This is a retrospective cohort study with data from the Marketplace for Medical Information in Intensive Care III (MIMIC-III) database, a longitudinal single-center database that contains information related to patients admitted to the intensive care unit at Beth Israel Deaconess Medical Center (Boston, MA, United States) between 2001 and 2012 (11). The database is maintained by the Massachusetts Institute of Technology (MIT) Computational Physiology Laboratory. The project was approved by the institutional review boards of MIT and the Beth Israel Deaconess Medical Center (BIDMC) and was granted a waiver of informed consent. After successfully completing the National Institutes of Health (NIH) web-based training course and the Protecting Human Research Participants examination (no. 41897755), permission was given to extract data from MIMIC-III.

The MIMIC-III database contains a total of 7,870 patients admitted to the NICU and patients who didn’t develop RDS were excluded. A total of 1,281 patients with RDS were included in the final study cohort, grouped according to nSOFA score level and survival status, respectively. The flow chart of patient screening is shown in Supplementary Figure 1. The diagnosis of RDS was carried out according to the Canadian neonatal network (CNN) (4) and defined as babies requiring respiratory support exceeding 24 h, intubation, surfactant administration (but not for meconium aspiration, pneumonia, or pulmonary hemorrhage), fraction of inspired oxygen (FiO₂) exceeding 25% for a minimum of 24 h or according to the international statistical classification of diseases and related health problem ninth revision (ICD-9) codes including code 769. For patients who were admitted to the NICU more than once, only the first NICU stay was included for analysis.

Clinical data were extracted from the MIMIC-III database for the first 24 hours of the patient’s NICU admission, including demographics, vital signs, laboratory tests, diagnostic codes, medications and survival data. Clinical data required to calculate the nSOFA score were also collected and included the receipt of intubation and mechanical ventilation, the FiO₂ to achieve the peripheral oxygen saturation (SpO₂), the platelet count and any requirement for glucocorticoid, inotropic, or vasoactive drugs. The SpO₂/FiO₂ were converted using the partial pressure of arterial oxygen (PaO₂)/FiO₂ ratio conversion where SpO₂/FiO₂ = 64 + 0.84 × (PaO₂/FiO₂) (12). Missing values in the nSOFA score were assumed to be in the normal range as in previous studies (13, 14) and if a variable was recorded more than once in the first 24 h, the most serious record was used. Comparison of baseline characteristics of patients with missing and non-missing SpO₂/FiO₂ is shown in Supplementary Table 1. Comorbidities identified based on documenting the ICD-9 codes included hemolytic disease of the newborn (HDN), congenital heart disease (CHD), acidosis, anemia, bleed, pneumonia, jaundice, and septicemia. Acute kidney injury (AKI) was defined according to the Kidney Disease: Improving Global Outcomes (KDIGO) guidelines as an increase in serum creatinine (Scr) of 0.3 mg/dL or more from baseline within 48 h (15).

The primary outcome of this study was classified as all-cause mortality. Inpatient mortality information was obtained from the Hospital Information System, and mortality information for discharged patients was obtained from the US Social Security Death Index.

The nSOFA score uses categorical scores with a total score range from 0 as best to 15 as worst to objectively describe (1) receipt of mechanical ventilation and oxygen to maintain a physiologic peripheral saturation which was scored 0 to 8; (2) inotropic or vasoactive drug support, including the use of corticosteroids for presumed adrenal insufficiency or catecholamine-resistant shock scored 0 to 4; and (3) the presence and severity of thrombocytopenia based on the most recent platelet measure scored 0 to 3 (9) (Table 1).

Values are presented as the means ± standard deviations or medians as interquartile ranges (IQRs) for continuous variables and categorical variables are presented as total numbers and percentages. Comparisons between groups were made using the Student’s t-test or the Mann-Whitney U test for continuous variables and χ²-test or Fisher’s exact test for categorical variables.

The area under the receiver operating characteristic curve (AUROC) was calculated to assess the accuracy of the nSOFA score and its subscales in predicting the primary outcome in patients with RDS. The cut-off values of the receiver operating characteristic (ROC) curve were identified by calculating the Youden index, which divided the high and low nSOFA groups. The Kaplan-Meier survival analysis were performed to evaluate the incidence rate of primary outcome events among groups according to various levels of the nSOFA score and discrepancies among groups were evaluated by log-rank test. Cox proportional hazards models were used to estimate the relationships between nSOFA score (per 1 score)/nSOFA groups (high nSOFA groups and low nSOFA groups) and primary outcomes. Characteristic variables with significant baseline differences or clinical significance were used as candidate predictors in the multivariate Cox regression model. Variables with more than 20% missing data were excluded from the analysis. The details of missing data are summarized in Supplementary Table 3.

