Variables with time-varying effects and the Cox model: some statistical concepts illustrated with a prognostic factor study in breast cancer.
Bellera CA, MacGrogan G, Debled M, de Lara CT, Brouste V, Mathoulin-Pélissier S. · 2010
Read Variables with time-varying effects and the Cox model: some statistical concepts illustrated with a prognostic factor study in breast cancer. by Bellera CA, MacGrogan G, Debled M, de Lara CT, Brouste V, Mathoulin-Pélissier S. (2010) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc2846954.txt.
Introduction
Bellera and colleagues warned in 2010 that a workhorse survival model in cancer research can mislead when a risk factor's effect changes over time. The Cox model assumes each factor shifts risk by a steady proportion throughout follow-up. With patients now living far longer, that assumption fails more often, and constant-effect summaries can hide early harm or late benefit.
The authors show the trap using nearly a thousand women treated for breast cancer and followed a median of fourteen years. Tumor grade's effect faded yet stayed strong. Hormone receptor status flipped sign: negative status raised the risk of spread early on, then turned protective later, which the standard model averaged into a misleading nothing. They urge routine checks for time-varying effects.
Catalog record
| CREATOR | Bellera CA, MacGrogan G, Debled M, de Lara CT, Brouste V, Mathoulin-Pélissier S. |
|---|---|
| YEAR | 2010 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Variables with time-varying effects and the Cox model: Some statistical concepts illustrated with a prognostic factor study in breast cancer By Carine A Bellera, Gaëtan MacGrogan, Marc Debled, Christine Tunon de Lara, Véronique Brouste, Simone Mathoulin-Pélissier Published in: BMC Medical Research Methodology DOI: 10.1186/1471-2288-10-20 ABSTRACT Background The Cox model relies on the proportional hazards (PH) assumption, implying that the factors investigated have a constant impact on the hazard - or risk - over time. We emphasize the importance of this assumption and the misleading conclusions that can be inferred if it is violated; this is particularly essential in the presence of long follow-ups. Methods We illustrate our discussion by analyzing prognostic factors of metastases in 979 women treated for breast cancer with surgery. Age, tumour size and grade, lymph node involvement, peritumoral vascular invasion (PVI), status of hormone receptors (HRec), Her2, and Mib1 were considered. Results Median follow-up was 14 years; 264 women developed metastases. The conventional Cox model suggested that all factors but HRec, Her2, and Mib1 status were strong prognostic factors of metastases. Additional tests indicated that the PH assumption was not satisfied for some variables of the model. Tumour grade had a significant time-varying effect, but although its effect diminished over time, it remained strong. Interestingly, while the conventional Cox model did not show any significant effect of the HRec status, tests provided strong evidence that this variable had a non-constant effect over time. Negative HRec status increased the risk of metastases early but became protective thereafter. This reversal of effect may explain non-significant hazard ratios provided by previous conventional Cox analyses in studies with long follow-ups. Conclusions Investigating time-varying effects should be an integral part of Cox survival analyses. Detecting and accounting for time-varying effects provide insights on some specific time patterns, and on valuable biological information that could be missed otherwise. BACKGROUND Survival analysis, or time-to-event data analysis, is widely used in oncology since we are often interested in studying a delay, such as the time from cancer diagnosis or treatment initiation to cancer recurrence or death. Thanks to the improvement of cancer treatments, and the induced longer life expectancy, we observe an increasing number of studies with long follow-up periods. Statistical models to analyze such data should thus adequately account for the increasing duration of follow-ups. The Cox proportional hazards (PH) model allows one to describe the survival time as a function of multiple prognostic factors [1]. This model relies on a fundamental assumption, the proportionality of the hazards, implying that the factors investigated have a constant impact on the hazard - or risk - over time. If time-dependent variables are included without appropriate modeling, the PH assumption is violated. As a result, misleading effect estimates can be derived, and significant effect in the early (or late) follow-up period may be missed. Checking the proportionality of the hazards should thus be an integral part of a survival analysis by a Cox model. The assumption, however, is not systematically verified. In a 1995 review of cancer publications using a Cox model, Altman et al. reported that most studies did not report verifying this assumption [2]; similar findings were reported recently by one of the co-authors of the present work [3]. Although the Cox model has been widely used (more than 25 000 citations since the publication of the original paper by Cox [4]), recent publications suggest a growing interest in the quality of its applications. Special papers in statistics have been published in the oncology literature providing general introductions to survival analysis [5-8]; topics covered included summarizing survival data, testing for a difference between groups, presenting existing statistical models, or assessing the adequacy of a survival model. Others works focused on providing definition of specific survival endpoints [9], or on the quality of reporting of survival events [3]. Assessing whether the assumption of proportional hazards is a central theme in survival analysis, and as such is discussed in several statistical textbooks [10-14] as well as in the general statistical literature [15-18]. To our knowledge however, this topic has been discussed in few medical journals. Importantly, this strong assumption does not seem to be systematically assessed. For illustration, a recent review of clinical trials with primary analyses based on survival end points showed that only one of the 64 papers that used a Cox model mentioned verifying the PH assumption [3]. Our objective is to inform clinicians, as well as those who read and write manuscripts in medical journals, about the importance of the underlying PH assumption, the misleading conclusions that can be inferred if it is violated, as well as the additional information provided by verifying it. After a theoretical introduction, we describe techniques to assess if this assumption is violated, and model strategies to account for, and describe time-dependency. We illustrate our discussion with a study on prognostic factors in breast cancer. METHODS AND RESULTS Survival analysis In many studies, the primary variable of interest is a delay, such as the time from cancer diagnosis to a particular event of interest. This event may be death, and for this reason the analysis of such data is often referred to as survival analysis. The event of interest may not have occurred at the time of the statistical analysis, and similarly, a subject may be lost to follow-up before the event is observed. In such case, data are said to be censored at the time of the analysis or at the time the patient was lost to follow-up. Censored data still bring some information since although we do not know the exact date of the event, we know that it occurred later than the censoring time. Both the Kaplan-Meier method and the Cox proportional hazards (PH) model allow one to analyze censored data [1,19], and to estimate the survival probability, S(t), that is the probability that a subject survives beyond some time t. Statistically, this probability is provided by the survival function S(t) = P (T > t), where T is the survival time. The Kaplan Meier method estimates the survival probability non-parametrically, that is, assuming no specific underlying function [19]. Several tests are available to compare the survival distributions across groups, including the log-rank and the Mann-Whitney-Wilcoxon tests [20,21]. The Cox PH model accounts for multiple risk factors simultaneously. It does not posit any distribution, or shape for the survival function, however, the instantaneous incidence rate of the event is modeled as a function of time and risk factors. The instantaneous hazard rate at time t, also called instantaneous incidence, death, or failure rate, or risk, is the instantaneous probability of experiencing an event at time t, given that the event has not occurred yet. It is a rate of event per unit of time, and is allowed to vary over time. Just as the risk of events per unit time, one can make an analogy by considering the speed given by a car speedometer, which represents the distance travelled per unit of time. Suppose, that the event of interest is death, and we are interested in its association with n covariates, X1, X2, ..., Xn, then the hazard is given by:(1) The baseline hazard rate h0(t) is an unspecified non-negative function of time. It