If you were to report survival and other outcomes among infants with a very high risk of dying or having long-term impairments, why would you include babies for whom a decision was made to let them die?
Let me put it this way, if 1000 babies are born in each of 2 epochs, and 900 are left to die, and the survival rate was 40% in the first epoch, and 60% in the second, among the 100 babies who were treated, then this is either not significantly different p=0.051 or highly statistically significant p=0.0072, depending on whether you analyze the data using the denominator as all live births, or only the live births who received active care with an attempt to have them survive.
In a brand-new report in the FPNEJM, almost all of the data regarding survival and long-term outcomes are presented as proportions of live births. The denominator used for almost all of the analyses was the 4274 live births, of whom over 1000 did not receive active care, leaving 3158 for whom neonatal intensive care was instituted.
I can see reasons for doing some of the analyses like this: if the decision not to intervene was made based on an analysis of risks, and only the very highest risk babies were not actively treated, then leaving them out could skew the data, and make them look more positive. But in reality we know that the major determinant of whether you get intensive care in this gestational time period (in the NICHD NRN, and I am sure in many other places also) is the hospital that you are born in, not their risk profile necessarily.
However, in those hospitals that are selective in treating the most immature babies, if the babies who were not treated did indeed have a higher risk of mortality, then leaving them out would make the data look better than they would be if all infants received active treatment.
That is indeed what the previous NRN data seem to show. In the paper from 2015, examining data from 2006 to 2011, it was the centers that treated all the babies who had the best survival when expressed as a proportion of live births, ranging from 10 to 20% at 22 weeks, for example. But if you look at survival among all those who received active treatment (including the babies from the universal treatment hospitals) at 22 weeks 23% survived, which is a little better than the survival in the centers that treated all the babies. Those hospitals that treated none of the 23 week infants had no survivors.
So how should we calculate survival rates? If there are many babies not receiving active treatment, then a shift to treating more babies might decrease the proportion of survivors among those treated, but increase the total survival among the live born.
I think that both numbers should be reported, as well as the numbers not actively treated, that is the only way you can really understand what is happening.
The new publication from the FPNEJM (Younge N, et al. Survival and Neurodevelopmental Outcomes among Periviable Infants. The Formerly Prestigious New England journal of medicine. 2017;376(7):617-28.) concentrates on survival among all live births of less than 25 weeks gestation, and barely reports survival and outcomes among those infants who received active care: only the Odds Ratios for those outcomes being reported in one section at the end of table 4.
It is possible to calculate some of the other outcomes, with the proviso that the exact numbers could be slightly different to the numbers I present below, depending on rounding errors, and other sources of variation.
The article reports outcomes from 3 non-overlapping epochs, infants born in 2000-2003, 2004-2007, and 2008 to 2011. They include data from 11 centers that were part of the Neonatal Research Network (the NRN) in all of those years. The previous study I mentioned had data from 25 centers that were members of the NRN from 2006 to 2011, so these new data include a subset of the data from Rysavy MA, et al. (Between-Hospital Variation in Treatment and Outcomes in Extremely Preterm Infants. The New England journal of medicine. 2015;372:1801-11) and add to that data from earlier years, and give more information about outcomes at about 2 years, the Bayley scores from version 2 for the earlier epoch, changing to Bayley 3 during the second Epoch. They use different thresholds for developmental delay for the different versions of the BSID, and concentrate on the cognitive composite from version 3.
The data show an improvement in survival (a small improvement but not likely to be due to chance) and an improvement in survival without the famous NDI (which, from here on, I will call neurological impairment or developmental delay, NDDI. I continue to insist on the fact that a low Bayley score is NOT an impairment, but a screening test showing a potential delay in development). For example at 23 weeks the frequency of that outcome increased from 7% to 11% to 13%, when calculated based on all live births, but increased from 9% to 16% to 19% when calculated based on babies who received active treatment.
Overall survival at 23 weeks is reported as 20%, 20% and 24% in the 3 epochs as reported in the article, but, when based only on those who received active treatment, it is 27%, 28% and 35%.
I have seen comments that these data show absolutely no improvement at 22 weeks, but in fact, expressed as survival among those who received active treatment, survival increased from 10% to 21 % and 17%, which may not be statistically significant, but is about a doubling of survival from the first epoch to the 2 later epochs.
Survival does seem therefore to be improving, the proportion receiving active treatment has not changed, however; in this study the improvement in survival is therefore probably a real improvement in our capacity to look after these babies, rather than a change in who we select for intensive care.
Among survivors, the proportion with NDDI has decreased somewhat, the discussion of the article puts it like this
the rate of survival without neurodevelopmental impairment and the rate of survival with neurodevelopmental impairment increased similarly (adjusted relative risk, 1.27; 95% CI, 0.99 to 1.65).
I guess ‘increased similarly’ is the way that is stated because the lower 95% CI is 0.99, I think you could put that differently and state that, among survivors, the Odds of not having NDDI increased from the 1st to the 3rd epoch, by a factor of 1.27. Although the CI include 1.0, I think that is very reassuring.
With this improvement in survival, I think there should be a reconsideration of hospital policies, and a lower threshold for intervening, an overall survival of about 1 in 5 at 22 weeks about 1 in 3 at 23 weeks (among babies who received active care) would both seem to make intervention more reasonable for more infants; not necessarily for everyone, as always, family values and wishes are extremely important in these decisions, but as survival improves, it makes sense that our willingness to try for survival should also improve.
The most encouraging thing about these data is that there is no evidence at all that increased survival increases the proportion of impairment among survivors, with the limitations of the data presented, the opposite is much more likely to be true.