Identifying Clinical Phenotypes in Extremely Low Birth Weight Infants – A Machine Learning Approach

DOI: https://doi.org/10.21203/rs.3.rs-869916/v1

Abstract

There is increasing evidence that patient heterogeneity significantly hinders advancement in clinical trials and individualized care. This study aimed to identify distinct phenotypes in extremely low birth weight infants. We performed an agglomerative hierarchical clustering on principal components. Cluster validation was performed by cluster stability assessment with bootstrapping method. A total of 215 newborns (median gestational age 27 [26–29] weeks) were included in the final analysis. Six clusters with different clinical and laboratory characteristics were identified: the “Mature” (Cluster 1; n = 60, 27.9%), the mechanically ventilated with “adequate ventilation” (Cluster 2; n = 40, 18.6%), the mechanically ventilated with “poor ventilation” (Cluster 3; n = 39, 18.1%), the “extremely immature” (Cluster 4; n = 39, 18.1%%), the neonates requiring “Intensive Resuscitation” in the delivery room (Cluster 5; n = 20, 9.3%), and the “Early septic” group (Cluster 6; n = 17, 7.9%). In-hospital mortality rates were 11.7%, 25%, 56.4%, 61.5%, 45%, and 52.9%, while severe intraventricular hemorrhage rates were 1.7%, 5.3%, 29.7%, 47.2%, 44.4%, and 28.6% in clusters 1,2,3,4,5, and 6, respectively (p < 0.001). Conclusions: Our cluster analysis in extremely preterm infants was able to characterize six distinct phenotypes. Future research should explore how better phenotypic characterization of neonates might improve care and prognosis.

What Is Known

What is new:

Introduction

Although considerable progress to reduce neonatal morbimortality has been made in recent decades, prematurity remains the greatest direct cause of neonatal death(1). In addition, survivors are more prone to develop severe morbidities that affect long-term neurodevelopment(2). Unfortunately, the incidence of some of these morbidities, such as bronchopulmonary dysplasia(3), necrotizing enterocolitis(4), and persistent ductus arteriosus(5) has not been decreasing in recent years. In addition, many clinical trials on preterm infants have not yielded positive results, creating a challenge in discovering new therapies and interventions. This apparent “stagnation” is not limited to the neonatal population. In adults with acute respiratory distress syndrome (ARDS), most of clinical trials had negative results and mortality remains high (6).

The heterogeneity of the patients investigated, particularly in the intensive care unit, is thought to be one of the reasons for this challenge in clinical research. Critically ill patients represent the most heterogeneous population in the hospital, with a high morbimortality rate(7). Patients with the same diagnosis frequently receive similar treatment and interventions but have varying outcomes. Although there is still no definitive explanation for why this occurs, it is becoming clear that organizing patients into shallow disorder-based categories may result in patients with distinct illness mechanism, responses to interventions, and outcomes being grouped together(7).

Better phenotyping of preterm infants is urgently needed in order to properly target clinical research and therapy. The aim of this study was to use machine learning approaches to identify and classify distinct characteristics of extremely preterm neonates.

Material And Methods

We used data from a retrospective study which analyzed newborns with birth weight less than 1000 grams, born between January 2012 and December 2017, admitted to a single-center tertiary neonatal intensive care unit in São Paulo, Brazil (8). Briefly, we included all neonates admitted to the neonatal intensive care unit, with birth weight less than 1000 grams, who did not have severe congenital malformation and had no missing data.

We recorded a total of 62 variables for each study participant that were used in the analysis:

General characteristics: Gestational age (weeks), birth weight (grams), CRIB II score (Clinical Risk Index for Babies) (9), SGA (Small for gestational age defined as birth weight < p10 Fenton growth scale), gender, 5-minute APGAR score, delivery type (vaginal or c-section), single or twin birth, antenatal corticoid, number of endotracheal intubation trials in the delivery room, lowest temperature in the first 12 hours of life and epinephrine necessity in the delivery room.

Hemodynamics: inotropic therapy necessity in the first 7 days of life, lowest pH in the first 3 days of life, lowest bicarbonate level in the first 3 days of life, lowest base excess level in the first 3 days of life, lowest and highest systolic/diastolic and mean blood pressure in the first 3 days of life, hypotension (defined as mean blood pressure lower than gestational age), persistent ductus arteriosus (PDA), PDA size, hemodynamically significant PDA (hsPDA) (defined as PDA > 1.5mm), and fluid bolus necessity.

Respiratory: Invasive mechanical ventilation necessity and its duration in the first 3 days of life, highest and lowest pCO2 in the first 3 days of life, pneumothorax with chest drainage necessity, highest Ppeak, PEEP, and Mean Airway Pressure in the first 3 days of life in mechanically ventilated patients.

