Preterm Infants Follow Three Distinct Early Feeding Trajectories

Preterm infants who began oral feeding in a Chinese neonatal intensive care unit followed three statistically distinct patterns of feeding-ability development during their first 14 days, according to a prospective observational study. The trajectories, identified by latent class growth modeling, separated infants with a high starting score and rapid improvement, those with a moderate start and later catch-up, and those with a low start and slow rise. Among the 213 infants who completed all five assessments, the high-start rapid-rise group accounted for 31.5%, the moderate-start accelerated catch-up group for 44.6%, and the low-start slow-rise group for 23.9%. The findings indicate substantial variation hidden by the overall upward trend in feeding ability, but the researchers cautioned that the exploratory, single-center results do not establish a clinical prediction tool.

Infants were eligible if they were born at 28 to 32 weeks of gestation, weighed less than 2,500 grams at birth, entered the study neonatal intensive care unit within 24 hours with stable vital signs, and initially received nasogastric or orogastric tube feeding, with or without parenteral nutrition. The investigation took place at the Affiliated Hospital of Shandong Medical and Pharmaceutical University in Binzhou, China, from May 2025 through April 2026. It deliberately enrolled only infants who received a uniformly prepared preterm formula, a design choice intended to reduce variation in feeding type rather than a description of all feeding practices in the unit. The researchers excluded infants with major congenital anomalies or neonatal complications including intraventricular hemorrhage, necrotizing enterocolitis, and sepsis.

Of 245 enrolled infants, 27 were excluded before oral feeding began because their condition worsened, they were transferred, or treatment was withdrawn, while five were lost during follow-up. The final cohort had a mean gestational age of 30.7 weeks and a mean oral-feeding-ability score that rose from 22.03 with a standard deviation of 2.95 on the first day after initiation to 34.41 with a standard deviation of 3.67 on Day 14. Generalized estimating equations, which accounted for correlation among repeated measurements, confirmed a significant overall increase across the five assessment points. The assessments occurred on Days 1, 3, 7, 10, and 14, with each daily score calculated from three scheduled feeding sessions.

Feeding ability was measured with the Chinese version of the Oral Feeding Ability Scale for Preterm Infants, a 23-item instrument covering pre-feeding condition, sucking, swallowing, and physiological stability during feeding. Each item was scored from 0 to 2, producing a total range of 0 to 46, with higher scores indicating greater ability. The scale’s internal consistency in this population was reported as acceptable, with a Cronbach’s alpha of 0.784. Because no validated score thresholds classify feeding ability into clinically defined categories, researchers analyzed the totals as a continuous variable. A session could still be scored when sucking or swallowing could not be elicited, in which case a zero was recorded for the affected dimension rather than excluding the observation.

Latent class growth modeling was used to determine whether infants followed different longitudinal patterns hidden by the group average. Models containing one through five classes were compared using information criteria, class separation, statistical tests, parsimony, clinical interpretability, and class size. The three-class solution was selected because it produced lower Akaike, Bayesian, and sample-size-adjusted Bayesian information criteria than the two-class model, while the Vuong–Lo–Mendell–Rubin and bootstrap likelihood-ratio tests supported the addition of a third class. Entropy was 0.949, indicating good separation, although the two-class model had a slightly higher value of 0.987. Models with four and five classes were not adopted because the additional classes were very small and the VLMR test did not support them.

The high-start rapid-rise group contained 67 infants and began with an estimated score of about 26. Its trajectory rose to approximately 28 on Day 3, 32 on Day 7, 36 on Day 10, and 38 on Day 14. The 95 infants in the moderate-start accelerated catch-up group started near 21, then reached roughly 23, 26, 30, and 34 at the successive assessments. The low-start slow-rise group consisted of 51 infants. Its estimated score was about 20 initially, dipped slightly to 19 on Day 3, and then increased to 23, 26, and 30. Thus, the low-start group remained below the other two throughout follow-up, even though its trajectory ultimately turned upward. The researchers described the slight early dip as consistent with, but not proof of, immature coordination of sucking, swallowing, and breathing.

