Hourly Kinetics of Critical Organ Dysfunction in Extremely Preterm Infants

Abstract

Rationale

Use of severity of illness scores to classify patients for clinical care and research is common outside of the neonatal ICU. Extremely premature (<29 weeks’ gestation) infants with extremely low birth weight (<1,000 g) experience significant mortality and develop severe pathology during the protracted birth hospitalization.

Objectives

To measure at high resolution the changes in organ dysfunction that occur from birth to death or discharge home by gestational age and time, and among extremely preterm infants with and without clinically meaningful outcomes using the neonatal sequential organ failure assessment score.

Methods

A single-center, retrospective, observational cohort study of inborn, extremely preterm infants with extremely low birth weight admitted between January 2012 and January 2020. Neonatal sequential organ failure assessment scores were calculated every hour for every patient from admission until death or discharge.

Measurements and Main Results

Longitudinal, granular scores from 436 infants demonstrated early and sustained discrimination of those who died versus those who survived to discharge. The discrimination for mortality by the maximum score was excellent (area under curve, 0.91; 95% confidence intervals, 0.88–0.94). Among survivors with and without adverse outcomes, most score variation occurred at the patient level. The weekly average score over the first 28 days was associated with the sum of adverse outcomes at discharge.

Conclusions

The neonatal sequential organ failure assessment score discriminates between survival and nonsurvival on the first day of life. The major contributor to score variation occurred at the patient level. There was a direct association between scores and major adverse outcomes, including death.

At a Glance Commentary

Scientific Knowledge on the Subject

Extremely low birth weight, extremely preterm infants experience high rates of mortality and critical illness. However, the prevalence, severity, and duration of organ dysfunction are not routinely reported.

What This Study Adds to the Field

The neonatal sequential organ failure assessment score, an adaptation of the sequential organ failure assessment and pediatric sequential organ failure assessment, is an operational definition of organ dysfunction applicable to preterm infants that successfully captured dynamic organ dysfunction kinetics among extremely preterm infants with adverse outcomes and identified those at the highest risk of mortality.

In adult and pediatric ICUs, severity of illness scores such as the acute physiology and chronic health evaluation, mortality prediction model, pediatric logistic organ dysfunction, sequential organ failure assessment (SOFA), and pediatric SOFA (pSOFA) are used to classify patients, describe kinetics of disease, and characterize therapeutic responses, complications, or changes in disease trajectory (1–5). Many of these scoring systems have been successfully integrated into the electronic health record (EHR) and are used to trigger best practice alerts or as clinical decision support tools, which are then used for risk stratification, standardization, benchmarking, research, and counseling families (1, 2).

There are numerous neonatal mortality risk assessment scores including the clinical risk index for babies, score for neonatal acute physiology, and neonatal therapeutic intervention scoring system, which use antenatal, perinatal, and laboratory data for early-life prognostication (6–8). Unlike their adult and pediatric counterparts, neonatal mortality risk assessment scores are neither commonly used in clinical practice nor included as standard in clinical research (9–12). Limited use of these metrics may reflect cumbersome data collection and limited validity after 24 hours from patient admission (13). The Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) extremely preterm birth outcomes tool (https://www.nichd.nih.gov/research/supported/EPBO/use) uses five variables known at birth (two continuous: gestational age [GA] and birth weight [BW]; and three binary: singleton birth, sex, and antenatal steroids) to generate a population-based range of risk for in-hospital mortality and neurodevelopmental outcomes among survivors at discharge for infants 401–1,000 g and less than 26 weeks’ gestation. The NICHD neonatal bronchopulmonary dysplasia (BPD) outcome estimator (https://neonatal.rti.org/index.cfm?fuseaction=BPDCalculator.start) incorporates seven variables (four at birth [GA, BW, sex, and race/ethnicity] and three evolving measures at a specific time point, including day of life [1, 3, 7, 14, 21, or 28], mode of respiratory support, and Fi_O2) to predict BPD risk for infants at 23–30 weeks’ gestation with BW of 501–1,249 g (14). Through integration of the five birth variables from the extremely preterm birth outcomes tool with the presence of intraventricular hemorrhage, respiratory support, and infection incidence at two discrete time points (on Day 7 of life, Day 28 of life, or 36 wk postmenstrual age), the NICHD outcomes trajectory tool (https://neonatal.rti.org/index.cfm?fuseaction=OTEstimator.start) predicts death or death/neurodevelopmental impairment at 18 to 22 months for infants of 22–32 weeks’ gestation and 401–1,000 g BW (15). Collectively, these tools demonstrate that individual mortality risk and poor outcomes in this population are strongly impacted by the hospital course (14–16), which underscores the need to objectively characterize illness severity frequently and longitudinally in neonates.

