Factors associated with the increased incidence of necrotising enterocolitis in extremely preterm infants in Sweden between two population-based national cohorts (2004–2007 vs 2014–2016)

Objective To investigate potential risk factors behind the increased incidence of necrotising enterocolitis (NEC) in Swedish extremely preterm infants. Design Registry data from two population-based national cohorts were studied. NEC diagnoses (Bell stage ≥II) were validated against hospital records. Patients All liveborn infants <27 weeks of gestation 2004–2007 (n=704) and 2014–2016 (n=895) in Sweden. Main outcome measures NEC incidence. Results The validation process resulted in a 28% reduction of NEC cases but still confirmed a higher NEC incidence in the later epoch compared with the earlier (73/895 (8.2%) vs 27/704 (3.8%), p=0.001), while the composite of NEC or death was lower (244/895 (27.3%) vs 229/704 (32.5%), p=0.022). In a multivariable Cox regression model, censored for mortality, there was no significant difference in early NEC (0–7 days of life) between epochs (HR=0.9 (95% CI 0.5 to 1.9), p=0.9), but being born in the later epoch remained an independent risk factor for late NEC (>7 days) (HR=2.7 (95% CI 1.5 to 5.0), p=0.001). In propensity score analysis, a significant epoch difference in NEC incidence (12% vs 2.8%, p<0.001) was observed only in the tertile of infants at highest risk of NEC, where the 28-day mortality was lower in the later epoch (35% vs 50%, p=0.001). More NEC cases were diagnosed with intramural gas in the later epoch (33/73 (45.2%) vs 6/26 (23.1%), p=0.047). Conclusions The increase in NEC incidence between epochs was limited to cases occurring after 7 days of life and was partly explained by increased survival in the most extremely preterm infants. Misclassification of NEC is common.


INTRODUCTION
Necrotising enterocolitis (NEC) is a devastating disease in preterm infants, 1 2 with severe implications for the infant, family and healthcare system. 3 46][7] Survivors have an increased risk for neurodevelopmental impairment. 8 9Low gestational age (GA) is the primary risk factor for NEC. 7 10The NEC incidence usually remains invariable over time in populations with an unchanged mean GA. 6 The NEC incidence in Sweden tripled between 1987 and 2009, likely attributed to the increased survival of extremely preterm (EPT) infants. 11 The 1-year survival was significantly higher in the latter cohort, while the incidence of other neonatal morbidities was unchanged or

WHAT IS ALREADY KNOWN ON THIS TOPIC
⇒ Low gestational age is the primary risk factor for necrotising enterocolitis (NEC).⇒ A recent Swedish national cohort study of extremely preterm infants showed an increased NEC incidence over 10 years.⇒ During the same period, mortality and the incidence of other major neonatal morbidities decreased.

WHAT THIS STUDY ADDS
⇒ Survival of the most premature infants seems to be the major driving factor behind the increase in NEC incidence.⇒ Early and late NEC have different risk factors, and only late NEC has increased over time.⇒ The misclassification of NEC in healthcare records and registers remains a concern.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
⇒ Future studies are needed to explore how risk factors and early and late NEC pathogenesis differ.⇒ Future studies are needed to address how to prevent NEC in the most preterm infants.⇒ Healthcare providers and registers have to improve the classification and validation of the NEC diagnosis.

Original research
decreased, except for NEC, which increased significantly among 1-year survivors from 6% to 10%. 13 The aim of this study was to identify factors explaining the increased incidence of NEC in EPT infants in Sweden between 2004-2007 and 2014-2016.Our primary hypothesis was that improved survival in the most preterm infants would drive the rise in NEC incidence.Furthermore, we hypothesised that there might be an overestimation of NEC incidence due to misclassification in register data.Therefore, we decided first to perform a validation of the NEC diagnosis.

