Ethics, Data, and Policy in Newborn Intensive Care

Background

Ethics in the neonatal intensive care unit (NICU), as in all clinical contexts, starts with the traditional triangular framework of principles. Autonomy means to do what the patient, or in this case the parent, thinks is right; paternalism means to do what the doctor thinks is best; beneficence and nonmaleficence mean to do what is most beneficial and least harmful. These principles are widely accepted as the basis of ethical medical practice. Their applications occur daily. The problem with using these widely accepted principles to guide practice is they sometimes conflict with one another. Then one must decide quickly which compromises are best. What is the right thing? What facts should be brought to bear in the decision? What weight should be given to each fact? Whose opinion should count?

Many decisions in the NICU evoke and reflect these principles. Sometimes, neonatologists make relatively straightforward decisions not to resuscitate babies born at 21 weeks, or to provide obligatory support to babies born at 28 weeks, or to not perform cardiopulmonary resuscitation on infants with lethal anomalies. In more difficult cases, when ethical principles are in conflict, their application to a particular case is not straightforward. Faced with a difficult case, it is rare that simply applying principles will help to devise a solution.

Applying ethical principles requires attention to the specific details of each case. We must ground our concerns in context, take data into account, and be sympathetic to patient preferences when the balance of benefits and burdens is not clear. In the NICU, health professionals are constantly and anxiously aware that the burdens of treatment are real, immediate, and significant, whereas the benefits of NICU interventions are distant, statistical, and unpredictable. Babies must undergo months of painful procedures such as intubation, ventilation, and intravenous catheterization in the hope that everything will turn out well in the end. However, sometimes outcomes are ambiguous. Babies are left with lifelong sequelae. How should we decide whether we did the right thing? How should we decide whether, in similar circumstances, we should again do the same thing?

To make good decisions, we need good data. We need to know (1) where to get data; (2) how to decide which data matters; (3) the implications of limited data; and (4) the implications of increasingly available data. These questions are the focus of this chapter.

Getting Data

Neonatologists have a number of repositories through which data are collected and analyzed. The Vermont Oxford Network (VON) and the National Institute of Child Health and Human Development (NICHD) Neonatal Research Network Generic Database use prespecified data elements to describe outcomes of preterm infants. Similar databases exist to describe outcomes of other neonatal populations. These include the Extracorporeal Life Support Organization database and the Congenital Diaphragmatic Hernia Study Group. For all these biorepositories, data are gathered using trained abstractors and ongoing quality assurance regarding data collection. Each has discrete variables related to the population of interest. Secondary analysis of electronic health record (EHR) data is a more feasible option than previous manual chart review, following meaningful use incentives for implementation of EHRs. Administrative data from billing claims are another commonly used source of outcomes information; examples include the Pediatric Health Information System from children’s hospitals, the National Inpatient Sample and its associated Kids Inpatient Database from the Agency for Healthcare Research and Quality, or the Kaiser Permanente Neonatal Minimum Data Set. From the standpoint of understanding outcomes to inform decision-making for sick infants, each type of data has benefits and drawbacks. Benefits of abstracted data include ongoing quality assurance for variables related to the disease of interest. Limitations include differences in definitions between data sources, and the way that individual data are aggregated into discrete variables. EHR data have the benefit of extreme detail and potential flexibility in defining measures of interest, but they are limited by the accuracy of the recording clinician. Claims data have the benefit of understanding costs of care but are limited by the accuracy of coding and discrepancies between clinical diagnosis and coding terminology for common diseases.

These multiple potential sources of information offer opportunities for those who want to better understand the determinants of outcomes, but the plethora of information is a mixed blessing. Databases can be local, regional, national, or international. They can gather data on all babies or only on specific subpopulations of babies. These variations lead to different outputs. Caregivers must then decide which database to use and how to interpret and apply the results.

Getting Data that Matters

What kind of information would parents, physicians, or judges need to decide when treatment should be obligatory and when it should be optional?

