Introduction

Clearly, at the corrected age of 4–6 months, preterm infants enter a process when breast milk or formula alone is not sufficient for further nutritional requests such that complementary foods are needed [1, 2]. The appropriate timing of the introduction of complementary foods [3,4,5] and macronutrients and micronutrients provided by complementary foods [6,7,8,9] have positive effects on children’s (including preterm children’s) physical growth and cognitive development. In addition to its nutritional effect, food-related behaviours (such as satiety responsiveness, food fussiness), skills, and attitudes acquired during the complementary feeding (CF) period have long- and short-term health effects [10, 11]. Generally, the CF period is necessary for nutritional and developmental reasons.

However, preterm children encounter more feeding problems than their term counterparts in the CF period, such as improper use of nutritional fortifiers [12, 2.

Table 2 Demographic data of the recruited participants
Full size table

Models to predict nutritional risks

According to the results of univariate analysis (Additional file 4: Appendix 4, Additional file 5: Appendix 5 and Additional file 6: Appendix 6), binary logistic regression analysis (Additional file 4: Appendix 4, Additional file 5: Appendix 5 and Additional file 6: Appendix 6), references and our group discussions, models to predict nutritional risk were developed as follows.

For preterm infants at the corrected age of 5–7 months

The model to predict underweight included factors of the z-scores for birth weight, the volume of milk intake per day, nutritional fortifier usage, the amount of cereal and animal food intake, food energy density, and recent poor weight gain. The model to predict stunting included factors of the z-scores for birth length, the volume of milk intake per day, nutritional fortifier usage, vitamin D and calcium supplementation, hours spent outdoors per week, and recent poor body length growth. The model to predict microcephaly included factors of the z-scores for birth weight and birth head circumference, the volume of milk intake per day, nutritional fortifier usage, the amount of animal food intake, vitamin D supplementation, hours spent outdoors per week, and recent poor head circumference growth.

For preterm infants at the corrected age of 8–11 months

The model to predict underweight involved factors of the z-scores for birth weight and birth length, the volume of milk intake per day, nutritional fortifier usage, the amount of cereal and animal food intake, food energy density, perceived eating difficulty, and recent poor weight gain. The model to predict stunting involved factors of the z-scores for birth weight and birth length, the volume of milk intake per day, nutritional fortifier usage, the amount of animal food intake, vitamin D, vitamin A and calcium supplementation, hours spent outdoors per week, and recent poor body length growth. The model to predict microcephaly involved factors of the z-scores for birth head circumference, the volume of milk intake per day, iron rich food (red meat, egg and yolk, and animal viscus) intake frequency, the amount of animal food intake, perceived eating difficulty, vitamin D supplementation, hours spent outdoors per week, and recent poor head circumference growth.

For preterm children at the corrected age of 12–36 months

The model to predict underweight covered factors of the z-scores for birth weight and birth length, the volume of milk intake per day, red meat intake frequency, the amount of cereal and animal food intake, food energy density, perceived eating difficulty, and recent poor weight gain. The model to predict stunting covered factors of the z-scores for birth length, the volume of milk intake per day, the amount of animal food intake, perceived eating difficulty, vitamin D and calcium supplementation, hours spent outdoors per week, and recent poor body length growth. The model to predict microcephaly covered factors of the z-scores for birth head circumference, egg and yolk intake frequency, the amount of cereal and animal food intake, perceived eating difficulty, vitamin D supplementation, hours spent outdoors per week, and recent poor head circumference growth.

Reliability

The κ coefficients of the interrater reliability and the test-retest reliability of the NRSP were all above 0.600, which meant that the reliability of the NRSP was moderate to substantial. The data are outlined in Table 3.

Table 3 The Reliability and Validity of the Nutritional Risk Screening Tool for Preterm Children
Full size table

Validity

The NRSP exhibited relatively higher efficiency in predicting underweight and stunting, with AUCs, accuracies, specificities, and NPVs near to or greater than 0.900, sensitivities above 0.600, PPVs above 0.400, LR + s near to or greater than 10, and rss above 0.400. On the other hand, the NRSP manifested a weaker ability in predicting microcephaly, with most of the values of validity indicators lower than those of underweight and stunting prediction. Nevertheless, the LR-s of all the predictive models were above 0.1, which suggests less satisfactory results. The data are displayed in Table 3.

