Study design and participants

Data on infants born between 2003 and 2007 were obtained from the prospective cohort of the Neonatal Research Network of Japan (NRNJ). All participating centers of the network are designated tertiary NICUs by the local government as well as the national government. The database contains information on the morbidity and mortality of very low birth weight infants (VLBWIs) with birth weight (BW) ≤1500 g. Considering the nation-wide population of VLBWIs born in Japan during the study period, more than 50% of VLBW infants in Japan were in the registry [8]. Our selection criteria were surviving infants with BWs between 501–1000 g at discharge, at institutions with follow-up rates of over 60% at the age of 3 years. Infants who died before discharge, had major congenital malformations, those who were gestational age <22 weeks and had a length of stay <30 days were excluded. Length of stay was used as an exclusion criterion because growth velocity during the first month could not be calculated. The inclusion and exclusion criteria used herein are the same as those described by Patel et al [6, 7].

Weight growth velocity (WGV)

WGV was calculated using the EM method reported by Patel et al [6, 7]. WGV is commonly used as a growth measure in VLBW infants, as it summarizes infant weight gain over a specific time interval, smoothing the variability that is inherent in daily weight measures. However, calculating growth velocity (GV) from daily weight measures by averaging these values over the desired time interval is extremely labor-intensive. The EM method allows accurate estimation of postnatal GV in ELBW infants during hospitalization in a NICU without the influence of length of stay, BW, or the presence of chronic lung disease. Comparisons of the EM method with four other methods (2-point BW model, 2-point average weight model, linear BW model and linear average weight model) have confirmed the superiority of the EM method in terms of accuracy [6, 7], with mean magnitudes of error of <0.1% regardless of starting point or time interval, and no errors exceeding 5% of the accurate standard GV for any of the infants analyzed.

The EM method is easy to use from the clinical perspective as it only requires information on weight and age at two time points. WGV (g/kg/day) was calculated with the EM method as follows: wherein, W1 is BW, Wn is weight at discharge, D1 is 0 and Dn is day at discharge. Patel’s EM assumes that growth occurs at a fraction of the previous weight (otherwise known as “first-order kinetics,” with the unit of g/kg/day), and average growth velocity is estimated based on the fact that growth in biological systems is generally nonlinear, often following an exponential pattern.

The WGV score, which was categorized by per one increase in WGV, was used in the categorical analysis. A WGV score of 6 represented WGV <7, a WGV score of 7 represented 7≤ WGV <8, and so on up to a WGV score of 13. A WGV score of 14 represents WGV ≥14.

Confounding factors

The following were adjusted for as confounding factors: maternal age, primiparous, maternal diabetes, pregnancy-induced hypertension, clinical chorioamnionitis, preterm rupture of membrane (PROM), antenatal corticosteroid, cesarean section, sex, gestational age (GA), BW, Apgar score (AS) (1 and 5 min), respiratory distress syndrome (RDS), chronic lung disease (CLD), patent ductus arteriosus (PDA), necrotizing enterocolitis (NEC) and intraventricular hemorrhage (IVH) [1, 5, 9, 10]. Clinical chorioamnionitis was diagnosed based on the clinical findings, such as maternal fever, leukocytosis and local pain during pregnancy, labor or delivery [11]. Antenatal corticosteroid use was defined as administration of at least 1 dose of corticosteroid to the mother at any time before delivery to accelerate fetal lung maturity. GA was determined by the best estimate based on early prenatal ultrasound examination, the last menstrual period and the physical examination of infants at birth. Small for gestational age (SGA) was defined as a BW less than the 10th percentile of the standard BW for gestational age published by the Japan Pediatric Society [12]. RDS was defined as RDS diagnosed by clinical and radiographic findings. CLD was defined as the oxygen requirement at corrected 36 weeks. PDA was diagnosed clinically or echocardiographically. NEC was defined using Bell classification stage II or greater [13]. The grade of IVH was diagnosed with cranial echography according to the classification of Papile, with grades III and IV corresponding to severe IVH [14].

Outcomes

The primary outcome was death or neurodevelopmental impairment (NDI) at the age of 3 years (i.e., 36–42 months). NDI was defined as any one of the following: cerebral palsy, unilateral or bilateral blindness, severe hearing impairment or developmental delay.

NDI was determined with a comprehensive follow-up protocol to evaluate signs and symptoms of cerebral palsy (CP), sensory abnormality and cognitive function [15]. Cerebral palsy was defined as a non-progressive central nervous system disorder characterized by abnormal muscle tone in at least one extremity and abnormal control of movement and posture [16]. Visual impairment included unilateral or bilateral blindness based on a diagnosis by ophthalmologists. Severe hearing impairment was defined based on the need of a hearing aid.

