 |
|
Letters to the Editor Indian Pediatrics 2004; 41:1182-1183 |
|
Indomethacin Prophylaxis for Intraventricular Hemorrhage in Very Low Birth Weight Babies |
|
The article on indomethacin prophylaxis for intraventricular hemorrhage by Nair, et al. in the June 2004 issue of Indian Pediatrics has some serious statistical and methodological errors that need clarification(1). The calculated sample size with the assumptions of the authors (15% to 5% reduction, alpha error 5% and beta error 85%) using Epi Info 6 yields a requirement of 180 subjects per limb, rather than 154 as stated. The authors’ calculation of post hoc power of the study after the interim analysis is also erroneous. The authors have to explain how they arrived, post hoc, at a figure of "70% power". Epi Info 6 program shows that to detect a 15% to 5% reduction in the key outcome, the interim sample size (n = 115) had a power of only 30%. To detect the difference in major IVH that was actually present in the study (10.7% vs 6.7%), as being significant with a 5% error, the interim sample size had a power as low as 6%. For the sub-group analyses where p value was <0.05, post hoc power calculation was anyway meaningless. The relative risk calculation of IVH grade III and IV in the birth weight category of 750-999 g is faulty. Since RR is defined as the incidence among exposed divided by incidence among the unexposed, it works out to be 6/24 divided by 1/26 (Table II). This gives an RR of 6.5 (95% CI of 0.8 to 50.1). This is quite different from what the authors have calculated: RR 2.05 (95% CI 1.29-3.26). In fact, there are many more errors in the relative risk calculations. For instance, among the 750-999 gms group, the authors claim that the RR for chronic lung disease is 1.84 (95% CI 1.08-3.11). But, the actual RR is 10/24 divided by 4/26, which is 2.7 (95% C10.98-7.5)(Table III). Next instance: in the 1000-1250 g group, the authors claim that the RR for chronic lung disease is 1.91 (95% CI 1.26-2.9). But, the actual RR is 6/32 divided by 1/33, which is 6.2 (95% CI 0.8-48.5). These are all gross deviations and there is no way that one could explain them away as being close approximations. An important drawback in design was that the study was unblinded. Thus, a measurement bias on the part of the person doing the ultrasounds cannot be excluded. This would be a major source of error. Sourabh Dutta, Assistant Professor, Division of Neonatology, Department of Pediatrics, Postgraduate Institute of Medical Education and Research, Chandigarh 160 012, India. E-mail: [email protected] |
|
References |
|
|