Renal: Acute kidney injury presence (defined as increase in serum creatinine > 0.3mg/dL), creatinine clearance according to Cockcroft-Gault Eq. (10), lowest urine output in the first 3 days of life and highest fluid overload in the first 3 days of life.

Hematologic/Infectious: Positive blood culture in the first 3 days of life, chorioamnionitis, highest c-reactive protein in the first 3 days of life, lowest platelet and hemoglobin levels in the first 3 days of life. Thrombocytopenia defined as platelet count < 50,000mm3.

Outcomes: Early death defined as death in the first 7 days of life, in-hospital mortality, length of stay in survivors patients, moderate or severe bronchopulmonary dysplasia classified at 36 weeks of corrected gestational age in survivors patients, domiciliary oxygen necessity, ibuprofen for PDA clinical closure, PDA surgical ligation necessity, intraventricular hemorrhage and respective grade, severe intraventricular hemorrhage was defined as IVH grade III and IV according to Papile-Burnstein classification(11).

Statistical methods

We performed an agglomerative hierarchical cluster analysis (HC) in 3 different models. In this algorithm, a distance metric is employed to calculate similarity (or dissimilarity) between two patients (or groups of patients), where the smaller the distance between 2 patients, the more similar they are between them. This algorithm was selected due to the lack of knowledge about the number of existing subgroups. In the first model we analyzed all 62 features and performed a dimensionality reduction technique. We used FAMD (Factor analysis of mixed data) to reduce dimensionality because of the presence of mixed (categorical and continuous) variables. We then carried out a hierarchical clustering using principal components with cumulative percentage of variance with at least 85%, with euclidean distance and Ward’s linkage criteria. In the second model, all features were reviewed by clinical experts to identify variables which may drive cluster allocation. We then created a new data set with those selected variables and, without dimensionality reduction, we created a distance matrix using Gower’s distance method (because of mixed data) and proceeded with HC with Ward’s linkage criteria. Cluster analysis was carried out using the following 8 variables: 1) highest pCO2 level; 2) Lowest base excess level; 3) Invasive mechanical ventilation necessity; 4) Inotrope necessity; 5) Positive blood culture; 6) Epinephrine necessity in the delivery room; 7) hemodynamically significant persistent ductus arteriosus; 8) Gestational age < 26 weeks.

In the third model, we used the same features from model 2 and applied FAMD to reduce dimensionality, using principal components with cumulative percentage of variance with at least 85%, with Euclidean distance and Ward’s linkage criteria.

The cluster stability was assessed with bootstrapping methods. The data were resampled and the Jaccard similarities of the original clusters to the most similar clusters in the resampled data were computed(12). The mean of the similarities was used as an index of stability, and only a mean greater than 0.75 was considered stable.

Continuous variables were tested for normality using Kolmogorov-Smirnov test. To compare results among clusters, we used chi-square for categorical variables and Kruskal-Wallis for continuous variables. All data with missing values were excluded from the analysis. Data were standardized before cluster analysis. All analyses were performed in RStudio software and the following packages were used: “factoextra”, “FactoMineR”, “stats”, “fpc”, “ggplot2”, and “tidyverse”. The study protocol was approved and informed consent was waived by the institutional ethics committee.

Results

The initial dataset included 258 extremely low birth weight infants. The overall population’s median gestational age was 27 (26–29) weeks and the median birth weight was 760 (600–880) grams. The hierarchical clustering analysis after feature selection and dimensionality reduction (FAMD) provided the most stable and clinically interpretable results (eTable 1). Eigenvalues and cumulative percentage variance are presented in eTable 2. This analysis included 215 patients who had complete data for the 8 selected variables. The general characteristics are presented in Table 1. Characteristics related to respiratory/hemodynamics and renal/hematologic/infectious are respectively presented in Table 2 and 3.

Cluster analysis

We used agglomerative cluster approach after FAMD dimensionality reduction, and a dendrogram (Fig. 1) was generated. Six clusters were identified with index of stability higher than 0.85 for all clusters.

Cluster 1 included 60 newborns (27.9% of the cohort) and could be considered as a “more mature” phenotype. Cluster 1 is the largest group, with a median gestational age (GA) of 28.8 weeks, which is higher than the other five clusters. This cluster had the lowest invasive mechanical ventilation (IMV) necessity in the first 3 days of life (48.3%). This group reported the least early death (3.3%), death during hospitalization (11.7%), severe intraventricular hemorrhage (1.7%), length of stay (71 days), moderate or severe BPD (41.5%), PDA surgical necessity (1.9%), and clinical PDA closure necessity rates (1.9%).

Cluster 2 included 40 newborns (18.6% of the cohort). Cluster 2 has a similar birth weight (median of 764g) compared to cluster 1, but has a lower GA (median 27 weeks). All patients in this group needed IMV in the first 3 days of life, however with good respiratory status: adequate pCO2 levels (median 39.3 mmHg) and lower pressure-related ventilatory settings (Ppeak, PEEP, and MAP). This group is characterized by a low rate of early death (7.5%) and severe intraventricular hemorrhage (5.3%).

