- Open Access
Response to: Practical methods for incorporating summary time-to-event data into meta-analysis
© Wang and Zeng; licensee BioMed Central Ltd. 2013
- Received: 19 April 2013
- Accepted: 1 November 2013
- Published: 19 November 2013
A response to and comment on Practical methods for incorporating summary time-to-event data into meta-analysis, by Jayne F Tierney et al.
- Practical Method
- Individual Trial
- Excel Spreadsheet
- Trial Report
- Estimate Hazard Ratio
In the example of the present article, the log-rank P-value of 0.075 gave a z-score of 1.78 according to the latter part of Tierney’s equation 14. Nevertheless, we found 2-sided P-value of 0.075 gave a z-score of 1.78 by using the net tools . But the z-score for the 2-sided P-value should be divided by 2 as described by Tierney and colleagues. Thus we can speculate that the z-score for a P-value of 0.075 should be 0.89, which was produced from 1.78/2.
We also advertently found that a right-tailed P-value of 0.0375 gave a z-score of 1.78 by using the previous mentioned net tools , and 0.0375 by chance equals 0.075/2. So we speculated that the latter part of equation 14 required a more accurate representation of z-score and P-value as: z-score for (P-value/2).
Tierney and colleagues reported that if a 1-sided P-value was reported, it can be used directly to calculate the z-score without dividing by 2. According to Li , a 1-sided P-value was used in log-rank test or Cox regression, and the exact P-value was given in the table for Chi-square (right-tailed test). However, we verified that a right-tailed P-value 0.075 gave a z-score of 1.44 by using the previous mentioned net tools .
Therefore, we can conclude that a 2-sided P-value can be used to directly to obtain the z-score. The latter part of equation 14 need to be modified into: z-score for P-value. Otherwise, a 1-sided P-value divided by 2 is required to obtain the z-score and the latter part of equation 14 should be expressed more exactly as: z-score for (P-value/2).
Jayne F Tierney, Lesley A Stewart, Davina Ghersi, Sarah Burdett and Matthew R Sydes.
and the related explanatory text altered accordingly: “As well as the events on each arm and overall, the z-score for half the two-sided P-value is required”
However, we disagree with the other points raised. It is suggested that one-sided P-values are used in log-rank tests and Cox regression models , whereas we find that it is standard practice to present two-sided (or two-tailed) P-values. Also, to our knowledge, all major statistical packages output two-sided P-values by default. It is therefore reasonable to assume that a P-value quoted in a trial publication will be two-sided unless otherwise stated, and to use this in equation 14. However, as described in the text : “If a one-sided P-value is reported it can be used directly to obtain the z-score.”
This is justified by equation 14 being derived algebraically from the definition of the log-rank statistic as a normally-distributed random variable, and by the fact that a one-sided P-value (assuming the most extreme direction of effect) is half the magnitude of the corresponding two-sided P-value.
The correspondents go on to suggest that either a two-sided P-value or a one-sided P-value divided by 2 can be used directly in equation 14. In fact, this would produce an incorrect z-score and hazard ratio. Using the example in Tierney and colleagues, the z-score for the reported P-value of 0.075/2 (= 0.0375) is 1.78, as the correspondents themselves found using internet tools. This produces an O-E of 19.57 and hazard ratio of 0.85; the latter being identical to that reported in the trial publication . If, as the correspondents suggested, we had used the reported P-value directly in equation 14, we would have obtained a z-score of 1.44, O-E of 15.82 and an incorrect hazard ratio of 0.88.
Finally, researchers wishing to estimate hazard ratios from published time-to-event data need not rely on deriving them manually using the equations provided , but instead can use the Excel spreadsheet that accompanies the paper. Moreover, they can use this to cross-check hazard ratios derived from different methods of estimation.
We thank all the people who have been involved in our study.
- Tierney JF, Stewart LA, Ghersi D, Burdett S, Sydes MR: Practical methods for incorporating summary time-to-event data into meta-analysis. Trials. 2007, 8: 16-10.1186/1745-6215-8-16.View ArticlePubMedPubMed CentralGoogle Scholar
- Z Score to P Value Calculator: [http://www.easycalculation.com/statistics/p-value-for-z-score.php]
- Li XS: Medical Statistics(version 2). 2008, Beijing: Higher Education PressGoogle Scholar
- Parmar MK, Torri V, Stewart L: Extracting summary statistics to perform meta-analyses of the published literature for survival endpoints. Stat Med. 1998, 17: 2815-2834. 10.1002/(SICI)1097-0258(19981230)17:24<2815::AID-SIM110>3.0.CO;2-8.View ArticlePubMedGoogle Scholar
- International Collaboration of Trialists on behalf of the Medical Research Council Advanced Bladder Cancer Working Party, EORTC Genito-urinary Group Australian Bladder Cancer Study Group, National Cancer Institute of Canada Clinical Trials Group, Finnbladder, Norwegian Bladder Cancer Study Group and Club Urológico Español de Tratamiento Oncológico (CUETO) group: Neoadjuvant cisplatin, methotrexate, and vinblastine chemotherapy for muscle-invasive bladder cancer: a randomised controlled trial. Lancet. 1999, 354: 533-540.View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.