Relative to all other ICUs, neonatal ICU (NICU) patients are unique because they are routinely admitted not only because of critical illness, but also physiologic immaturity that must resolve before discharge home. As a consequence, NICU stays often last weeks to months (5, 17–21). The inability to characterize longitudinal severity of illness spanning the entire NICU hospitalization results in BW and GA becoming the predominant metric to convey disease severity. Although important for population-level outcomes and overall risk, BW and GA cannot convey patient-level or disease-level illness severity, especially over a protracted hospitalization. Extremely premature (<29 wk), extremely low birth weight (ELBW) (<1,000 g) infants 1) experience high mortality despite advances in perinatal and neonatal care, 2) are at high risk for adverse outcomes and organ dysfunction, 3) are admitted to the NICU because of profound physiologic immaturity, and 4) experience both new and evolving pathology during the months-long birth hospitalization.

Similar to the adult SOFA and pSOFA, which were developed specifically to characterize patients with sepsis (5, 22), the neonatal SOFA (nSOFA) was developed and validated as an operational definition of organ dysfunction associated with mortality risk in preterm infants with late-onset infection that was subsequently validated for all-cause mortality in a multicenter study of 20,152 NICU patients (23–25). These studies measured the utility of the nSOFA score for mortality at selected time points within a discrete window relative to sepsis evaluation or the maximum single score within the first 28 days of life. We aimed to quantify organ dysfunction from birth to death or discharge via nSOFA score calculation at hourly intervals among inborn infants of ELBW and less than 29 weeks’ gestation to measure the kinetics that occurred by GA group and time and among patients with and without adverse outcomes.

Methods

This was a single-center, retrospective, observational cohort study approved by the University of Florida Institutional Review Board prior to the collection of any data. This study was deemed exempt from patient consent by the University of Florida Institutional Review Board because it posed minimal risk and is secondary research use of identifiable private information for which consent is not required. All clinical data in the EHR for all infants of ELBW and less than 29 weeks’ gestation admitted to the University of Florida Health level IV NICU between January 2012 and January 2020 were extracted and deposited into an integrated data repository for this work. Infants born at outside hospitals, those who survived less than 12 hours, those with confirmed severe congenital anomalies, or those who completed less than 22 weeks’ gestation were excluded as previously published (26). Clinical variables that were unavailable in the integrated data repository were collected via chart review. Average risk for in-hospital mortality was calculated using the NICHD neonatal research network (NRN) extremely preterm birth outcomes tool (https://www.nichd.nih.gov/research/supported/EPBO/use) for infants of 401–1,000 g and less than 26 weeks’ gestation and reported as “NRN calculator average survival.”

Definitions

Demographic variables and outcomes were defined as previously reported (17). Pregnancy-induced hypertension was a maternal diagnosis of preeclampsia or gestational hypertension. Preterm prolonged rupture of membranes was rupture more than 18 hours before preterm delivery. Preterm labor was labor before 37 weeks. Antenatal steroids was the receipt of steroids at any time prior to delivery. Chorioamnionitis was histologic evidence of chorioamnionitis or funisitis. Small for gestational age was defined as previously reported (27). All-cause neonatal mortality, death due to any cause after the first 12 hours of life, was reported. Prolonged early antibiotics was at least 5 days of parenteral broad-spectrum antimicrobial treatment started in the first 3 days of life. Spontaneous intestinal perforation (SIP) was intestinal perforation at less than 7 days of life without evidence of necrotizing enterocolitis (NEC). Bacteremia was stratified by the timing of onset after birth (early-onset: ⩽3 days of life; late-onset >3 days of life). Severe intraventricular hemorrhage (SIVH) was grade 3–4 IVH (28). NEC was modified Bell’s stage at least 2 (29). BPD was the need for respiratory support at 36 weeks’ postmenstrual age (30). Severe retinopathy of prematurity (SROP) was unilateral or bilateral ROP in infants who received laser therapy or bevacizumab in at least one eye prior to discharge. The maximum vasoactive-inotropic score (VIS^max) was calculated as previously described (26).

Neonatal Sequential Organ Failure Assessment

The nSOFA score ranges from 0–15 and is calculated using the presence of mechanical ventilation and Sp_O2:Fi_O2 ratio for respiratory dysfunction, presence of vasoactive medications and/or corticosteroids for cardiovascular dysfunction, and platelet count for hematologic dysfunction (Table 1) (23–25). Scores were derived for each patient every hour beginning at admission until death or discharge.

Table 1.

Neonatal Sequential Organ Failure Assessment Score

Respiratory score		0	2	4	6	8
Criteria	Not intubated OR Intubated, SpO2/FiO2⩾300	Intubated, SpO2/FiO2 < 300	Intubated, SpO2/FiO2 < 200	Intubated, SpO2/FiO2 < 150	Intubated, SpO2/FiO2 < 100
Cardiovascular score	0	1	2	3	4
Criteria	No inotropes AND No systemic steroids	No inotropes AND Systemic steroid treatment	One inotrope AND No systemic steroids	Two or more inotropes OR One inotrope AND systemic steroid treatment	Two or more inotropes AND Systemic steroid treatment
Hematologic score	0	1	2	3
Criteria	Platelet count* ⩾150 × 103	Platelet count 100–149 × 103	Platelet count <100 × 103	Platelet count <50 × 103

^*Most recent measurement.

