Cumulative incidence function estimation using population-based biobank data

Malka Gorfine, David M. Zucker, Shoval Shoham

TL;DR

Researchers have developed a new statistical estimator for the Cumulative Incidence Function (CIF) specifically designed for large-scale biobank data. Unlike standard methods that discard "prevalent cases" (individuals already diagnosed at the time of recruitment), this novel approach incorporates them to significantly increase estimation efficiency and allow researchers to estimate disease risk for ages before the study's recruitment window began.

What Problem Does This Paper Address?

Population-based biobanks, like the UK Biobank (UKB), are goldmines for medical research, but they come with a major statistical headache: left truncation.

Biobanks typically recruit participants within a specific age range (e.g., ages 40 to 69). This creates two issues:

  1. Survival Bias: The study only includes people who survived long enough to be recruited.
  2. The "Prevalent Case" Dilemma: Many participants enter the study already having the disease being studied.

Standard tools, like the Aalen-Johansen estimator, handle survival bias by using "risk-set correction." However, this method essentially ignores anyone who was diagnosed before they walked through the recruitment door. This is a massive waste of data—in the UKB, prevalent cases can make up 20% or more of total disease instances. Furthermore, because these tools only look at what happens after recruitment, they cannot estimate disease incidence for younger ages (e.g., what is the risk of leukemia at age 30 if recruitment starts at 40?).

Methods

The authors propose a novel CIF estimator based on an illness-death model (also known as a semi-competing risks model).

The key innovation lies in how they handle the math:

  • Conditional Distribution: Instead of just looking at the time to disease, they focus on the conditional distribution of the age at diagnosis given the age at death.
  • Efficiency via Inclusion: By modeling this relationship, they can "back-calculate" the probability of disease for prevalent cases.
  • Robustness: Unlike previous attempts to use Inverse Probability Weighting (IPW) which can become highly biased if the study's follow-up is too short, this new estimator remains stable.
  • Inference: The paper provides a rigorous mathematical proof (using empirical process theory and the functional delta method) to show the estimator is consistent and provides valid confidence intervals and "confidence bands" (uncertainty ranges across the entire age spectrum).

Datasets

The paper validates the method using two primary sources:

  1. Extensive Simulations: 16 different scenarios varying disease hazards, recruitment distributions, and censoring ages to prove the estimator works across different "shapes" of data.
  2. UK Biobank (UKB): Real-world application focusing on two types of cancer: Acute Myeloid Leukemia (AML) and Brain Cancer. They analyzed data for over 500,000 participants, specifically looking at the differences between males and females.

Key Insights and Ideas

  • Prevalent Cases are Data, Not Noise: The "aha moment" is realizing that prevalent cases aren't just a source of bias to be corrected; they are valuable observations that contain information about disease incidence at younger ages.
  • Breaking the $c_L$ Barrier: Most survival analysis is "blind" to anything that happened before the minimum recruitment age ($c_L$). This method effectively peers into the past, allowing for a CIF curve that starts at age 0 (or the earliest observed onset), providing a more complete picture of lifetime risk.
  • The "Combination" Strategy: The authors suggest a "combination estimator" that averages their new method with the traditional Aalen-Johansen. This provides a "best of both worlds" approach, maintaining high efficiency while staying robust.

Key Results

  • Increased Precision: In the UK Biobank analysis of AML in males, the new method produced confidence bands that were up to 27% narrower than the standard Aalen-Johansen estimator.
  • Bias Reduction: Simulations showed that while older IPW methods (like Chang-Zheng) failed spectacularly when follow-up was short, the new estimator remained unbiased and accurate.
  • Interpretation: For non-lethal diseases, the new approach provides a CIF that is much easier for clinicians to interpret: "Given the person is alive at age 40, what was their risk of having developed the disease by age 35?"

Limitations and Future Work

The authors acknowledge one significant hurdle: Low Fatality Diseases. Because the estimator relies on the relationship between diagnosis age and death age, it requires a sufficient number of "death events" among those diagnosed. For diseases with very high survival rates and young onset (like certain breast cancers), there may not be enough data on deaths to accurately estimate the necessary conditional distributions.

Future work may involve finding ways to incorporate other types of "terminal events" or auxiliary data to make the estimator more applicable to low-mortality conditions.

Summary updated

Abstract

Many countries have established population-based biobanks, which are being used increasingly in epidemiological and clinical research. These biobanks offer opportunities for large-scale studies addressing questions beyond the scope of traditional clinical trials or cohort studies. However, using biobank data poses new challenges. Typically, biobank data are collected from a study cohort recruited over a defined calendar period, with subjects entering the study at various ages falling between $c_L$ and $c_U$. This work focuses on biobank data with individuals reporting disease-onset age upon recruitment, termed prevalent data, along with individuals initially recruited as healthy, and their disease onset observed during the follow-up period. We propose a novel cumulative incidence function (CIF) estimator that efficiently incorporates prevalent cases, in contrast to existing methods, providing two advantages: (1) increased efficiency and (2) CIF estimation for ages before the lower limit, $c_L$.
BibTeX
@inproceedings{gorfine2025cumulative,
  author = {Malka Gorfine and David M. Zucker and Shoval Shoham},
  title = {{Cumulative incidence function estimation using population-based biobank data}},
  booktitle = {Biometrics},
  year = {2025},
  doi = {10.1093/biomtc/ujaf049},
}

Loading related content...

Figure