The propensity score matching (PSM) was used to adjust for ethnicity, sex, and gestational age to ensure the comparability across of high and low nSOFA groups and between survival and non-survival groups in the analysis of baseline characteristics. Baseline characteristics of the original and matched cohorts were presented separately. Multivariate Cox regression analysis was also performed on the matching cohort. Subgroup analyses was used to investigate whether the risk was modified by sex and septicemia. All analyses were first unadjusted as model one, then adjusted for sex, ethnicity, gestational age, weight, and heart rate to get model two and finally for anemia, bleed, septicemia, and pulmonary surfactant to get model three.

All data analyses were performed using R software (version 4.0.4; R Foundation for Statistical Computing, Vienna, Austria) and SPSS statistical software (SPSS Statistics 24.0). Bilateral p-values < 0.05 for all analyses were considered statistically significant.

There were 1,281 patients with RDS who were analyzed between 2001 and 2012 of whom 733 (57.2%) were male and 1,127 (93.1%) were less than 37 weeks of gestational age. There were 40 (3.1%) patient deaths in this cohort. The optimal cut-off value for nSOFA score associated with death was defined as 2.5 and this was used to divide the cohort into low and high nSOFA groups.

There was a total of 347 (27.1%) patients with a high nSOFA score, of whom 37 (10.7%) died, compared with 3 (0.3%) of the 934 (72.9%) patients with a low nSOFA score. Patients in the high nSOFA group had more gestational age less than 28 weeks, smaller height, weight and head circumference, lower SpO₂/FiO₂, less urine output, additional complications such as acute kidney injury, bleeding and infection and had higher rates of ventilator and vasoactive drug use on the first day of admission (all p < 0.05). Baseline information for the original cohort and matched cohort of RDS patients with high and low nSOFA scores is shown in Table 2. Baseline characteristics of grouping according to the occurrence of follow-up deaths is shown in Supplementary Table 4.

Compared with survivors, non-survivors had higher nSOFA scores (6.20 vs. 1.58, p < 0.001) and subscales (respiratory score: 3.45 vs. 1.09; cardiovascular score: 2.45 vs. 0.38; hematologic score: 0.30 vs. 0.11). The mortality rate of RDS patients increased with increasing nSOFA score (0–2: 0.4%; 2–4: 1.6%; 4–6: 7.0%; 6–8: 12.2%; 8–10: 29.0%; 10–15: 42.1%). The area under curve (AUC) for the nSOFA score predicting death in patients with RDS was 0.89 (95% CI, 0.84–0.95, p < 0.001), which was higher than the predictive efficacy of the subscale [respiratory score: 0.80 (95% CI, 0.72–0.86, p < 0.001); cardiovascular score: 0.85 (95% CI, 0.79–0.91, p < 0.001); hematologic score: 0.55 (95% CI, 0.49–0.61, p = 0.690)] (Figure 1). The optimal cut-off value of the nSOFA score for predicting death in patients with RDS was 2.5, a sensitivity of 92.5% and a specificity of 75.0% (Supplementary Figure 2). We added gestational age to the original nSOFA score for scoring and analysis. The results showed that the AUC of the modified nSOFA score was 0.91 (95% CI, 0.86–0.95, p < 0.001), which was higher than that of the original nSOFA score and gestational age alone [0.73 (95% CI, 0.62–0.84, p < 0.001)].

The prognostic impact of nSOFA scores on patients who with RDS was studied. The Kaplan-Meier survival analysis curves for assessing the mortality between groups based on low and high nSOFA groups are shown in Figure 2. There was a statistically significant difference in mortality between low nSOFA and high nSOFA groups (log-rank p < 0.001). Univariate Cox proportional risk analysis showed that each unit increase in nSOFA scores was significantly associated with mortality (HR: 1.59, 95% CI: 1.46–1.73; p < 0.001). Even after adjusting for sex, ethnicity, gestational age, weight, heart rate, anemia, bleed, septicemia, and pulmonary surfactant, each unit score increase in nSOFA score remained significantly associated with mortality in patients with RDS (aHR: 1.48, 95% CI: 1.32–1.67; p < 0.001). Similarly, the high nSOFA group was significantly associated with higher mortality in RDS patients compared with the low nSOFA group (aHR: 19.35, 95% CI: 4.41–84.95; p < 0.001; Table 3).

Stratified and interaction analyses were used to evaluate whether the association differed by sex and septicemia. The results showed that the association between nSOFA score and mortality in RDS patients did not vary with sex and septicemia status (Supplementary Table 5). PSM did not change any of the conclusions (Table 3).