Using a large population dataset, we developed a risk adjustment model for NICU LOS for ELBW infants, to estimate individual patient LOS.
Birth weight is one of the most important determinants of the length of neonatal ICU stay. Darlow et al., in their study on outcomes of preterm babies in New Zealand, reported that median length of stay was significantly high for ELBW babies compared to VLBW babies.[5] Role of birth weight as an important determinant of the length of stay has been reported in multiple other studies as well. [24-28]
Gestational age is generally considered as another important determinant of NICU stay. Patel et al. reported that the length of stay is inversely proportional to the increasing GA. They reported that only about 75% of the babies ≤28 weeks were discharged by the predicted date (EDD) while close to 99 % of patients born at GA ≥29 weeks were discharged by the predicted date. There was also a greater variability in length of stay among lower GA categories compared to higher GA categories.[8] Hinchliffe et al also reported that birth weight and GA significantly influenced the LOS even after accounting for the deaths. [29] Manktelow et al. also described clinical variables that have the most consistent influence on the length of stay in addition to the birth weight and gestation. [30] The impact of GA in the length of stay is also reported by several other researchers. [14, 24-26, 31, 32] At the same Eichenwald et al. had reported that the GA at birth had no effect on the PMA at discharge, but they looked at babies 30-34 6/7 weeks of GA only in their study. [7]
We found that SGA babies (defined as birth weight <10th percentile for gestational age) have an advantage in reducing the length of stay. Rawlings et al. reported that growth retarded LBW babies usually meet the discharge criteria earlier, as they are more mature than their counterpart. This is important while predicting discharge, as we should consider BW and gestation age simultaneously. [33] Lee et al also have reported survival advantage for SGA neonates of < 31 weeks [34] At the same time Altman et al. has reported that small for gestational age status was associated with longer hospital stay in moderate preterm infants [35] We used only birth weight in our analysis because birthweight generally represents the GA for appropriately grown babies. We accounted for SGA babies separately in our analysis
The sex and race of the babies also may play role in their NICU stay. Female infants had lower LOS in our study. There are a couple of other studies that also have reported similar advantage for female sex, especially among ELBW babies. [9, 27, 36] We found that African Americans had a shorter hospital stay compared to white/Caucasian infants. Aly et al. also has reported longer hospital stay among white babies. [36]
Neonatal ICU stay is often complicated by a variety of conditions like RDS, BPD, late onset sepsis, NEC, IVH, post hemorrhagic hydrocephalus and ROP. The presence and severity of these conditions may significantly alter the NICU stay and prognosis for these babies. Respiratory morbidities and need for respiratory support may significantly alter the length of stay. Manktelow et al. have reported that, in addition to GA and birth weight, initial reason for admission (need for respiratory support) showed most consistent association with longer hospital stay. [30] BPD is the most common respiratory morbidity in surviving preterm infants. [32, 37]. Presence and severity of BPD has been shown to affect the length of stay. [38] Presence of pulmonary hypertension, lower birth weight, presence of G tube, tracheostomy, and need for mechanical ventilation or supplemental oxygen were associated with longer duration of hospitalization in babies with BPD. [16, 39-42] We found that presence of G tube or need for tracheostomy may have major role in prolonging the hospital stay as based on the scores of 5 and 6 respectively. At the same time, Grey et al has reported that need for mechanical ventilation or supplemental oxygen does not affect the LOS, however majority of those were normal birth weight infants (>2.5 kg). [43] In our analysis, presence of surgical NEC, need for creation of stoma and complications such as short bowel were major factors affecting the length of stay independently. Cotton et al reported that surgical NEC was associated with prolonged hospital stay in extreme preterm infants after examining the data for about 3900 infants from the NICHD centers.
We found that presence of sepsis significantly prolonged the length of stay. In line with our results, presence of sepsis has been reported as a significant factor in determining the length of hospital stay by many other researchers as well. [24, 42].
Retinopathy of prematurity continues to be one of the most common morbidity among preterm infants and its incidence and severity are inversely proportional to birth weight and gestational age. [8, 42, 44] We found that presence of stage 1 -2 ROP, severe ROP and need for ROP surgery will prolong the length of stay. Hintz et al from their analysis of NICHD center data and Beeby from their Australian data have reported that grade 3 or worse ROP significantly prolonged the LOS. [13, 42]
Development of cholestasis was directly proportional to the (Score for Neonatal Acute Physiology) SNAP II scores and inversely proportional to GA, the same factors that predicted LOS, in Gastroschisis as per Canadian Pediatric Surgery Network (CAPSNet) data base[45] We also found that Cholestasis significantly contributed the length of stay.