One essential truth at the intersection of NICU epidemiology and ethics is that gestational age—particularly at the margins of viability—is a powerful predictor of survival. In the United States, as in virtually all the industrialized world, infants born after 24 weeks’ gestation have survival rates greater than 80%. This is high enough so that treatment is generally considered obligatory. For these infants, the ethical principle of best interest requires their resuscitation, in the same way that sick children born at term deserve resuscitation. Conversely, for infants born before 22 weeks’ gestation, survival is essentially zero. Consequently, these infants and their parents deserve our compassion but not necessarily our interventions, on the ethical grounds of strict futility.

In between, spanning approximately 3 weeks of gestational age, from 22 to 24 weeks, we require not only data but also its interpretation. Tyson et al., using the NICHD Neonatal Research Network, attempted to quantify additional risk factors for both mortality and neurologic morbidity among infants born on the cusp of viability. Their analysis revealed that, for babies at the same gestational age, singleton status, birth weight, antenatal steroids, and female gender all improve the likelihood of survival and intact neurodevelopmental outcome. By considering these factors—which are available at the time of birth—caregivers can more accurately estimate the chances that a baby will survive or that those survivors will have neurodevelopmental impairment. This predictive model has been turned into a publicly available and widely used calculator, which is used in many locations to counsel families at risk of delivering a baby in this gestational age range. The model has been updated recently to include more modern neonatal and obstetric outcomes and to provide a range to account for the wide center variability in outcomes.

However, two problems remain. First, for many infants, the algorithm’s predictive value is still not very good—some lower-risk patients will die, and some of the highest-risk patients will survive. Second, by only using data available at the time of birth, the algorithm ignores a potentially important feature of NICU care—time. It does not account for prognostic features that might become available as the infant’s course unfolds in the NICU. This is an important limitation that has ethical implications.

There are distinct advantages to making decisions over the first few days of an infant’s NICU stay rather than in the delivery room at the time of birth. The first is emotional. Parents often appreciate the opportunity to get to know their baby as an individual, as opposed to making decisions based only on anonymous population-based prognostications that are available at the time of birth. Second, there is valuable information to learn while the baby is in the NICU. Several time-sensitive prognostic features have been evaluated for infants born at the border of viability. These include illness severity scores such as the Score for Neonatal Acute Physiology (SNAP), cranial ultrasound results in the first week or two of life, and caregiver intuitions that the patient would “die before NICU discharge.” Unfortunately, although illness severity scores on the first day of life have good prognostic power for death or survival, their power diminishes over time. Intriguingly, serial intuitions that an individual baby will die before discharge—offered by medical caregivers for infants who require mechanical ventilation and for whom there is an ethical alternative to continued ventilation, namely extubation and palliative care—are remarkably accurate in predicting a combined outcome of either death or survival with neurodevelopmental impairment (mental developmental index [MDI] or psychomotor developmental index [PDI] <70). Babies with abnormal results from a cranial ultrasound examination whose doctors agree with one another that the babies are likely to die have a less than 5% chance of surviving with both MDI and PDI greater than 70 at 2 years, independent of their gestational age. The predictive power of these data, acquired over time during an individual infant’s NICU course, although not perfect, is greater than any algorithm available at the time of birth.

What do prospective parents consider when asked to decide whether to resuscitate their micropreemie? They may not want the precise prognostic estimates that we try to offer. For many parents, the death of their baby in the NICU is not necessarily the worst outcome. Instead, for many parents, it may be emotionally more difficult to not try to save their baby than it is to try and fail. The decision not to try may leave parents with a lifetime of self-doubt about whether, had they only tried, their baby might have survived. The old adage, “It is better to try and fail than not to try at all” seems to summarize these attitudes.