Analytical Methods

Tests of normality were performed on all continuous data. Nonparametric continuous variables were summarized as medians with quartiles (25th and 75th percentiles). Categorical variables were presented as percentages. For descriptive data comparisons, we used the Kruskal-Wallis test for continuous data with Dunn’s multiple comparisons test and the χ² test for categorical data. The threshold for statistical significance was less than 0.05 for two-sided tests. Calculation of hourly nSOFA scores was performed with R version 3.6.2 (R Foundation for Statistical Computing) using the raw data from the integrated data repository with random audits to ensure score accuracy. When calculation of more than one nSOFA score was possible within a single hour for a single patient, the maximum nSOFA score (nSOFA^max) based on simultaneously calculated nSOFA components was used. The area under the receiver operating characteristics curve (AUC) for mortality using the maximum nSOFA^max was calculated. Infants who survived the time interval of interest but died later were counted as survivors for time interval mortality-based AUCs. Likelihood ratios for mortality based on nSOFA^max scores were calculated for each GA group. Spearman’s rank correlation coefficient was calculated with a two-tailed P value and 95% confidence intervals, and a correlation matrix was generated. Row means of timed nSOFA scores with 95% confidence intervals were calculated from raw hourly nSOFA values for each patient. Cumulative nSOFA scores were the sum of the total or component nSOFA scores during the interval described. The average weekly sum of nSOFA scores over the first 28 days of life was defined as the 28-day nSOFA index (nSOFA^index). Among survivors with or without an adverse outcome, nSOFA scores were compared using repeated-measures ANOVA. Linear regression was performed to measure the relationship between the nSOFA^index and the sum of adverse outcomes among survivors. All analyses were performed using GraphPad Prism version 9.

Results

Patients

We identified 559 infants of ELBW and less than 29 weeks’ gestation during the study period. After excluding those who did not meet our inclusion criteria (89 were outborn, 11 died after <12 h of life, 7 had severe congenital anomalies, 7 died in the delivery room, 1 transferred to an outside hospital at 24 d of life, and 8 remained hospitalized at the time of data extraction), 436 infants were included and categorized into six GA groups for comparison of demographics and common outcomes of interest (Table 2). Significant and expected differences in the frequency of maternal risk factors for extremely preterm birth, including race, pregnancy-induced hypertension, preterm prolonged rupture of membranes, preterm labor, chorioamnionitis, and mode of delivery, were present between the GA groups. Based on study design, expected differences between groups were present for continuous variables, including BW, GA, 5-minute Apgar, and length of stay, as well as the frequency of categorical variables, including small for gestational age, prolonged early antibiotics, SIVH, SIP, late-onset bacteremia, NEC, ROP, BPD, and death.

Table 2.

Cohort Characteristics

	⩽23 wk (n = 63)	24 wk (n = 84)	25 wk (n = 84)	26 wk (n = 89)	27 wk (n = 61)	28 wk (n = 55)	P Value
Maternal
Age, median (quartiles)	28 (23–33)	26 (23–33)	26 (23–33)	27 (23–33)	29 (23–33)	26 (23–33)	0.34
Race, n (%)							0.04
Black	30 (48)	43 (51)	36 (43)	33 (37)	18 (30)	30 (55)
White	27 (43)	30 (36)	40 (48)	37 (42)	33 (54)	22 (40)
Other	6 (9)	11 (13)	8 (9)	19 (21)	10 (16)	3 (5)
Pregnancy-induced hypertension, n (%)	12 (19)	15 (18)	24 (29)	23 (26)	30 (49)	30 (55)	<0.001
PPROM, n (%)	23 (37)	35 (42)	29 (35)	37 (42)	14 (23)	9 (16)	0.008
Preterm labor, n (%)	48 (76)	63 (75)	58 (69)	62 (70)	28 (46)	14 (25)	<0.001
Antenatal steroids*, n (%)	56 (89)	81 (96)	78 (93)	83 (93)	57 (93)	54 (98)	0.35
Vaginal delivery, n (%)	34 (54)	36 (43)	27 (32)	29 (33)	17 (28)	6 (11)	<0.001
Chorioamnionitis†, n (%)	39/61 (64)	51/82 (62)	47/83 (57)	47/88 (53)	23/60 (38)	11/55 (20)	<0.001
Neonatal
GA, median (quartiles)	23.3 (23.1–23.6)	24.4 (24.1–24.7)	25.3 (25–25.7)	26.3 (26.1–26.6)	27.3 (27–27.6)	28.3 (28–28.6)	<0.001
BW, median (quartiles)	550 (500–610)	625 (531–690)	717 (629–822)	830 (740–908)	870 (765–943)	860 (730–916)	<0.001
SGA, n (%)	5 (8)	13 (16)	9 (11)	9 (10)	7 (11)	20 (36)	<0.001
Male, n (%)	30 (48)	40 (48)	45 (54)	56 (63)	25 (41)	31 (56)	0.12
5 min Apgar score, median (quartiles)	4 (3–6)	6 (3–7)	6 (4–7)	6 (4–7)	7 (5–8)	7 (6–8)	<0.001
Early bacteremia, n (%)	2 (3)	4 (5)	3 (4)	1 (1)	1 (2)	0	0.48
SIP, n (%)	12 (19)	7 (8)	6 (7)	3 (3)	0	2 (4)	<0.001
SIVH‡, n (%)	34/60 (57)	23/81 (28)	15/81 (19)	12/85 (14)	1/61 (2)	4/55 (7)	<0.001
Prolonged early antibiotics, n (%)	53 (84)	65 (77)	55 (65)	63 (71)	29 (48)	26 (47)	<0.001
Late bacteremia§, n (%)	27/48 (56)	21/73 (29)	22/76 (29)	26/86 (30)	6/60 (10)	3/55 (5)	<0.001
NEC, n (%)	8 (13)	10 (12)	13 (15)	9 (10)	6 (10)	3 (5)	0.58
ROP (any stage)‖, n (%)	24/24** (100)	52/57 (91)	52/70 (74)	50/81 (62)	23/59 (39)	24/52 (46)	<0.001
SROP‖, n (%)	14/24** (58)	21/57 (37)	17/70 (24)	18/81 (22)	3/59 (5)	3/52 (6)	<0.001
BPD‖ (O2 at 36 wk), n (%)	19/24** (79)	35/57 (61)	45/70 (64)	49/81 (60)	28/59 (47)	22/52 (42)	0.02
BPD-DC on O2‖, n (%)	14/24** (58)	25/57 (44)	26/70 (37)	31/81 (38)	18/59 (31)	11/52 (21)	0.03
Death, n (%)	38 (60)	27 (32)	14 (17)	8 (9)	2 (3)	3 (5)	<0.001
Length of stay‖, median (quartiles)	128** (116–156)	120 (105–145)	100 (88–118)	96 (85–115)	82 (73–92)	77 (64–99)	<0.001

