Use of the expanded Apgar score for the assessment of intraventricular and intraparenchymal haemorrhage risk in neonates

OBJECTIVES
Preterm birth is a key factor contributing to haemorrhage incidence in neonates. This study focused on defining relevant parameters for the assessment of intraventricular and intraparenchymal haemorrhage risks in neonates.


MATERIAL AND METHODS
Chi-square automatic interaction detection was used to analyse the Apgar score (AS), the Apgar max score, and the course of resuscitation documented according to the expanded AS in 696 infants born between 2009 and 2011 in the Neonatal and Intensive Care Department of the Medical University of Warsaw.


RESULTS
Gestational age was the most relevant discriminating variable for the prediction of intraventricular III degree and intraparenchymal haemorrhage incidences. Infants born before the 31st week of pregnancy made up 80% of the intraventricular or intraparenchymal haemorrhage cases. Additionally, a fraction of inspired oxygen > 0.8 at ten minutes after birth was a better discriminating variable in the youngest neonates than an Apgar max score ≤ 5, identifying 31.6% and 20.6% of infants with intraventricular and intraparenchymal haemorrhage, respectively.


CONCLUSIONS
Consideration of the oxygen concentration supplied during resuscitation significantly improves the prognosis of intraventricular and intraparenchymal haemorrhages in preemies compared to the use of the classical AS.


Results
Gestational age was the most relevant discriminating variable for the prediction of intraventricular and intraparenchymal haemorrhage incidences. Infants born before the 31st week of pregnancy made up 80% of the intraventricular or intraparenchymal haemorrhage cases. Additionally, a fraction of inspired oxygen > 0.8 at ten minutes after birth was a better discriminating variable in the youngest neonates than an Apgar max score ≤ 5, identifying 31.6% and 20.6% of infants with intraventricular and intraparenchymal haemorrhage, respectively.

Conclusions
Consideration of the oxygen concentration supplied during resuscitation signi cantly improves the prognosis of intraventricular and intraparenchymal haemorrhages in preemies compared to the use of the classical AS.

Background
In 1952, in response to a student who asked her how to estimate the health status of a newborn immediately after birth, Virginia Apgar wrote down ve points on a piece of tissue paper. A year later, the Apgar score (AS) was made public, and two years later it was o cially published. In 2015, the American Academy of Paediatrics (AAP) and the American College of Obstetricians and Gynaecologists (ACOG) accepted the AS as a reliable and convenient measure of an infant's health status and the reaction to resuscitation, if needed [1,2].
The AS was never intended to serve as a prognostic tool, and according to the AAP and the ACOG guidelines, it should not be used to predict individual neonatal mortality or neurologic outcome. However, the correlation between the AS and a newborn's further development became of great interest in clinical practice [1][2][3][4][5]. A low AS in the fth minute after birth has been linked to abnormal development later in life [6,7]. Numerous studies also indicate an impact of the procedures performed immediately after birth on the incidence of retinopathy of prematurity (ROP), bronchopulmonary dysplasia (BPD), and intraventricular or intraparenchymal haemorrhage (IVH/IPH), particularly in preemies [8][9][10]. In fact, the immaturity of preemies is the main cause of IVH and IPH [11,12]. IVH and IPH substantially contribute to further complications and neonatal mortality. They are also key causes of developmental disorders of motor, cognitive, and behavioural functions in schoolchildren [12]. Therefore, we investigated if the interventions performed in the labour ward and described in the expanded AS can reliably predict the incidence of early complications in infancy, in particular IVH and IPH. We analysed the following parameters: i) the AS at 1, 3, 5, 10, and 15 minutes after birth, ii) the maximum value from the AS measured at 1, 3, 5, and 10 minutes after birth (AS max), and iii) the course of resuscitation documented according to the expanded AS (ventilation using a mask ventilator, intubation, oxygen concentration used, cardiac massage, and use of adrenaline).

Statistical analysis
The chi-square automatic interaction detection (CHAID) algorithm was used for data analysis (SPSS 15.0). Data were compared using Pearson's chi-square test with P set at 0.05 (Bonferroni adjusted for multiple comparisons) and a cut-off selected automatically for all the parameters. The algorithm starts with the discretisation of the continuous variables. Next, the variable that best discriminates between decision classes is chosen based on the results of the chi-square test. In this way, the whole set of observations (i.e. root node) is split into subsets (i.e. internal nodes) that are as homogeneous as possible (regarding decision classes),with the branches extending from the root to the nodes describing the rules of division. The internal nodes are then further divided based on the results of the chi-square test. The division continues until nodes containing only one decision class (i.e. leaf nodes) are obtained, the desired depth of the tree is reached, or when the outcome of the next division does not signi cantly change the structure of the decision classes. Optionally, the decision tree can be trimmed. The branches (rules of division) that do not signi cantly in uence the accuracy to distinguish between decision classes in regard to preceding nodes can be shortened.
The branches linking the nodes, going from the top (root) to the bottom (leaves) of the tree, create decision rules in the form of implications: if A and if B and if…and if Z, then the ratio of the decision class (e.g. incidence of IVH) equals X. We aimed to de ne the relevant parameters affecting IVH/IPH incidence and to assess the risk of IVH/IPH based on medical ndings and their interactions. Therefore, the classes of our binary variable, namely the presence or lack of IVH/IPH, were marked as 1 or 0, respectively.

