Does Progesterone Treatment Prevent Preterm Birth? A Case of “Skim Milk Masquerades as Cream”
This is off my usual beat, but a relative asked me to investigate progesterone treatment to prevent preterm birth. In her first pregnancy, membranes ruptured at 31 weeks and the baby was born a week later. (The baby was fine thanks to her mother taking good care of herself in pregnancy, steroids to mature her lungs, and her mother’s colostrum and breastmilk.) Pregnant for the second time, this woman had been told that progesterone injections would be given weekly beginning at 20 weeks to prevent a second occurrence. Being a member of my family, she wasn’t quite so willing as the average pregnant woman to automatically agree to this plan.
I started with the Cochrane Database of Systematic Reviews, the granddaddy of systematic review collections and a highly respected source. (A systematic review is a study of studies. Reviewers collect and analyze research on a specific issue using prespecified criteria. If the studies are sufficiently alike, statistical techniques can be used to pool their data.) The Cochrane had a review current as of 2008 that included four trials of progesterone treatment for women who had a preterm birth in a prior pregnancy. Progesterone was administered either as weekly injections or as a daily vaginal suppository, depending on the trial. Here is what the study summary said:
Progesterone was associated with a statistically significant reduction in the risk of preterm birth less than 34 weeks’ gestation (one study; 142 women; risk ratio (RR) 0.15; 95% confidence interval (CI) 0.04 to 0.64); [and] preterm birth less than 37 weeks’ gestation (four studies; 1255 women; RR 0.80; 95% CI 0.70 to 0.92).
For those of you unaccustomed to interpreting statistical lingo, let me translate: The reviewers’ statistical analysis revealed that compared with untreated women, one small study found that treated women were much less likely to deliver before 34 weeks’ gestation, and when data in four studies were pooled, women were somewhat less likely to deliver before 37 weeks’ gestation (the demarcation for full term). In neither case does the CI (confidence interval) cross one, which means differences between groups was probably not due to chance. In other words, progesterone treatment appears to be highly effective. The reviewers also found no short-term or longer-term differences in adverse outcomes for mother or child, but deemed the body of evidence on safety inadequate primarily because there is only one small follow-up study of 274 children. (We’ll come back to this study in a minute.)
This sounded extremely promising, but appearances can be deceiving. I knew that the Cochrane rules for conducting a systematic review do not capture all the possible problems with an individual trial, and reviewers, however well-intentioned, unconsciously will bring their biases along with them. Accordingly, I decided to get the four trials and evaluate them myself. A very different picture emerged when I did:
Trial #1 was published in 1975 and included only 43 women. I wasn’t able to get a copy of it, but we can ignore it as it didn’t contribute much to the review’s conclusion.
Trials #2 and #3 were published in 2003, and when I was searching for the papers, I found they had been critiqued in a commentary by Marc Keirse. Dr. Keirse is one of the authors of the bible of evidence-based maternity care, Guide to Effective Care in Pregnancy and Childbirth, and a professor at an Australian medical school. I verified his criticisms with my own copies of the two trials.
Trial #2 took place in Brazil. It was a vaginal suppository trial of progesterone versus placebo in 157 women. Dr. Keirse pointed out a number of serious problems that indicated that the investigators either did not know how to analyze their data properly or deliberately manipulated the statistics to get the “right” answer. Chief among them is that investigators collected data on 157 women, even though they had calculated before the trial that 90 women would be sufficient, and getting all 157 participants involved collecting data for a full two years longer than planned. One reason they might have done this is that their findings were not statistically significant when they reached 90 participants, so they kept collecting data until differences emerged (a process that is not ethical in research). This is akin to flipping a penny as many times as it takes for “heads” to show up with the desired frequency. Also important, investigators excluded women from the results if they had a preterm birth for reasons other than spontaneous preterm labor, mostly (10 of 15 cases) for preterm rupture of membranes. But preterm birth is preterm birth. If the study was supposed to determine whether progesterone prevented it, those women should not have been excluded. Add them back in, and the gap between treatment and placebo groups closes.
Trial #3 was of weekly progesterone versus placebo injections. It was carried out at 19 U.S. medical centers and included 463 women. This trial was conducted and reported properly; however, Dr. Keirse points out that the difference between groups was not because progesterone treatment reduced preterm births. Preterm birth rates in the treatment group were identical with that anticipated in women not having treatment based on an earlier study of similar women conducted at hospitals in the same medical center network. The difference lay in women in the placebo group having much higher preterm birth rates than anticipated. Furthermore, the theory behind administering progesterone is that it will prevent contractions, but similar percentages in both groups made hospital visits for preterm labor and tocolytic therapy. It seems that whatever was going on, progesterone wasn’t preventing preterm birth.
Investigators in trial #3 also noted an adverse effect not considered in the Cochrane review: the weekly injections were painful and could result in soreness, swelling, and bruising. This would be trivial if progesterone injections prevented preterm birth, but does it?
Trial #4 was published after Dr. Keirse’s critique. This trial, so far as I can judge, was a well-conducted international multicenter trial of progesterone in vaginal suppositories. The largest trial, it included 659 women allocated to either progesterone gel or placebo. It reported no difference in rate of preterm births (42% progesterone vs. 41% placebo), preterm labor, or any other adverse outcomes.
The follow-up study, of 194 treated children and 84 untreated children (the trial allocated to groups on a 2 to 1 basis) from trial #3 at around age 4, is too small to detect differences in uncommon adverse effects. Even if the study were big enough, age 4 is far too young to evaluate potential effects on sexual maturation and reproductive capacity. We should not be complacent about the safety of exposing fetuses to weeks of excess levels of a sex hormone. Keirse points out that the use of another hormone confidently prescribed to pregnant women, DES, became notorious for causing devastating effects in the offspring. The damage did not become apparent until children reached their teens, but continues to severely impact the health of the grandchildren of the women prescribed DES.
We’re not done yet. Had the trials shown unequivocally that progesterone treatment was effective, we do not know if their results can be generalized to all women with prior preterm birth. The trials included both women who had prior preterm births preceded by preterm labor and women whose preterm births were preceded by preterm rupture of membranes. Those situations overlap in that one can lead to the other, but the mechanism behind the two is different. In my family member’s case, rupture of membranes occurred out of the blue and was attributed to a silent infection. Leaving aside that the trials did not find that progesterone prevented preterm contractions, prior preterm labor was not her problem. How, then, even if it worked, would progesterone treatment benefit her?
My family member plans to decline progesterone treatment. Unfortunately, clinicians at her hospital and others like it will go on prescribing it in the belief that the evidence shows that it works and is harmless.
The progesterone review illustrates a general problem: Conscientious clinicians who want to practice evidence-based care cannot possibly review the research on every pertinent topic. They must rely on recommendations derived from systematic reviews, but the “black box” nature of these reviews makes it impossible to distinguish “cream” from “milk.” Until such time as systematic reviewers universally engage in critical thinking, the danger will remain as Mark Twain put it: “Researchers have already cast much darkness on the subject and if they continue their investigation, we shall soon know nothing at all.”
Citation: Dodd JM, Flenady V, Cincotta R, & Crowther CA (2006). Prenatal administration of progesterone for preventing preterm birth. Cochrane database of systematic reviews (Online) (1) PMID: 16437505