Definition of abbreviations: BPD = bronchopulmonary dysplasia; BPD-DC on O₂ = bronchopulmonary dyplasia discharged on oxygen; BW = body weight; GA = gestational age; NEC = necrotizing enterocolitis; PPROM = preterm premature rupture of the membranes; ROP = retinopathy of prematurity; SGA = small for gestational age; SIP = spontaneous intestinal perforation; SIVH = severe intraventricular hemorrhage; SROP = severe ROP.

^*Any exposure.

^† Missing data on 6 patients.

^‡ Missing data on 13 patients.

^§ Among survivors >72 h.

^‖ Among survivors to discharge.

^**One infant was discharged to an outside hospital prior to completion of 36 wk and was excluded from these calculations.

Mortality and the nSOFA score

Ninety-two (21%) patients died; 42 (46%) were female and the median GA was 24 (interquartile range [IQR], 23–25) weeks, median BW was 550 (IQR, 485–665) g, and median time of death was 5 (IQR, 2–14) days. nSOFA scores (n = 891,643) and components (n = 2,673,129) were derived for each patient (n = 436) for each hour of life beginning at admission until death or discharge. Group total nSOFA score and component nSOFA score profiles were distinguishable between those who died versus those who survived to discharge (Figures 1A–1C and Figure E1 in the online supplement).

Figure 1.

Longitudinal neonatal sequential organ failure assessment (nSOFA) score trajectory in the neonatal period by survival group. (A) Consecutive q1-hour nSOFA score mean (internal line) and 95% confidence intervals (surrounding band) among all survivors to discharge (n = 344) and all nonsurvivors (n = 92) over the first 4 weeks after birth. (B) Consecutive q1 hour nSOFA score mean (internal line) and 95% confidence intervals (surrounding band) among survivors (all or only infants ⩽24 wk gestation at birth; the median birth gestational age among all who died) to discharge (n = 344) and nonsurvivors who died at no more than 120 hours (50th percentile of mortality timing; n = 49). (C) Consecutive q1 hour nSOFA score mean (internal line) and 95% confidence intervals (surrounding band) among survivors (all or only infants ⩽24 wk gestation at birth) to discharge (n = 344) and nonsurvivors who died after 120 hours (50th percentile of mortality timing; n = 43).

Overall, the discrimination for mortality by the nSOFA^max was excellent (AUC, 0.91; 95% CI, 0.88–0.94) (Table E1). Similarly, discrimination in each quartile of time when death occurred was excellent (range of AUCs, 0.90–0.96). Discrimination between all nonsurvivors and survivors within specific GA groups in each quartile of time when mortality occurred were very good to excellent (range of AUCs from each GA group, 0.83–0.99), except the comparison of all nonsurvivors to infants of no more than 23 weeks’ gestation in the 49–120-hour quartile (0.69; 0.56–0.82). Likelihood ratios for mortality were proportional to the nSOFA^max score cutoff in each GA group and in each postnatal mortality timing quartile (Figure E2).

A cumulative measure of hourly nSOFA scores in the first 24 hours of life revealed significant differences between survivors and nonsurvivors (⩽5 d): 1) 6 hours after birth (median for >24-week survivors, 2 [IQR, 0–10] vs. died ⩽5 days 14 [IQR, 5–25]; P < 0.001); 2) 6 hours for ⩽24-week survivors (6 [IQR, 0–13] vs. died ⩽5 days 14 [IQR, 5–25]; P < 0.05); 3) most pronounced at 24 hours of life (>24-week survivors 12 [IQR, 0–38] vs. died ⩽5 days 92 [IQR, 48–177]; P < 0.001) (Figure 2 and Figure E3).

