Mortality in the neonatal intensive care setting: Do benchmarks tell the whole story?

Australian neonatal mortality data are collected and shared within collaborative networks. Individual unit outcomes are benchmarked between units and presented in quarterly or yearly reports. Low mortality is commonly interpreted as optimal performance. However, current collected data do not differentiate between death due to severe illness and death following treatment limitation. This study aims to explore the physiological condition immediately before death, and the proportion of deaths attributed to treatment limitation.

• Mortality is compared between units in Australia and used as a benchmark outcome.• Redirection of care or treatment limitation is common in the neonatal setting but is varied dependent on ethical, religious and family or clinician practices and is not universally documented.

What this paper adds
• 43% of total mortality in our unit were in a scenario of physiologically stable infants where palliative care was discussed and intensive care treatment was ceased.• Benchmarking mortality without the documentation of local treatment limiting practices is not reflective of an individual neonatal intensive care units performance.• Optimal benchmarking of mortality between neonatal units requires data on physiological stability around the time of death, and whether palliative care was offered as care option.
2][3] Australia has seen a decline in the neonatal mortality rate from 4.6 per 1000 live births in 1990 to 2.9 in 2000 and 2.2 in 2018.Similar decline in mortality has been reported in both high and low mortality countries, ranging between 9% and 67% reduction. 1,4,5Efforts have been made to compare or benchmark inter-hospital mortality using risk adjustment models; however, there are significant limits to their precision and ability to provide an accurate comparison. 6he variation in decline in mortality rates has been attributed to demographic variability, the provision of active treatment at extremely low gestations, and the willingness to offer palliative care and withdrawal of intensive care treatment. 3,7,8he approach to resuscitation and active treatment of very sick infants is a complex decision-making process considering the risk and burden of hospitalisation and predicted prognosis with potential permanent neurologic impairment.It is well documented that intraventricular haemorrhage (IVH), hypoxic ischaemic encephalopathy (HIE), and congenital anomalies are associated with adverse neurodevelopmental outcomes. 9,102][13][14][15] However, studies do not often report on details of the clinical condition at the time of withdrawal of treatment, for example in a stable infant or an infant who rapidly deteriorates and dies while receiving complete intensive care support.When withdrawal of intensive care treatment is described, rarely is the distinction reported between infants who were likely to die regardless of intensive intervention (i.e.unstable infants extubated in the presence of their families) and stable infants who were extubated due to a predicted neurodevelopmental outcome or expected poor quality of life.The absence of these details would suggest that benchmarking mortality could be a misleading concept.
The primary aim of this study was to explore our local mortality and the physiological condition in the immediate period before death, and how many cases could be attributed to withdrawal of intensive care to benchmark more accurately against other neonatal intensive care units.We hypothesised that a significant portion of mortality in our unit were infants who were physiologically stable enough to survive.A secondary aim was to describe the progression of illness towards the time of death, and whether clinicians would alter their approach towards offering withdrawal of intensive care treatment.

Methods
We conducted a single centre retrospective chart review of 100 consecutive neonatal deaths from 2015 onwards in the John Hunter Children's Hospital neonatal intensive care unit (NICU).Our unit is a tertiary referral centre that services a region with an estimated birth rate of 170 infants <32 weeks gestation per year and a total birth rate of 10 500 newborns per year.All infants born within the time period, who were ≥23 weeks gestation and admitted for more than 6 h to the NICU were included in this study.Infants were excluded when they died in the delivery suite (stillborn, unable to be resuscitated) and those transferred or discharged before death due to incomplete medical records.A convenience sample of 100 consecutive deaths was chosen and data collection was retrospectively started from the 1st of January 2015.Patient characteristics were described by measures of central tendency and measures of dispersion.Data on maternal and infant demographics, antenatal history and complications, mode and location of delivery, gestational age at birth, birth weight, time of birth and death, the relevant clinical status of the infant before death, and the cause of death as listed on the death certificate or otherwise specified by autopsy results was collected from the medical records.
Illness severity scores have been developed over time in neonatology in parallel to those of the paediatric and adult population.The CRIB-II and SNAPPE-II are scores that have been used to predict mortality when scored within the first 12 h of admission to NICU.These scores, however, do not give a clear representation of an infant's physiological status in the dying phase. 6An appropriate validated score for this purpose could not be identified hence we created an easy-to-use illness score describing common aspects of NICU care associated with abnormal physiology and mortality.Six clinical features commonly associated with physiological stability were used in the score.They included pH, oxygen saturation index (OSI = MAP Â FiO 2 Â 100/SpO 2 ), 16 use of inotropes and how many, use of inhaled nitric oxide, whether blood products were given or whether antiepileptic medications were given (Table 1).To be classified as physiologically stable, infants required a score of 4 or below out of a possible 13 points.Infants were scored at 12 h prior to death using the day charts and medication charts, and the clinical notes were explored for whether a treatment limiting discussion (TLD) had taken place.A TLD was recorded if there was evidence of a discussion with the medical team regarding withholding resuscitation, removing an endotracheal tube and respiratory support or redirection of treatment with the goal of comfort palliative care.
For our secondary aim, we explored the illness score at both 48 h and 12 h before death, and explored the clinical notes for when and how clinical progress and option for TLD had taken place and whether a change in clinical condition altered the content of treatment limitations.

