A new RCT published in JAMA pediatrics compared growth and other clinical outcomes between infants <33 weeks gestation and <1250g who were managed with routine gastric residual measurements or without. (Parker LA, et al. Effect of Gastric Residual Evaluation on Enteral Intake in Extremely Preterm Infants: A Randomized Clinical Trial. JAMA Pediatr. 2019) I was quite interested to read this when I saw the title, unfortunately the way it was analyzed, and the way it is presented make it nearly impossible to interpret. In addition there is at least one major error in the data presented.
The first problem is that although the “standard care” group has residuals measured, there is no indication of how they were interpreted. In the protocol which is provided as a supplemental file, the only mention of the feeding standards is as follows :
In addition, the nurse assesses the infant for any signs or symptoms of feeding intolerance or NEC (i.e., abdominal distension and/or tenderness, increased abdominal girth, visible bowel loops, presence of emesis, and visible blood in the stool). It is standard protocol to aspirate RGC prior to each feeding. However, for this study, this will only occur in infants randomized to Group 1.
What was done with any of this information is not described. Was the volume considered important? The colour? Presumably they didn’t aspirate prior to each feed in order to ignore the findings.
The next big problem is the primary outcome: “weekly enteral nutrition measured in mL/kg for 6 weeks after birth”, I am not sure what that means. Did they add all the intake over 6 weeks and compare between groups? Did they compare after each week, and so do 6 comparisons? Apparently, from the protocol, the plan was to do a t-test, designed for groups with unequal variance (“Welch adjusted” they call it). But the analysis which is presented is a Generalized Linear Mixed Model, which is a term that doesn’t tell me anything, but it seems to have been some sort of repeated measures test, which therefore should account for the multiple comparisons.
So what did they find? What were the primary outcome data for the two groups? I don’t know. Nowhere in the manuscript are the primary outcome results given. They do give a p-value however! In table 2 the first group of numbers are for weekly feedings in mL/kg/d and the p-value for Treatment is 0.048, but the actual numbers are written as NA. The next group of numbers are for the “simple main effect” and give some numbers which are not consistent with anything else they have written, i.e. for week 6 the numbers are “128.4 (119.9 to 136.9)” and “141.6 (133.2 to 150.0)”, according to the methods this should be the weekly feeding volume which seems quite unlikely. I presume this is either the daily volume on the last day of week 6, or the averaged daily volume over the 6th week. And I have to guess that the figures in parentheses are mean plus or minus 1 standard deviation, but that is never specified.
As far as I can tell then, by week 6 the babies were receiving inadequate feeds if they didn’t measure gastric residuals, and even more inadequate feeds if they did! To only achieve 140 mL/kg/d after 6 weeks of feeds in a group of babies with a mean of about 27 weeks and 900 grams seems to be well below what we should be achieving. As a result the growth outcomes are very poor, a 27 week baby weighing 900 grams, should by 6 weeks of age be weighing about 1400g, but, from one of the few results that are presented as interpretable data, both groups weighed just over 1100g (which I think are means adjusted for covariates)
Many of the results are presented as “least square means” which is an SAS (that is a particular stats software package) jargon for means, adjusted for covariates. Which again makes them difficult to interpret. Some of them are presented as the “mean estimated log weights” in the abstract, and sometimes in the abstract they are completely unexplained: “the no residual group were discharged 8 days earlier (4.21 [95% CI, 4.14-4.28] vs 4.28 [95% CI, 4.19-4.36]; P = .01)” 4.21 what? (I could have written WTF? but I am too polite).
It is not really surprising that not measuring aspirates would accelerate feed progression, even though here the weekly increase is from a desperately slow 18 mL/kg/d to an extremely slow 21 mL/kg/d. The big question is, is it safe?