Figure 2.

The cumulative neonatal sequential organ failure assessment (nSOFA) score on first day of life by survival group. The cumulative nSOFA score among survivors (⩽24 wk gestation at birth and >24 wk gestation at birth) and nonsurvivors (median gestational age 24 wk; median timing of death 5 d). Violin plots with all patient values including median and quartiles are shown. No mortality occurred before 12 hours of life based on study design. Comparisons were made by Kruskal-Wallis. Maximum possible nSOFA score was 15 for any hour. Maximum possible cumulative nSOFA was 360. *P < 0.05 and **P < 0.001.

Impact of GA on nSOFA Profiles and nSOFA^max Scores

We measured the extent and duration of organ dysfunction from birth to 12 weeks of age hourly (n = 693,504 nSOFA scores) among survivors (n = 344) classified by GA group (Figure 3). Weekly comparisons of nSOFA scores between survivors in all GA groups over the first 4 weeks of life revealed minimal but statistically significant sources of score variations, including time (percent variation by week: 1%, 0.1%, 1%, and 0.2%) and the interaction between time and GA (1%, 1%, 1%, and 0.4%); all P < 0.001. GA contributed to nSOFA score variation (7.5%, 22%, 24%, and 22%; all P < 0.001). However, in all weeks examined, the majority of the nSOFA score variation between GA groups occurred at the individual patient level (54%, 56%, 51%, and 59%; all P < 0.001).

Figure 3.

Longitudinal neonatal sequential organ failure assessment (nSOFA) score trajectory among survivors in the first 12 weeks of life by gestational age group. Consecutive q1 hour neonatal sequential organ failure assessment score mean (internal line) and 95% confidence intervals (surrounding band) among survivors to discharge over the first 12 weeks of life by birth gestational age (A) 23 or fewer weeks, (B) 24 weeks, (C) 25 weeks, (D) 26 weeks, (E) 27 weeks, and (F) 28 weeks.

We measured the categorical distribution of nSOFA^max scores among survivors overall, for each GA group, and within the relevant quartiles when mortality occurred (Table 3). The most common nSOFA^max score category at each time point was 0–3 (range, 47–55%) and the least common was 12 or more (range 4–5%). Differences within a time interval between categorical score representation among GA groups were found for each time interval examined (P < 0.001). Differences in categorical score representation within GA groups across time intervals were found for 23, 24, and 27-week groups (all P = 0.05), but not 25-, 26-, or 28-week groups.

Table 3.

Characterization of Survivors by Maximum Neonatal Sequential Organ Failure Assessment Score

	0–3	4–7	8–11	⩾12	Death in Interval, n (%)
0–48 h
⩽23	10 (19%)	22 (42%)	14 (26%)	7 (13%)	10 (26%)
24	22 (28)	31 (40%)	21 (27%)	5 (6%)	5 (19%)
25	37 (47%)	28 (36%)	10 (13%)	3 (4%)	6 (43%)
26	51 (59%)	14 (16%)	21 (24%)	—	3 (38%)
27	34 (57%)	18 (30%)	7 (12%)	1 (2%)	1 (50%)
28	41 (75%)	6 (11%)	8 (15%)	—	—
Sum	195 (47%)	119 (29%)	81 (20%)	16 (4%)	25 (27%)
49–120 h
⩽23	7 (17%)	12 (29%)	15 (36%)	8 (19%)	11 (29%)
24	29 (43%)	30 (44%)	7 (10%)	2 (3%)	11 (41%)
25	38 (50%)	24 (32%)	11 (14%)	3 (4%)	2 (14%)
26	55 (64%)	20 (23%)	10 (12%)	1 (1%)	—
27	44 (73%)	12 (20%)	3 (5%)	1 (2%)	—
28	40 (73%)	11 (20%)	2 (4%)	2 (4%)	—
Sum	213 (55%)	109 (28%)	52 (12%)	17 (4%)	24 (26%)
121–336 h
⩽23	1 (3%)	9 (25%)	15 (42%)	11 (31%)	6 (16%)
24	14 (22%)	22 (34%)	25 (39%)	3 (5%)	4 (15%)
25	26 (36%)	24 (33%)	20 (28%)	2 (3%)	4 (29%)
26	55 (66%)	14 (17%)	11 (13%)	3 (4%)	3 (38%)
27	50 (85%)	5 (8%)	4 (7%)	—	1 (50%)
28	46 (87%)	5 (9%)	2 (4%)	—	2 (67%)
Sum	192 (52%)	79 (22%)	77 (21%)	19 (5%)	20 (22%)
⩾337 h
⩽23	2 (8%)	3 (12%)	15 (60%)	5 (20%)	11 (29%)
24	12 (21%)	17 (30%)	25 (44%)	3 (5%)	7 (26%)
25	26 (37%)	19 (27%)	21 (30%)	4 (6%)	2 (14%)
26	50 (62%)	11 (14%)	19 (23%)	1 (1%)	2 (25%)
27	52 (88%)	3 (5%)	4 (7%)	—	—
28	42 (81%)	5 (10%)	5 (10%)	—	1 (33%)
Sum	184 (53%)	58 (17%)	89 (26%)	13 (4%)	23 (25%)

