TL;DR
This paper introduces a new statistical method for analyzing survival data in case-control family studies, specifically addressing correlated failure times within families. The method uses a flexible "shared frailty" model and a novel "pseudo full likelihood" approach to estimate risk factors and the degree of familial dependence, backed by rigorous theoretical guarantees.
What Problem Does This Paper Address?
Analyzing survival data, especially in family studies, is complex. When individuals within a family share common genetic or environmental factors, their failure times (e.g., age of disease onset) are correlated. This correlation, known as "frailty," needs to be accounted for to accurately estimate the effects of risk factors. Traditional case-control studies, which are effective for rare diseases, face additional challenges when extended to family structures due to the retrospective sampling and the complex dependencies. Existing statistical methods often lack established large-sample theory for these specific case-control family study designs, making it difficult to assess the reliability of their estimates. This paper aims to fill that gap by providing a robust and theoretically sound method for this scenario.
Methods
The core of this paper is a novel estimation technique for a general semiparametric shared frailty model. This model allows for flexible assumptions about the distribution of the "frailty" (the unobserved factor causing familial dependence) as long as it has finite moments. The authors propose a "pseudo full likelihood" approach, which is a clever way to handle the complexity of the data by approximating the full likelihood function.
Key methodological contributions include:
- A novel estimation procedure: This procedure is designed to be non-iterative and easily computed, which is a significant advantage over some existing methods that require complex iterative algorithms.
- A nonparametric estimator for the cumulative baseline hazard function: This is a crucial component for understanding the underlying risk over time, and the proposed estimator is simple and efficient.
- Rigorous large sample theory: The paper provides theoretical guarantees (consistency and asymptotic normality) for their proposed estimators, which is essential for validating the method and understanding its behavior with larger datasets.
Datasets
The authors' method is illustrated using a real-world dataset from a breast cancer study conducted at the Fred Hutchinson Cancer Research Center. This study involved case and control probands and their female relatives (mothers). The paper also uses simulated data to evaluate the performance of their proposed method under various scenarios.
Key Insights and Ideas
- Bridging the gap for complex family data: The paper provides a much-needed statistical tool for analyzing correlated survival times in case-control family studies, a scenario that has been challenging for existing methods.
- Flexibility in frailty modeling: The ability to accommodate arbitrary frailty distributions with finite moments makes the proposed method broadly applicable.
- Computational simplicity: The development of a non-iterative, easily computed estimator is a significant practical advancement.
- Theoretical rigor: Providing large-sample theory builds confidence in the reliability of the estimates obtained from the proposed method.
Key Results
The simulation studies presented in the paper demonstrate that the proposed method performs well. The estimators show minimal bias and confidence intervals with appropriate coverage rates across various censoring levels. The paper also compares their method to existing approaches (Hsu et al., Shih and Chatterjee) and finds that their proposed method is either comparable in efficiency or slightly more efficient, while being computationally simpler. The analysis of the breast cancer data suggests a significant dependency between the ages of onset for mothers and daughters, even after adjusting for first full-term pregnancy, indicating the importance of accounting for familial factors.
Limitations and Future Work
The authors acknowledge that the asymptotic theory for the case of "left-restricted" sampling (where proband observation times are restricted to a certain range) becomes algebraically very complicated. They express hope to develop this asymptotic theory in future work. Additionally, while the paper focuses on the gamma frailty model for simulations, the proposed method is designed for a more general class of frailty distributions.
Summary updated