Results
To evaluate the relationship between IVH/IPH incidence and gestational age, we rst prepared a frequency plot (Fig. 1). IVH cases were observed most frequently in the youngest neonates. Because 80% of IVH/IPH cases were observed in the infants born before the 31st week of GA, only these neonates were chosen for further analysis.
Arti cial intelligence algorithms are currently the tool of choice in medical diagnostics and in decisionmaking systems. We used one of these modern statistical tools -a CHAID decision tree -to create a clear knowledge system based on decision rules. Next, we constructed a CHAID decision tree. When we used the expanded AS and a fraction of inspired oxygen (FiO 2 ) > 0.8 at 10 minutes after birth as discriminating variables, we identi ed 12 infants with IVH/IPH in a group of 38 neonates (31.6%) (Fig. 2). When we used the classical AS and an AS max ≤ 5, we identi ed 13 neonates with IVH/IPH in a group of 63 infants (20.6%) (Fig. 3).
These results indicate a higher discriminative power of the FiO 2 at 10 minutes after birth (based on the expanded AS) than the AS max. In the key leaf node, the discriminative power of the expanded AS to identify infants with IVH/IPH was 10% higher than that of the AS max. Additionally, using the expanded AS resulted in a more speci c (. smaller) subgroup of neonates (38 vs. 63), in which a similar number of IVH/IPH cases was identi ed (12 vs. 13).
The above analysis was conducted on the group of youngest infants (≤ 31 weeks GA), who experienced IVH/IPH the most often. In this group, an FiO 2 > 0.8 at 10 minutes after birth was a discriminating parameter. Therefore, we decided to investigate this parameter further. We analysed the whole group of infants and added the GA as a new parameter to our decision tree. In this way, the tree reached a depth of 2. In the node of ≤ 26 weeks of GA, a new division was generated, creating two subgroups based on the FiO 2 at 10 minutes after birth as a decision parameter: ≤0.3 and > 0.3. An FiO 2 > 0.3 was a discriminating value in this analysis and identi ed 29.1% of the group as having experienced IVH/IPH (Fig. 4). Lastly, using an FiO 2 ≥ 21% as a reference point, we analysed the concentration of oxygen that was given to the infants with IVH/IPH (Fig. 5). An FiO 2 > 80% resulted in the highest IVH/IPH incidence. An FiO 2 > 30% identi ed IVH/IPH cases less precisely, but it included a larger group of infants. Therefore, even though we suggest the FiO 2 as a powerful discriminative parameter, a direct correlation between the oxygen concentration and IVH/IPH incidence should be evaluated carefully and might require further analysis of, for example, the partial pressures of carbon dioxide (pCO 2 ) and oxygen (pO 2 ).