Cluster 3 included 39 newborns (18.1% of the cohort). Cluster 3 is represented by neonates with same gestational age (median 27.5 weeks) and no statistically significant difference on birth weight (median 682g), compared to cluster 2. Severity score (CRIB) and 5-minute APGAR score were also similar between cluster 2 and 3. However, cluster 3 is characterized by a worse respiratory status, with the highest pCO2 levels among all clusters (median 64.8 mmHg) and higher pressure-related ventilatory settings. It is also the group that received inotrope the most, had lower pH, HCO3, and BE levels. Moreover, this group had lower systolic blood pressure and mean blood pressure compared to cluster 2. This group evolved with significantly more death rate (56.4%) and severe intraventricular hemorrhage (29.7%).

Cluster 4 included 39 newborns (18.1% of the cohort). Cluster 4 represents the most immature group, where 100% were born with gestational age lower than 26 weeks (median of 25.1 weeks). This group also had very high pCO2 levels (median 56.6 mmHg) and pressure-related ventilatory settings. This cluster evolved with the highest death (61.5%) and severe intraventricular hemorrhage (47.2%) rates.

Cluster 5 included 20 newborns (9.3% of the cohort) who required “Intensive Resuscitation” at delivery room. Cluster 5 represents neonates with worse delivery room conditions. In this group, all patients received epinephrine in the delivery room. They had the lowest APGAR score at 5th minute and antenatal corticoid rate.

Cluster 6 included 17 newborns (7.9% of the cohort) and could be named “Early septic” group. Cluster 6 is the smallest cluster and represents mainly neonates with positive blood culture within postnatal day 3 (94.1%). This cluster represents neonates with GA, BW, 5-minute APGAR, CRIB score, pCO2 level, pressure-related ventilatory settings, pH, HCO3, BE, and blood pressure similar to Cluster 2. However, they evolve with higher c-reactive protein level (median of 20.3 mg/dL), lower platelet level (median of 54000/mm3), higher death (52.9%), and higher severe intraventricular hemorrhage (28.6%) rate, compared to Cluster 2.

Discussion

Our cluster analysis identified six distinct phenotypes of extremely preterm infants who exhibited specific clinical and laboratorial characteristics: the “Mature” (Cluster 1), the mechanically ventilated with “adequate ventilation” (Cluster 2), the mechanically ventilated with “poor ventilation” (Cluster 3), the “extremely immature” (Cluster 4), the neonates requiring “Intensive Resuscitation” in the delivery room (Cluster 5), and the “Early septic” group (Cluster 6).

Despite medical improvements, many clinical conditions remain with limited therapeutic alternatives and high mortality rates(13). A classic example in adult patients is acute respiratory distress syndrome (ARDS). According to the Berlin Definition, ARDS is defined by a PaO2/FiO2 ratio of less than 300mmHg and bilateral opacities on chest radiography that cannot be explained by cardiac etiology(14). Despite the importance of this definition in identifying and including individuals with ARDS in clinical trials, these findings are prevalent in critically ill patients and are not exclusive to those with ARDS(15). As a result, a wide range of disease-mechanism are equally treated, and enrolled in clinical trials to receive the same intervention. This could be one of the reasons why so many ARDS trials have failed and mortality remains high. Heterogeneity is becoming more widely acknowledged as a cause of clinical trial failure(6).

Clinical research is increasingly utilizing machine learning (ML) algorithms, particularly in critically ill patients (7). Cluster analysis is a key ML’s application that computes data similarities (or dissimilarities) and divides them into subgroups. Typically, data with similar aspects are grouped together. Within a heterogeneous group, this is beneficial for identifying patients with comparable characteristics.

To illustrate, many algorithms have been applied to ARDS patients and they primarily discovered two distinct clusters: “hypoinflammatory” and “hyperinflammatory” (15). Results from clinical trials showed that patients with a hyperinflammatory phenotype had lower mortality rates with a higher PEEP strategy (16) and benefited from a conservative fluid therapy (17). Those benefits were not observed in patients with hypoinflammatory phenotype. These analysis are not exclusive to patients with ARDS. Clustering methods, in principle, can aid with almost any heterogeneous disease. Recently, Geri et al. identified five distinct clusters in adults with septic shock, each with its own set of characteristics (18). Several studies have also used cluster analysis to identify asthma subtypes(19).