METHODS Participants
Epoch 1, the Extremely Preterm Infants in Sweden Study (EXPRESS) cohort (infants born April 2004 to March 2007), included 704 liveborn infants, of which 26 had a validated NEC diagnosis. 14Epoch 2, the EXPRESS 2 cohort (January 2014 to December 2016), included 895 liveborn infants.The two cohorts have previously been described in detail. 12 13During epoch 1, neonatal and perinatal data were prospectively collected by local investigators.During epoch 2, the Swedish Neonatal Quality Register 15 was used for the primary data collection.In a second step, mortality and major morbidity data were cross-checked against medical records. 13

NEC diagnosis validation
NEC was defined as Bell stage II or higher in both epochs. 16Clinical data, including macroscopic and biopsy results from surgery and autopsy, were used to confirm NEC diagnosis.Spontaneous intestinal perforation (SIP) was recorded as a separate entity.
NEC cases in epoch 1 were previously validated against hospital records. 14At that time, nine cases of suspected NEC were reclassified because they did not fulfil the X-ray criteria.These nine cases were revalidated in the present study to achieve a uniform NEC classification in the two cohorts, resulting in one of these cases being reclassified from no NEC to NEC.For epoch 2, a similar validation was performed using a predefined standardised protocol (online supplemental table S1).For each reported NEC case, medical records were scrutinised.Diagnostic criteria for NEC included: typical findings of NEC at surgery or autopsy, NEC confirmed by biopsy, suggestive abdominal symptoms together with intramural or portal gas on abdominal X-ray or ultrasound.In addition NEC treatment and diagnostic characteristics were collected.Four undiagnosed NEC cases were added to the original epoch 1 study after examining the

Original research
healthcare records of infants with significantly reduced enteral feeds. 14This was not feasible for epoch 2 due to a paucity of feeding data.
The NEC cases for epoch 1 were independently validated by a senior neonatologist and a resident in neonatology. 14In contrast, a senior neonatologist and coauthor independently validated the NEC cases for epoch 2 at each of the six study centres.
Background characteristics, mortality and morbidity have been previously described. 13Birth weight z-score was calculated from an intrauterine growth chart. 17Sepsis was confirmed by positive blood culture.Patent ductus arteriosus was diagnosed if requiring medical or surgical treatment.Retinopathy of prematurity (ROP) stages 3-5 was defined according to International Classification of Retinopathy of Prematurity-2, 18 intraventricular haemorrhage (IVH) grades 3-4 according to Papile et al 19 and severe bronchopulmonary dysplasia (BPD) as treatment with ≥30% oxygen at 36 weeks' postmenstrual age. 20

Missing data
Both data sets have previously been manually completed for missing data using hospital records. 12 13Data completeness was high, 98% for epoch 1 and 95% to 99% for epoch 2. 15 Due to the small amount of missing data, listwise deletion was used and no imputation was performed.

Statistics
Welch's unequal variance t-test was used to compare means, and the Wilcoxon rank-sum test was used for medians.The Pearson's χ 2 test was used to compare categorical variables.The logrank test was used for group comparison in the survival model.Cox regression was used to analyse the effects of risk factors on NEC while considering variable observation periods due to early mortality.
Propensity score stratification was used to determine if the increased NEC incidence was confined to infants with the highest risk of NEC.The score was created using a Cox regression model with NEC as the dependent variable.Perinatal risk factors for NEC were used as covariates for the Cox model: GA, birth weight z-score (SD), Apgar score at 5 min, caesarean delivery, prenatal corticosteroid treatment and multiple pregnancy.The propensity score was used to classify newborns into three risk groups and the rate of NEC occurrence was compared between epochs for each risk tertile.
Statistical analyses were conducted in R V.4.2.1 (R Project for Statistical Computing).