Discussion
The objective evaluation of an infant's health status is important not only for the care the newborn requires but also for scienti c research. The need for an unambiguous and comparable record of such an assessment led to the invention of a clear numerical scale. Virginia Apgar was the rst to propose a measure of an infant's health status that is quantitative, and therefore applicable in medical research.
To assess the health status of infants born prematurely and those requiring medical interventions, Rudiger et al. suggested a modi cation of the AS (the speci ed AS). The modi cation aimed to relate the AS to the GA, but it was independent of the requirements to achieve the score [13,14]. Hence, an infant born full-term or prematurely, who has no problems adapting to extrauterine life or who reacts favourably to resuscitation or other interventions, obtains the maximum speci ed AS.
The rapid progress observed in the eld of neonatology in the last decades, the expanding knowledge base, and technological developments have resulted in increased survival of newborns. This has led to a signi cantly greater number of medical procedures undertaken immediately after birth. These procedures and their outputs should not be omitted in the health assessment of newborns [13,15,16]. Our results con rm the link between the interventions undertaken immediately after birth and the health status later in life. In 2006, the AAP Committee on Foetuses and Newborns and the ACOG Committee on Obstetric Practice proposed an expanded AS to account for these interventions and their effects [2,16]. This scale contains, along with the classical AS, a listed course of resuscitation procedures. In use since 2008 in the Neonatal and Intensive Care Department of the Medical University of Warsaw, the expanded AS has signi cantly improved the precision of the documentation of the resuscitation procedures. It has also decreased the subjectivity of the health status evaluation of newborns. Therefore, this score was used in our study.
Further developments in the health assessment of newborns include the combined AS developed by Rudiger and Aguar, which merges the speci ed and expanded AS and assigns numeric values to the descriptive components of the expanded AS [14]. Dalili et al. compared the four types of AS and concluded that only a low score according to the combined AS predicts the incidence of IVH and its neurological complications in infants with perinatal hypoxia [17]. A low combined AS at 5 minutes after birth has been independently linked to IVH incidence. However, it does not indicate the severity of IVH. These results agree with the ACOG and AAP, our results, and other published studies showing the low prognostic value of the classic AS on prospective neurological status or unfavourable health prognosis [18][19][20][21][22][23].
Our study was based on decision trees, which have several advantages over other decision algorithms. Their simple structure allows for an easy interpretation of the results. They enable the assessment of the importance of the variables and attributes and do not require assumptions regarding the speci c distribution of the variables. The missing data do not cause problems in the analysis, and the classi cation based on decision trees is characterised by high accuracy. Additionally, the possibility of modelling dependencies of nonlinear phenomena and of analysing various sets of variables (nominal, ordinal, and continuous) allows for greater precision and detail in the description of the infants' health status and the undertaken interventions. The decision tree algorithm suited our study goals and setup because we required i) a simple data representation structure, ii) the possibility to analyse various sets of variables (nominal, ordinal, and continuous) with non-normal distribution, and iii) a high-accuracy classi cation system.
Our CHAID-based analysis shows the importance of the concentration of the oxygen supplied after birth in the prognosis of IVH/IPH incidence in premature infants. In score-only-based systems this information is lost. The combined AS only considers the presence or absence of oxygen supply. In our analysis, we de ned the thresholds for the selection of the groups of newborns for whom the health status prognosis can be made. Nevertheless, this is only a preliminary study that focuses on the prognosis of IVH and IPH incidence based on an infant's health status after birth. The GA and the birth body weight are the most important risk factors for complications such as IVH, ROP, and BPD. Although these diseases have complex pathogenesis, the states of blood supply and oxygenation of cells and tissues are the key in uencing factors [24,25].

Conclusions
The concentration of oxygen used during resuscitation is an important factor in the prognosis of IVH and IPH incidence. In follow-up studies, we plan to de ne the elements of the expanded AS that facilitate predictions of other early complications in preemies, like ROP and BPD. This information will not only improve the assessment of a newborn's health status but also allow doctors in labour wards to choose the most appropriate and effective interventions. The Bioethical Committee of the Medical University of Warsaw approved the study (KB/11/2009).

Consent for publication
Not applicable.

Availability of data and materials
The datasets used and analysed during the current study are available from the corresponding author on reasonable request.  The frequency of IVH/IPH is shown in relation to gestational age (in weeks). For every gestational week, the number of infants with (1, blue) or without (0, red) IVH/IPH is shown. n = 696 Figure 2 The decision tree based on the expanded Apgar score (fraction of inspired oxygen 10 minutes after birth) for infants ≤31st week of gestational age (GA). The group consisting of the youngest infants (≤31st week of GA) was split into three subgroups based on the fraction of inspired oxygen (FiO2) at 10 minutes after birth: ≤0.25, 0.25-0.8, and >0.8. The decision classes referring to the presence or lack of IVH/IPH were marked as 1 or 0, respectively. n = 276; Pearson's chi-square test with P inset at 0.05 (Bonferroni adjusted for multiple comparisons) was used to assess statistical signi cance.

Figure 3
The decision tree based on Apgar max score. The group consisting of the youngest infants (≤31st week of GA) was split into two groups based on the Apgar max score (apg_max): ≤5 and >5. The decision classes referring to the presence or lack of IVH/IPH were marked as 1 or 0, respectively. n = 276; Pearson's chi-square test with P inset at 0.05 (Bonferroni adjusted for multiple comparisons) was used to assess statistical signi cance. The decision tree relating IVH/IPH incidence to the gestational age (GA) and the fraction of inspired oxygen (FiO2) ten minutes after birth. The study group (n = 696) in the root node was split into three subgroups based on GA: ≤26th, 26th -36th, and >36th week of GA. The node ≤26th week of GA was further divided into subgroups according to the FiO2 ten minutes after birth: ≤0.3 and >0.3. The decision classes referring to the presence or lack of IVH/IPH were marked as 1 or 0, respectively. Pearson's chisquare test with P inset at 0.05 (Bonferroni adjusted for multiple comparisons) was used to assess statistical signi cance.