Reporting exposure-time associations of adverse events
Lineberry et al (1) provide recommendations to improve adverse events reporting in clinical trial publications. We agree that this topic is of high importance and especially welcome these recommendations, because most publications of clinical trials report safety data but ignore timing and duration of adverse events and report only relative frequencies instead. In situations in which the time on treatment for both treatment groups is similar, the reporting of relative frequencies can be sufficient. But especially in oncology trials time on treatment between treatment groups often differs, therefore methods which incorporate time on treatment have to be used.
These methods are described by the authors as exposure-time associations and illustrated (Fig. 1) by an example taken from a publication of Nozawa et al (2). We feel, however, that these plots presented in Figure 1 should not serve as a “best-practice” example. According to the legend, the cumulative chance of experiencing particular adverse events over time is displayed. It can easily be seen that there must be something wrong with this figure since all five “cumulative chances” approach 100% although the crude proportions are 44% for stomatitis (A), 31% for thrombocytopenia (B), 22% for anemia (C), 17% for hyperglycemia (D) and 22% for pneumonitis (E) as taken from Table 2 in Nozawa et al (2). The reason for this discrepancy is that Nozawa et al (2) plotted the cumulative distribution functions of five particular adverse events only for those patients in whom such a particular adverse event was observed. This also explains the somewhat mysteriously increasing numbers at risk that simply represent the observed number of a particular adverse event over time. Therefore the number of patients which is artificially set equal to 100% is different for different adverse events and depends on the time on treatment. Thus, these plots do not provide any information on the probability (chance) of suffering from a particular adverse event. The only thing that can be derived is whether a particular adverse event occurs early or late in time.
Alternatively we would recommend to analyse adverse events either as single or composite endpoints and to consider death or progression as competing events. The cumulative probabilities can then be calculated using the Aalen-Johansen estimator for the cumulative incidence functions for both competing events. In our opinion the graphical display of the Aalen-Johansen estimator is a more adequate method to report exposure-time associations, because additionally to the information on the occurrence of a specific adverse event it displays the probability of experiencing this specific adverse event. An application of this method to safety data was reported in Proctor et al (3).
Tanja Proctor (a), Martin Schumacher (b)
(a) Institute of Medical Biometry and Informatics, University of Heidelberg, Heidelberg, Germany
(b) Institute for Medical Biometry and Statistics, Faculty of Medicine and Medical Center, University of Freiburg, Freiburg, Germany
References
1 Lineberry Neil, Berlin Jesse A, Mansi Bernadette, Glasser Susan, Berkwits Michael, Klem Christian et al. Recommendations to improve adverse event reporting in clinical trial publications: a joint pharmaceutical industry/journal editor perspective BMJ 2016; 355 :i5078
2 Nowaza M, Nonomura N, Ueda T, et al. Adverse event profile and dose modification of everolimus for advanced renal cell carcinoma in real-world Japanese clinical practice. Jpn J Clin Oncol 2013; 43:1132-8. doi:10.1093/jjco/hyt121
3 Proctor, T., and Schumacher, M. (2016) Analysing adverse events by time-to-event models: the CLEOPATRA study. Pharmaceut. Statist., 15: 306–314. doi: 10.1002/pst.1758
Competing interests:
No competing interests
25 November 2016
Tanja Proctor
Research Fellow
Martin Schumacher
Institute of Medical Biometry and Informatics, University of Heidelberg, Heidelberg, Germany
Im Neuenheimer Feld 130.3, D- 69120 Heidelberg, Germany
Rapid Response:
Reporting exposure-time associations of adverse events
Lineberry et al (1) provide recommendations to improve adverse events reporting in clinical trial publications. We agree that this topic is of high importance and especially welcome these recommendations, because most publications of clinical trials report safety data but ignore timing and duration of adverse events and report only relative frequencies instead. In situations in which the time on treatment for both treatment groups is similar, the reporting of relative frequencies can be sufficient. But especially in oncology trials time on treatment between treatment groups often differs, therefore methods which incorporate time on treatment have to be used.
These methods are described by the authors as exposure-time associations and illustrated (Fig. 1) by an example taken from a publication of Nozawa et al (2). We feel, however, that these plots presented in Figure 1 should not serve as a “best-practice” example. According to the legend, the cumulative chance of experiencing particular adverse events over time is displayed. It can easily be seen that there must be something wrong with this figure since all five “cumulative chances” approach 100% although the crude proportions are 44% for stomatitis (A), 31% for thrombocytopenia (B), 22% for anemia (C), 17% for hyperglycemia (D) and 22% for pneumonitis (E) as taken from Table 2 in Nozawa et al (2). The reason for this discrepancy is that Nozawa et al (2) plotted the cumulative distribution functions of five particular adverse events only for those patients in whom such a particular adverse event was observed. This also explains the somewhat mysteriously increasing numbers at risk that simply represent the observed number of a particular adverse event over time. Therefore the number of patients which is artificially set equal to 100% is different for different adverse events and depends on the time on treatment. Thus, these plots do not provide any information on the probability (chance) of suffering from a particular adverse event. The only thing that can be derived is whether a particular adverse event occurs early or late in time.
Alternatively we would recommend to analyse adverse events either as single or composite endpoints and to consider death or progression as competing events. The cumulative probabilities can then be calculated using the Aalen-Johansen estimator for the cumulative incidence functions for both competing events. In our opinion the graphical display of the Aalen-Johansen estimator is a more adequate method to report exposure-time associations, because additionally to the information on the occurrence of a specific adverse event it displays the probability of experiencing this specific adverse event. An application of this method to safety data was reported in Proctor et al (3).
Tanja Proctor (a), Martin Schumacher (b)
(a) Institute of Medical Biometry and Informatics, University of Heidelberg, Heidelberg, Germany
(b) Institute for Medical Biometry and Statistics, Faculty of Medicine and Medical Center, University of Freiburg, Freiburg, Germany
References
1 Lineberry Neil, Berlin Jesse A, Mansi Bernadette, Glasser Susan, Berkwits Michael, Klem Christian et al. Recommendations to improve adverse event reporting in clinical trial publications: a joint pharmaceutical industry/journal editor perspective BMJ 2016; 355 :i5078
2 Nowaza M, Nonomura N, Ueda T, et al. Adverse event profile and dose modification of everolimus for advanced renal cell carcinoma in real-world Japanese clinical practice. Jpn J Clin Oncol 2013; 43:1132-8. doi:10.1093/jjco/hyt121
3 Proctor, T., and Schumacher, M. (2016) Analysing adverse events by time-to-event models: the CLEOPATRA study. Pharmaceut. Statist., 15: 306–314. doi: 10.1002/pst.1758
Competing interests: No competing interests