Peter Vedsted

BACKGROUND: Whether longer time to diagnosis (diagnostic delay) in patients with cancer symptoms is directly and independently associated with poor prognosis cannot be determined in randomised controlled trials. Analysis of observational data is therefore necessary. Many previous studies of the influence of delay on survival have been subject to great uncertainties due to unmeasured confounding and measurement error. Systematic evaluation of confounding is needed.

AIM: The purpose of this project is to assess confounding in studies of delay and survival in breast cancer patients. We aim to: 1) Clarify which factors are considered confounders or intermediate variables in the literature. 2) Assess how and to what extent these factors bias survival estimates.

CONSIDERATIONS: As illustrated in Figure 1, symptoms of cancer may alert patients, GP's, and hospital doctors differently and influence both delay and survival time in different ways. We therefore assume that the impact of confounding factors depends on the type of delay studied (e.g., patient delay, GP delay, referral delay, or treatment delay).

MATERIALS & METHODS: The project includes systematic review and methodological developments based on empirical data.

STUDY DESIGN 1: To identify scientific similarities and discrepancies in choice of co-variables in studies of delay and breast cancer survival we conduct a systematic review according to CRD guidelines with special focus on statistical model building. Our study is based on a previous work by Richards et al.: Influence of delay on survival in patients with breast cancer: a systematic review (Lancet 1999;353:1119-26).

STUDY DESIGN 2: To assess the importance of potential confounders and intermediate variables in observational studies of delay and survival we test commonly used co-variables on empirical data. The Danish Breast Cancer Cooperative Group Database (DBCG) has a suitable size and sufficiently detailed data to develop statistical models. Co-variables are gathered by linking DBCG data with the National Patient Registry, the Danish Cancer Registry, the National Pathology Registry, and Statistics Denmark. Among other statistical methods the empirical study will involve use of sensitivity analysis. The statistical analyses will be performed using Stata 9.

PERSPECTIVES: A systematic evaluation of confounding will improve our ability to interpret observational studies of delay and cancer survival and enable us to make a more appropriate and reasonable choice of statistical models when analysing delay and survival data.