Several other researchers have investigated the association between the length of stay and severity of illness in NICU babies. Bender et al. developed a prediction model for NICU LOS using as predictors birth weight, gestational age, and two severity of illness tools, the Score for Neonatal Acute Physiology (SNAPPE) and the Morbidity Assessment Index for Newborns (MAIN), based on the infant characteristics from a single center in USA. They concluded that length of stay prediction is improved by accounting for severity of illness in the first week of life beyond factors known at birth. [46] Berry et al reported that length of stay >/= 21 days was predicted by SNAPPE-II score and surgery. [47] Similarly de Courcy-Wheeler et al. has reported that another clinical severity score, Clinical Risk Index for Babies (CRIB) score being associated with increased length of stay. [31] Hintz et al reported that the presence of late onset sepsis, BPD, need for postnatal steroids, ROP stage 3 and surgery were significantly associated with Late Discharge in full models. They also found that compared to infants who had none or only 1 risk factor, those with 3 risk factors had a 7- to 8-fold increase in the odds of Late Discharge, while those with 4-5 factors had an 8- to 11-fold increase. [42]
The peculiarities of the hospital can also be a significant factor in the length of stay. Attrition caused by transfer out of the hospital can significantly alter the LOS data between the hospitals. [14, 26]
Presence of congenital anomalies, especially major ones that require surgery, is another important factor in determining the length of stay. [26, 43, 47] We found that presence of minor congenital anomalies also can increase the length of stay.
The sex and race of the babies also may play role in their NICU stay. Females infants and African Americans had lower LOS in our study. There are a couple of other studies that also have reported similar advantage for female sex, especially among ELBW babies. [9, 27, 36]. Aly et al. also has reported longer hospital stay among white babies. [36]
Several researchers also have reported that birth weight [5, 26-28] and gestational age [14, 24-26, 32, 43] as important determinants of length of hospital stay. Our model incorporates the birthweight and sum of the comorbidity scores into the calculation as LOS=5.5+44.5(BW1) +27(BW2) +11(BW3) +BW4+4.3(PCCI), aR2=0.76. A few authors have endeavored into this task and have developed prediction models. Beeby et al, developed a regression model as LOS= 79.1 – (22.7if back transferred, or 61.3if died) – (2.6 x(GA–23)) + (5.1xweeks of IPPV) + (7.9 if ROP grade >2)+ (11.6 if in oxygen at 36 weeks post –conceptual age) from their analysis of Australian data.[13] Bannwart et al from his Brazilian data and built a mathematical model, with risk factors present during the first 3 days of life and also another model with factors present during the entire hospitalization period.[48] Zernikow and co researcher studied the accuracy of two LOS prediction models – multiple linear regression model (MR) and artificial neural network model (ANN). They reported that predicted and actual LOS highly correlated (for MR, r = 0.85 to 0.90; for ANN, r = 0.87 to 0.92). [49] Lee et al developed a LOS prediction model based on antenatal and perinatal risk factors scores which was stratified according to the birth weight groups. Their model predicted LOS better in the lower birth weight strata compared to higher birth weight strata. [9]
Our model helps in predicting the possible length of stay more accurately which will help in managing the resource allocation appropriately, insurance billing, adhering to treatment strategies that can reduce complications which can prolong the hospital stay, training and counseling parents more appropriately.
There are several limitations to our model. Not all needed items of data are always available in large national database. In order to produce a model that can reliably predict the length of stay, we have focused on clinical comorbidities. However, it appears that length of stay can be strongly influenced by other factors, including organizational factors as a range of approaches are possible with regard to both inpatient care and discharge policies. Other important factors that might influence length of stay include whether the unit has a specified weight or gestational age criteria before discharge can occur, policies on home tube feeding and home on oxygen therapy and the presence and extent of a community support including home nursing support. We could not account for these factors in our analysis. Though data used for modeling is old, our aim for this study was to conceptualize the modeling which predicts length of stay by taking consideration of comorbidities in addition to gestational age and birthweight. External validation is necessary to determine a prediction model's reproducibility and generalizability to new and different patients.