With advancements in perinatal medicine and the development of neonatal intensive care units (NICUs), neonatal mortality has declined in most parts of the world (1–3). The survival rates of extremely preterm infants have also increased from 76.0% in 2012 to 78.3% in 2018 (4–6). At the same time, the number of newborns requiring intensive care has increased, with NICU admissions rising from 6.4% in 2007 to 7.2% in 2018 (7, 8); therefore, prolonged length of stay (LOS) of these high-risk newborns admitted to the NICU has become a concern.

Prolonged LOS-NICU causes adverse effects on the newborn, the family, the medical staff, and even the hospital. For newborns, prolonged LOS-NICU exposes them longer to the hospital environments, including the noise, bright light, hospital-acquired infections, and so on, which would lead to a higher incidence of neonatal complications (9, 10). Some studies found that these exposures could affect the future development of newborns, such as developmental retardation, predisposition to chronic illness, impairment of cognitive function, and neurodevelopmental disorders (11–13). Prolonged LOS-NICU could also prevent the establishment of parent–newborn interactions and increase the cost of hospitalization (14, 15), which may cause severe emotional and financial stress to families (16). Moreover, parents are concerned about how long their newborn needs to stay in the NICU. Accurate information about the LOS may alleviate unnecessary anxiety for parents. However, it is often difficult for medical staff to make accurate predictions when parents consult with them. From the healthcare system perspective, a prolonged LOS-NICU could reduce the utilization rate of beds and exacerbate the problem of inadequate healthcare resources (17). To reduce unnecessary LOS and avoid prolonged LOS-NICU, we must determine the risk factors affecting the LOS-NICU of newborns.

Determining the risk factors for the LOS-NICU of newborns is conducive to improve the ability of predicting LOS-NICU accurately, which is critical for planning hospital resources, counseling families, stimulating quality improvement initiatives, and effectively avoiding prolonged LOS-NICU of newborns (18, 19). However, there is little evidence related to this issue. Moreover, most of the available evidence comes from observational studies, which may lead to selection bias. In addition, some important risk factors identified in single-center studies might be found only by chance because of the level of medical institutions [there are currently four categories of NICUs (20): Level I NICU: well newborn nursery, which offers regular nursery care for healthy newborns; Level II NICU: special care nursery, which cares for premature and sick newborns; Level III NICU: neonatal intensive care unit, which cares for seriously ill newborns; Level IV NICU: regional neonatal intensive care unit, which cares for the most critically ill newborns and babies), which might not apply to other hospitals. Therefore, the purpose of this review was to determine the risk factors for the LOS-NICU of newborns from multiple studies and to evaluate the evidence that currently exists.

The review followed the PRISMA reporting guidelines, a 27-item list designed to improve the reporting of systematic evaluations (21), and was registered with PROSPERO (registration number: CRD42022370357). All relevant analyses were based on previously published studies and did not require ethical approval or patient consent.

The literature search of the databases identified 4,713 potentially relevant articles, and 1,295 duplicates were excluded, leaving 3,418 articles to be screened for title and abstract. After the eligibility assessment, the full text of 102 articles was retrieved, of which 21 met the inclusion criteria. We searched the list of all the references included in the study and conducted a full text search of 10 articles, among which 2 original texts could not be found. According to the strict inclusion and exclusion criteria, two articles were determined to meet the requirements. Ultimately, 23 articles were identified for inclusion in this review. Study identification is summarized in Figure 1.

Of these studies, four were prospective studies (24–27) and the remaining 19 studies were retrospective (1, 9, 18, 28–43), of which 14 were multicenter studies (9, 24, 25, 27–29, 31, 33–35, 37–39, 42) and 9 were single-center studies (1, 18, 26, 30, 32, 36, 40, 41, 43). The sample size ranged from 191 to 7,60,037. Two studies with the largest sample sizes included 23,551 newborns (34) and 7,60,037 newborns (38). Six of these studies constructed clinical prediction models (9, 18, 27, 34, 42, 43). The basic characteristics of the included literature are given in Table 1.

