Archived Comments for:
Different methodological approaches to the assessment of in vivo efficacy of three artemisinin-based combination antimalarial treatments for the treatment of uncomplicated falciparum malaria in African children
Neal Alexander, London School of Hygiene and Tropical Medicine
5 September 2008
This article is a welcome step towards establishing a consensus for analytical methods for trials of antimalarials. However, I would like to add a word of caution to the recommendation of survival analysis, made here and previously in this journal[1].
The problem is that the event times are often observed rather imprecisely. In particular, parasitological failure may not be detected until the next scheduled visit. In that case, we do not know precisely when the parasitological failure occurred, only that it was between the current and previous examinations. Analysis methods exist to allow for this problem, which is known as interval censoring[2].
Defining the time of event to be the time of examination, as recommended by Ashley et al, does not avoid the problem. The fact remains that more frequent examinations result in more accurate estimation of the event rate (assuming the events are not notified by the patient on occurrence). Ignoring uncertainty in the observed time causes the accuracy of the analysis to be over-estimated. In most cases this effect is likely to be unimportant. However, in some trials, 90% of patients have treatment failure[3]. In such circumstances, measuring the outcome at, say, two time points rather than continuously, and then ignoring the resulting loss of information, is equivalent to crediting the trial with a sample size more than 10% greater than it really has[4]. Most trials will not have such high failure rates. However, this is a potential drawback of standard survival analysis and should be borne in mind.
Neal Alexander
[1] Stepniewska K, White NJ. Some considerations in the design and interpretation of antimalarial drug trials in uncomplicated falciparum malaria. Malar J 2006;5:127.
[2] Lindsey JC, Ryan LM. Tutorial in biostatistics: methods for interval-censored data. Stat Med 1998;17(2):219-38.
[3] Durrani N, Leslie T, Rahim S, Graham K, Ahmad F, Rowland M. Efficacy of combination therapy with artesunate plus amodiaquine compared to monotherapy with chloroquine, amodiaquine or sulfadoxine-pyrimethamine for treatment of uncomplicated Plasmodium falciparum in Afghanistan. Trop Med Int Health 2005;10(6):521-9.
[4] Alexander N. Precision of rate estimation under uniform interval censoring. Stat Med 2008;27(17):3442-3445.
Competing interests
No competing financial interests. The author has published analysis of antimalaria trial data using methods different to the ones advocated by Ashley et al (Dunyo et al 2006 PLos Clinical Trials 1:e14).
Effect of imprecision of observed failure times
5 September 2008
This article is a welcome step towards establishing a consensus for analytical methods for trials of antimalarials. However, I would like to add a word of caution to the recommendation of survival analysis, made here and previously in this journal[1].
The problem is that the event times are often observed rather imprecisely. In particular, parasitological failure may not be detected until the next scheduled visit. In that case, we do not know precisely when the parasitological failure occurred, only that it was between the current and previous examinations. Analysis methods exist to allow for this problem, which is known as interval censoring[2].
Defining the time of event to be the time of examination, as recommended by Ashley et al, does not avoid the problem. The fact remains that more frequent examinations result in more accurate estimation of the event rate (assuming the events are not notified by the patient on occurrence). Ignoring uncertainty in the observed time causes the accuracy of the analysis to be over-estimated. In most cases this effect is likely to be unimportant. However, in some trials, 90% of patients have treatment failure[3]. In such circumstances, measuring the outcome at, say, two time points rather than continuously, and then ignoring the resulting loss of information, is equivalent to crediting the trial with a sample size more than 10% greater than it really has[4]. Most trials will not have such high failure rates. However, this is a potential drawback of standard survival analysis and should be borne in mind.
Neal Alexander
[1] Stepniewska K, White NJ. Some considerations in the design and interpretation of antimalarial drug trials in uncomplicated falciparum malaria. Malar J 2006;5:127.
[2] Lindsey JC, Ryan LM. Tutorial in biostatistics: methods for interval-censored data. Stat Med 1998;17(2):219-38.
[3] Durrani N, Leslie T, Rahim S, Graham K, Ahmad F, Rowland M. Efficacy of combination therapy with artesunate plus amodiaquine compared to monotherapy with chloroquine, amodiaquine or sulfadoxine-pyrimethamine for treatment of uncomplicated Plasmodium falciparum in Afghanistan. Trop Med Int Health 2005;10(6):521-9.
[4] Alexander N. Precision of rate estimation under uniform interval censoring. Stat Med 2008;27(17):3442-3445.
Competing interests
No competing financial interests. The author has published analysis of antimalaria trial data using methods different to the ones advocated by Ashley et al (Dunyo et al 2006 PLos Clinical Trials 1:e14).