Epidemiology and the Limits of Medical Ethics in the NICU

William Meadow, MD, PhD., The University of Chicago, August, 2006

          Traditional ethical discussions in medical contexts usually invoke tensions between autonomy (do what the patient, or in the case of pediatrics the parents, want), paternalism (do what the doctors want), and beneficence (do the right thing).  Perhaps these concepts are so often invoked because they are timeless, articulating universal perspectives that transcend local variations.  Alternatively, these concepts may be invoked so often because, when it comes right down to it, they are unhelpful, unconvincing, and if they worked better there wouldn't be nearly so many tensions. 

            In this LIT-UP, I begin by embracing the latter view (that these concepts are not helpful), and then, I hope, move beyond its apparent cynicism to get to a better place.  I will invoke not theology or philosophy, but epidemiology to get there.

            Two levels of authority are traditionally articulated when considering the "ethics" of any particular medical dilemma.  At one extreme, the broadest view, which I will call distributive justice, asks about the "right-ness" of a course of action from the perspective of society as a whole -- can we afford to care for 500 gram-ers?  as opposed to spending money on what alternative -- immunizations for one year olds?  ICU care for 80 year olds? estate tax relief?  is it fair that some, but not all, of our patients can get access to brand-name NICUs?  or high-quality follow-up during childhood? or decent nursing homes in what has been charmingly called one's 'third life'?

The other extreme, the narrowest perspective, I will call personal justice.  This concept describes the traditional medical dyad -- one physician/ one patient -- with its familiar ethical imperative to do what's best for this patient, right here, right now, and never mind the broader social implications (which we probably can't do much about anyway, or so goes the unstated coda).

I will concentrate primarily on the ethical dilemmas surrounding care of extremely premature infants.  Premies are the paradigm case for NICU ethics -- they are common, their diseases are well described, and their overall outcomes are fairly well known.  It is my job to weave these threads together.  I begin with the distributive justice thread.

Distributive Justice in the NICU

Approximately 1 baby in 100 (40,000/yr in the U.S.) is born 'extremely low birth weight' (ELBW; <1.0 kg or 2.2 pounds).  All go to the NICU, more than half survive.  One baby in 500 (8,000/yr) is born alive with birth weight < one pound (~450g) -- some go to NICUs, almost all die. 

The first "law" of NICU ethics is summarized neatly then -- birth weight determines mortality.  But the entire phenomenon of 'birth-weight specific mortality' is quite limited in scope.  Currently, survival is so good above 750g/26wks (well over 80% in most big centers) that there are no credible grounds to support parental requests for non-intervention.  Below 450 grams/23wks, survival is so bad that there are no credible grounds to support physicians overriding parental requests for withholding intensive intervention and providing comfort care.  What can we make of the "middle ground", after noting that it is only 300 grams and 3 gestational weeks (from 23-25 wks) wide?

            One of the more interesting aspects of birth weight specific mortality is that it used to be a moving target -- and it no longer is.  Consider the first three decades of neonatology (the modern era of neonatology began, for all intents and purposes, in the mid 1960s, with the advent of mechanical ventilation and CPAP/PEEP).  Published articles in the mid-1970s with titles like "1500 grams, how small is too small?" were followed by mid-1980s by titles like "1000 grams, ….?" and,  by the mid-1990s, "800 grams …..?"   There was every expectation that the lower limit of birth weight specific survival would continue to be rolled back.  But it stopped.  Since 1997 or so, there has been no significant improvement in BW-specific survival for ELBW infants.  This observation has two consequences -- one, the unbridled optimism of the first three decades of neonatology should probably get bridled.  Second, published outcome studies now, and for the first time, have clinical relevance.  Before roughly 10 years ago, any outcome data set was hopelessly out of date before it was even published (the roughly 5 year time lapse between data acquisition and appearance in print was a recipe for obsolescence).  Now, since birth weight specific survival is standing still, published data acquired 5 years ago are relevant today, and will probably remain relevant 5 years from now.

            The second law of NICU ethics responds to a simple question.  If the smallest babies are least likely to survive, when do they die?  The answer is just as simple -- doomed infants die early, and the smallest and the sickest die the quickest.  In the mid-1990s, the median day of death for ELBW non-survivors was day of life (DOL) 3, implying that half of all doomed NICU babies would die in the first three days.  This phenomenon has morphed a bit, though unimportantly from the ethical perspective.  Overall, the median DOL for NICU non-survivors is a bit longer now (7 to 10 days, depending on the study population).  However, for the most ethically relevant population, extremely premature infants at the threshold of viability, the median day of death remains roughly 3 days.

            This observation carries two interesting epidemiologic consequences, each speaking to a commonly held ethical misperception.  If half of the smallest NICU non-survivors die in the first three days, then the likelihood of survival for infants on DOL 4 is radically different from that on DOL 1, even for those babies at the margin of viability.  Indeed, at all birth weights, the likelihood of survival for babies who survive for even a few days increases dramatically.   This knowledge informs the apparently irrational behavior of physicians and parents who choose NICU intervention for the most extremely premature infants despite the low likelihood of survival.  We can tell parents that if they 'hold their breath' for a few days (3 days, 5 days, at most a week), their baby will either be dead, or have a greatly improved chance of going home.

