Ethics at the Limit of Viability: A Premie’s Progress

Abstract

After completing this article, readers should be able to: Progress in neonatology is generally portrayed as inexorable—doing better and better with smaller and smaller. For approximately the first 30 years of the specialty, this was true. A succession of manuscripts published between 1960 and 1990 bore witness to the success, with titles such as “1,500 g: How Small is Too Small?” that subsequently were followed by: 1,000 g:?, 800 g:?, and 500 g:?.By 1990, virtually all neonatal intensive care units (NICUs) had survival rates of 90% or greater for infants whose birthweights (BWs) were greater than 1,000 g. Consequently, for individual infants whose BWs were greater than 1 kg, parental refusal of intervention was precluded in the absence of other, nonBW-related circumstances. At the other end of a relatively narrow BW spectrum (approximately <450 g), survival was dismal. At a minimum, parental requests for nonresuscitation of infants who weighed less than this limit seemed supportable, under the broad rubric of futility. Thus, the ethical debate surrounding NICU care was played out along a birthweight dimension of roughly 1 lb.These epidemiologic truths were recognized by the early 1990s. However, much has changed in NICU care in the past decade. Exogenous surfactants are administered uniformly for respiratory distress. High-frequency oscillation and inhaled nitric oxide are widely available. Antenatal corticosteroids have become standard therapy for women in whom preterm delivery is threatened.In this brief article, we consider how these medical advances affected both the epidemiology and ethics of life and death for extremely low-birthweight (ELBW) infants in the NICU during the past 10 years. In parallel with Newton’s three laws for dealing with the motions of small bodies, we will develop three laws for dealing with the ethics of small bodies. However, by the end of the article, we are forced to recognize the uncertain application of Newtonian mechanics to the smallest of our bodies. Consequently, each law is modified by its own caveat.Three groups of infants die in the NICU. One group is born with congenital anomalies incompatible with life. Most of these babies receive comfort care for their entire short lives; few present ethical dilemmas. A second subset of doomed infants is born at term, but is extremely ill. These infants usually have some combination of pulmonary insult (meconium aspiration/persistent pulmonary hypertension) and hemodynamic collapse (septic shock). They receive “full support” for most of their NICU existence, and ethical controversies are rare.The third subset of doomed NICU infants is born preterm. As a group, preterm babies account for about 50% of all NICU deaths. Further restricting the BW/gestational age category to the “limits of viability” (broadly interpreted), infants born at less than 700 g comprise approximately 5% of all NICU admissions, but account for almost one third of all NICU deaths. Most NICU ethical controversy surrounds the lives and deaths of these tiny babies.Figure 1 displays the risk of dying as a function of BW for 1,142 infants whose BWs were less than 1,000 g when born at the University of Chicago over the past 10 years. At all times, smaller babies were less likely to survive than larger ones, which is expected. Moreover, the largest improvement in BW-specific mortality was seen for the smallest infants, who were at the greatest risk to die. Infants whose BWs were 800 to 1,000 g were likely to survive during the entire decade, and improvement was hardly possible. In contrast, infants whose BWs were 450 to 700 g were unlikely to survive in 1990, but much more likely to live by 2001.However, embedded in the figure is another phenomenon. The improvement in BW-specific mortality may be slowing or stopping. Figure 2 recasts the same data set, displaying survival for all 1,142 ELBW infants as a function of birth year. The curve appears to be flattening, or even flat, after 1997. Although this might be considered a statistical aberration, comparable observations were reported for a much larger cohort by Horbar and colleagues from the Vermont-Oxford network. Despite (or because of) the clinical advances of the past decade, success with this group of infants born at the threshold of viability appears to have reached a plateau. Time will tell whether this is a temporary pause or a more permanent barrier.Observations thus far have referred only to data obtained on the day of birth. However, time represents another ethically relevant dimension. Is there an accentuation or decay of BW-specific mortality with time? Do such changes make a difference?Figure 3 plots the day of death (DOD) for ELBW infants who died in the NICU at the University of Chicago between 1991 and 2001. The median DOD for the decade was 3 days. However, the median DOD has risen steadily since 1995 from 2 days to almost 10 days after birth.Why does it matter when doomed babies die? This question can be answered on two distinct levels, one of which might be considered to reflect justice on a personal level and the other reflecting a more distributive justice. From the personal perspective, the longer a baby lives before dying, the greater the potential for suffering by both the infant and the