We read with interest the results of the randomized trial on
Ringer’s lactate (RL) vs normal saline in children
with acute diarrhea and severe dehydration [1]. The study
authors had used a rigorous methodology to address a
pertinent question, and found no difference in the outcomes
between the two groups. We wish to highlight a few
methodological issues, which, if addressed, could have
further improved the quality of the study:
The authors mention that the primary
outcome variable was ‘change in pH from baseline’. However,
they possibly used the difference in post-intervention
pH between the groups and not the magnitude of ‘change from
baseline’ for calculating the sample size. There is no
mention of the mean or SD of the change in pH from baseline
in the study from which the authors estimated the sample
size. The sample size could have been very different if the
standard deviation of this outcome was large (or small!)
from the one used in the sample size calculation.
At least four different approaches can be
employed to analyze a continuous outcome that is measured at
two time points (i.e. baseline and after treatment)
in a RCT: post-treatment, change between baseline and
post-treatment, percentage change between baseline and
post-treatment, and analysis of covariance (ANCOVA) with
baseline value as a covariate [2]. The authors chose to use
a slightly different approach using the change from baseline
as the outcome but used ANCOVA to adjust for a few
covariates other than the baseline pH. Compared to the
change from baseline analysis, ANCOVA with baseline as the
covariate has higher statistical power, particularly if
correlation coefficient between baseline and follow-up
values is <0.8 [2,3]. More importantly, the latter analysis
has the advantage of being unaffected by baseline
differences between the groups (it adjusts each patient’s
follow up score for his/her baseline score) [3]. In
contrast, the change from baseline analysis takes the
pretest difference too seriously and might produce biased
results in the presence of imbalance in baseline scores
between the two groups [4]. Though not statistically
significant, the baseline pH was higher in the RL group [1].
Instead of providing only the P
value, the authors should have provided the results of the
‘ANCOVA’ model in a more detailed way - Vickers, et al.
[3] have provided an excellent model for depicting the
results of the analysis using ANCOVA model (albeit, with
baseline as covariate). The unadjusted and adjusted
mean difference of change from baseline along with 95% CI
would have given the readers some idea about the precision
of the results and the magnitude of confounding caused by
the two covariates.
The term ‘repeated measures’ usually
implies that the analysis involved an interaction term, i.e.
‘group*time’ in the model. It is not clear if the P
value mentioned in the study refers to the P value of
this interaction term.
The authors adjusted only for baseline
serum sodium and chloride - the two factors found to be
significant on bivariate analysis - in the ANCOVA model.
Many researchers have effectively demonstrated the
inappropriateness of this approach, i.e. adjustment
for only ‘significant’ variables [5]. Moreover, the clinical
relevance of adjusting for serum chloride when baseline
serum pH had already been accounted for in the change from
baseline analysis is not clear. The better approach would be
to use pre-specified ANCOVA where a few a priori
selected important baseline variables are used as covariates
[6]. An important variable that had to be adjusted was the
time interval between the baseline and the time to achieve
primary end point, as the latter was not fixed in the two
groups. Not including it in the model because of lack of
significant result is not valid as the insignificant result
is more likely be due to lack of power rather than due to
true absence of difference between the groups.
1. Mahajan V, Sajan SS, Sharma A, Kaur J.
Ringers lactate vs normal saline for children with
acute diarrhea and severe dehydration- A double blind
randomized controlled trial. Indian Pediatr. 2012;49:963-8.
2. Vickers AJ. The use of percentage
change from baseline as an outcome in a controlled trial is
statistically inefficient: a simulation study. BMC Med Res
Methodol. 2001;1:6.
3. Vickers AJ, Altman DG. Statistics
notes: Analysing controlled trials with baseline and follow
up measurements. BMJ. 2001;323:1123-4.
4. Van Breukelen GJ. ANCOVA versus change
from baseline: more power in randomized studies, more bias
in nonrandomized studies [corrected]. J Clin Epidemiol.
2006;59:920-5.
5. Begg CB. Suspended judgment.
Significance tests of covariate imbalance in clinical
trials. Control Clin Trials. 1990;11:223-5.
6. Senn S. Testing for baseline balance in clinical
trials. Stat Med. 1994;13:1715-26.