Article type
Year
Abstract
Background: Kaplan-Meier (KM) analyses are frequently applied in studies analysing survival data. These analyses consider patients lost to follow up as censored data. In several previous evaluations of publications, we found discrepancies with regard to information on patient numbers presented in the KM curve and in the text.
Objectives: The aim of this review was to assess the consistency between information presented in KM curves on patients lost to follow up and patients at risk and the respective information presented in the text of publications.
Methods: We conducted a search for randomized controlled trials that included at least one KM curve and were published in 2003 by screening journals with a high impact factor (BMJ, JAMA, Lancet, New England Journal of Medicine). We searched for details about reasons for censoring, planned and actual duration of observation, number of patients at risk, and number of patients without complete follow up. We then compared the information obtained from the KM curves with the information in the text.
Results: We identified 76 publications that included KM curves. Twenty-seven (36%) publications provided insufficient information, mainly because no details were provided on the number of patients at risk, but also because no details were included on competing risks or on the planned or actual duration of observation. Inconsistencies were identified in 12 (16%) studies. In total, 51% of the publications presented incomplete, unclear or inconsistent information.
Conclusions: Half the publications that included KM curves provided incomprehensible or inconsistent information on patients lost to follow up. Reviewers should therefore assess the information presented in the text and figures to identify any inconsistencies. Due to the potential effects on study results, this issue should also be considered in the overall evaluation of the quality of a study.
Objectives: The aim of this review was to assess the consistency between information presented in KM curves on patients lost to follow up and patients at risk and the respective information presented in the text of publications.
Methods: We conducted a search for randomized controlled trials that included at least one KM curve and were published in 2003 by screening journals with a high impact factor (BMJ, JAMA, Lancet, New England Journal of Medicine). We searched for details about reasons for censoring, planned and actual duration of observation, number of patients at risk, and number of patients without complete follow up. We then compared the information obtained from the KM curves with the information in the text.
Results: We identified 76 publications that included KM curves. Twenty-seven (36%) publications provided insufficient information, mainly because no details were provided on the number of patients at risk, but also because no details were included on competing risks or on the planned or actual duration of observation. Inconsistencies were identified in 12 (16%) studies. In total, 51% of the publications presented incomplete, unclear or inconsistent information.
Conclusions: Half the publications that included KM curves provided incomprehensible or inconsistent information on patients lost to follow up. Reviewers should therefore assess the information presented in the text and figures to identify any inconsistencies. Due to the potential effects on study results, this issue should also be considered in the overall evaluation of the quality of a study.