Similarly, it is feasible that we assume that all extremely preterm neonates behave the same and overlook possible heterogeneity in this population. We hypothesize that this heterogeneity may be a major reason why many clinical trials in neonates have negative results or are difficult to interpret. To illustrate, Mitra et al(20) conducted a meta-analysis of PDA interventions under the assumption of transitivity, which subjectively assumes that the characteristics of the populations were similar across studies (21). However, in this study, the mean gestational age ranged widely from 25.5 (1.2) to 33.6 (2.1) weeks. To date, we still do not have a definitive answer as to what is the best approach to PDA. Another controversial topic is whether and to whom postnatal corticosteroids are beneficial. Results from clinical trials are conflicting and have low quality of evidence(22).

In our study, we found remarkably different clusters of preterm infants. In-hospital mortality, severe intraventricular hemorrhage, and hsPDA rates ranged from 11.7–61.5%, 1.7–47.2%, and 11.7–100%, respectively, between clusters. Clusters 2 and 3 exhibit comparable characteristics immediately after birth (GA, BW, CRIB score, 5-minute APGAR score, and antenatal corticosteroid rates). Nevertheless, they behave and have completely different outcomes, highlighting the worst respiratory status of cluster 3. Patients in cluster 1 represent a subset with the lowest hsPDA rates (only 11.7%). In fact, only one neonate in this cluster received ibuprofen required PDA surgical closure. As a result, prophylactic PDA closure would be unlikely to help this population. Furthermore, in this subgroup, any intervention that analyzes PDA closure as an endpoint would potentially be considered a “success”. Another example would be to focus on interventions to prevent severe intraventricular hemorrhage in neonates from clusters 3,4,5 and 6, as rates of severe IVH are low in clusters 1 and 2. Our discoveries have a wide range of applications, and we have highlighted just a few of them.

It is important to note that even in homogeneous datasets, cluster algorithms can find patterns and generate clusters. Moreover, clustering is frequently an exploratory analysis, and the patterns discovered may not be meaningful. Therefore, cluster validation is an important step, and one technique to validate them is to assess their stability (12). Hence, we performed an assessment of cluster stability by randomly reassembling the data. If the patterns are meaningful, then a cluster should not simply disappear when the dataset is randomly changed. When we tested the stability in model 1 (Dimensionality reduction without feature selection) and model 2 (Feature selection without dimensionality reduction), we did not find stable clusters. However, in model 3 (Feature selection with dimensionality reduction) all clusters had a high index of stability (> 0.85), which represents meaningful patterns (eTable 2).

It is worth noting that our study has some limitations. First, our study population is from a tertiary single-center in Brazil’s southeastern region that is a reference for high-risk pregnancies. We believe that the clusters identified in our study cannot be immediately translated to clinical practice until further studies elucidate the exact characterization of each cluster. Nonetheless, our analysis suggests that there are clinically relevant subgroups in extremely preterm infants, and identifying well-defined phenotypes in this cohort could improve our understanding of disease mechanisms and proper management. Second, principal component analysis showed low explained variance, suggesting that potential features that could better explain these phenotypes were not included. We suggest that further studies should include features related to echocardiographic and lung ultrasound parameters (23, 24). Finally, while feature selection was done by experienced physicians, it was primarily subjective, and we cannot exclude the possibility that there are other variables of greater significance to each phenotype.

In conclusion, our unsupervised machine learning algorithm in a cohort of extremely preterm neonates identified six distinct phenotypes that could help clinicians and researchers individualize clinical support. Future research should explore how better phenotypic characterization of neonates might affect management and prognosis.

Abbreviations

Declarations

Authors contributions:

FYM developed the theory, contributed to the design and implementation of the research; led data acquisition; analyzed data and to the writing of the manuscript. VLJK and WBC contributed to the design and implementation of the research and revised the manuscript. All authors approved the final manuscript as submitted.

Acknowledgments:

            We thank all the colleagues at the NICU staff at the Faculty of Medicine of the University of São Paulo

Conflict of Interest

            The authors declare that they have no conflict of interest and no funding was received for this study.