We further explored the correlations between the scores of each dimension of the NRSP and malnutrition. We found that the scores of anthropometric assessment were positively correlated with malnutrition in all age groups, except in stunting and microcephaly for preterm infants with a corrected age of 5–7 months. Feeding practices gradually manifested their significantly positive effect as age increased. However, nutrient supplementation did not have a significant correlation with underweight, stunting, or microcephaly. Data are depicted in Table 4.

Table 4 Correlations of scores of the Nutritional Risk Screening Tool for Preterm Children with underweight/stunting/microcephaly
Full size table

Finally, we compared anthropometric, biochemical, and intellectual developmental indicators between the high and low nutritional risk groups. We found that the z-scores for body weight, body length, and head circumference 1 or 3 months after the first interview were all greater in the low-risk groups versus the high-risk groups. Notwithstanding, there were no significant differences with respect to levels of haemoglobin, RBC count, MCV, MCH, MCHC, serum iron, and vitamin D between the high- and low-risk groups. The full-scale DQs of the high-risk groups were all lower than those of the low-risk groups. Gross motor and social communication DQs were lower in the high-risk group for preterm children with a corrected age of 12–36 months. Fine motor and adaptability DQs were lower in high-risk groups for preterm infants at the corrected age of 5–7, and 8–11 months. Verbal DQ was lower in the high-risk group for preterm infants at the corrected age of 8–11 months. Data are outlined in Table 5.

Table 5 The anthropometric parameters, biochemical levels, and intellectual development quotients between the high- and low-nutritional-risk groups
Full size table

Discussion

The NRSP was designed for routine clinical use for health care staff when following up with preterm children. The data from this study indicate that the NRSP has acceptable reliability and validity.

The NRSP has moderate to substantial reliability

We used interrater reliability and test-retest reliability to assess the ability of the NRSP in yielding the same nutrition outcome on the same individual. The interrater reliability and the test-retest reliability of the NRSP were all above 0.600, which implies that the reliability of the NRSP is moderate to substantial [34]. It is higher than the reliability of the Paediatric Yorkhill Malnutrition Score (PYMS, κ = 0.53) [35], the Screening Tool for Risk on Nutritional Status and Growth (STRONGkids, κ = 0.483), and the Paediatric Nutrition Screening Tool (PNST, κ = 0.601) [20], but lower than that of the Screening Tool for the Assessment of Malnutrition in Paediatrics (STAMP, κ = 0.882) [21].

