Variable Selection in Functional Linear Cox Model
Pith reviewed 2026-05-19 11:31 UTC · model grok-4.3
The pith
A group MCP penalty on spline coefficients selects which functional and scalar covariates matter in Cox survival models.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The authors propose a spline-based semiparametric estimator for the functional coefficients in a linear Cox model, regularized by a group minimax concave penalty that simultaneously enforces smoothness within each functional effect and sparsity by zeroing out entire groups of spline coefficients for irrelevant covariates. An efficient group descent algorithm solves the resulting optimization, and a data-driven procedure chooses the smoothing and penalty parameters.
What carries the argument
Group minimax concave penalty applied to blocks of spline basis coefficients, which enforces both intra-function smoothness and inter-function sparsity in the estimated hazard contributions.
If this is right
- The joint penalty produces accurate variable selection and unbiased effect estimates when the true signals are both smooth and sparse.
- Application to the NHANES cohort identifies the temporally varying activity distributions and demographic factors tied to all-cause mortality.
- Automated selection of smoothing and sparsity parameters removes the need for manual cross-validation in high-dimensional sensor data.
- Scalar covariates and multiple functional covariates are handled inside the same penalized framework.
Where Pith is reading between the lines
- The same spline-plus-group-penalty structure could be tried on other time-to-event models such as accelerated failure time or frailty models.
- Wearable monitoring systems might use the selected functions to focus alerts on only the most predictive signals rather than processing every channel.
- Testing the selected activity patterns on independent cohorts would show whether the mortality associations hold beyond the original sample.
Load-bearing premise
Functional coefficients are smooth enough to be represented accurately by a fixed spline basis without introducing bias into the hazard relationships.
What would settle it
On simulated data with known sparse but non-smooth functional effects, the method would select the wrong variables at a higher rate than an unpenalized spline fit or an alternative penalty.
read the original abstract
Modern biomedical studies frequently collect complex, high-dimensional physiological signals using wearables and sensors along with time-to-event outcomes, making efficient variable selection methods crucial for interpretation and improving the accuracy of survival models. We propose a novel variable selection method for a functional linear Cox model with multiple functional and scalar covariates measured at baseline. We utilize a spline-based semiparametric estimation approach for the functional coefficients and a group minimax concave type penalty (MCP), which effectively integrates smoothness and sparsity into the estimation of functional coefficients. An efficient group descent algorithm is used for optimization, and an automated procedure is provided to select optimal values of the smoothing and sparsity parameters. Through simulation studies, we demonstrate the method's ability to perform accurate variable selection and estimation. The method is applied to 2003-06 cohort of the National Health and Nutrition Examination Survey (NHANES) data, identifying the key temporally varying distributional patterns of physical activity and demographic predictors related to all-cause mortality. Our analysis sheds light on the intricate association between daily distributional patterns of physical activity and all-cause mortality among older US adults.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript proposes a novel variable selection method for a functional linear Cox model with multiple functional and scalar covariates measured at baseline. It employs a spline-based semiparametric estimation approach for the functional coefficients combined with a group minimax concave penalty (MCP) to enforce both smoothness and sparsity, an efficient group descent algorithm for optimization, and an automated procedure for selecting the smoothing and sparsity parameters. Performance is assessed via simulation studies demonstrating accurate variable selection and estimation, and the method is applied to the 2003-06 NHANES cohort to identify key temporally varying physical activity patterns and demographic predictors associated with all-cause mortality.
Significance. If the central claims hold, the work could provide a practical tool for variable selection and interpretation in survival analysis involving high-dimensional functional covariates from wearables and sensors. The integration of spline smoothness with group MCP sparsity addresses a relevant need in biomedical time-to-event modeling, and the NHANES application illustrates potential utility for understanding distributional physical activity patterns and mortality risk in older adults.
major comments (2)
- [Methods (spline expansion and group MCP)] The central modeling assumption that the spline basis expansion of each functional coefficient β_j(t), when combined with the group MCP penalty, simultaneously enforces smoothness and sparsity without materially distorting the underlying hazard relationships or partial likelihood contributions lacks supporting approximation error bounds. This is load-bearing for the claim of accurate selection, especially since NHANES physical activity data can exhibit sharp peaks or non-smooth features that a fixed low-order spline may smooth away prior to penalization. No section derives bounds linking spline truncation order or knot placement to the Cox partial likelihood.