References

  1. Hug L, Alexander M, You D, Alkema L (2019) National, regional, and global levels and trends in neonatal mortality between 1990 and 2017, with scenario-based projections to 2030: a systematic analysis. Lancet Glob Heal 7(6):e710–e720
  2. Blencowe H, Cousens S, Chou D, Oestergaard M, Say L, Moller AB et al (2013) Born Too Soon: The global epidemiology of 15 million preterm births. Reprod Health 10(SUPPL. 1):1–14
  3. Zysman-Colman Z, Tremblay GM, Bandeali S, Landry JS (2013) Bronchopulmonary dysplasia - Trends over three decades. Paediatr Child Heal 18(2):86–90
  4. Alsaied A, Islam N, Thalib L (2020) Global incidence of Necrotizing Enterocolitis: A systematic review and Meta-analysis. BMC Pediatr 20(1):1–15
  5. Ngo S, Profit J, Gould JB, Lee HC (2017) Trends in patent ductus arteriosus diagnosis and management for very low birth weight infants. Pediatrics 139(4):1–9
  6. Matthay MA, McAuley DF, Ware LB (2017) Clinical trials in acute respiratory distress syndrome: challenges and opportunities. Lancet Respir Med 5(6):524–534
  7. Castela Forte J, Perner A, van der Horst ICC (2019) The use of clustering algorithms in critical care research to unravel patient heterogeneity. Intensive Care Med 45(7):1025–1028
  8. Matsushita F, Krebs V, Ferraro A, de Carvalho W (2020) Early fluid overload is associated with mortality and prolonged mechanical ventilation in extremely low birth weight infants. Eur J Pediatr 179(11):1665–1671
  9. Parry G, Tucker J, Tarnow-Mordi W. CRIB II: an update of the clinical risk index for babies score For personal use. Only reproduce with permission from The Lancet Publishing Group. 2003;361:1789–91
  10. Cockcroft DW, Gault MH (1976) Prediction of creatinine clearance from serum creatinine. Nephron 16(1):31–41
  11. Papile LA, Burstein J, Burstein R, Koffler H (1978) Incidence and evolution of subependymal and intraventricular hemorrhage: A study of infants with birth weights less than 1,500 gm. J Pediatr 92(4):529–534
  12. Hennig C (2007) Cluster-wise assessment of cluster stability. Comput Stat Data Anal 52(1):258–271
  13. Wilson JG, Calfee CS (2020) ARDS Subphenotypes: Understanding a Heterogeneous Syndrome. Crit Care 24(102):1–8
  14. Ranieri VM, Rubenfeld GD, Thompson BT, Ferguson ND, Caldwell E, Fan E et al (2012) Acute respiratory distress syndrome: The Berlin definition. JAMA - J Am Med Assoc 307(23):2526–2533
  15. Sinha P, Calfee CS (2019) Phenotypes in acute respiratory distress syndrome: Moving towards precision medicine. Curr Opin Crit Care 25(1):12–20
  16. Brower RG, Lanken PN, MacIntyre N, Matthay MA, Morris A, Ancukiewicz M et al (2004 Jul) Higher versus lower positive end-expiratory pressures in patients with the acute respiratory distress syndrome. N Engl J Med 351(4):327–336
  17. Famous KR, Delucchi K, Ware LB, Kangelaris KN, Liu KD, Thompson BT et al (2017) Acute respiratory distress syndrome subphenotypes respond differently to randomized fluid management strategy. Am J Respir Crit Care Med 195(3):331–338
  18. Geri G, Vignon P, Aubry A, Fedou AL, Charron C, Silva S et al. Cardiovascular clusters in septic shock combining clinical and echocardiographic parameters: a post hoc analysis. Intensive Care Med. 2019
  19. Deliu M, Sperrin M, Belgrave D, Custovic A (2016) Identification of Asthma Subtypes Using Clustering Methodologies. Pulm Ther 2(1):19–41
  20. Mitra S, Florez ID, Tamayo ME, Mbuagbaw L, Vanniyasingam T, Veroniki AA et al (2018) Association of placebo, indomethacin, ibuprofen, and acetaminophen with closure of hemodynamically significant patent ductus arteriosus in preterm infants a systematic review and meta-analysis. JAMA - J Am Med Assoc 319(12):1221–1238
  21. Cipriani A, Higgins J, Geddes JR, Salanti G (2013) Research and Reporting Methods Annals of Internal Medicine Conceptual and Technical Challenges in Network Meta-analysis. Ann Intern Med 159(8):130–137
  22. Ramaswamy VV, Bandyopadhyay T, Nanda D, Bandiya P, Ahmed J, Garg A et al. Assessment of Postnatal Corticosteroids for the Prevention of Bronchopulmonary Dysplasia in Preterm Neonates: A Systematic Review and Network Meta-analysis. JAMA Pediatr. 2021 Jun 7;175(6):e206826–e206826
  23. Singh Y, Tissot C, Fraga MV, Yousef N, Cortes RG, Lopez J et al (2020) International evidence-based guidelines on Point of Care Ultrasound (POCUS) for critically ill neonates and children issued by the POCUS Working Group of the European Society of Paediatric and Neonatal Intensive Care (ESPNIC). Crit Care 24(1):1–16
  24. Martino L, De, Yousef N, Ben-Ammar R, Raimondi F, Shiv S-A, Luca D, De. Lung ultrasound score predicts surfactant need in extremely preterm neonates. Pediatrics. 2018;142(3)

Tables

Table 1

General characteristics

 

Cluster 1

N = 60

“Mature”

Cluster 2

N = 40

“Adequate ventilation”

Cluster 3

N = 39

“Poor ventilation”

Cluster 4

N = 39

“Extremely immature”

Cluster 5

N = 20

“Intensive Resuscitation”