Our study was the first systematic review of risk factors for the LOS-NICU of newborns. Among the 23 studies that we included, birth weight, gestational age, sepsis, NEC, BPD, and ROP were the critical risk factors that were most widely studied and consistently associated with LOS-NICU. The results of our systematic review provide an up-to-date comprehensive summary of the latest evidence. They will inform the development of interventions to reduce LOS-NICU and prevent prolonged LOS-NICU of newborns.

Our results were consistent with previous findings that birth weight and gestational age are the most important risk factors affecting LOS-NICU (18, 24–26, 30, 32, 40, 41). Although another study found no significant association between LOS-NICU and gestational age (37), this study only assessed preterm infants with birth weight less than 1,500 g in Israel, which may not be applicable to other neonatal populations. Prevention of prolonged LOS due to gestational age and low-birth-weight starts with a healthy pregnancy. Preterm parturition is a syndrome that is triggered by multiple factors. Clinicians and researchers play a key role in improving biochemical knowledge on preterm delivery, identifying risk factors, and developing interventions that can address this complex syndrome (44). The birth weight and gestational age, as inherent factors, have the advantage of being measured directly and objectively at birth. Therefore, some studies have used birth weight and gestational age to predict LOS-NICU (45–47). However, Bender et al. (18) and Hintz et al. (9) found that predictions of LOS-NICU of newborns would not be very accurate if only these inherent factors were considered and not combining them with other factors. Bender et al. established a predictive model of LOS-NICU including birth weight, gestational age, and two disease severity tools as predictors. The result showed that the addition of the first-week disease severity improved the prediction accuracy of LOS-NICU compared with including only birth weight and gestational age. In addition, male infants generally have a longer NICU stay than female infants (27, 31, 32, 34), and the LOS-NICU of newborns varies in ethnicity (27, 34, 42). However, there are currently no studies to explain the causes of these phenomena.

Congenital anomalies were often used as one of the exclusion criteria in some studies, while four of the other studies that considered congenital anomalies showed that it had a significant effect on length of stay (24, 28, 30, 37). Some congenital anomalies are unlikely to have an impact on LOS-NICU, while some severe anomalies or those requiring surgery may have a significant impact on LOS-NICU, such as gastroschisis. However, none of the studies that we included gave a clear definition of congenital anomalies, and there was no accepted list. In the available studies, we were unable to identify which specific congenital anomalies were the risk factors for LOS-NICU. Therefore, we should consider this issue in our future studies.

There is a controversy about whether antenatal treatment and maternal factors are risk factors for LOS-NICU. Five studies found that newborns delivered by the cesarean section were weaker, sicker, and born earlier than those born vaginally, leading to a longer LOS-NICU (31–34, 41). Nevertheless, three studies showed that they were not associated with a longer LOS-NICU despite the proportion of cesarean sections being high (30, 36, 40). They argued that cesarean sections are the result of unsatisfactory fetal conditions but not an individual confounding factor for LOS-NICU. Early evidence suggests that antenatal corticosteroids confer benefits on fetal lung maturation, and they are widely recommended for women at risk of preterm birth (48), but there is no certainty about their effect on LOS-NICU (27, 31, 34). Moreover, several maternal factors associated with LOS-NICU were revealed in a multicenter study in China (33), including primigravida, maternal hypertension, and cesarean section; however, these impacts were minimal. Therefore, more research is needed to determine the association between antenatal treatment or maternal factors and the LOS-NICU of newborns.

Neonatal diseases and adverse conditions are not only important risk factors for LOS-NICU but are also a leading cause of neonatal death. Fifteen million premature infants are born every year, and more than one million of them die from diseases caused by the prematurity itself (49). Infection is one of the most common adverse events in hospitalized newborns, which poses a threat to all newborns (50–52). The most common infections affecting LOS-NICU in our study included sepsis, NEC, and pneumonia. These diseases may require prolonged symptomatic treatment, anti-infective therapy, antishock therapy, nutritional support, and surgical treatment and pose a high risk of complications, resulting in a longer treatment time and LOS. A study from Poland showed that the median LOS-NICU of all infected newborns was twice as long as that of uninfected newborns (25). At the same time, the occurrence of one or more clinical infections, as well as the site of infection, had a different kind of impact on LOS-NICU. Studies related to the incidence, risk factors, causative microorganisms, and impact on LOS of neonatal infections are necessary to increase and maintain awareness of the impact of infections, to help develop local and international diagnostic and treatment guidelines, to facilitate adequate and appropriate resource allocation, and to help design multicenter interventional studies. In addition, neonatal acute kidney injury (AKI), ROP, and BPD are also essential diseases affecting LOS-NICU. A multinational multicenter study found that the overall incidence of AKI was 29.9%, and newborns with AKI had an independent mortality rate four times higher than those without AKI, with a longer independent LOS-NICU (29), which is consistent with the result of Charlton et al. (35).