- [Simulation studies and NHANES application] Simulation studies and the NHANES application are presented as empirical support for accurate variable selection and estimation, yet the manuscript provides no details on error bars, data exclusion rules, or whether tuning parameter selection was pre-specified versus post-hoc. This weakens grounding of the performance claims and leaves open questions about robustness to the distributional features of the functional covariates.
minor comments (2)
- [Notation and penalty definition] Clarify the precise definition and implementation of the group MCP penalty across the spline basis coefficients for each functional covariate j, including any scaling or normalization steps.
- [Parameter selection procedure] Add explicit discussion of how the automated procedure for selecting smoothing and sparsity parameters avoids overfitting or data-dependent bias in the reported NHANES results.
Simulated Author's Rebuttal
We thank the referee for their detailed and constructive comments on our manuscript. We address each major comment below, indicating planned revisions where appropriate.
read point-by-point responses
-
Referee: [Methods (spline expansion and group MCP)] The central modeling assumption that the spline basis expansion of each functional coefficient β_j(t), when combined with the group MCP penalty, simultaneously enforces smoothness and sparsity without materially distorting the underlying hazard relationships or partial likelihood contributions lacks supporting approximation error bounds. This is load-bearing for the claim of accurate selection, especially since NHANES physical activity data can exhibit sharp peaks or non-smooth features that a fixed low-order spline may smooth away prior to penalization. No section derives bounds linking spline truncation order or knot placement to the Cox partial likelihood.
Authors: We appreciate the referee's point regarding the need for theoretical approximation error bounds. Deriving rigorous bounds for the spline truncation error in conjunction with the group MCP penalty and the Cox partial likelihood is a non-trivial theoretical endeavor that lies beyond the primary scope of this paper, which emphasizes methodological development and empirical performance. In functional data analysis, B-splines are commonly used to enforce smoothness, and the group penalty promotes sparsity across the entire functional coefficient. Our simulation studies, which include scenarios with varying smoothness, support that the method performs well without material distortion. To address the concern about NHANES data potentially having sharp peaks, we will add a discussion in the Methods section on the choice of spline basis order and knot placement, along with sensitivity to these choices. We will also note this as a limitation in the Discussion. revision: partial
-
Referee: [Simulation studies and NHANES application] Simulation studies and the NHANES application are presented as empirical support for accurate variable selection and estimation, yet the manuscript provides no details on error bars, data exclusion rules, or whether tuning parameter selection was pre-specified versus post-hoc. This weakens grounding of the performance claims and leaves open questions about robustness to the distributional features of the functional covariates.
Authors: We agree that providing these details will strengthen the manuscript. In the revised version, we will add standard error bars or standard deviations from the simulation replications to the tables reporting selection accuracy and estimation errors. For the NHANES application, we will specify the data exclusion criteria used, such as participants with insufficient valid accelerometer data or missing key covariates. Additionally, we will clarify that the tuning parameters (smoothing and sparsity) were selected using the automated cross-validation procedure described in the methods, which was pre-specified. These additions will better demonstrate robustness to the distributional features of the covariates. revision: yes
Circularity Check
No significant circularity; derivation uses standard spline basis and group MCP with independent simulation validation
full rationale
The paper's core contribution is a new estimation procedure that expands functional coefficients in a spline basis and applies a group MCP penalty to induce both smoothness and sparsity in the functional linear Cox model. This is a standard semiparametric construction (spline representation of β(t) plus known MCP penalty) rather than a self-referential definition. The optimization algorithm, parameter selection procedure, and performance claims are evaluated through separate simulation studies that generate data under known models and measure selection/estimation accuracy; these are not forced by the fitting equations themselves. The NHANES application is an external data demonstration. No load-bearing step reduces a claimed result to a fitted quantity renamed as a prediction, nor does the argument rest on self-citations whose content is unverified. The derivation chain therefore remains self-contained against external benchmarks.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.