Cluster 6

N = 17

“Early septic”

P value

Gestational age (weeks), median (IQR)

28.8 (27.1–30.4)

27 (26.7–28.5)

27.5 (26.5–29.1)

25.1 (24.3–25.7)

26.5 (25.4–29.8)

26.5 (25.3–29.1)

< 0.001a

Gestational age < 26 weeks, n (%)

0 (0)

0 (0)

1 (2.6)

39 (100)

8 (40)

5 (29.4)

< 0.001b

Birth weight (grams), median (IQR)

829 (665–950)

764 (680–908)

682 (600–886)

750 (580–840)

609 (515–780)

708 (600–790)

0.006a

Birth weight < 750g, n (%)

21 (35)

18 (45)

23 (59)

19 (48.7)

14 (70)

11 (64.7)

0.039b

CRIB score, median (IQR)

10 (9–12)

11 (10–13)

11 (10–13)

14 (12–15)

13 (12–15)

11 (10–14)

< 0.001a

Epinephrine at delivery room, n (%)

0 (0)

0 (0)

0 (0)

0 (0)

20 (100)

0 (0)

< 0.001b

SGA, n (%)

39 (65)

15 (37.5)

19 (48.7)

6 (15.4)

11 (55)

10 (58.8)

< 0.001b

Female gender, n (%)

35 (58.3)

21 (52.5)

19 (48.7)

15 (38.5)

9 (45)

10 (58.8)

0.472b

5-minute APGAR score, median (IQR)

8 (8–9)

8 (8–9)

8 (7–9)

7 (5–8)

5 (1–6)

8 (6–9)

< 0.001a

APGAR < 7, n (%)

6 (10)

8 (20)

7 (17.9)

16 (41)

17 (85)

5 (29.4)

< 0.001b

Vaginal birth, n (%)

3 (5)

3 (7.5)

3 (7.7)

15 (38.5)

5 (25)

6 (35.3)

< 0.001b

Twin birth, n (%)

11 (18.3)

12 (30)

12 (30.8)

15 (38.5)

5 (25)

3 (17.6)

0.292b

Antenatal corticoid, n (%)

39 (65)

20 (50)

20 (51.3)

17 (43.6)

4 (20)

8 (47.1)

0.021b

ET number of trials, median (IQR)

0 (0–2)

1 (0–2)

1 (0–2)

1 (1–2)

1.5 (1–2)

1 (0–2)

< 0.001a

Lowest temperature, median (IQR)

35 (34.5–35.9)

35 (34.2–35.5)

34.9 (34–35.6)

34.8 (34–35.6)

34.5 (34–34.9)

35 (33.9–35.9)

0.069a

Early death, n (%)

2 (3.3)

3 (7.5)

7 (17.9)

9 (23.1)

5 (25)

5 (29.4)

0.010b

In-hospital mortality, n (%)

7 (11.7)

10 (25)

22 (56.4)

24 (61.5)

9 (45)

9 (52.9)

< 0.001b

Intraventricular hemorrhage, n (%)

20 (34.5)

10 (26.3)

20 (54.1)

23 (63.9)

10 (55.6)

4 (28.6)

0.005b

Severe intraventricular hemorrhage, n (%)

1 (1.7)

2 (5.3)

11 (29.7)

17 (47.2)

8 (44.4)

4 (28.6)

< 0.001b

IVH grade, median (IQR)

0 (0–1)

0 (0–1)

1 (0–3)

2 (0–3)

1 (0–3)

0 (0–3)

< 0.001a

Pneumothorax, n (%)

2 (3.3)

2 (5)

2 (5.1)

5 (12.8)

1 (5)

2 (11.8)

0.458b

Length of stay (days), median (IQR)

71 (62–92)

88 (71–114)

113 (93–125)

95 (89–119)

112 (90–132)

88 (81–104)

0.004a

Death or BPD, n (%)

46 (76.7)

39 (97.5)

39 (100)

39 (100)

20 (100)

16 (94.1)

0.002b

Death or severe BPD, n (%)

28 (46.7)

29 (72.5)

36 (92.3)

34 (87.2)

18 (90)

15 (88.2)

0.007b

Moderate or Severe BPD, n (%)

22/53 (41.5)

18/30 (60)

14/17 (82.4)

10/15 (66.7)

9/11 (81.8)

6/8 (75)

0.015b

Domiciliar oxygen, n (%)

9/53 (17)

6/30 (20)

8/17 (47.1)

4/15 (26.7)

3/11 (27.3)

3/8 (37.5)

0.192b

PDA surgical, n (%)

1/53 (1.9)

6/30 (20)

5/17 (29.4)

6/15 (40)

2/11 (18.2)

1/8 (12.5)

0.004b

Ibuprofen, n (%)

1/53 (1.9)

18/30 (60)

8/17 (47.1)