The NRSP has moderate to high validity

The AUCs of the NRSP in predicting underweight, stunting, and microcephaly were all above 0.700, which suggests that the effectiveness of the NRSP in malnutrition prediction is relatively high [36]. The AUCs of NRSP were greater than those of the PYMS and the STAMP in predicting wasting (0.717 and 0.657, respectively) and stunting (0.628 and 0.643, respectively) [37]. Sensitivities were all above 0.600, and specificities were all above 0.700 for NRSP, which indicates a moderate to high extent, and they were similar to the sensitivities of the PYMS and STAMP for predicting wasting (0.878 and 0.776, respectively) and stunting (0.724 and 0.759, respectively) [37]. Additionally, the PNST had an approximate sensitivity of 0.88 and a specificity of 0.78, while STRONGkids had a higher sensitivity of 0.94 but a lower specificity of 0.44 [20]. The accuracies of the NRSP in predicting underweight and stunting were near to or above 0.900, which were higher than those of the Subjective Global Nutritional Assessment (SGNA, 67.07%) and the STAMP (45.12%) [38]. The PPVs of the NRSP in predicting underweight and stunting were similar to the SGNA (64.86%) and STAMP (47.06%), while the NPVs of the NRSP were clearly higher than the SGNA (68.89%) and STAMP (47.06%). At the same time, LR + s were higher than the SGNA (2.14) and STAMP (0.93). On the other hand, LR-s were similar to the SGNA (0.52), but lower than the STAMP (1.33) [38]. For a disease with a 10% prevalence, the ideal sensitivity is 90%, specificity is 80%, PPV is 33%, NPV is 98%, LR+ is more than 10, and LR- is less than 0.1 for a diagnostic test [39]. In our study, the total prevalence of malnutrition was 11.43%. Hence, the NRSP has ideal specificity, PPV, NPV, and LR+ in predicting underweight and stunting; however, the NRSP’s potential to predict microcephaly is weaker; moreover, sensitivity and LR- are less favourable for the NRSP. The data revealed that the NRSP classifications were moderately correlated with underweight, stunting and microcephaly, with correlation coefficients varying from 0.355 to 0.558 [40]. This was slightly stronger for the NRSP associated with underweight than stunting or microcephaly, which was the same as reports of the STRONGkids (r = − 0.16 for weight for age, W/A; r = 0.03 for height for age, H/A) [41], SGNA (r = 0.440 for W/A, r = 0.278 for H/A) [42], and PNST (r = 0.66 for W/A, r = 0.19 for H/A) [43]. Further, the NRSP had a stronger correlation with the anthropometry than the STRONGkids, SGNA and PNST. The somewhat weaker validity and correlation of the NRSP with stunting or microcephaly versus underweight might be because body length or head circumference growth are significantly affected by genetic factors, the social and economic environment, cerebral development, and skull thickness compared to mere nutrition factors [44,45,46].

The z-scores for anthropometric parameters and intellectual DQs were significantly higher in the low-risk groups than in the high-risk groups, which indicates that the classification by the NRSP is valid and reasonable. Health care staff should shed light on improving the feeding practices of preterm children with high nutritional risk to facilitate their physical growth and intellectual development. There were no discrepancies with respect to haemoglobin, RBC count, MCV, MCH, MCHC, serum iron, and vitamin D levels between the high- and low-risk groups, which might be because this study was a single-centre investigation, and the participants basically followed the same advice on nutrient supplementation.

Foetal growth status and feeding practices were critical factors for predicting malnutrition

We found that the anthropometric assessment score was positively correlated with malnutrition. The main component of the anthropometric assessment dimension was foetal growth status; therefore, we posited that foetal growth status was of paramount effect on extrauterine growth. A high score for anthropometric assessment indicates worse foetal growth status, which further signals less nutrient storage and a greater probability of disease occurrence, leading to extrauterine growth retardation.

The feeding practice score had a positive correlation with malnutrition only in the corrected 12–36-month age group, and even showed a negative association with malnutrition in the corrected 5–7-month age group. Reasons for the contrary relationship in the early stage might be that in the early stage of preterm birth, because of their low birth weight or length, they are probably regarded as having malnutrition, and the lower their birth weight is, the more likely they are to use nutritional fortifiers. When using nutritional fortifiers, preterm infants obtain lower scores in feeding practices, resulting in a false negative association between the score of feeding practices and malnutrition. As age increases, the effect of foetal growth status might be attenuated, and the positive effect of feeding practices gradually appears.

In the NRSP, some items (such as current diseases and nutrient supplementation, which are recognised in the literature as important factors of nutritional risk prediction) had no significant association with malnutrition in our study, which is contrary to our knowledge. We suspect that this was due to the relatively small sample size from a single centre. We decided to retain these items in the NRSP and anticipate further investigation.

There are limitations in this study. A major limitation was the relatively small sample size. This may have led to the second limitation, which was the moderate reliability and validity of the NRSP. Hence, a large-scale multicentre study should be conducted to broadly promote the NRSP models.

Conclusion

The present study shows that the NRSP has moderate to substantial reliability and validity in predicting underweight, stunting, and microcephaly. Health care staff should shed light on improving the feeding practices of preterm children with high nutritional risk classified by the NRSP to facilitate their physical growth and intellectual development. However, more research is needed to promote the NRSP models.