Developmental delay was defined as a developmental quotient (DQ) <70 measured using the Kyoto Scale of Psychological Development (KSPD) test or judged by physicians for infants who did not undergo the test [15, 17]. The assessors of the KSPD test at each center were blinded to the perinatal details but were informed of gestational age. The physicians were not always blinded to perinatal and neonatal morbidities and interventions.

The KSPD test was administered by psychologists working in each participating center. The latest version of the KSPD was standardized in 2001 for 2677 Japanese children, and the mean and one standard deviation of DQ were 100.6 and 13.4, respectively [18]. Developmental function of infants whose DQ were not measured by the KSPD was evaluated by physicians based on whether the infants were able to say meaningful words, say their own name or age and distinguish between circles with diameters of 4 cm and 6 cm at each center [17, 19].

Statistical Analysis

Clinical characteristics of the subjects were stratified based on the following WGV distributions: WGV <9, 9≤ WGV <12 and WGV ≥12. Cutoff values of WGV 9 and 12 were used because the 25th and 75th percentiles of WGVs were 9.4 and 11.9, respectively. Univariate logistic regression analyses were performed to assess the effects of WGV score on death and NDI. Odds ratios (ORs) and their 95% confidence intervals (95% CIs) were calculated using simple descriptive analysis. Multivariate logistic regression analyses were performed to adjust for confounding factors. These factors included: primiparous, pregnancy-induced hypertension, clinical chorioamnionitis, antenatal corticosteroid, male sex, GA, BW, AS at 5 min <7, RDS, CLD, PDA, NEC and IVH, which have been known to affect the neurodevelopmental outcome in VLBWIs [1, 20, 21]. Adjusted odds ratios (AORs) and their 95% CIs were also calculated. We calculated that the planned sample size should give over 80% power to detect the difference in all variables after adjustment, with an OR of 1.5.

Categorical scales were used to analyze WGV scores because the relationship between WGV and outcomes may not be linear (i.e., the categorical scales help to understand the non-linear relationship). Validity was confirmed using descriptive approaches. In the categorical analysis, WGV scores were set as categorical variables, with the reference WGV score of 10. This value was chosen because the mean WGV was 10.4 and median WGV was 10.5. In the categorical analysis, ORs and AORs are based on WGV scores versus the reference score. Subgroup analyses were also performed for extremely preterm infants: GA <28 weeks vs. GA ≥28 weeks [22, 23], BW <750 g vs. BW ≥750 g[15, 24] and SGA vs. non SGA [25]. Consistent with Patel’s indication that EM should only be used for infants with a length of stay <160 days [7], sensitivity analysis was performed by excluding infants with a length of stay ≥160 days.

In order to avoid selection bias from missing data for WGV and outcomes, we performed multiple imputations (MI) with the assumption that data were missing at random [26]. MI is a Monte Carlo technique in which the missing values are replaced by stochastic and multiple simulated data by modeling with assumption of the missing mechanism. The MI model was based on year of birth, center, plurality, antenatal steroid, sex, out-born, birth weight, GA, CLD at 36 weeks, IVH III-IV, PVL, sepsis, NEC and intestinal perforation, which were found to correlate to the adverse outcome in the previous study [27]. This model allowed us to draw valid inferences from the repeated imputed data in the analysis of incomplete data. Statistical analyses were performed using JMP 10.0.2 (SAS Institute, Inc., Cary, NC, USA) and SAS 9.2 (SAS Institute, Inc., Cary, NC, USA).

The study protocol was approved by the Ethics Review Committee of Jichi Medical University and is available to the public on the Internet [28]. Parents provided written informed consent for their child’s participation.

Study limitations

This study has several limitations worth noting. First, the KSPD test evaluates neurodevelopment, and thus our results may be difficult to compare with studies that used other measures, such as the Bayley Scales of Infant Development III [27]. Second, our data included outcomes that were judged by pediatricians without the KSPD test. As previously discussed [27], infants at 22–23 weeks of age are more likely to be examined by pediatricians, and are more likely to have other conditions such as blindness and CP. Thus, this might have introduced some bias. Third, there were missing data for some infants who were not followed up. To address this issue, we used MI to draw inferences with the assumption that data were missing at random. However, if this assumption was inappropriate, our estimation could be deemed invalid. Fourth, we had no data about the feeding. A recent study showed that breastfeeding lead to better neurodevelopment in spite of suboptimal initial weight gain in very preterm infants [29]. In order to generalize our results, it will be necessary to conduct a study with large population-based data. We are currently working to identify interventions for managing WGV in order to achieve and support satisfactory neurodevelopmental outcomes.