8/15 (53.3)

6/11 (54.5)

2/8 (25)

0.157b

aKruskal-Wallis; bChi-square; CRIB: Clinical Risk Index for babies; SGA: Small for gestational age; ET: endotracheal tube; IQR: Interquartile Range; BPD: Bronchopulmonary Dysplasia; PDA: Persistent Ductus Arteriosus

Table 2

– Respiratory and hemodynamic characteristics

 

Cluster 1

N = 60

“Mature”

Cluster 2

N = 40

“Adequate ventilation”

Cluster 3

N = 39

“Poor ventilation”

Cluster 4

N = 39

“Extremely immature”

Cluster 5

N = 20

“Intensive Resuscitation”

Cluster 6

N = 17

“Early septic”

P value

IMV, n (%)

29 (48.3)

40 (100)

38 (97.4)

39 (100)

20 (100)

12 (70.6)

< 0.001a

IMV duration (% of time), median (IQR)

0 (0–62.5)

92.7 (58.3–100)

100 (87.5–100)

100 (100–100)

100 (100–100)

100 (0–100)

< 0.001b

Highest pCO2 (mmHg), median (IQR)

39.3 (33.4–46.4)

45.3 (37.7–55.1)

64.8 (46.3–86)

56.6 (46.3–67.6)

49.5 (38.7–70.4)

49.3 (37.1–62.9)

< 0.001b

Lowest pCO2 (mmHg), median (IQR)

30.2 (26.3–34.9)

30.1 (24.8–34.4)

30.9 (27.4–36.2)

31.8 (23.5–37.1)

28.6 (26.3–38.4)

28.4 (26–33.7)

0.865b

Highest Ppeak (cmH2O), median (IQR)

18 (16–18)

18 (17–19)

20 (18–22)

19 (18–21)

20 (18–22)

19 (18–25)

< 0.001b

Highest PEEP (cmH2O), median (IQR)

6 (6–7)

6 (6–7)

7 (7–8)

7 (6–7)

7 (6–8)

7 (6–8)

< 0.001b

Highest MAP (cmH2O), median (IQR)

8.5 (8–9)

8.9 (8.4–9.5)

11.9 (10–14)

10.5 (9.3–13.5)

10.8 (9.2–11.6)

10 (8.8–13)

< 0.001b

Inotrope therapy, n (%)

0 (0)

0 (0)

33 (84.6)

19 (48.7)

9 (45)

7 (41.2)

< 0.001a

Lowest pH, median (IQR)

7.3 (7.24–7.34)

7.24 (7.15–7.31)

7.05 (6.94–7.16)

7.12 (7.03–7.23)

7.15 (7.04–7.28)

7.21 (7.08–7.33)

< 0.001b

Lowest HCO3 (mEq/L), median (IQR)

16.9 (14.3–18.5)

15.7 (14.3–17.4)

14 (12.1–17.7)

14.5 (11.4–16.6)

14.6 (12.9–16.9)

14.7 (11.1–18)

0.003b

Lowest BE, median (IQR)

-7.5 (-10.3 to -5.6)

-8.7 (-11.4 to -7)

-13.9 (-18.9 to -10)

-11.6 (-15.7 to -9.3)

-11.5 (-12.8 to -9.8)

-9.4 (-16.4 to -7.9)

< 0.001b

Lowest systolic blood pressure (mmHg), median (IQR)

47 (40–52)

40.5 (36–48)

39 (34–41)

36 (32–42)

36 (31–42)

37 (33–46)

< 0.001b

Lowest diastolic blood pressure (mmHg), median (IQR)

23 (20–26)

20.5 (19–25)

20 (18–21)

18 (16–21)

18.5 (17–21)

19 (18–24)

< 0.001b

Lowest mean blood pressure (mmHg), median (IQR)

31 (27–36)

30 (26–33)

26 (25–30)

26 (22–29)

26 (22–29)

27 (25–30)

< 0.001b

Hypotension, n (%)

23 (39)

15 (37.5)

24 (61.5)

15 (40.5)

10 (50)

6 (37.5)

0.228a

MBP < 30mmHg, n (%)

23 (39)

19 (47.5)

29 (74.4)

28 (75.7)

16 (80)

12 (75)

< 0.001a

Highest systolic blood pressure (mmHg), median (IQR)

69 (60–80)

63 (59–71)

68 (57–78)

61 (53–70)

64 (56–69)

66 (60–78)

0.012b

Highest diastolic blood pressure (mmHg), median (IQR)

42 (35–51)

36.5 (30–43)

41 (33–48)

38 (30–42)

36 (30–44)

39 (34–47)

0.024b

Highest mean blood pressure (mmHg), median (IQR)

51 (44–59)

46 (41–51)

50 (41–60)

46 (38–50)

42 (39–53)

45 (42–56)

