Source: Library of Congress

Source: Library of Congress

A new study coming out in the journal Anesthesia & Analgesia[1] in August claims that having an epidural during childbirth can cut both the pain of childbirth and postpartum depression.  Obviously epidurals can reduce pain (if they work for you; there are a small subset of women who may experience more pain or other problems), but the idea that they reduce postpartum depression (PPD)?  This is something that ought to be looked into given the prevalence of PPD and the impact it can have on mother-child outcomes.

The study comes from China where 214 women either received an epidural or not during labour and were then assessed for various outcomes at three days and six weeks.  Women self-selected into their groups (i.e., they decided if they wanted the epidural or not) and thus demographic variables were also measured to ensure no differences between the two groups or to control for any inherent differences.  (It seems women opted into the epidural early on and it’s unclear if women were able to select an epidural after a certain point.)  Women’s pain was assessed using a 1-10 scale; all women reported on the pain once they had reached 2cm or more dilation (which is fuzzy given pain at 2cm and pain at 7cm would be very different) and again at 10cm dilation and women who did receive the epidural also reported on pain at both 10min and 30min post-epidural.

Most generally, the authors found no demographic differences between the women who selected to have an epidural versus those who did not.  There were, however, a few birth differences with more women in the epidural group (66.4%) receiving oxytocin before delivery than those in the non-epidural group (50.5%), though the use of oxytocin was unbelievably common.  More women in the non-epidural group ended up with a c-section (36.4% vs. 15%) and were less likely to have a vaginal delivery (50.5% vs. 70.1%).  The duration of the first stage of labour was significantly shorter in the non-epidural group (300min versus 510min); however, the other stages were of similar duration.  Pain at 10cm was rated significantly worse for the non-epidural group versus the epidural group (with all mothers who received a c-section removed from the analysis).

Postpartum measurement and analyses on day three found significant differences between the two groups on both Marital Well-Being and Maternal Anxiety, with women in the non-epidural group rating lower on both variables, but found no differences in Depression.  Postpartum measurement and analyses on day forty-two (six weeks) found that women in the non-epidural group were less likely to be breastfeeding (49.5% vs. 70.1%) and more likely to suffer clinical levels of depression measured using the Edinburgh Postnatal Depression Scale (34.6% vs. 14%).  The authors used Stepwise Logistic Regression to test the viability of each of the 10 variables that were significantly related to PPD at six weeks – satisfied with habitation condition, PMS pre-pregnancy, attendance at childbirth classes, NRS pain score at 10-cm cervical dilation, c-section, satisfied with overall medical service, marital satisfaction score at 3d postpartum, anxiety score at 3d postpartum, depression at 3d postpartum, and breastfeeding at 42d postpartum – and receiving an epidural still remained significant in reducing the risk, along with attending childbirth classes, depression at day three, and breastfeeding at six weeks.

This research brings to mind another piece of “research” (quotations there for good reason which I’ll get to) from 2012 that reported a link between post-traumatic stress disorder (PTSD) type symptoms and natural birth[2].  In this study they found that of the nearly 26% of women who experience PTSD-type symptoms, 80% had a natural, drug-free delivery.  The assumption therein was that it was the pain of childbirth that was causing the PTSD symptoms; however, and this is a big however, the “research” was so poorly analyzed, no such thing could be determined.  Yes, the statistic that 80% of the group who experienced these symptoms had a natural birth does hold, however, there were many other variables that were also significant, including previous “traumatic” births, depression and anxiety, medical complications during pregnancy, and they anticipated more fear and pain during childbirth.  The researchers controlled for none of these other variables and thus we simply can’t say that the issue is with natural birth per se, or if it is simply related to PTSD symptomology via a relationship with another variable.

But doesn’t this new study control for that, you ask?  Well, yes and no.  The first problem is that the new study doesn’t include interactions in their model (this is also a problem for the research on PTSD).  They have multiple independent variables that predict PPD, but they don’t test if or how these interact together to predict PPD and as anyone who works in the field of studying birth trauma or PPD knows, it is never a single issue (see [3] for a fuller discussion).  For example, the fact that there were significantly more c-sections in the non-epidural group leads one to question if c-sections interacted with other elements like pain or anxiety at day three to cause PPD at six weeks.  Another question that pops up when looking at the data is the role of depression at day three: Although there was no significant difference between the two groups on depression on day three, it was a significant predictor of PPD at six weeks.  This would suggest that it was interacting with another variable to result in the increase in the non-epidural group, perhaps having had a c-section or lack of marital support at home, to offer a couple suggestions; however, without interactions, we cannot say.

When looking at a complex issue like PPD, it is often a variety of factors interacting that lead to the outcome of PPD and although looking at independent predictors is helpful, it doesn’t give us close to the entire story (and given some of the findings here, which I will discuss next, this becomes even more important).  In addition to the event-level variables, there was no attempt to examine women’s fear of pain prior to labour, mental state prior to labour, or any psychological variable that would have a large influence on a woman’s reaction to events in labour.  The focus prenatally was on the physical side of pregnancy.