Here again there are problems, in the abstract and in the text it is stated that the Odds for developing NEC in the intervention vs control group are 0.58 [95% CI, 0.18-0.19] vs 0.026 [95% CI, 0.006-0.109]). Which would be a 22-fold increase in the Odds of NEC, or an Odds Ratio of 22. But of course an Odds of NEC in the intervention group of 0.58 would mean that there were 25 cases of NEC and 44 without NEC, so that isn’t likely either, especially as the odds doesn’t lie between its 95% confidence intervals, which is impossible.
There is some potential clarification from the body of the article, in table 5 it is noted that the “odds” of NEC was 0.058 (0.018, 0.19) and in the results at the end of the section describing the subjects it is noted that 4 patients in the intervention group were withdrawn for NEC. Four out of the 69 intervention patients makes an incidence, a rate, or a frequency of 5.8% or 0.058. But it does not make an Odds of 0.58, the Odds of NEC is 4/65 (NEC/no NEC) which is 0.061. It looks like there were probably 2 cases of NEC among the 74 standard care group, for an incidence of 2.7%, and an odds of 0.0278.
After slogging my way through all these results it appeared that there were about twice as many cases of NEC in the intervention group as in the controls. I thought I was getting this all clear when I looked at the flow chart, the CONSORT figure, which states that there were 7 cases of NEC in the intervention group, and 4 cases in the controls. Which completely messes up all my attempts to understand this article. If there were 7 cases of NEC, then the incidence of NEC among the intervention babies is actually 10.1%, and the odds is 0.012, compared to 4 controls. with a frequency of 5.4% and an odds of 0.057.
In the discussion the authors state “we found no differences in incidence of NEC” which is clearly untrue, the incidence of NEC was quite different between groups. A true statement would have been “the difference in incidence of NEC that we found has very wide compatability limits, which include a possibility of a large reduction or a major increase in NEC”.
I think this paper is a complete failure of the review and editorial process of JAMA pediatrics (and of galley editing), how this could have been published in this form I don’t understand. It could have been a nice little RCT adding a bit more data to the question of measuring residuals, and should most clearly have stated that there was inadequate power to determine safety, and that the confidence intervals for the incidence of NEC are extremely wide. (If we assume that there were 4 cases of NEC in the intervention/no residuals group, and 2 in the controls, then the relative risk of NEC is 2.15 with 95% compatibility limits of 0.4 and 11. If there were 7 cases vs 4 cases, the RR is 1.99, 95% CL 0.6-6.5). We should also note that there were 6 deaths in the standard/measured residual group, and only 1 in the intervention/no residual group; which gives an RR of 0.19, 95% CL 0.02 to 1.5).
As it is we still are not clearly any the wiser, after a trial where it is not clear what was done or what was found.
I don’t take note of residual volumes, I have worked at one place which had not measured them for 15 years, and in 2 other places we stopped routinely measuring residuals completely while I was there. All that observational data suggests no benefit, and potential nutritional harms from measuring gastric residuals, but some stronger data, to convince other units to stop the practice if it is indeed safe, would have been helpful to improve nutritional outcomes of our very preterm babies.
Neonatal Research 
We would like to address several of the comments made regarding this article.
1. What was done with any of the information regarding gastric residuals not described. Was the volume considered important? The colour? Presumably they didn’t aspirate prior to each feed in order to ignore the findings.
The aim of this study was not to evaluate gastric residual color or volume but to determine differences between routine pre-feed aspiration and evaluation of gastric residuals and omission of gastric residuals. We have previously published a general set of guidelines (Li YF et al. (2014). Gastric residual evaluation in preterm neonates: a useful monitoring technique or a hindrance? Pediatrics and Neonatology, 55(5), 335-340) that included a suggested response algorithm used in our NICU for responding to gastric residuals.
2. The next big problem is the primary outcome: “weekly enteral nutrition measured in mL/kg for 6 weeks after birth”, I am not sure what that means. Did they add all the intake over 6 weeks and compare between groups? Did they compare after each week, and so do 6 comparisons?
