancer clinical trial data that is appended with US Census and other administrative data sources containing detailed information from patients, their neighborhoods and treatment centers, and their clinical trial outcomes;
  
  
• evaluate for the presence of supraindividual variation in clinical trial outcomes through basic hierarchical statistical methods (i.e., “null models”) where variance is partitioned across two levels;
  
  
• undertake hypothesis-driven analyses regarding specific sources of variation in outcomes within each of the two conceptual levels in both models described in the previous aim.