0.008b

PDA, n (%)

29 (48.3)

40 (100)

37 (94.9)

35 (89.7)

19 (95)

12 (70.6)

< 0.001a

hsPDA, n (%)

7 (11.7)

40 (100)

33 (84.6)

31 (79.5)

18 (90)

10 (58.8)

< 0.001a

PDA size, median (IQR)

0 (0–1.3)

2.5 (2.1–3)

2.1 (1.9–2.7)

2.4 (1.6–2.6)

2.4 (2.2–3)

2 (0–2)

< 0.001b

Highest lactate (mg/dL), median (IQR)

47.5 (30–62)

46 (34–68)

56.5 (37–91)

50 (36–70)

62 (27–84)

51 (42–61)

0.571b

Fluid bolus, median (IQR)

0 (0–0)

1 (0–1)

2 (0–4)

1 (0–3)

1.5 (0–4)

2 (0–4)

< 0.001b

aChi-square; bKruskal-Wallis; IQR: Interquartile Range; IMV: Invasive Mechanical Ventilation; Ppeak: Peak inspiratory airway pressure; PEEP: Positive end-expiratory pressure; MAP: Mean Airway Pressure; BE: Base excess; MBP: Mean Blood Pressure; PDA: Persistent Ductus Arteriosus; hsPDA: Hemodynamically significant persistent ductus arteriosus

Table 3

– Renal, Hematologic and Infectious characteristics

 

Cluster 1

N = 60

“Mature”

Cluster 2

N = 40

“Adequate ventilation”

Cluster 3

N = 39

“Poor ventilation”

Cluster 4

N = 39

“Extremely immature”

Cluster 5

N = 20

“Intensive Resuscitation”

Cluster 6

N = 17

“Early septic”

P value

Acute kidney injury, n (%)

9 (15)

16 (40)

17 (43.6)

16 (42.1)

8 (40)

6 (35.3)

0.005a

Creatinine clearance (mL/min), n (%)

26.5 (17.6–39.6)

17.4 (13–23.4)

13.2 (10.5–18.8)

13.4 (8.4–15.7)

10.9 (8.4–14.7)

19.4 (15–25.3)

< 0.001a

Lowest urine output (mL/kg/h), median (IQR)

2.5 (1.7–3.4)

2.8 (2.2–3.9)

1.7 (0.9–3.4)

2 (1.2–4.1)

3.1 (2–4.2)

1.6 (1–2.9)

0.044b

Highest fluid overload, median (IQR)

13.2 (7.2–19.7)

9.9 (5.5–16.8)

15.6 (9–24.8)

15.9 (8–25.3)

12.2 (7.6–21.4)

20.1 (11.4 -

0.092b

Positive blood culture, n (%)

0 (0)

0 (0)

0 (0)

0 (0)

1 (5)

16 (94.1)

< 0.001a

Highest C-reactive protein (mg/dL), median (IQR)

4.6 (1.5–10.6)

5.4 (1.8–14.2)

12.7 (6.2–22.1)

7.2 (2.1–13.5)

8.5 (3.2–22.6)

20.3 (7–51.6)

0.001b

Chorioamnionitis, n (%)

4 (6.7)

2 (5)

5 (12.8)

8 (20.5)

2 (10)

4 (23.5)

0.126a

Lowest Hemoglobin (g/dL), median (IQR)

15 (13–19.7)

13 (11.7–14.8)

11.6 (9.1–13.4)

10.3 (8–13.2)

9.8 (9.3–11.7)

11.2 (9.2–16.4)

< 0.001b

Lowest Platelet (mm3), median (IQR)

100000 (66000–154000)

109500 (62000–160000)

79000 (44000–102000)

104000 (64000–165000)

67500 (47000–105000)

54000 (27000–103000)

0.001b

Thrombocytopenia < 50.000mm3, n (%)

8 (13.3)

6 (15)

11 (28.2)

5 (12.8)

6 (30)

8 (47.1)

0.016a

Platelet transfusion necessity, n (%)

26 (43.3)

21 (52.5)

30 (76.9)

21 (53.8)

15 (75)

11 (64.7)

0.013a

Blood transfusion necessity, n (%)

7 (11.7)

15 (37.5)

27 (69.2)

27 (69.2)

14 (70)

10 (58.8)

< 0.001a

Number of Platelet transfusion, median (IQR)

0 (0–1)

1 (0–1)

1 (1–2)

1 (0–2)

1 (1–3)

1 (0–2)

< 0.001b

Number of Blood transfusion, median (IQR)

0 (0–0)

0 (0–1)

0 (1–2)

1 (0–2)

1 (0–2)

1 (0–1)

< 0.001b

aChi-square; bKruskal-Wallis; IQR: Interquartile Range; Fluid overload was calculated by: (Total input in mL – Total output in mL)/Birth weight