Results
One hundred patients died in our unit from January 2015 until April 2020 out a total of 4207 admissions during that period.The median (IQR) gestation of the infants that died was 26 weeks (24-35), a birth weight of 940 g (680-2345), and the postnatal age of death was 8 days (2-19).The primary causes of death as described on the death certificate were sepsis, congenital anomalies, extreme prematurity/respiratory insufficiency and HIE.

108
Cause of death and physiological stability scores are presented in Table 2.
Forty-eight infants showed a score of 4 or below 12 h before death and were classified as physiologically stable.In 43 (90%) of these cases, TLD had taken place and the parents opted for withdrawal of intensive care support (Table 3).Causes of death in the 43 stable infants with withdrawal of treatment were HIE (n = 12), congenital anomalies (n = 10), IVH (n = 6), extreme prematurity (n = 7), sepsis/NEC (n = 7) and other (n = 1).In these cases, the common theme regarding redirection of treatment as documented in the medical records was the consideration of prognostication with poor neurodevelopmental impairment or significant disability.Five infants who were considered physiologically stable at 12 h before death suffered an acute event and died despite cardiopulmonary resuscitation, none of these patients had a TLD documented in their medical record.
Fifty-two infants were physiologically unstable at 12 h before death.In 35 (67%) cases TLD occurred and in 29 of these 35 patients withdrawal of treatment was opted for by the parents (Table 4).Six unstable infants died despite all intensive care efforts while being on a ventilator.Illness progression for each diagnostic category from 48 h to 12 h before death is presented in Figure 1.
Four infants received an unstable classification at 48 h and stable at 12 h.These infants showed a clinical improvement and none of these infants had a TLD documented in their medical record.Conversely, 17 infants who were physiologically stable at 48 h but showed a clinical deterioration (unstable at 12 h) had a TLD documented after this deterioration and all went on to have intensive care support withdrawn.