parents. In this context, our observations should be comforting. Although not as brief as in the past, dying still is not protracted for most doomed NICU babies. The tiniest doomed babies are unlikely to survive immediately after birth, receive what might be called a brief “trial of NICU therapy,” and succumb within days (as opposed to weeks or even months).The phenomenon of early death for NICU nonsurvivors has another, unexpected consequence (Fig. 4). The likelihood of survival to discharge as a function of an infant’s birthweight with the added dimension of time shows that by day of life (DOL) 4, overall survival for the entire group of ELBW infants was very high (>80%). Further, survival to discharge for this population of infants did not depend significantly on BW. Once an infant had survived to DOL 4, the likelihood of subsequent survival to discharge was greater than 50%, even if the BW was less than 600 g. Accordingly, it appears that the first NICU law—BW-specific mortality—becomes less ethically relevant with every passing day.These data also can be analyzed on the level of “distributive justice,” a term applied to the claim of society at large on the resources used by its members considered individually. Does NICU care for preterm infants cost too much or use scarce resources inefficiently? As opposed to what?If most doomed extremely preterm infants die in the first few days after birth, they do not use up many health care dollars. Moreover, if most extremely preterm survivors have a long NICU stay before they are discharged, NICU dollars are disproportionally diverted to survivors (which, we would argue, generally is good).Figure 5 presents the percentage of NICU bed-days devoted to nonsurviving infants at the University of Chicago for 1,142 ELBW infants admitted between 1991 to 2001. For every year, the percentage of bed-days was less than 10%. Even when the BW spectrum is restricted to infants whose BWs were less than 600 g (whose overall mortality was 74%), the percentage of bed-days occupied by nonsurvivors was only 13%.This represents the third important ethical implication to the phenomenon of early death for NICU nonsurvivors. NICU bed-days and NICU dollars are very precisely targeted to survivors independent of whether the likelihood of overall survival is high or low.As a final point, these data stand in sharp contrast to comparable analyses performed on patients admitted to an adult medical intensive care unit (MICU). At the University of Chicago, 29% of overall beds in the MICU were devoted to patients who would not survive to discharge. For adults who required mechanical ventilation, more than 50% of bed-days were devoted to nonsurvivors. Translated into economic terms, roughly 95 cents of every NICU dollar is spent on infants who will go home to their families compared with more than 50 cents of every adult ICU dollar spent on patients who will not leave the hospital alive.Moving from population-based inquiry to analysis of the courses of individual patients involves not only accurately diagnosing a particular syndrome (or accurately determining population statistics for a group of infants at a particular BW/gestational age). Instead, it involves trying to determine which individuals among a larger group that has a similar diagnosis (or BW/gestational age) are the most likely to die.Scores of illness severity have been created in an attempt to predict the likelihood of survival or nonsurvival for patients at the time of admission to an ICU. Each of these scores represents an algorithmic assessment of physiologic stability (eg, heart rate, blood pressure, base excess) based on the assumption that patients who have the most deranged physiology are least likely to survive. Generally, such scores work for large groups. However, the scores have proven less valuable for individual patients. Even at the highest end of APACHE, PRISM, or SNAP scores, the predictive power of nonsurvival rarely is greater than 50%.However, one dimension of ICU care is consistently omitted from these illness severity scores: time. ICU care usually lasts longer than a single day, and decisions (both medical and ethical) should be revisited as time passes and patient responses or nonresponses to therapeutic interventions are observed.Does the predictive power of algorithmic assessments of illness severity improve or worsen with time? One might hypothesize that as time passes, survivors and doomed patients should diverge along the dimension of illness severity. With this view, survivors’ illness severity scores should decrease as they become more physiologically stable, and nonsurvivors’ scores should increase as they become less stable (blood pressure falls, oxygen requirements increase, kidneys shut down, patients lapse into coma). Alternatively, one might hypothesize that many physiologic disturbances leading to NICU admission are at least transiently correctable. If this were true, illness severity scores for survivors and nonsurvivors might converge after NICU admission, leaving the ultimate fate of the infant less clear as time passes.Figure 6 presents the average Score for Neonatal Acute Physiology (SNAP) scores as a function of DOL for populations of intubated survivors and nonsurvivors cared for in our NICU. Two important points emerge. On DOL 1, SNAP distinguished the