An Introduction to Survival Statistics: Kaplan-Meier Analysis

Authors' disclosures of potential conflicts of interest are found at the end of this article. S tudies of how patients respond to treatment over time are fundamentally important to understanding how therapies influence quality of life and progression of disease during survi-vorship. When investigators examine change over time in continuous variables (e.g., patient self-reports of pain, fatigue, or nausea) in the same individuals, repeated measures are typically analyzed using analysis of variance (ANOVA) or perhaps latent growth curve modeling Other studies—particu-larly those that compare the long-term effects of new drugs or other therapeutic regimens to some " standard " therapy—focus on time to binary (yes/no) disease-related events of interest, such as death (time to event). Such studies are particularly apropos to generating improvements in cancer therapies, in which new treatments are compared to " standard " regimens, and are shown or disproved to extend progression-free survival (PFS), time to progression, or overall survival (OS) in patients with a particular cancer. Time-to-event studies typically employ two closely related statistical approaches, Kaplan-Meier (K-M) analysis and Cox proportional hazards model analysis (sometimes abbreviated as proportional hazards model or Cox model). K-M is a uni-variate approach, while Cox analysis is multivariable. Both use many familiar aspects of parametric and nonparametric statistical techniques (e.g., independent and dependent variables, null hypothesis testing, and confidence intervals). On the other hand, survival analyses employ other analytical techniques, terms, and computations that some oncology advanced practitioners (APs) may be less familiar with. No published research that addressed oncology APs' knowledge and ability to interpret statistical tests was found, but a study of medical residents examined their knowledge within the context of statistical procedures used in medical studies (Windish, Huot, & Green, 2007). Results proved that there was a mismatch between statistical procedures used and these clini-cians' understanding, and therefore the ability to judge the quality and ve-racity of published research. That is, more than 81% correctly interpreted relative risk, but only 10.5% understood K-M, and 11.9% could interpret the 95% confidence interval (CI) and statistical significance. Given that APs' knowledge deficits may be some