RESULTS
A total of 704 infants were included in epoch 1 and 895 in epoch 2, all liveborn at less than 27 weeks of gestation (table 1).
Significant risk factors from the univariate univariable Cox regression analyses (online supplemental table S2) were selected for a univariate multivariable model (table 2).The risk for NEC

Original research
was not proportional over time, so Cox regression could not be used for the entire period.There was a steep risk increase in NEC in epoch 2 in the survival analysis (figure 1), suggesting the possibility of different risk factors, hence the division into early NEC (0-7 days) and late NEC (>7 days).In the multivariable model, there was no significant difference in early NEC between epochs (HR=0.9(95% CI 0.5 to 1.9), p=0.9) (table 2), while the risk of late NEC was higher in epoch 2 (HR=2.7 (95% CI 1.5 to 5.0), p=0.001) (table 2).In addition to epoch, IVH remained an independent risk factor for late NEC.A low Apgar score at 5 min was the only significant risk factor for early NEC (table 2).Among NEC cases, GA and Apgar score at 5 min were lower in epoch 2 versus epoch 1 (online supplemental table S3), while birth weight, rate of small for gestational age or other neonatal outcomes did not differ (online supplemental table S3).The proportion of NEC infants diagnosed with intramural gas on X-ray was higher in epoch 2 (45.2% vs 23.1%, p=0.047), but there was no difference in macroscopic NEC (table 3).
NEC cases had lower GA and had a higher incidence of IVH, ROP stages 3-5 and sepsis compared with infants without NEC (online supplemental table S4).Mortality was also higher, NEC versus no NEC (37.0% vs 24.9%, p=0.007) (online supplemental table S4).
In the propensity score analysis, the NEC incidence was significantly higher in epoch 2 versus epoch 1 only in the tertile of infants with the highest risk of NEC (34/291 (12%) vs 6/217 (2.8%), p<0.001), while there were no significant differences in the medium and low-risk tertiles (table 4).In the tertile of high-risk infants, who had an average GA of 23.6 weeks, the 28-day mortality was significantly lower in epoch 2 (35% vs 50%, p<0.001) (table 4).

DISCUSSION
The present study explored possible factors underlying the increase in NEC incidence in Sweden between 2004-2007 and 2014-2016.Even though the Cox regression model, censored for mortality, showed a remaining epoch difference in the incidence of late NEC, the propensity score analysis showed that the increase in NEC was limited to the high-risk group with the lowest GAs.Furthermore, the mortality in the high-risk group was significantly lower in epoch 2, and the overall risk of the composite outcome NEC or death was lower in epoch 2 in the whole study population.This suggests that increased survival of the most immature infants was an important contributing factor to the increased incidence of NEC.This is supported by the lack of an expected increase in NEC incidence in the lowest gestational weeks in epoch 1 (online supplemental figure S3, panel A).
The NEC incidence among infants <27 weeks was 8.2% in epoch 2, which was more than twice as high as in epoch 1.[23][24]