population of future survivors from nonsurvivors. SNAP values for nonsurvivors (24±8.7 [SD]) were significantly higher than values for survivors (13±6.1; P<.001). However, this difference diminished steadily over time. By DOL 10, there no longer was a significant difference between SNAP values for infants who would survive to be discharged and those who would subsequently die.Analyzing infants individually, SNAP scores of all survivors improved over time, with values lower on the day they were extubated (and consequently exited the study) than on the day of their admission. A more surprising finding was that SNAP scores for 85% of the nonsurvivors also improved over time, with final SNAP scores on the day they died being lower than scores on DOL 1. Moreover, only 9 of 45 nonsurvivors had a “U-shaped” course in which SNAP initially fell, then rose again prior to death.Despite its intuitive appeal, the hypothesis that survivors would improve, nonsurvivors would not, and doomed infants would become increasing apparent was not proven. It is true that most patients who die do so early and are sick their entire (short) lives. However, these findings suggest one of the most counterintuitive observations of NICU dynamics:Most NICU infants who die late are NOT sick their whole lives. Their prospects for survival become progressively less –not more –apparent with each passing day.Thus, illness severity scores are not helpful to ethicists. Perhaps this is due to physiologic algorithms being too “cold,” impersonal, and removed from the “feel” of the bedside to be the best vehicle for predicting life and death in the NICU. Perhaps experienced caretakers, calling on inchoate but nonetheless real bedside intuitions, can do better.Clinical intuitions have been shown to be predictive of outcomes for ICU patients of all ages. Most people who work in ICUs and people who have known patients there have experienced caretakers’ prognostications: “I think your loved one will live” or “Things look pretty bleak.” In the context of NICU infants, how accurate are these estimates at the time of admission and do they become more or less accurate over time? As before, we might hypothesize that as the infant’s response or nonresponse to NICU therapy becomes clear, the clinical course of survivors and doomed patients would diverge. If so, NICU patients would “declare” themselves ever more stridently over the course of their hospital stays.To test this hypothesis, we asked doctors and nurses caring for 254 ventilated infants in our NICU two questions on every day of mechanical ventilation: “Do you think this baby will die in the NICU or survive to discharge?” and “If you think the baby will survive, will there be none, mild, moderate, or severe permanent neurologic damage?”The dominant finding of this analysis is portrayed in Figure 7. A total of 192 of 254 infants (79%) ventilated NICU babies survived, and almost all surviving infants were predicted by all (or almost all) observers to survive on all (or almost all) days of mechanical ventilation. At least 90% of all NICU ventilation days for survivors were characterized by virtually unanimous prediction of survival. In contrast, 15% of survivors survived despite at least one caretaker predicting their death. Indeed, 4% of all survivors had at least one hospital day in which all caretakers predicted death.Prediction profiles for nonsurvivors split both along the dimension of time and the dimension of accuracy. Sixty percent of nonsurvivors in the study died before DOL 10. As a group, their prediction profiles were remarkably homogeneous—dismal and accurate. On every day, every observer predicted that the baby would die (Fig. 7B).In contrast to the homogeneity characterizing profiles of infants who died early, those who died later were a heterogeneous group. Approximately 75% of late-dying infants were predicted to live by many (if not all) observers on many (if not all) hospital days. Fifty percent of the infants suffered a fatal medical catastrophe with little advance warning (eg, necrotizing enterocolitis, pneumonia). The rapid and unexpected nature of their demise is emphasized by the observation that for many of these late-dying infants, not even 1 day of their hospital stay prior to the actual day of death was marked by 0% prediction of survival. Other late-dying infants had prediction profiles categorized by great uncertainty both across respondents and across days and even weeks. Several periods were characterized by “pessimism” (ie, low, even 0% predictions of survival), alternating with periods of “optimism” that were characterized at times by up to 100% prediction of survival (Fig. 7C). These nonsurvivors often lingered for many weeks prior to death.Although most nonsurvivors had at least some predictions of death before they died, so did some survivors (there were many more survivors than nonsurvivors). Consequently, although approximately one third of all the ventilated infants in the study had at least 1 day characterized by a prediction of death, almost 50% of these were wrong. Thus, almost 50% of the children predicted to die survived to be discharged. More stringent criteria for prediction of nonsurvival improved predictive power only slightly; 15% of ventilated patients had at least 1 day characterized by unanimous predictions of death, yet 