The Water Birth Series Part Two Audio File
Yesterday we talked about the types of evidence available for water birth, but what changes about interpreting the studies on water birth if this wide collection is mostly non-randomized studies?
First, it does NOT mean these studies are invalid or poor quality any more that conducting a study as an randomized controlled trial means the study is valid and high quality. There are many decisions a researcher makes beyond whether or not the study participants will be randomized to an intervention. And it is how a researcher makes these decisions that determine the validity and quality of a study. But it does mean that we, as consumers of this research, must be aware of the biases of each study we look at. Most importantly, we need to understand how those biases may make an estimate of the “truth” about water birth different from the actual truth about water birth.
This is hard to do for the average researchers – and even harder for the average clinician may not have had a good research class. This is why it is so common when you read systematic reviews that they limit the study to randomized controlled trials or the review will be limited to studies that meet certain “quality” criteria. This makes it easier for a broader group of interdisciplinary researchers to participate in writing systematic reviews. But, when we do we artificially limit the “available evidence” and this type of artificial limiting leads people to believe there is little evidence about neonatal safety with water birth.
For the meta-analysis I led, we used a tool that helps us separate different types of bias in a paper. This is important because different types of bias can have different types of effects on what the paper finds.1 The rest of this post is dedicated to talking about those types of bias.
Confounding is a special problem that happens in some studies that doesn’t actually depend on the other types of bias. Confounding happens when something NOT measured is associated with both the intervention and the outcome. So in our case, anything that may determine if a woman deliver
s in water AND may determine an infants status after delivery was a potential confounder. I’m sure you can think right away of some potential confounders, like comparing water birth women to women being treated for gestational hypertension who were not allowed to use the water. Most studies addressed the issue of confounding by restricting the sample to low risk pregnancies and demonstrating that the two groups had similar maternal and gestational age, some demonstrated similar birth weights.
We thought at first that there may be risk for confounding if the control group used uterotonics or analgesics, but our review of the latest systematic reviews gave us no reason to suspect that this would confound the results; though it was interesting that neonatal safety of uterotonics and analgesics was determined based on limited outcomes of APGARs and NICU admission. My meta-analysis reported on nine different measures of neonatal well being reported with water birth
Selection bias is a problem that happens when the method you use to select people for your study creates unequal groups. This was a problem for studies that looked back at their medical records to select women who had a water birth, and then made a control group of “low risk women”. Why? Because giving birth in water requires that you not experience a reason to be removed from the water – so not funny fetal heart tones, no meconium, whatever the criteria were that would make a woman leave the water. None of the studies reported applying these criteria to their control group, so they most likely ended up with a group of the best water birth labors, and just normal labors of low risk women. If this happened – if there was a difference in the way they selected the women for the groups – the study would be biased in favor of the water births.
Measurement bias was an interesting one for water birth because measurement bias occurs when a study doesn’t do a good job of sorting people into the experimental and control groups. This may seem unnecessary in water birth studies – it is obvious who delivered in the water. But we counted measurement bias when the control group included women who used hydrotherapy during labor, but did not deliver in water, and there was no explanation of why. When the explanation was missing, we were concerned the study may be including all women who were removed from the water for exclusion criteria in the control group. If this happened, a study would be biased in favor of water birth.
Attrition bias is that problem I explained earlier – when you have uneven dropout from a study between the control and experimental groups. We were worried about attrition bias in a few studies because they never told us the number of women who started the study, so there was no way for us to determine if unequal attrition occurred. Attrition bias would most likely bias the study in favor of which ever side had the highest attrition.
Missing data causes bias because there are often important reasons why data is missing – like if the group that gets one medicine all dies before the final assessment. It was rarely a problem because these studies were relying on standard medical record data. But we did have a few studies that told us there were “no differences” but never gave us the raw numbers to be sure all the women were included in the analysis.
Outcome Measurement Bias
Outcome measurement bias happens when the person evaluating the outcome can influence how the outcome is counted. For Most studies was not a risk because, for the most part we were dealing with standard assessments that were included in the medical record. But there were a few studies that were testing water birth as something new at their facility. In these cases we were aware the medical team may have a lower threshold for intervention in babies born into water.
Reporting bias happens when a researcher decides only to publish the “significant” results. This was not a problem in the water birth studies because, as I have said, they were all reporting basic neonatal assessments from the medical record. And almost all of the studies declared what they were going to report in the methods section and were faithful to reporting those outcomes.
We cast a wide net to be sure that if there was a study that reported evidence of poor outcomes for with water birth, we wanted to find it. So we included studies that were only reported at conferences so all we had was an abstract. We included studies that were in different languages and used Google Translate to ensure we had the right numbers for the right data – but we didn’t want to use Google Translate to try to score the risk of bias because the risk of errors in translation was too high. Some of our studies were publish during the 80s and early 90s when reporting standards were different, they didn’t give us the information we needed to assess these issues. This means we were not able to assess the risk of bias for all studies.
But there is something we can do to estimate if bias is changing the results of our study. It is called a sensitivity analysis, and we can use it to see if the results we found were stable when we removed any studies at risk of bias from the analysis. If we find that the answer to the question is different in the sensitivity analysis, we assume the true answer is closer to the sensitivity analysis than the full analysis — remember, we don’t know for sure these studies had bias, only that they were at risk. If you check out the study, you will see that the results of the sensitivity analysis did move some estimates that favored water birth to the null — which means that there was no difference between water birth and conventional delivery.