A note on competing risks in survival data analysis.

Satagopan JM, Ben-Porat L, Berwick M, Robson M, Kutler D, Auerbach AD. · 2004

Read A note on competing risks in survival data analysis. by Satagopan JM, Ben-Porat L, Berwick M, Robson M, Kutler D, Auerbach AD. (2004) 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-pmc2410013.txt.

Introduction

Satagopan and colleagues explained a trap in survival statistics in 2004. The trap is competing risks: a patient may meet a different event first, which removes any chance of the event being studied. Treating that patient as merely censored, the standard Kaplan-Meier habit, distorts the answer.

Their paper shows how to estimate cumulative incidence, the chance of the event by a given time, while handling rival events properly. They work through two published data sets, computing incidence both ways. The Kaplan-Meier figures overshot the true incidence, because they pretend rival-event patients could still have the studied event later. The corrected method counts them as permanently out of the running. The authors addressed the note to clinical researchers, stressing that the choice of method changes reported risks in real studies.

Catalog record

CREATORSatagopan JM, Ben-Porat L, Berwick M, Robson M, Kutler D, Auerbach AD.
YEAR2004
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC2410013

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A note on competing risks in survival data analysis By J M Satagopan, L Ben-Porat, M Berwick, M Robson, D Kutler, A D Auerbach Published in: British Journal of Cancer DOI: 10.1038/sj.bjc.6602102 ABSTRACT Survival analysis encompasses investigation of time to event data. In most clinical studies, estimating the cumulative incidence function (or the probability of experiencing an event by a given time) is of primary interest. When the data consist of patients who experience an event and censored individuals, a nonparametric estimate of the cumulative incidence can be obtained using the Kaplan–Meier method. Under this approach, the censoring mechanism is assumed to be noninformative. In other words, the survival time of an individual (or the time at which a subject experiences an event) is assumed to be independent of a mechanism that would cause the patient to be censored. Often times, a patient may experience an event other than the one of interest which alters the probability of experiencing the event of interest. Such events are known as competing risk events. In this setting, it would often be of interest to calculate the cumulative incidence of a specific event of interest. Any subject who does not experience the event of interest can be treated as censored. However, a patient experiencing a competing risk event is censored in an informative manner. Hence, the Kaplan–Meier estimation procedure may not be directly applicable. The cumulative incidence function for an event of interest must be calculated by appropriately accounting for the presence of competing risk events. In this paper, we illustrate nonparametric estimation of the cumulative incidence function for an event of interest in the presence of competing risk events using two published data sets. We compare the resulting estimates with those obtained using the Kaplan–Meier approach to demonstrate the importance of appropriately estimating the cumulative incidence of an event of interest in the presence of competing risk events. Survival analysis is the analysis of data measured from a specific time of origin until an event of interest or a specified endpoint (Collett, 1994). For example, in order to determine the incidence of death due to breast cancer among breast cancer patients, every patient will be followed from a baseline date (such as date of diagnosis or date of surgery) until the date of death due to breast cancer or study closing date. A patient who dies of breast cancer during the study period would be considered to have an ‘event’ at their date of death. A patient who is alive at the end of the study would be considered to be ‘censored’. Thus, every patient provides two pieces of information: follow-up time and status (i.e., event or censored). However, a patient can experience an event different from the event of interest. For example, a breast cancer patient may die due to causes unrelated to the disease. Such events are termed competing risk events. Survival data are often summarised using the cumulative incidence function for an event. The goal of this paper is to illustrate to the clinical investigator the nonparametric estimation of the cumulative incidence for an event of interest in the presence of competing risk events. GENERAL CONCEPTS IN SURVIVAL ANALYSIS Clinical studies often focus on estimating the survivor function or the overall survival probability. This is the probability of being event-free at least up to a given time. The event is any specific event of interest. The overall survival probability is estimated using the person follow-up time and event status. The survival at a given time is the conditional probability of surviving to a specific time given that the individual is at risk for the event (such as mortality) at that time. This is estimated as the ratio of the number of individuals that are event-free at that time to the number of individuals that lived event-free at least up to that time. Henceforth, when referring to survival at a given time, we in fact mean survival conditional upon being at risk at that time. For example, let us consider the event of interest to be death. Suppose 100 breast cancer patients lived for at least 1 year following surgery and five patients died at the end of the first year. The estimated survival at 1 year is 95%. This estimate is the survival probability conditional upon the fact that the 95 surviving patients were all at risk at 1 year. Suppose at 2 years, 10 more patients die. The estimated survival at 2 years is 85/95=89.5%. The estimated overall survival up to and including 2 years is the probability of having survived the first year and the second year. This is calculated as 95*89.5=85%. Therefore, the overall survival or the probability of surviving up to 2 years is 85%. This is the basis for the Kaplan–Meier estimate of survival probability (Kaplan and Meier, 1958). Equivalently, we can consider estimating the incidence of mortality among patients with breast cancer. In the above example, the estimated incidence of mortality at 1 year is five out of 100 patients, that is, 5%. The probability of dying at 2 years is the probability of living past the first year and dying during the second year. Note that 95 of 100 patients survived past the first year, and 10 of these 95 patients died in the second year. Therefore, the estimated incidence of mortality at 2 years is 95/100*10/95=10%. The cumulative incidence of mortality up to a given time is the probability that an individual dies by that time. This is the sum of mortality incidences occurring up to that time. Therefore, the cumulative incidence of mortality at 2 years is the sum of mortality incidences in the first and second years, that is, 5+10=15%. This is simply the converse of survival. In other words, the cumulative incidence of an event at a given time is one minus the overall survival probability at that time. An investigator may be interested in examining outcomes other than mortality, such as incidence of disease recurrence or incidence of a second primary cancer. In this scenario, a patient is followed from a baseline time (such as date of diagnosis or date of treatment) until the outcome of interest. In order to estimate the incidence of ovarian cancer among breast cancer patients, any patient who does not develop ovarian cancer is treated as censored. If a woman has not developed ovarian cancer by the study closing date, she would be considered censored. If a woman dies of breast cancer prior to developing ovarian cancer, she would be considered censored at her date of death when evaluating the incidence of ovarian cancer. However, the latter type of censoring is informative since this patient is censored due to the occurrence of an intervening event (mortality). Such intervening events are known as ‘competing risk events’. Events other than death can also be competing risk events. Suppose a breast cancer patient undergoes a prophylactic oophorectomy after surgery for breast cancer. This prophylactic treatment substantially reduces the probability of developing ovarian cancer, and hence is treated as a competing risk event when calculating ovarian cancer incidence. Thus, a competing risk event may preclude the onset of the event of interest, or may modify the probability

Read the complete text