25% of them survived.Perhaps even more depressing is the deterioration of positive predictive value (PPV) for intuitions of nonsurvival as a function of the day of life on which the predictions were obtained. During the first 10 days of life, a single day of unanimous intuitions of nonsurvival retained impressive predictive power, with 80% of patients who had at least 1 day of unanimous prediction of death actually dying. However, after DOL 10, the PPV of 1 day of unanimous prediction of death decreased sharply, and beyond DOL 21, the PPV of 1 day of unanimous intuitions of nonsurvival decreased to less than 50%.Perhaps our caretakers are sensing something, but are not sufficiently sensitive about what they are sensing. Perhaps our respondents are predicting “die” when, in fact, they mean “bad” that encompasses some combination of “die or if they survive, they will not be close to neurologically normal.” That might be a more relevant ethical outcome for many parents and caretakers; informed consent should inform the parents not only about life or death, but offer some sense of the quality of that life.Although caretaker intuitions for ventilated NICU infants are so imperfect as predictors of a baby’s death as to be virtually useless, changing the outcome variable only slightly may make a substantial difference. As an example, the question may be asked of how well these same intuitions predict either death or significant neurologic morbidity after discharge.We have obtained follow-up data at 1 year of corrected age for 120 ELBW infants who were admitted to our NICU (760 g BW and 26 weeks’ gestation at birth). Only 44 (37%) were alive and neurologically normal at 1 year corrected age, which is comparable to other published data. We ask the most important question we can envision from the perspective of personal ethics: How good were we at predicting long-term outcome (good or bad) for individual patients within this overall group while we were still caring for them in the NICU? For any child who was sufficiently sick to require mechanical ventilation, but for whom no caretaker ever predicted “die before discharge,” the likelihood of surviving and having normal neurologic examination results at 1 year was 70%. Consistently favorable intuitions doubled the likelihood that the babies would be normal. In contrast, if even one infant day was characterized by unanimous predictions of nonsurvival, the likelihood of being alive and normal at 1 year was only 4%.Recent epidemiologic observations suggest the need to revise our three laws of NICU ethics: Epidemiology appears to have resolved two issues (more or less). First, distributive justice concerns in the NICU are a canard. Any people who thought they were important for any area of medicine should close down adult ICUs. Second, if accurate life-or-death prediction for individual infants is the goal, we have yet to find the right dousing stick, and we have used up many of the more obvious ones. We may have found an excellent dousing stick for “life without significant neurologic impairment,” but only time will tell.There also are several unresolved issues that may never be resolved. For example, how dismal must the likelihood of “survival without severe neurologic morbidity” be to allow parents to elect not to provide or continue life-prolonging intervention? No matter how large the data-set on which inferences are based, inevitably there is some statistical uncertainty about the point estimate determined for the outcome of any predefined group of infants. More importantly, though, is how low a number should “count” in this discussion. At the extreme, no infant ever has been reported to survive at 19 weeks’ gestation, which places this time point beyond the penumbra of parental discretion for now. However, there are scattered reports of survival at 350 to 400 g and 21 to 22 weeks’ gestation, and by 400 to 450 g and 23 weeks’ survival, the number is countable (although substantially less than 5%). This “penumbra of ethical discretion” issue is literally a “baby and bath water” problem. Some (hopefully few) potentially “good” babies will be “thrown out” in return for allowing parents to forestall the possibility that most (but not all) doomed/damaged infants will be required to suffer extensively for no ultimate purpose.Another unanswered question is: What type of morbidity is determined as sufficiently worthless? Adolescent NICU graduates who clearly are handicapped by most “normal” standards self-report a score on their quality of life that is significantly higher than the self-reported quality of life for “normal,” nonhandicapped teens. Certainly, many NICU graduates are so impaired that they cannot self-report anything. On the other hand, perhaps cognitive dissonance is real and these children (most kids? these families? most families?) are much more flexible than we who are not required to adapt to hardship can anticipate. That would be very good from the child’s point of view, but very difficult for an ethicist to impose on a parent who wanted to incorporate apparently dismal quality of life into the benefit-burden calculus of NICU intervention.Ethicists (and judges) always have the easy part because they simply create internally consistent rules. Physicians and parents have it harder because the real world is always messier.