Jim’s AsPredicted.org prediction

I like to predict what trials will show before I see the results. For my reasons click here.

The PHOENIX trial in this week’s Lancet (click here) tested the effect of immediate or delayed delivery for women with late preterm (34-37 weeks) pre-eclampsia.  The researchers randomised 450 women (471 infants ) to planned delivery and 451 (475) to expectant management. The primary maternal outcome was a composite of death, morbidity or a systolic blood pressure of at least 160 mm Hg, and the primary fetal one a composite of death or neonatal unit admission.  Nottingham was a participating centre but, apart from recruiting a few participants, I had no involvement.

I favoured delay. I thought planned delivery would reduce trivial adverse maternal events such as episodes of high blood pressure, but nothing that mattered, and that it would harm the baby. In May of this year, before I’d seen any results, I wrote on aspredicted.org (click here):

1. “The primary maternal outcome will favour immediate delivery. This will be statistically significant at the P<0.05 level. However after exclusion of the component “recorded systolic blood pressure ≥160 mmHg” from the primary maternal composite outcome the difference will no longer be nominally significant. I appreciate that this could be judged a data driven analysis, which is why I am registering my prediction here.”

2. “The primary short term baby outcome will favour expectant management. This will be statistically significant at the P<0.05 level.”

I was partially correct. The primary maternal outcome was reduced by early delivery, 289 (65%) v 338 (75%), relative risk 0·86, 95% CI 0·79–0·94; p=0·0005, and the primary fetal one increased, 196 (42%) v 159 (34%), RR 1·26, 1·08–1·47; p=0·0034). We can be confident that both effects are real. The trial was registered, the outcomes pre-defined, the sample size large, everyone was followed-up and the differences are unlikely to have occurred by chance.

However I was wrong to predict that the reduction in adverse maternal outcomes would disappear when raised BP was excluded. The top row of table 3 “maternal morbidity composite outcome” i.e. the composite without the raised BP component, was 68 (15%) v 90 (20%), RR 0.76; 0.59-0.98. The intervention really does reduce maternal morbidity. The authors argue that this strengthens the argument in favour of early delivery. But let’s look at what the morbidity consisted of.  Supplemental appendix table 3.  Not easy to access, so I’ve tidied it up below.

*The maternal death occurred unexpectedly, 5 days after delivery in a woman with medical co-morbidities, and was judged to be unrelated to the trial. **The numbers don’t add up because some women had multiple morbidities. No participants had a Glasgow coma score <13, stroke or reversible ischaemic neurological deficit, transient ischaemic attack, cortical blindness or retinal detachment, posterior reversible encephalopathy, hepatic haematoma or rupture, or acute renal failure (creatinine >200 µmol/L).

Pretty much all the difference was in worsening tests revealing liver or renal damage or low platelets. These are the tests which, alongside BP measurement and the fetal heart rate pattern, we use to monitor pre-eclampsia, and to judge the timing of delivery with expectant mangement. In well organised hospitals we should be able to prevent really serious adverse events such as stroke, heart attack, permanent organ damage or death. There were few such events and no obvious excess in the expectant management group.

Fetal outcomes are in table 4. No babies died and the only differences were in admission rates to various levels of neonatal unit admission. Here’s the relevant section.

There are more details in the supplementary appendix table 5. Serious adverse baby events were very rare and did not differ between groups.

Here’s my final take. The “harms” to the mother from waiting a bit were limited to abnormal blood or blood pressure tests to which the doctors responded correctly, and from which no long-term harm ensued. The “harms” to the baby from planned delivery were a bit of additional monitoring and oxygen therapy from which all the babies also emerged healthy. The authors conclude that “this trade-off should be discussed with women”. I agree, but I still favour delay.

I made a third prediction on Aspredicted.org:

3. The primary long term baby outcome (PARCA-R at 2 years) will not show any statistically significant difference (at the 5% level) between the groups. However I predict that the point estimates for those measures which had been predefined in the analysis plan e.g. mean or median scores, or rates of scores below various cut offs, will all favour expectant management.

Babies are resilient. They recover pretty well from all the stress that nature and doctors throw at them.  But they prefer to not deliver preterm without a good reason. I’ll have to wait a little longer to find out if I’m right, and the PHOENIX trial is unlikely to be big enough to prove it, but I still think there will be subtle long-term harms from planned early delivery.

Jim Thornton