Enhanced secondary analysis of survival data: reconstructing the data from published Kaplan-Meier survival curves.
Guyot P, Ades AE, Ouwens MJ, Welton NJ. · 2012
Read Enhanced secondary analysis of survival data: reconstructing the data from published Kaplan-Meier survival curves. by Guyot P, Ades AE, Ouwens MJ, Welton NJ. (2012) 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-pmc3313891.txt.
Introduction
Guyot, Ades, Ouwens, and Welton presented a method in 2012 for recovering patient level survival data from published trial charts. Trials of time until an event, such as death, usually publish only summary figures and a survival curve. Those summaries are not enough for combining trials or judging whether a treatment is worth its cost, which blocks deeper follow up analysis.
Their algorithm works backward from a digitized survival curve, using reported numbers of events and patients still at risk, to rebuild data very close to the original. Several observers repeating the process got nearly identical results, with survival figures and median times matching closely. The authors recommend that trials always publish counts at risk and total events beside their curves, so others can reuse the data.
Catalog record
| CREATOR | Guyot P, Ades AE, Ouwens MJ, Welton NJ. |
|---|---|
| YEAR | 2012 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 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:
Enhanced secondary analysis of survival data: reconstructing the data from published Kaplan-Meier survival curves By Patricia Guyot, AE Ades, Mario JNM Ouwens, Nicky J Welton Published in: BMC Medical Research Methodology DOI: 10.1186/1471-2288-12-9 ABSTRACT Abstract Background The results of Randomized Controlled Trials (RCTs) on time-to-event outcomes that are usually reported are median time to events and Cox Hazard Ratio. These do not constitute the sufficient statistics required for meta-analysis or cost-effectiveness analysis, and their use in secondary analyses requires strong assumptions that may not have been adequately tested. In order to enhance the quality of secondary data analyses, we propose a method which derives from the published Kaplan Meier survival curves a close approximation to the original individual patient time-to-event data from which they were generated. Methods We develop an algorithm that maps from digitised curves back to KM data by finding numerical solutions to the inverted KM equations, using where available information on number of events and numbers at risk. The reproducibility and accuracy of survival probabilities, median survival times and hazard ratios based on reconstructed KM data was assessed by comparing published statistics (survival probabilities, medians and hazard ratios) with statistics based on repeated reconstructions by multiple observers. Results The validation exercise established there was no material systematic error and that there was a high degree of reproducibility for all statistics. Accuracy was excellent for survival probabilities and medians, for hazard ratios reasonable accuracy can only be obtained if at least numbers at risk or total number of events are reported. Conclusion The algorithm is a reliable tool for meta-analysis and cost-effectiveness analyses of RCTs reporting time-to-event data. It is recommended that all RCTs should report information on numbers at risk and total number of events alongside KM curves. Keywords: Survival analysis, Individual Patient Data, Kaplan-Meier, algorithm, life-table, Cost-Effectiveness Analysis, Health Technology Assessment BACKGROUND Normal practice in the reporting of results from RCTs is to publish the sufficient statistics for each arm: means and standard deviations for continuous outcomes, numerators and denominators for binary outcomes. CONSORT guidelines recommend that for each primary and secondary outcome "study results should be reported as a summary of the outcome in each group, together with the contrast between the groups, known as the effect size" [1]. The publication of sufficient statistics facilitates the inclusion of the trial in subsequent meta-analysis or economic assessments. However, reporting results of trials with survival time outcomes almost never follows these principles [2]. Due to censoring, time-to-event outcomes are not amenable to standard statistical procedures used for analysis of continuous outcomes: the average survival time is a biased estimate of expected survival in the presence of censored observations. Consort guidelines recommend instead, that for survival time, the measure of effect could be the hazard ratio or difference in median survival time. The Cochrane handbook also advises that the effect measure for time-to-event outcomes should be expressed as a hazard ratio. This severely limits the way in which RCTs with survival time data can be included in any kind of secondary data analyses, whether as part of a cost-effectiveness analyses (CEA) or in an analysis of treatment efficacy. In the case of CEA, what is required is an estimate of the difference between arms in expected survival: this cannot be reconstructed from the reported, or pooled, hazard ratio or medians without further assumptions. This demands that the survival curves are extrapolated for each treatment with a lifetime horizon, which is clearly impossible based solely on a hazard ratio or a median estimate. For the evidence syntheses, i.e. meta-analyses or network meta-analyses, pooling the treatment effect over several trials must either use estimates of median survival, which has been shown to be unsatisfactory [3], or fall back on estimates of the hazard ratio. This is also limiting and unsatisfactory, as it requires proportional hazards, an assumption that is seldom checked and sometimes implausible on inspection. These issues are examined further in the discussion. As a response to the poor, limited, and inconsistent reporting of results from survival data, several authors have attempted to extract data from the published Kaplan-Meier (KM) curves in order to carry out meta-analysis [4-11]. However, in this earlier work the survival probabilities were extracted from the graphs or the text at a relatively small number of follow-up times in order to approximate aggregate or life-table data, and they did not use all of the information reported to help identify the censoring pattern. In this paper, we develop an algorithm that attempts to reproduce the sufficient statistics in detail, by reconstructing the Kaplan-Meier data on which the survival curves are based. The KM curves are in effect pictorial representations of these data. We use digital software to read in the coordinates of the KM curves from the published graph and we use the information on numbers at risk, often published at four or five time points under the x-axis of the KM graph, and total number of events, where available, to reconstruct the Kaplan-Meier data for each arm. A key feature of our approach is the use of iterative numerical methods to solve the inverted KM equations, which is necessary to obtain consistent results and make the best use of the information available. The paper is organised as follows. In the illustration section, we apply the algorithm to one published Kaplan-Meier curve. In the reliability and accuracy section, we compare the summary measures reported in the original publications with those obtained by the analysis of the reconstructed Kaplan-Meier data on six pairs of KM curves reconstructed by three observers each on two occasions. We conclude with a discussion of the potential of this technique in meta-analysis and health technology assessment based on cost-effectiveness analysis. In the method section, we briefly describe the Kaplan-Meier estimation method. We then describe the inputs required for the algorithm, before presenting the algorithm itself. The general principle of the algorithm is developed as well as solutions to some particular pitfalls that may be encountered in practical applications. The R-code for the algorithm is provided in the Additional file 1. RESULTS Illustration The illustrative example uses KM curves, reproduced in Figure 1, on locoregional control events in head and neck cancer [12]. The numbers at risk were available every 10 months, from zero to fifty months. There was no total number of events reported in the publication. [Figure: Example of published Kaplan-Meier curves [13].] The tables 1 and 2 illustrate the input data needed to run the algorithm for this figure. Table 1 shows the extracted co-ordinates for the arm 'radiotherapy' between 0 and 10 months using the DigitizeIt software. Table 2 presents the format required for the numbers at risk provided in the publication. The first column is the interval, the second column shows the time, the third column shows the row of the extracted co-ordinates that the time corresponds to, the fourth column is the upper row of the extracted co-ordinates for which the time is less than the following time at which we have a number at risk and the last column is the number at risk. Time 0 always corresponds to the 1st row. In this example, row 30 corresponds to the largest time extracted from the graph-reading