Longitudinal nSOFA Profiling by Pathology

A group profile comparison over the first week of life between all patients with SIVH to those without revealed minimal but significant sources of nSOFA score variation over time (0.96%; P = 0.001) as well as the interaction between time and SIVH (0.36%; P = 0.001). SIVH contributed to nSOFA score variation (6.78%; P = 0.001) (Figure 4A). However, the majority of the nSOFA score variation between those with and without a diagnosis of SIVH occurred at the individual patient level (60.8%; P = 0.001). Restricted comparisons performed between survivors of no more than 24 weeks’ gestation with SIVH or no more than grade 2 IVH revealed similar results in nSOFA score variation over time (1.55%; P = 0.001), the interaction between time and SIVH (0.43%; P = 0.06), SIVH (4.62%, P = 0.005), and the individual patient level (56.9%; P = 0.001) (Figure 4B).

Figure 4.

Longitudinal neonatal sequential organ failure assessment (nSOFA) score trajectory among survivors in the neonatal period by severe intraventricular hemorrhage group. (A) Consecutive q1 hour nSOFA score mean (internal line) and 95% confidence intervals (surrounding band) over the first week of life among survivors with severe intraventricular hemorrhage (grade 3–4; SIVH) and those without any intraventricular hemorrhage. (B) Consecutive q1 hour nSOFA score mean (internal line) and 95% confidence intervals (surrounding band) over the first week of life among survivors born at 24 or fewer weeks’ gestation with severe intraventricular hemorrhage (grade 3–4; SIVH) and those without SIVH (grade ⩽2). G2 = grade 2; IVH = intraventricular hemhorrhage.

Weekly comparisons over the first 4 weeks of life between those with BPD who were discharged home on oxygen (BPD-DC on O₂) and those without BPD revealed minimal but significant nSOFA score variation over time (percent variation by week: 0.5%, 0.1%, 0.3%, and 0.1%), as well as the interaction between time and BPD-DC on O₂ (0.5%-NS, 0.1%, 0.1%-NS, and 0.1%; all P < 0.001 except NS [not significant]) (Figure 5A). BPD-DC on O₂ contributed to nSOFA score variation (7.4%, 13%, 15%, and 13%); however, in all weeks examined, the majority of the nSOFA score variation between those with and without BPD-DC on O₂ occurred at the individual patient level (54%, 66%, 63%, and 69%; all P < 0.001). Comparisons restricted to infants of no more than 24 weeks’ gestation showed similar contributions to nSOFA score variation at the patient level for all 4 weeks examined (56%, 54%, 52%, and 67%; all P = 0.001) and revealed a group peak nSOFA score in the second week of life for those with BPD-DC on O₂ (Figure 5B). In contrast, infants of more than 24 weeks’ gestation with BPD-DC on O₂ showed a group peak nSOFA score in the first week of life (Figure 5C).

Figure 5.

Longitudinal neonatal sequential organ failure assessment (nSOFA) score trajectory in the neonatal period by bronchopulmonary dysplasia with discharge on oxygen (BPD-Discharged on O₂) group. (A) Consecutive q1 hour nSOFA score mean (internal line) and 95% confidence intervals (surrounding band) over the first week of life among survivors with BPD-Discharged on O₂ and those without BPD. (B) Consecutive q1 hour nSOFA score mean (internal line) and 95% confidence intervals (surrounding band) over the first week of life among patients born at 24 or fewer weeks’ gestation with BPD-DC on O₂ and those without BPD. (C) Consecutive q1 hour nSOFA score mean (internal line) and 95% confidence intervals (surrounding band) over the first week of life among patients born at more than 24 weeks’ gestation with BPD-DC on O₂ and those without BPD.

Weekly comparisons over the first 4 weeks of life between those with SROP and those without SROP revealed minimal but significant nSOFA score variation over time (percent variation by week: 0.6%, 0.4%, 0.3%, and 0.2%) and the interaction between time and SROP (0.4%, 0.2%, 0.2%, and 0.1%-NS; all P = 0.001) (Figure 6A). SROP contributed to nSOFA score variation (13%, 16%, 16%, and 14%); however, in all weeks, the majority of nSOFA score variation between those with and without SROP occurred at the individual patient level (43%, 61%, 60%, and 65%; all P = 0.001). Comparisons restricted to infants of no more than 24 weeks’ gestation showed similar contributions to nSOFA score variation at the individual patient level for all 4 weeks examined (44%, 65%, 63%, and 71%; all P = 0.001) (Figure 6B).

Figure 6.

