Abstract
A case-control family study is a study where individuals with a disease of interest (case probands) and individuals without the disease (control probands) are randomly sampled from a well-defined population. Possibly right-censored age at onset and disease status are observed for both probands and their relatives. Correlation among the outcomes within a family is induced by factors such as inherited genetic susceptibility, shared environment, and common behavior patterns. For this setting, we present a nonparametric estimator of the marginal survival function, based on local linear estimation of conditional survival functions. Asymptotic theory for the estimator is provided, and simulation results are presented showing that the method performs well. The method is illustrated on data from a prostate cancer study.
Introduction
In epidemiological research, studying rare diseases presents a significant sampling challenge. A simple random sample of the general population often yields too few "events" (disease onsets) to provide statistical power. To circumvent this, researchers frequently employ case-control family studies. In this design, individuals with the disease (cases) and without it (controls) are recruited as "probands," and data is subsequently collected on their relatives.
While efficient for data collection, this structure introduces two primary statistical hurdles: selection bias and clustering. Because the sample is intentionally weighted toward cases, it does not represent the general population's risk profile. Furthermore, family members share genetic predispositions and environmental factors, meaning their data points are not independent. Standard survival analysis tools, such as the Kaplan-Meier (KM) estimator, assume independent and identically distributed data; applying them to case-control family data results in significant bias.
The Methodological Approach
Zucker and Gorfine propose a new, fully nonparametric estimator designed to address these complexities without making rigid assumptions about the shape of the survival distribution. This work builds upon and improves a 2017 kernel-smoothed approach by Gorfine et al., which, while innovative, relied on less efficient median-based calculations and lacked a formal proof of asymptotic properties.
Local Linear Estimation
The core of the new method is the use of local linear estimation to determine conditional survival functions. Unlike simple kernel smoothing, the local linear approach is particularly effective at reducing bias near the boundaries of the data. The authors derive a mathematical relationship that expresses the marginal hazard function—the instantaneous risk of the disease—in terms of the survival of relatives conditioned on the status of the proband.
Optimization and Bandwidth Selection
To ensure the estimator performs reliably across different datasets, the authors introduced two key refinements:
- Time Transformation: A transformation is applied to make the observation times of the probands approximately uniform, which stabilizes the estimator's performance.
- Bootstrap Bandwidth Selection: The authors developed a specific bootstrap-based procedure to select the optimal bandwidth ($h$), the parameter that controls the degree of smoothing applied to the data.
Empirical Validation
Simulation Results
The estimator was rigorously tested through 1,024 replications across 32 distinct design scenarios. These scenarios varied the level of "frailty" (unobserved shared risk factors), event rates, and family sizes. The results demonstrated that the new estimator consistently outperformed previous methods in terms of efficiency, yielding lower standard errors. In specific contexts, such as gamma frailty models with low event rates, the reduction in bias was substantial.
Case Study: Early-Onset Prostate Cancer
The method was applied to a real-world dataset from a 1999 study on early-onset prostate cancer, involving 730 case probands, 693 control probands, and 7,316 relatives (fathers, brothers, and uncles).
The results highlighted the dangers of using standard tools: a "Naive" Kaplan-Meier estimator significantly overestimated survival because it failed to account for the case-heavy recruitment process. In contrast, the new nonparametric estimator produced results that aligned closely with the Surveillance, Epidemiology, and End Results (SEER) registry, a high-quality population benchmark. This alignment validates the estimator's ability to correct for selection bias in practical applications.
Theoretical Contributions and Limitations
A major contribution of this work is the derivation of the asymptotic normality of the estimator. This provides the mathematical proof of consistency required for researchers to apply the method with confidence in large-scale studies.
However, some limitations remain. The current model assumes symmetry—that the relationship between a proband and all relatives is identical. In practice, a brother may share more risk factors than an uncle, suggesting that future iterations could benefit from relationship-specific weighting. Additionally, the bootstrap methods used for bandwidth selection and confidence intervals are computationally intensive, which may pose challenges for exceptionally large datasets.
Conclusion
By leveraging local linear estimation and providing a rigorous theoretical framework, Zucker and Gorfine have developed a robust tool for survival analysis in case-control family studies. Their method effectively bridges the gap between complex, biased data collection and the need for accurate, population-level survival estimates.
Summary updated