Landmark Analysis at the 25-Year Landmark Point

Abstract

Abstract-This statistical primer presents the landmark analysis method, exploring its appropriate use and interpretation while recognizing its limitations. This observational method is used for comparing time-to-event outcome between groups determined during study follow-up. The goal of the landmark method is to estimate in an unbiased way the time-to-event probabilities in each group conditional on the group membership of patients at a specific time point, the landmark time. The need that led to its development, the impact of the method, and its pros and cons, along with available alternative approaches, are presented. Simulations explore its performance, using realistic parameters from a recent cardiovascular study. 1 in an article addressing the issue of bias in the analysis of survival of "responders" and "nonresponders," a popular analysis in oncology. In the erroneous "naive" analysis, "responders" and "nonresponders" were compared with regard to survival from the start of the study, although their response status was not determined at baseline but later during follow-up. The landmark method has since been used extensively in medical research to correct for the bias inherent in the analysis of time-to-event outcome between groups determined during study follow-up.In a recent study, landmark analyses were performed to explore the association of extended clopidogrel use and long-term clinical outcomes of patients receiving drugeluting stents (DES) and bare-metal stents (BMS) for treatment of coronary artery disease. 2 The article, considered to have recently popularized the landmark approach in cardiovascular studies, particularly of coronary stents, is used later in this review in simulations as a case example.The goal of this statistical primer is to present the landmark method and explore its appropriate use and interpretation while recognizing its limitations.

Endovascular Repair of Type B Aortic Dissection

“…17 Nevertheless, with Landmark analysis, findings become observational because some confounders may not be accounted for, and early events may be omitted, thus rendering the analysis susceptible to time point-specific results and loss of power. 38 The risk of dependence on any lack of proportionality of hazards, however, was partly defused by additional Landmarking at 12 months and at 1 month of follow-up with consistent results with late benefit of TEVAR ( Figure IIA-IIC in the online-only Data Supplement).…”

New-Onset Diabetes Mellitus After Heart Transplantation ― Incidence, Risk Factors and Impact on Clinical Outcome ―

“…Also, landmark survival analysis was performed to separately analyze the early and late mortality outcomes across the 3 groups. 16 All reported P values are 2-sided, and the conventional hypoglycemic agents and/or insulin therapy initiated postoperatively. Oral glucose tolerance test or glycated hemoglobin level were obtained as necessary to confirm the diagnosis of new-onset DM.…”

Extended Adjuvant Temozolomide with cis-Retinoic Acid for Adult Glioblastoma

“…Although not randomized, the comparison is reasonable given the similarity of the cohort demographics. It provides a relatively unbiased comparison of these otherwise similar groups 25 , although it uses specific time points for comparison and therefore has less statistical power than that of a standard log-rank test. Our results suggest that the effectiveness of extended-duration temozolomide with cis-retinoic acid provided in a provincial cancer centre is at least similar to the clinical trial efficacy observed with 6 months of temozolomide.…”

“…This test allowed us to compare the proportions of patients living and dead in each cohort, while controlling for the fact that the rate may change over time. This methodology effectively compares patients who might have been eligible for more than 6 cycles of temozolomide in the eortc/ncic study with the patients in our cohort who received it 25 . Multivariate Cox proportional hazards modelling was used to determine the effect of individual prognostic variables on survival.…”