Longitudinal neonatal sequential organ failure assessment (nSOFA) score trajectory in the neonatal period by severe retinopathy of prematurity (ROP) group. (A) Consecutive q1 hour nSOFA score mean (internal line) and 95% confidence intervals (surrounding band) over the first week of life among all survivors with severe ROP (received intervention prior to discharge) and those without ROP. (B) Consecutive q1 hour nSOFA score mean (internal line) and 95% confidence intervals (surrounding band) over the first week of life among all survivors with severe ROP (received intervention prior to discharge; median gestational age 25 wk) and those without ROP who completed 24–25 weeks’ gestation at birth (no surviving infant at 23 or fewer wk in the cohort was without a diagnosis of ROP).

Outcomes and nSOFA^index

Expected positive correlations with GA included BW (0.56) and NRN calculator average survival (0.77); negative correlations included ROP (−0.40), length of stay (−0.60), nSOFA^max (−0.46), and 28-day nSOFA^index (−0.52; all P < 0.001) (Figure E4). The 28-day nSOFA^index was positively associated with length of stay (0.60), VIS^max (0.55), nSOFA^max (0.84), and BPD (0.42), and negatively associated with 5-minute Apgar (−0.27), NRN calculator average survival (−0.46), and BW or GA (−0.52). The sum of adverse outcomes (SIVH, SROP/ROP-I, BPD-DC on O₂, and infection [early-onset bacteremia, late-onset bacteremia, SIP, and NEC]) at discharge among survivors was inversely proportional to GA, and BPD-DC on O₂ was the most common (Figures E5A and E5B). The 28-day nSOFA^index among survivors was inversely proportional to GA (maximum, ⩽23 wk; median, 590; IQR, 348–689; and minimum, 27 wk; median, 36; IQR, 1–97) (Figure 7A) and associated with the sum of adverse outcomes at discharge (r² = 0.30; P < 0.001) (Figure 7B).

Figure 7.

Relationship between the neonatal sequential organ failure assessment index (nSOFA^index), gestational age, and the number of adverse outcomes among survivors. (A) nSOFA^index values by birth gestational age among survivors. Histograms represent medians. Error bars represent interquartile ranges. Comparisons by Kruskal-Wallis. (B) The relationship between the nSOFA^index and the sum of adverse outcomes at an individual patient level (maximum of 7 including severe intraventricular hemorrhage [SIVH], infection [early-onset bacteremia, late-onset bacteremia, spontaneous intestinal perforation, or necrotizing enterocolitis], bronchopulmonary dysplasia with discharge on O₂, and severe retinopathy of prematurity). All patient nSOFA^index values are shown. *P < 0.05 and ** P < 0.001.

Discussion

Differences in the severity and duration of critical illness between patients of different GA during their NICU hospitalization are expected. The hourly nSOFA scores calculated in this cohort of comprehensively studied extremely premature infants with ELBW from birth to death or discharge, over 3.5 million (total and component) individualized illness severity metrics, demonstrate 1) excellent discriminatory ability for death within the first 24 hours of life, 2) the major contributor to nSOFA score variation occurred at the patient level, and 3) a direct association between nSOFA scores and major adverse outcomes, including mortality, SIVH, BPD, and SROP.

Individualized, objective, and reproducible measurement of death risk for extremely premature infants with ELBW remains relevant (15, 31, 32). Mortality risk assessment scores (score for neonatal acute physiology, clinical risk index for babies, and neonatal therapeutic intervention scoring system) that utilized the most severe physiologic derangements exhibited, most frequently on the first day of admission, have demonstrated utility, to predict all-cause in-hospital mortality among NICU patients (33). However, there has been limited integration of these specific measures of illness severity into clinical decision support. The average length of stay for extremely premature nonsurvivors with ELBW has increased from 5 to 17 days and highlights the need for mortality risk stratification to go beyond the first 24 hours of life, when many mortality risk assessment scores have been studied and validated (31, 32, 34). The NICHD extremely preterm birth outcomes tool offers a robustly validated estimate of in-hospital mortality and survivor neurodevelopmental impairment for the most premature infants based on data available at birth but does not integrate evolving measures of illness severity or the presence of pathology. Use of static population-level data coupled in the setting of significant interhospital variation in outcomes for extremely premature infants with ELBW further complicates neonatal prognostication (15, 31, 35–38).