Univariate comparisons identified six characteristics that varied significantly across trajectory groups: gestational age at birth, birth weight, postmenstrual age when oral feeding began, mechanical ventilation, illness severity as measured by the Neonatal Medical Index, and pre-feeding behavioral state. Multinomial logistic regression then examined whether these variables were independently associated with group membership, using the high-start rapid-rise group as the reference. Compared with that group, lower gestational age and lower birth weight were associated with membership in both less favorable groups. The low-arousal and moderate-arousal behavioral categories were also associated with the moderate-start catch-up group. For the low-start slow-rise group, the analysis additionally found associations with later oral-feeding initiation, mechanical ventilation, and a Neonatal Medical Index score of at least 4.

Some regression estimates were large but imprecise. Relative to the high-start group, the low-start group had an odds ratio of 2.258 for each additional week of postmenstrual age at feeding initiation, 4.875 for mechanical ventilation, and 5.497 for an NMI score of at least 4. Low arousal had an odds ratio of 50.908 and moderate arousal 16.056 for membership in the low-start group, but the confidence intervals were extremely wide: 7.889 to 328.497 and 3.280 to 78.611, respectively. These magnitudes should therefore not be treated as stable risk estimates. Sparse cells contributed to the uncertainty; only four infants in the low-start group had been mechanically ventilated, and only four were classified as high arousal before feeding. The authors explicitly characterized the findings as associative and hypothesis-generating rather than predictive.

Several design features limit how far the results can be generalized. Oral feeding began only after infants reached at least 32 weeks of postmenstrual age, had stable vital signs, and had been weaned from mechanical ventilation for more than 48 hours. Nevertheless, the low-start group began oral feeding at a later median postmenstrual age than the other groups, which likely reflected a more adverse clinical profile rather than greater developmental maturity. The unit used scheduled oral feeding attempts under a protocol that did not primarily follow behavioral cues, and it also used side-lying positioning and a specific preterm formula. The trajectories may therefore reflect interactions between infant readiness and this feeding protocol, not maturation alone, and their frequencies may differ in neonatal units using predominantly cue-based practices.

Other important information was unavailable. Follow-up stopped on Day 14, so the study did not determine whether trajectory classes predicted time to full oral feeding, postmenstrual age at full feeding, length of hospital stay, or later growth and development. Ventilation history was recorded only as a binary variable; ventilation duration, bronchopulmonary dysplasia, and respiratory support at feeding initiation were not measured. Actual daily enteral and parenteral intake, feeding interruptions, parenteral-nutrition indications, and subsequent weight gain were also not recorded, so the prescribed nutritional targets could not be treated as observed exposure. Pre-feeding state was averaged across all five assessment days rather than measured only at initiation, and illness severity was assessed at discharge. Formal inter-rater reliability testing was absent, the behavioral categories were created operationally without external validation, and the study excluded medically complex and human-milk-fed infants. Larger multicenter cohorts with longer follow-up and prospective nutritional measurements are therefore needed to determine whether the three patterns reproduce elsewhere and add clinical value beyond information available when oral feeding begins.

Subject of Research: Longitudinal oral-feeding trajectories and associated clinical factors in preterm infants

Article Title: Three distinct trajectories of early oral feeding ability in preterm infants and their associated clinical factors

Article References: Tang, X., Zhu, Y., Wu, Y., Sun, S., Zhou, M., & Yu, X. (2026). Three distinct trajectories of early oral feeding ability in preterm infants and their associated clinical factors. Frontiers in Pediatrics, 14, Article 1924274. https://doi.org/10.3389/fped.2026.1924274

Image Credits: AI Generated

DOI: 10.3389/fped.2026.1924274

Keywords: preterm infants, oral feeding, feeding ability, latent class growth modeling, neonatal intensive care, gestational age, illness severity, mechanical ventilation, behavioral state, prospective cohort, Three, distinct