Among the scores developed specifically for neonatal medicine, six are now widely used to predict LOS-NICU, namely the Apgar score, the Score for Neonatal Acute Physiology (SNAPPE), the Morbidity Assessment Index for Newborns (MAIN), the Critical Risk Index for Babies (CRIB), the Transport risk index of physiologic stability (TRIPS), and the Hassan scale. In contrast, the Hassan scale is a composite outcome score for neonatal morbidity that assesses the number of morbidities diagnosed during the hospitalization period following delivery (53). Patil et al. (38) demonstrated the correlation between this scale and LOS-NICU in a study that included 7,037 neonates from 1997 to 2018 and concluded that the Hassan scale distinguished between long-stay and short-stay newborns better than other composite scores. In addition to neonatal scores, laboratory indicators during neonatal hospitalization also affected LOS-NICU. Ge et al. first explored the correlation between NT-proBNP7 levels and delayed discharge, showing that higher NT-proBNP7 levels were associated with a longer LOS-NICU (1). However, it is unclear whether heart function itself or disorders that affect heart function better explain this association.

Fewer studies have attempted to explain the effect of neonatal treatment factors and organizational factors on LOS-NICU. However, this is not easy, because even in the same country, different hospitals at the same level may provide different types of treatment and care to the newborn. Among limited studies on this topic, Zhang et al. compared LOS-NICU in independent children’s hospitals and perinatal prenatal centers having delivery facilities (33). Even after adjusting for neonatal morbidity, LOS-NICU remained significantly longer in children's hospitals than in prenatal centers, which may be related to the greater complexity of neonatal conditions at children’s hospitals and may also reflect differences in clinical care practices, healthcare policies, and availability of proper postdischarge medical care. Furthermore, two studies reached different conclusions about whether the transfer was a risk factor or not (9, 32), possibly emanating from different countries and different study populations.

This systematic review has the following limitations. First of all, while the search strategy was comprehensive and rigorous, it may still have missed including some studies. In addition, it is not feasible to perform a meta-analysis because of significant heterogeneity in the study design, study population, the definition of LOS-NICU, and statistical analysis methods. Finally, because most of the included studies were retrospective in nature, causal assertions could not be made regarding the risk factors for LOS-NICU.

In conclusion, we identified several of the most critical risk factors affecting LOS-NICU in published studies, including birth weight, gestational age, sepsis, NEC, BPD, and ROP. However, a few high-quality studies were available, and all included studies did not consider the effect of sociodemographic factors on LOS, such as family income, and parents’ psychological and health status. In addition to this, the level of NICU, home nursing, and antenatal and delivery room management may impact both severity of illness and LOS. Therefore, well-designed and more extensive prospective studies investigating the risk factors affecting LOS-NICU are still needed in the future. Applications of machine learning (ML) methods have been used extensively to solve various complex challenges in the field of medicine in recent years (54). ML methods are characterized by their ability to examine a lot of data and discover exciting relationships, provide interpretation, and identify patterns. ML can help enhance the reliability, performance, predictability, and accuracy of diagnostic systems for many diseases. Common machine learning classification models are Bayes network (BN) models, support vector machine (SVM), Radial basis function (RBF) tree, decision table, and naive Bayes. ML models have demonstrated high predictive performance in predicting LOS after pediatric heart transplantation and patients with COVID-19 (55, 56), which is a future direction for our research. In the meantime, studies should also focus on interventions to effectively reduce LOS-NICU and improve short- and long-term newborn outcomes.