A second problem is that the authors fail to provide information on the type of labour women had.  The mere fact that a significantly greater number of women in the non-epidural group ended up with an emergency c-section suggests something else was happening during these labours.  Given previous research that finds having an epidural often increases the risk of a c-section[4], one has to wonder what was going on here.  One possibility has to do with the use of oxytocin in the two groups.  If the epidural group received it after the administration of the epidural in order to speed things up again, the intense contractions and increased pain would be less noticeable.  However, for the 50.5% of women in the non-epidural group, the increased intensity and pain would be far more noticeable than if they had a natural delivery.  In many ways, the main question here becomes about the protocols in place at this hospital that are resulting in such high rates of intervention.

The third problem (and final one I’ll discuss herein although I’m positive there are more) is the use of stepwise regression to determine significance.  For those who don’t know stats too well, regression is a statistical technique where you enter various variables to see how well they predict some outcome individually and as a whole (the “model”).  The most common way to do this is to enter the ones that have theoretical or practical relationships (along with any interactions) and you get values that tell you how well they do at predicting the outcome as a whole and also how each individual variable does in predicting the outcome.  One of the methods you can use to enter your predictor variables is to do it “stepwise”.  In this technique, each variable is examined one at a time to the model and if it results in a significant increase in predictive power, it is included, otherwise it’s ignored so you get the variable that predicts the best in Step 1, then the one that predicts the remaining variance in Step 2, and so on.  At the end you have the variables that result in the best individual predictions.

Stepwise regression has been written of in scholarly articles as being the one type of regression that should NEVER be used (e.g., [5]).  One of the main problems is that stepwise overestimates significance, sometimes biasing coefficients up to 25%[6], and the methods used herein (i.e., logistical stepwise regression, small-moderate sample size for the number of variables) are ones that are prone to this overestimation.  This means that the significance of their variables may not be that significant.

A second problem is that stepwise will actively hide the third variable problem.  If two variables are highly related and both are related to the outcome variable, only the one that has the greatest prediction will be entered.  Why?  Imagine we have variables A, B, and C and we’re using A and B to predict C.  If A and B are both highly related and both are related to C, once the one with the highest predictive value is included, say B, the remaining variable A will not predict C above and beyond what B has done and will not be included leaving it to seem as though B predicts it alone when perhaps it is in part due to the relationship with A.  In the case herein, the final model they provide may reflect other relationships (such as the relationship between depression and breastfeeding duration or having had a c-section) that are masked by the method used.

[Notably, the authors say they did one regression that was not stepwise and that the use of an epidural remained significant as an independent predictor, but fail to report the significance of the other variables in the model and also failed to include any interaction terms to allow us to look at the relative significance given the focus on the epidural herein.  In this vein, it’s worth mentioning that although the authors focus on the use of the epidural as their centrepiece, even within their own flawed analyses, the epidural is the least significant predictor of the four, with breastfeeding being the largest and most significant independent predictor.]

Despite my problems with this study, it raises a very good point and one that we should consider: For some women, the use of an epidural may truly reduce the risk of PTSD symptoms or PPD given the rise in women who are becoming incredibly fearful of the process of birth.  In the UK alone, there is a surge in requests for c-sections in order to avoid labour from women who have tokophobia (the fear of childbirth)[7].  Tokophobia has multiple possible causes including previous negative experiences, fear of pain, depression, previous sexual assault, and so on.  Some believe that this increase is in part due to the dramatic and painful portrayal of birth in our society, but for others it’s likely from their own experiences of being mistreated, feeling helpless, and lacking support that leads to a fear of experiencing such things again.

If we enter a situation in which a woman is in labour and terrified of the pain for whatever reason, it is sensible and fair to offer her the choice to have an epidural.  However, if we start telling women that having an epidural will lower the risk of PPD based on this data alone, we run the risk of assuming a simplistic answer to a complex problem and harming more women.  If an epidural replaces support, care, and giving women the chance to feel in control of their labour (with or without pain relief), we will likely run into even more cases of PPD and PTSD symptoms.  The flaws in this study – especially the failure to consider the labour as a whole and to discuss the differences that led to a significantly greater number of women receiving emergency c-sections in the non-epidural group and the failure to consider interactions between their predictors – mean that the value of this study lies in highlighting the issue of pain in labour and how various things may contribute to result in problems post-partum and various things may help prevent it.  If anything, the study tells us what we already knew: PPD is a complex and multifaceted problem that requires individualized treatment and prevention.


[1]Ding T, Wang DX, Qu Y, Chen Q, Zhu SN.  Epidural labor analgesia is associated with a decreased risk of postpartum depression: a prospective cohort study.  Anesthesia & Analgesia 2014; 119: 383-92.

[2] Polachek IS, Harari LH, Baum M, Strous RD.  Postpartum post-traumatic stress disorder symptoms: the uninvited birth companion.  Israeli Medical Association Journal 2012; 14; 347-53.

[3]Kendall-Tackett KA.  Depression in New Mothers: Causes, consequences, and treatment alternatives, 2nd Edition.  Oxon, OX: Routledge, 2010.

[4]Morton SC, Williams MS, Keeler EB, Gambone JC, & Kahn KL. Effect of epidural analgesia for labor on the cesarean delivery rate. Obstetrics & Gynecology 1994; 83: 1045-1052

[5] Flom PL, Cassell DL.  Stopping stepwise: why stepwise and similar selection methods are bad, and what you should use.  [Full text here:]

[6]Steyerberg EW, Eijkemans MJC, Habbema JDF.  Stepwise selection in small data sets: a simulation study of bias in logistic regression analysis.  Journal of Clinical Epidemiology 1999; 10: 935-42.