If trying and failing is seen as a more acceptable option than not trying at all, that could help us figure out which outcome statistic is most relevant to parents. Many studies based on database information, including that of the NICHD, try to predict how many babies either die or survive with severe neurodevelopmental impairment. This approach implicitly considers both severe disability and death to be equivalent and equally bad outcomes. The implicit assumption is that, if there is a high likelihood of death, then we should not attempt resuscitation or provide life-sustaining interventions. By contrast, if trying and failing is better than not trying at all, then perhaps we should combine different variables—reporting the percentage of survivors who are neurologically intact while leaving the babies who die out of the denominator. Understanding outcomes as a fraction of infants who survive to discharge is often obscured in research reports of preterm infant outcomes, due to epidemiologic concerns about “competing outcomes.” Because an infant who does not survive to NICU discharge cannot have developmental delays at age 2, these two very different outcomes—death or severe disability—are reported together. This is problematic for two reasons. First, while combining competing outcomes makes sense from a statistical perspective, from a family or clinician’s perspective, death in the NICU and delays at age 2 are very different scenarios. For the parent who prefers that the NICU team attempt resuscitation, these combined outcomes do not offer a sense of what happens if the baby does survive. Second, presenting NICU outcomes data as combined outcomes obscures an important fact of neurologic morbidity in extremely preterm infants: the incidence of neurologic morbidity in NICU survivors is not very different when comparing infants at 23, 24, 25, and 26 weeks’ gestation. The essential epidemiologic difference for infants born in this gestational range appears to be whether the baby will survive at all. Once the baby leaves the NICU, the risk of severe morbidity is largely the same. This is true in single-center and multicenter studies, in the United Kingdom, Canada, Europe, and the United States.

It has been traditionally, and inaccurately, perceived that mortality follows morbidity, meaning that because infants born at 22 to 24 weeks’ gestation have the highest rate of mortality, survivors have the highest rates of neurologic morbidity. This belief is supported by the lack of parsing out outcomes of mortality versus morbidity among survivors. It could contribute to an inaccurately pessimistic view of outcomes. A fascinating insight has been offered by Janvier et al., who have done extensive surveys comparing responses to requests for resuscitation of sick micro-preemies with resuscitation of comparably sick patients at other ages (from term infants to 80-year-olds). Consistently, it appears that micro-preemies are devalued. For comparable likelihood of survival and comparable likelihood of neurologic morbidity in survivors, more respondents would not offer resuscitation or would allow a micro-preemie to die.

Finally, there is difficulty in assigning value to morbidity in surviving infants in the NICU. Traditional outcomes studies for survivors of NICU care are generally limited to Bayley scores at 18 to 26 months of age. This is problematic for multiple reasons. First, NICU success is often viewed as “all or none.” In most neonatal follow-up literature, a Bayley Mental Developmental Index (MDI) or Psychomotor Developmental Index (PDI) greater than 70 is classified as normal , whereas an MDI or PDI less than 70 is classified as an adverse outcome . A continuum of scores may be better interpreted as a range of strengths and areas that require additional attention, rather than as a single measure of success or failure. Whether individual clinicians think of developmental delays along a continuum or not, presenting outcomes data as pass/fail in the literature may contribute to the overpessimistic attitudes toward survivors of preterm birth prevalent in the medical community.

Furthermore, studies of former preterm infants have shown limited correlation between Bayley scores at 2 years of age and school readiness by 5 years of age. This has been interpreted in the neonatal community as a poor reflection on the Bayley assessment itself, yet in studies of infants with congenital heart disease there is a much tighter correlation between 2-year and later outcomes. It seems more likely that the problem with assigning value to morbidity is not the Bayley assessment itself but that former preterm infants have significant capacity to continue improving their developmental progress after 2 years of age. It also may be that Bayley scores at age 2 do not yet reflect the higher-order organizational and executive functioning issues that can become problematic by school age in survivors of prematurity. Verrips et al. have attempted to assess the effects of permanent residual disability for NICU survivors and their immediate families; they have demonstrated consistently that children with disabilities and their parents place a much higher value on their lives, and the quality of those lives, than do either physicians or NICU nurses. The vast majority of infants who survive the NICU, even those with significant permanent neurologic compromise, have “lives worth living,” as judged by the people most affected by those lives.

Policy Implications of Limited Data

Despite the wealth of available data on outcomes of sick infants after the NICU, we are still unable to predict future neurologic morbidity with the amount of detail needed for many clinicians and parents to make decisions regarding intervention. This, along with changes in the recognition of disability as a social construct as well as a medical one, has led to legal challenges with important implications for clinical care.