The need for longitudinal severity of illness assessment in the NICU population is shown by the inverse relationship between prognostic ability and likelihood of survival (15, 31, 39). The majority of infants with ELBW now survive to discharge, underscoring the need for longitudinal metrics that distinguish not only mortality risk but also early identification of the risk of adverse outcomes among survivors (17). BW, GA, and the individual patient’s NICU course (including the severity of illness) strongly influence long-term outcomes, including BPD (14, 15, 40, 41). Continuous measure of a single variable such as heart rate variability has been used to predict mortality risk and neonatal outcomes (42), but this approach may require specialized equipment that could limit widespread integration into clinical decision support or research. NICHD web-based tools that integrate evolving measures at up to four time points during the hospitalization (birth, 7 days of life, 28 days of life, and 36 weeks postmenstrual age) such as NICHD outcomes trajectory tool or the BPD outcome estimator are highly accessible. However, the frequency of use, whether these are the optimal times for assessment, and impact on clinical decision support are all unknown. The nSOFA score showed good to excellent discrimination of mortality across centers, BW, and time points after admission in a multicenter cohort of 20,152 infants including all GAs (25). Similarly, discriminative ability of the nSOFA^max for all-cause in-hospital mortality in our restricted cohort of inborn infants with ELBW and less than 29 weeks’ gestation was excellent (0.91; 95% CI, 0.88–0.94), and was maintained when assessed by postnatal age when death occurred (range of AUCs, 0.90–0.96). Taken together, individualized nSOFA trajectories discriminated for in-hospital mortality during the entire hospitalization and were dependent upon simple, real-time data routinely collected for all patients. The capacity for repeated, cumulative measures of the nSOFA is likely to be advantageous over static assessments or evolving measures performed at limited (<5) and selected (weekly) postnatal ages (13).

Our longitudinal assessments using the nSOFA in this cohort confirmed that timing, duration, and severity of illness affects outcomes in NICU patients (15, 26, 35, 43, 44). However, these factors are neither routinely measured in daily practice nor reported in clinical studies, likely owing to an inability to meaningfully integrate specific automated measures of neonatal disease progression into the EHR. The nSOFA is easily amenable to automation and EHR integration. nSOFA-based patient-level automated measures of variables known to be associated with high mortality, including mechanical ventilation on high supplemental O₂, use of vasoactive-inotropic medications, and severe thrombocytopenia, have the potential to change practice through standardization and benchmarking, improve research, and illustrate pathophysiology that may be unknown. As with the SOFA and pSOFA, the most likely use of the nSOFA as a metric in the NICU is in a research setting. However, the potential for meaningful application of this clinical tool for early decision making or future targets for surrogate endpoints (e.g., reduction in peak or cumulative disease severity) disease is also promising. For example, we found differences in nSOFA group profile kinetics by GA cutoff among those who developed BPD-DC on O₂; comparisons restricted to infants of no more than 24 weeks’ gestation infants revealed a group peak nSOFA in the second week of life for those with BPD-DC on O₂. In contrast, infants of more than 24 weeks’ gestation with BPD-DC on O₂ showed a group peak nSOFA in the first week of life.

Though we found significant associations between nSOFA scores and adverse outcomes and death, the predictive accuracy found in this study may be affected by center-specific practices, such as the threshold for initiation and cessation for invasive mechanical ventilation (as well as oxygen saturation targets), inotropic/vasoactive medications, and the frequency of platelet assessments. However, substantial variations in care and practices as well as outcomes exist between academic centers, including NRN centers (36). Therefore, any metric that reflects the extent of clinical assessment and intervention may be potentially affected by variation in practice between providers at a single center and between centers. Although site-specific variations in care exist that may affect the generalizability of our results and the results of every multicenter study, the utility of the nSOFA as demonstrated in the multicenter studies to date was not substantially influenced by site-specific variations (23–25).

Limitations

This study has limitations inherent to all retrospective discovery-oriented analyses. However, to our knowledge, this is the most comprehensive, longitudinal quantification of organ dysfunction in the largest cohort of extremely preterm infants. Our objective was to determine the relationship between the nSOFA score and adverse outcomes in this unique population. We acknowledge the rates of BPD and SROP in this cohort were greater than some centers, though within range of NRN centers (9, 45, 46). We did not have adequate patients in this cohort with neurodevelopmental assessments (Bayley III) at 18–26 months adjusted age to facilitate a meaningful assessment of the relationship between the nSOFA and neurodevelopmental impairment (NDI). However, based on the well-described association of NDI and poor outcomes with clinical entities we show are reflected by the nSOFA (47), it would be expected that the nSOFA would be associated with NDI, and we plan to perform this analysis when sufficient data become available. Our goal was to measure organ dysfunction in each infant from birth to death or discharge. Accordingly, outborn infants with ELBW were not included, and results from that population may be different owing to a greater risk of adverse outcomes as compared with inborn infants (38, 48, 49). Although the longitudinal nSOFA group profiling revealed important differences in nSOFA scores and timing of score differences, these retrospective data cannot inform care. Instead, the potential for early recognition of adverse outcomes and likelihood of progression toward specific diseases during a specific time window is a practical first step for research to attenuate the risk in this population. Additional metrics including the vasoactive-inotropic score may improve the utility of the nSOFA (26). The current study does not address if the variables in the nSOFA are the best variables or are ideally weighted to predict the risk of all adverse outcomes; however, they are representative of infants we “know are sick.” We have established a 12-center consortium where the nSOFA score is automatically calculated and recorded in the EHR for every patient in the NICU every 15 minutes that will serve as sites for prospective evaluation and refinement of the nSOFA score.

Conclusions

We measured organ dysfunction hourly from birth to death or discharge in 436 extremely premature infants with ELBW using the nSOFA. The nSOFA is a useful tool that can readily be integrated into the EHR to make longitudinal, robust, and objective assessments of illness severity to facilitate NICU patient classification and comparison within and between centers.