Discussion
In this single-centre study over a 5-year period, we found that 43% of mortality was attributed to a clinical scenario where the option for palliative care was discussed in physiologically stable infants.Had the unit taken a different approach to withdrawal of treatment and continued intensive care treatment to all infants who had withdrawal of treatment for quality-of-life reasons and they all survived, the unit mortality would have fallen from 2.4% to 1.4%.While this would represent a significant fall in neonatal mortality, it would not be considered a superior outcome since it that failed to respect and reflect parental wishes in the face of extremely challenging circumstances.What matters ethically is not necessarily to have the lowest mortality rate in neonatal unite, rather to have the lowest mortality infants where survival is desired and sought.For infants who may survive with very substantial long-term disability or burden of illness and treatment, what matters is that parents are informed and involved in assessments of the infants best interests and that, where this is consonant with the parents' wishes and infants interests, that the infant is provided with the best quality palliative and end of life care.This finding suggests that benchmarking mortality without the documentation of local treatment limiting practices is not reflective of an individual NICU's performance, but a reflection of the combined outcome of the clinician's willingness to offer palliative care and the parental decision to accept such a pathway of care.This study also showed that treatment limitation or palliative care was discussed with most but not all parents before death in our unit.
Data on neonatal mortality in Australian NICUs are collected in collaborative networks such as the NICUS, ANZNN and other international networks (Vermont Oxford network, International network for evaluation of outcomes of neonates) providing a platform aimed at improving the efficacy and efficiency of neonatal care.Various outcomes including mortality are benchmarked for comparisons between units and presented in quarterly unit reports in an effort to improve individual unit performance.Unit reports have shown marked variability in mortality between units.Despite having comparable or even lower baseline incidence of severe IVH, HIE and specific congenital abnormalities, our unit's mortality rate was higher than the median for the number of expected cases. 4While there are many possible causes for difference in mortality such as higher rates of sepsis, NEC or prematurity, an important consideration is whether mortality is attributable to treatment-limiting discussions and parental decisions for palliative care.
Common causes for mortality in infants with the lowest physiological scores were congenital abnormalities, HIE, and severe  IVH.Mortality rates secondary to these diseases have a wide range reported.][19] For infants with severe HIE, reported mortality rates varies between 9% and 30%. 20,21A recent meta-analysis found that mortality varies between 5% and 15% in infants with significant congenital abnormalities and variation was`noticeable according the anomaly. 17,22No study on IVH, HIE or congenital abnormalities reported the physiological condition around the time of death or unit-specific attitudes towards their willingness to discuss treatment limited care, limiting interpretation of the data.Previous studies reviewing mortality and treatment of the dying neonate have shown a considerable difference in the approach to treatment across the world.13][14][15]23 In Israel withdrawal of treatment has been found to be as low as 28% and in South America one large multi-centre study found that while an infant may not be resuscitated, withdrawal of treatment is not offered and that up to 54% of infants die following cardiopulmonary resuscitation. 24,25These data suggest that treatment-limiting discussions do not uniformly occur across countries reinforcing the importance of uniform documentation and data collection regarding practice in end of life care.Verhagen et al. performed a cross-cultural review of 4 neonatal units in the USA, Canada, and the Netherlands (Chicago, Milwaukee, Montreal and Groningen) and reviewed infants who died and their physiological stability at the time of death. 11The authors found a large variation between cultures.For example, in Chicago, elective extubation for quality-of-life reasons never occurred in contrast to 19%-35% in the other units.As a result, Chicago also had a higher rate of infants who died while receiving cardiopulmonary resuscitation (31%) compared to 4% to 12% in the other centres.Our data and that of Verhagen could be used to prompt neonatal collaborations to collect additional detail for the benchmark outcome of mortality.
In France, Boutillier et al.'s 2023 prospective study, reviewed the presence and outcome of life and death discussions with families regarding withdrawing or withholding life sustaining intervention. 23They found that in a single French NICU 37% of deaths occurred after withdrawal or withholding of life sustaining treatment in stable infants, similar to that found in our study.Interestingly, they followed 34 infants who survived to discharge despite withholding or withdrawing life sustaining treatments.Of these infants 10 died before the age of 2, of the remaining 24, at 2 years year of age all required ongoing subspeciality care with 9 requiring input from more than 3 specialists.Fourteen patients had moderate to severe functional limitations, and five with normal neurological development, the others with isolated delays.This outlines the complexity in prognostication, and uniform provision of end-of-life care.Our data showed that it can be difficult to use a single time point to provide an accurate prognosis, as some would improve and some would deteriorate in a short period of time.
We acknowledge that estimating the rates of treatment limitations between units can be difficult due to the variation in clinician approach to an infant with a life-limiting illness.Ethical decision-making in an infant with an end-of-life condition is challenging for all involved, especially in conditions with high rates of mortality such as extremely preterm infants, and coupled with prognostic uncertainty. 14To circumvent this difficult issue we suggest to add more detail surrounding the physiological condition shortly before death.

Limitations
There are several limitations to our data.Subjective issues such as clinician variation, decision making around end-of-life decisions and parental preferences can be difficult to appreciate from the clinical notes alone.Furthermore, the data of only one unit in the ANZNN network has been detailed, limiting external validity.
We acknowledge that the score we used to classify physiological stability and its cut point were arbitrary.However, using a cut point of 3 or 5 (instead of 4) would alter the primary outcome only slightly to 36% or 49% of mortality assigned to treatment limitation in physiologically stable infants and parental decision to withdraw intensive care treatment.Illness scores, however designed, cannot be validated against intact survival when a large portion of parents opt for withdrawal of intensive care treatment.The optimal illness score is likely best designed based on consensus of a large group of users using a Delphi process before it can be used in better benchmarking mortality.

Conclusion
Using mortality as a benchmark outcome has been integral in improving healthcare provision to neonates over time.Technological advances have allowed the most vulnerable infants to survive, but alongside this there is an increasing clinician and parental acceptance of withholding or withdrawing treatment in physiologically stable infants.Using mortality as a benchmark and performance outcome without considering the circumstances surrounding death is not reflective of an individual NICU's performance.For optimal benchmarking of mortality our study emphasises the need to consider standardised documentation across units with regards to the physiological status and circumstances around time of death as well as the well-being after discharge for those who survive.This could include recording a clinical stability score at time of death; uniform records of the presence and outcomes of TLDs; and long term follow up of those who survive despite TLDs including the impact of comorbidities and reduced functional abilities of those infants.In time, this will allow comparison of units at a broader level by including obstetric outcomes, mortality rates and neuro-disability rates with the hope to improve neonatal care delivery.

Fig. 1
Fig. 1 Median physiological stability scores.Score at 48 h and 12 h before death, classified by cause of death.

Table 1
Physiological stability score

Table 2
Cause of death and physiological stability score

Table 3
Physiological stability and treatment limitations discussions

Table 4
Outcome of treatment limitation discussion