            As a cross-cultural aside, this phenomenon is exactly the reverse in adult ICUs.  In the NICU, every day you don't die, you are more and more likely to survive to go home.  In the adult ICU, every day you don't get discharged reduces the chance that you will ever get out of the hospital alive  

            One final distributive justice point derives from the phenomenon of early-death for doomed infants.  Consider a population of premies with birth weight > 1 kg.  Since the vast majority of these infants will survive to be discharged, it is no surprise that the vast majority of NICU $$ expended on this population are devoted to infants who will survive to discharge, as opposed to those who will die in the the NICU.  Consider, now, infants whose likelihood of survival is quite low (say 500 - 600 g birth weight, overall survival roughly 1 in 3).  Of every 100 such infants, 66 will die, but half will be dead in 3 days, and the overall average length of stay is, not surprisingly, quite low (roughly 10 days).  Consequently, the population of 66 non-survivors will occupy roughly 66  x 10 or 660 NICU bed-days in total.  The 33 NICU survivors, on the other hand, will remain in the NICU at least 100 days on average), accounting for 3300 NICU bed days.  Consequently, even for a population whose overall survival is 33%, 3300/3900 (85%) of NICU bed-days (a very accurate proxy for overall NICU expenditures) will be devoted to survivors.  This, then, can be considered the third "law" of NICU ethics -- NICU $$ are remarkably well targeted to survivors as opposed to non-survivors, independent of the absolute risk of death. 

            Again, for cross-cultural comparison, there is no identifiable population in the adult ICU where even 50% of ICU $$ are devoted to patients who will survive to be discharged, as opposed to those who will die in the hospital.  For adults over 85 years (whose ICU mortality is roughly comparable to 600 gram-ers), approximately 85% of MICU bed-days/$$ are devoted to patients who will die in hospital during that admission.

Personal Justice in the NICU

            I move now to the issues of personal justice in the NICU -- ethical considerations of babies taken one at a time.  The first question to take up is the most pressing -- "is my baby going to die in the NICU?" (I will consider the issue of survival with brain damage later).  This question is central to any assessment of the ethics of informed consent for parents.  If it turned out that caretakers were clueless about which infants were doomed and which would survive to be discharged, then any "counseling" that might occur during the NICU stay would seem, at first glance, to lack validity, and 'informed consent' would morph from a central ethical tenet to a cruel joke.

            So, how good are we, doctors and nurses, at predicting who will live and who will die?  And, more to the ethical point, how good are we at these predictions for individual babies, considered one-at-a-time, WHILE THE DECISION IS ETHICALLY RELEVANT -- that is, while the baby is still on the ventilator, when there is an ethical alternative (withdrawal) that might be entertained.  It is easily shown that babies who have an abnormal neurologic exam at the time of their NICU discharge are highly likely to continue to have an abnormal neurologic exam as they grow older.  It is just as easily shown that this observation is ethically useless for the parents and caretakers of these infants -- it is simply too late to do anything but wait, and pray.   As I tell my epidemiology students, the positive predictive value (PPV) of an abnormal neuro exam at the time of NICU discharge for the outcome of an abnormal neuro exam at the time of discharge is 1.0 -- and its worth to the family is 1 - PPV!

            Two modalities come to mind when addressing prognostication for infants while they are still on a ventilator in the NICU -- algorithms and intuitions.  I'll take up algorithms first.  Sick babies can be "scored" -- that is, given points for derangements in physiologic stability.  The higher the score, the more deranged the physiology, the more likely the baby is to die.  This algorithmic approach (SNAP, SNAPII, SNAPPEII has been perfected by Richardson, Lee, and colleagues.  Infants with higher scores on admission to a NICU are, as a group, more likely to die than infants with lower SNAPs on day 1.  So far, so good.

            But we've already seen that admission to a NICU works a bit like a 'trial of therapy'.  That is, if parents just 'hold their breath' for three days or so, the likelihood of survival for a population of infants increases dramatically.  How does this impact the prognostic power of algorithms for individual babies?  That is, do the SNAP scores of survivors and non-survivors diverge over time, so that survivors 'declare' their ultimate discharge, while doomed infants 'declare' their impending demise?  Reasonable as that might seem, it doesn't work that way.  Indeed, the SNAP scores of non-surviving and surviving populations converge, not diverge, over time.  With every passing day, illness severity is LESS likely, not more likely, to distinguish babies who are doomed to die in the NICU from those who will go home to their families.

            How can this possibly be?  Two phenomena are at work here.  First, we NICU docs are good at what we do, and what we do is fix physiologic derangements.  So low blood pressure gets raised, and metabolic acid gets neutralized, and electrolytes get balanced, and SNAP scores go down, even for sick kids (especially for sick kids, whose scores were particularly high to begin with).  Second, the sickest kids, with the highest SNAP scores, die soonest (more than half in 72 hours). So they don't "count" in the SNAP score distribution of the DOL 4 population.  This is no statistical trick, but rather a NICU reality.  The sickest kids, who die the quickest, present few ethical dilemmas -- they get admitted, we try to save them, we fail, they die (quickly and cheaply).  It's the doomed infants who linger who present the ethical dilemmas.  Unfortunately, serial SNAP scores do not distinguish these doomed infants from their ultimately surviving confreres.  Doomed infants 'cloak' themselves with every passing day.