Original research
Increasing survival of preterm infants has been shown to increase the incidence of NEC. 1 A Dutch cohort study described an increase in NEC incidence from 6.4% to 16% after implementing a new guideline for active treatment management of EPT infants. 25n contrast, two recent US multicentre cohort studies of EPT infants found lower NEC rates over time, despite increased survival. 26 27Improved neonatal care and increased use of breast milk feeding instead of formula were proposed explanations. 26irth in epoch 2 remained a significant risk factor for late NEC in the Cox regression model after adjustment for GA and IVH.However, the Cox regression did not address whether NEC incidence and mortality differed between epochs in the group of most vulnerable infants; therefore, a propensity score analysis was added.In the propensity score analysis, the increased NEC incidence in epoch 2 was restricted to the high-risk tertile, where the 28-day mortality was significantly lower, suggesting that early mortality in high-risk infants with the lowest GAs during epoch 1 may have contributed to the difference in NEC incidence.However, we cannot exclude that residual confounding could explain part of the increased NEC incidence.
In Sweden, preterm practices have changed gradually between 2004-2007 and 2014-2016.However, these changes did not theoretically increase the risk of NEC as the rate of antenatal steroid treatment was consistently high, infants were routinely fed mother's or donor milk and formula was rarely given before 34 weeks' postmenstrual age. 28 29Probiotics and human milkbased fortifiers were not used for NEC prevention in Sweden during 2004-2016. 30he increased risk of NEC observed between the epochs was restricted to those who developed NEC after the first 7 days of life.It has been shown that age at the onset of NEC increases in a non-linear way with decreasing GA at birth, 31 32 suggesting that different survival ratios between GA groups could be a factor.This is supported by our finding that the composite outcome of NEC or death was lower in epoch 2.
In our study, the risk factors for early NEC (first 7 days) differed from those for late NEC (after 7 days).A low Apgar score at 5 min was the sole significant risk factor for early NEC in the Cox regression model, while it was not a significant risk factor for late NEC.4][35][36] However, findings have been inconsistent for NEC, and causality has been difficult to prove. 33ince NEC arises via different pathogenic pathways, 37 our observation of different risk factors in early and late NEC cases may be important.
Sepsis and severe IVH were significantly associated with NEC.Sepsis could not be assessed as a risk factor for NEC since our data set did not allow us to determine the temporal relationship between the two diagnoses.Sepsis and IVH have been previously described as risk factors for NEC. 6 34 38iagnostic accuracy for NEC improved in epoch 2 versus epoch 1, which might contribute to the higher NEC incidence during epoch 2 across all GAs.Positive findings were more frequent among NEC infants for the diagnostic modalities X-ray, ultrasound and biopsy in epoch 2. Ultrasound was seldom or never used for NEC diagnosis during epoch 1 but was commonly used during epoch 2, which may have contributed to improved diagnostics and may have affected the difference in NEC between epochs.The laparotomy rate was high in both epochs compared with other cohorts of very low birthweight infants. 6 26isclassified NEC cases, especially overdiagnosis, were common in both epochs.After Bell stage I, SIP ranked second  NEC, necrotising enterocolitis.

Original research
and third for the most common reason for misclassification in epoch 1 and epoch 2, respectively.Misclassifying SIP as NEC in clinical databases is well described, 39 40 especially among EPT infants. 39 41 strength of this study was that both cohorts were population based, and data were prospectively collected.Both cohorts were rigorously validated. However, this did not explain the differences in NEC incidence in this study.Even though both cohorts were population based and included births during two 3-year periods, they were not adequately powered to assess all risk factors associated with NEC.A limitation of the study is that data on feeding were unavailable for epoch 2, which may have resulted in a slightly lower incidence of NEC in epoch 2.

CONCLUSION
NEC has increased significantly during the observed decade in Sweden.The increased NEC incidence is partly caused by increased survival among infants with the lowest GAs, but improved diagnostics and other factors cannot be excluded.The increase in NEC incidence was restricted to infants with onset after 7 days of age, and different risk factors were observed for early (0-7 days) and late (>7 days) NEC.

Figure 1
Figure 1 Kaplan-Meier curves displaying estimated probability for no necrotising enterocolitis (NEC) for both epochs.Each vertical tick mark in the curve indicates death.The curves are right censored at 100 days.Log-rank test was used for the p value.
denotes p<0.05.Data are shown as mean or numbers and proportions (%).The propensity score, which includes gestational age, birth weight z-score (SD), Apgar score at 5 min, caesarean delivery, prenatal corticosteroid treatment and multiple pregnancy, is divided into three risk tertiles: low, medium and high.Covariates, mortality and NEC incidence are shown for each epoch and tertile.*Welch's two-sample t-test; Pearson's χ 2 test.

Table 1
Background characteristics, mortality and morbidity in extremely preterm infants in epoch 1 and epoch 2

Table 2
Univariate multivariable HRs for early and late NEC NS 2.0 (1.2-3.2),p=0.008Cox regression for each outcome, censored for death for the composite of epoch 1 and epoch 2. Bold denotes p<0.05.Only significant covariates from the univariate univariable model are included (online supplemental table S1).NEC, necrotising enterocolitis; NS, not significant in univariate univariable model.

Table 3
Clinical characteristics of NEC cases

Table 4
Risk factors, NEC and mortality among tertiles of infants based on NEC risk propensity score