Feeding intake is extremely difficult to measure and compare between infants on a daily basis. So trying to compare days before “full feedings” or day when” full feeding” is reached is difficult. Many babies will regress, thus there is a large problem with noise which has to be modulated for comparison purposes. Analysis for average daily intake for each week was thus used for the primary outcome.
3. Apparently, from the protocol, the plan was to do a t-test, designed for groups with unequal variance (“Welch adjusted” they call it). But the analysis which is presented is a Generalized Linear Mixed Model, which is a term that doesn’t tell me anything, but it seems to have been some sort of repeated measures test, which therefore should account for the multiple comparisons.
Generalized linear model (GML) encompasses linear models that appropriately handle a range of dependent variable types, including continuous, count, and dichotomous (McCullagh P & Nelder JA (1989). Generalized linear models (2nd edition). New York: Chapman & Hall/CRC publishers). A GLM for a continuous variable having 1 independent variable with 2 levels (1 numerator degree of freedom) and some number of error (denominator) degrees of freedom is exactly equivalent to a t-test with that same number of degrees of freedom. As t-tests do not provide the ability to include covariates, a GLM approach was used to evaluate models containing covariates.
4. Nowhere in the manuscript are the primary outcome results given. They do give a p-value however! In table 2 the first group of numbers are for weekly feedings in mL/kg/d and the p-value for Treatment is 0.048, but the actual numbers are written as NA.
The primary outcome was average daily intake in ml/kg for weeks 1-6. The least square means, along with the 95% CI are provided in the simple main effects listed for Weekly Feedings in Table 2. Those values are the results of the Simple main effects analysis, which are required as the treatment by time interaction was statistically significant, indicating that the pattern of differences between groups differed by week.
5. In table 2, the next group of numbers are for the “simple main effect”. I have to guess that the figures in parentheses are mean plus or minus 1 standard deviation, but that is never specified.
While, it may have been clearer had we labeled it Least Square Mean (95% CI) rather than Estimate, the table heading states the values in parentheses are the 95% CI for the value.
6. Many of the results are presented as “least square means” which is an SAS jargon for means, adjusted for covariates, which makes them difficult to interpret. Some of them are presented as the “mean estimated log weights” in the abstract, and sometimes in the abstract they are completely unexplained: “the no residual group were discharged 8 days earlier (4.21 [95% CI, 4.14-4.28] vs 4.28 [95% CI, 4.19-4.36]; P = .01)” 4.21 what?
In table 3, we specified that the gamma distribution was modeled for days to discharge. The gamma distribution uses log of the response as the link function, so the 4.21 and 4.28 values represent log transformed values. As with other transformed values, we provided raw (non-transformed) values as well, as those are more easily interpreted from a clinical standpoint.
7. In the text it is stated that the Odds for developing NEC in the intervention vs control group are 0.58 [95% CI, 0.18-0.19] vs 0.026 [95% CI, 0.006-0.109]).
In the abstract and text is it is stated that the Odds for necrotizing enterocolitis was (0.058 [95%CI, 0.018-0.19] vs 0.026 [95%CI, 0.006-0.109]). Table 5 states that these odds are for the control versus the intervention group. The error in the point estimate (0.58) and the CI (0.18-0.19) was discovered immediately after publication and this information was appropriately communicated to the journal editor.
8. In the discussion the authors state “we found no differences in incidence of NEC” which is clearly untrue, the incidence of NEC was quite different between groups. A true statement would have been “the difference in incidence of NEC that we found has very wide compatability limits, which include a possibility of a large reduction or a major increase in NEC”.
The purpose of this study was not to primarily evaluate NEC. It is very common to do secondary analyses in studies like this to look at potentially adverse outcomes. The number of NEC cases were more in the treatment group but this was not statistically different. To study this in a rigorous manner would require a much larger sample size.
9. As far as I can tell then, by week 6 the babies were receiving inadequate feeds if they didn’t measure gastric residuals, and even more inadequate feeds if they did!
The purpose of this study was not to determine whether the infants were having appropriate growth only to determine differences in growth between groups.