            What about intuitions?  Perhaps serial algorithms are too 'cold', too 'unfeeling' to identify doomed babies one-at-a-time.  After all, they weren't invented for that purpose (they were invented to convince administrators that Level III NICU docs weren't worse than Level II NICU docs, just because they had more deaths in their units -- the Level III babies were sicker, duh!).  Perhaps intuitions of devoted caretakers, watching the babies over hours and days, would allow us to distinguish survivors from non-survivors.

            We have performed the same simple 'experiment' daily over the past decade in the NICU at the University of Chicago.  We begin by identifying ventilated babies (the ethics of care for doomed non-ventilated infants [usually children with lethal congenital anomalies] are quite distinct, and I will not consider them further here).  On every day that the baby is on mechanical ventilation, we ask doctors (attendings, fellows, residents) and nurses (RNs, NNPs) whether they think the baby is going to survive to discharge or die in the NICU.  Then we correlate the intuitions with outcomes.  

            We have over 1500 ventilated infants enrolled in this ongoing study. And the findings continue to surprise us.  First, the vast majority of NICU babies, even ventilated NICU babies, have homogenous, and correct, prognostication patterns.  That is, every caretaker, on every NICU day, knows what's going to happen.  About 70% of the time, everyone knows the baby is going to live -- and they do.  About 5% of the time everyone knows the baby is going to die -- and they do.  So the entire phenomenon of ethical dilemmas in the NICU turns out to be restricted to at most 1/4 of the ventilated patients.  What can we say about these?

            Approximately 25% of NICU infants are predicted by at least one caretaker, on at least one vent day, to "die in the NICU".  Imagine counseling the parents of these babies?  "M'am, I'm Dr. Meadow, and I think your baby is very sick; indeed, I think your baby is going to die in the NICU, despite our best efforts."  "Well, doctor", you can imagine the parent responding, "how good are you when you make these predictions of death"? 

            It turns out that I, the other NICU attendings, the fellows, residents, nurses, and NNPs, are not very good at all at predicting death in the NICU.  Of every 100 ventilated ELBW infants predicted to die, nearly 40% will survive to discharge.  Now imagine my response to that parent.   "Well, m'am, you should know that even though I think your baby is going to die, I'm wrong almost half the time".  "Thank you, doctor"!

            More stringent prognostic criteria improve the power of intuitions of death only slightly.  Infants with corroborated predictions of "die" -- that is, more than one person predicting death, still survive roughly 25% of the time.  Even infants with unanimous predictions of die -- m'am, everyone on the team thinks your baby is going to die -- still survive almost 20% of the time. 

            In sum, and surprisingly, we cannot predict impending death in the NICU -- at least, not in any ethically relevant fashion.

            But perhaps death isn't the most ethically relevant outcome.  Perhaps what parents really fear is a 'burdensome outcome', defined as "either death or survival with significant neurologic morbidity".  How good are we caretakers at predicting that?

            Let's start with important epidemiologic outcome data, restricted to ventilated ELBW babies.  Of all ELBW infants, roughly two-thirds will survive.  Of the survivors, at least a third will have significant permanent neurologic damage (mental retardation and cerebral palsy are by far the most common).  So, two-thirds of two-thirds, or a little less than one half of these babies will be alive and neurologically normal at two years.  The ethically relevant question before us, at the level of "personal justice", is how good are we at predicting which half any individual baby will belong to?

            Algorithms don't work here -- they just don't.  SNAP scores comparing babies with a 'burdensome' outcome to those who will survive 'unscathed' don't start out all that different, and converge rapidly to become indistinguishable.  Whatever physiologic derangements portend, the ones we measure don't help us counsel families about impending death or future brain damage.

            Amazingly, intuitions may work -- particularly predictions of "die before discharge".  Our data thus far suggest that of every 100 ventilated ELBW babies with a prediction of "die", even for a single NICU day, 94 will NOT be alive and normal at two years of age.  Phrased in other words, the likelihood of an infant with a prediction of "die" while in the NICU being alive and normal at two years of age is less than 6 in 100.  Now that's a number that might be relevant when counseling a parent.  But these data remain preliminary.

I make one final request.  Epidemiologic data, such as those presented here for our NICU, can and should be developed to inform conversations about ethical decision-making in all NICUs.  Choose another algorithm, choose another method if you wish -- but gather the data (preferably while the babies remain ventilated and ethical alternatives to continued ventilation are available).  These numbers would provide parents with some ability to assess the appropriate weight to give to physicians' estimations of the likely outcomes for their babies.   The fact that we don't have good data describing the accuracy (or inaccuracy) of our prognostic assessments for individual ventilated NICU patients is a much more serious ethical failing than the fact that our conceptions of autonomy, beneficence, and paternalism remain incomplete.