RESUMO
BACKGROUND: Due to high risks of feeding intolerance, preterm infants often receive parenteral nutrition (PN) to ensure sufficient nutrition and energy intake. However, there is a lack of data on the status of clinical PN practice and barriers among neonatal care units in low- to middle-income countries like Vietnam. This extensive survey explores the status and barriers of PN practice for preterm infants in neonatal units across Vietnam and identifies the practical implications of enhancing nutritional outcomes in preterm infants. METHODS: A multicenter nationwide web-based survey on PN practice in preterm infants was conducted across 114 neonatal units from 61 provinces in Vietnam. RESULTS: Among 114 neonatal units receiving a request for surveys, 104 units (91.2%) from 55 provinces participated. Neonatal units were categorized as level I (2/104, 1.9%), II (39/104, 37.5%), III (56/104, 53.8%), and IV (7/104, 6.8%). We found that the initiations of PN within the first hour and the first two hours of life occurred in 80.8% (84/104) and 95.2% (99/104) of the units, respectively. The early provision of amino acids, or AA (within the first day of life) and lipids (within two days of life) were documented by 85% (89/104) and 82% (84/104) of the respondents, respectively. The initial dose of AA ranged from 0.5 to 3 g/kg/day; the dose of AA less than 1 g/kg/day was reported by 7.7% (8/104) of the respondents; the maximum dose of AA ranged from 2 to over 4.5 g/kg/day, with 4 g/kg/day reported by 47.1% (49/104) of the respondents. The initial dose of lipids was between 0.5 and 2 g/kg/day, frequently 1 g/kg/day, reported by 51.9% (54/104) of the respondents; the target lipid dose ranged from 3 to 4 g/kg/day in 93.3% (97/104) respondents; the maximum target dose for lipid was 4 g/kg/day in 36.5% (38/104) of the respondents. The initial glucose dose was distributed as follows: 46.2% of respondents (48/104) administered 4 mg/kg/minute, 21.2% (22/104) used 5 mg/kg/minute, 28.8% (30/104) used 6 mg/kg/minute, and 3.8% (4/104) used 3 mg/kg/minute. Additionally, 48.1% of respondents (50/104) reported a maximum glucose infusion rate above 13 mg/kg/min and 19.2% (20/104) above 15 mg/kg/min. Nineteen percent (20/104) of the respondents reported a lack of micronutrients. Barriers to PN initiation included difficulty in establishing intravenous lines, the absence of standardized protocols, the lack of lipids and micronutrients, infections, and unavailable software supporting neonatologists in calculating nutrition paradigms. CONCLUSION: This study's findings highlight the highly variable PN practice across neonatal units in Vietnam. Deviations from current practical guidelines can be explained by various barriers, most of which are modifiable. A monitoring network for nutritional practice status and a database to track the nutritional outcomes of preterm infants in Vietnam are needed.
RESUMO
Multilayer neural networks are among the most powerful models in machine learning, yet the fundamental reasons for this success defy mathematical understanding. Learning a neural network requires optimizing a nonconvex high-dimensional objective (risk function), a problem that is usually attacked using stochastic gradient descent (SGD). Does SGD converge to a global optimum of the risk or only to a local optimum? In the former case, does this happen because local minima are absent or because SGD somehow avoids them? In the latter, why do local minima reached by SGD have good generalization properties? In this paper, we consider a simple case, namely two-layer neural networks, and prove that-in a suitable scaling limit-SGD dynamics is captured by a certain nonlinear partial differential equation (PDE) that we call distributional dynamics (DD). We then consider several specific examples and show how DD can be used to prove convergence of SGD to networks with nearly ideal generalization error. This description allows for "averaging out" some of the complexities of the landscape of neural networks and can be used to prove a general convergence result for noisy SGD.
