Building a prediction model for iron deficiency anemia among infants in Shanghai, China

Abstract Iron deficiency anemia (IDA) is a common micronutrient deficiency worldwide in infants. Iron deficiency anemia, during infancy, can have long‐lasting detrimental effects on the immune and neural systems; the damage is irreversible. This study aimed to build a prediction model to predict the potential risk of IDA among infants. To collect relevant information for model building, we recruited 528 infants from Fenglin Community Health Service Center in Shanghai, China, and collected the information of infants and their parents by using a structured questionnaire. We also got the blood routine examination results of the infants. Then, we used a multilayer perceptron model (MLP) of the neural network model in IBM SPSS Modeler 18.0 to construct the prediction model. Of the 528 included infants, 80 (15.2%) of them had lower hemoglobin values (<110 g/L) and were finally diagnosed with IDA. Based on the accuracy of different models, the model with the highest accuracy rate (97.3%) was chosen, and all the preselected 26 variables were included in the model. After the modeling, the results indicated that the number of months of exclusive breastfeeding was the most important predictive variable, followed by the mother having anemia during pregnancy, and then the number of months of feeding the infant with iron‐fortified rice flour. The model has good sensitivity (100%) and specificity (100%). By using this model, we can predict the potential risk of an infant having IDA and can take the initiative to prevent iron deficiency through the improvement of feeding methods.

Iron deficiency anemia (IDA) is the most common form of anemia, especially among infants between 6-24 months, in China (Liu, Chen, & Zhao, 2011;Wang, Sun, & Chang, 2018). Iron is an essential nutrient for the development and cell growth of the immune and neural systems. It also helps in the regulation of energy metabolism and exercise (WHO, 2016a). Thus, IDA can have a significant impact on infant development (Baker, Greer,

& Committee on Nutrition
American Academy of P, 2010). For example, IDA can cause irreversible damage to an infant's long-term neurodevelopment and psychomotor development, especially if IDA is severe and persistent (Beard, 2008;Peirano et al., 2009). Prevention and control of an infant's anemia can improve the nutritional status of childhood and the health of infants. Thus, identifying the factors associated with IDA among infants and providing relevant interventions is a research priority. Based on findings of different studies, the American Academy of Pediatrics (AAP) concluded that the risk factors for ID/IDA in infants were prematurity, low birth weight, exposure to lead, exclusive breastfeeding beyond 4 months of age without supplemental iron, and weaning to whole milk or complementary foods that do not include iron-fortified cereals or foods naturally rich in iron (Baker et al., 2010). However, Chinese still have not an authoritative standard and guidance to prevent infant IDA. Thus, it is essential to develop a model that can predict IDA among Chinese infants. The results of the model can be used to guide the development of intervention strategies that can be used to prevent IDA in China. Furthermore, the plans for iron supplements for infants vary among different guidelines in China, and they do not often take the personal situation into consideration. Therefore, it is essential to evaluate the current condition of Chinese infants and provide individualized plans to improve their health.
This study aimed to build a predictive model of IDA among infants and to develop practical intervention strategies.

| K E Y ME SSAG E S
• Of the 528 included infants, 80 (15.2%) had lower hemoglobin values (<110 g/L) and were finally diagnosed as IDA.
• The final chosen predicting model has an accuracy rate of 97.3%, and all the preselected 26 variables were included in the model, with a sensitivity and specificity of 100%.
• The number of months of exclusive breastfeeding was the most important predictive variable, followed by mother having anemia during pregnancy, and then the number of months of adding iron-fortified rice flour to the infant diet.

| ME THODS
A cross-sectional study was conducted between October 2015 and August 2017 in Shanghai, China, to collect data that would be used in the prediction model.

| Participants' recruitment
Infants who were taken to Fenglin Community Health Center (CHC, a primary care center) to get their 6-month-old examination were recruited (during 6-7 months). The inclusion criteria included participants currently living in Shanghai, willing to come to the CHC for IDA treatment in the coming one months if identified as an IDA case, and those between 6 and 7 months. Infants who were 8 months or elder were excluded. The parents who were willing to participate in this study signed a consent form and filled out one questionnaire. The study was designed with the plan to recruit 500 participants based on the assumption that the prevalence of IDA is 13% among infants in Fenglin Community Health Center, with an absolute error of 0.03, and a type one error of 5%.

| Measures
A structural questionnaire was used to collect information from the parents of the infants (completed by one of the parents).
The information provided below was collected during the survey. For infants, we collected age (in days), gender (male vs. female), birth weight (2,500-3,000 g, 3,000-3,500 g, 3,500-4,000 g, or ≥4,000 g, based on birth certificate records), gestational age (<37 weeks, 37 weeks, 38-40 weeks, 41 weeks, or 42 weeks, based on birth certificate records), gravida (1, 2, 3 or 4 and above), para (1, 2, or 3), delivery mode (cesarean section, spontaneous delivery, or obstetric forceps), feeding patterns (exclusive breastfeeding, mixed feeding, artificial feeding), the age of complementary food (iron-fortified rice flour) added (in months), and the supplementation of vitamin AD (VitAD, Yes or No), vitamin D (VitD, Yes or No), Ca (Yes or No), and Fe (Yes or No). For parents, we collected age (in years), educational level (high school or below, 2-year college degree, bachelor's degree, master's degree or above), occupation, and annual income (in RMB). For the mothers, we further collected information on reproductive age (in years), anemia status during pregnancy (Yes or No), body mass index (BMI < 18.5 or >24), gestational hypertension (Yes or No), gestational diabetes (Yes or No), thyroid disease during pregnancy (Yes or No), passive smoking during pregnancy (Yes or No), and the supplementation of iron during pregnancy (Yes or No). The surveys were performed by self-report of the infants' parents.
After the parents completed the questionnaires, the physicians at the CHC checked the questionnaire before the parents left the center. If there were any inconsistencies or incomplete information, the physicians would check further with the parents. In addition, the physicians conducted the physical development evaluation of the infants and collected the results of blood routine examinations.
The physical assessment included weight, height, nutrition level, the physical development of the infant (moderate and severe malnutrition, mild malnutrition, normal weight, overweight, mild obesity, or moderate and severe obesity).

| Laboratory test
After the parents finished the questionnaire, peripheral blood was collected from the left ring finger of the infant. The collected blood was sent to the laboratory for testing; the testing results were obtained in 10 min. The SYSMEX XS-500i automatic blood cell analyzer was used to get hemoglobin values. During this procedure, medical services were provided to the IDA patients, and their respective IDA cases were verified.
In this study, we followed the WHO diagnostic criteria of anemia for a 6-month-old infant, which states as follows: An infant will be diagnosed with anemia if he/she has hemoglobin (Hb) < 110 g/L at the 6-month visit. If infants had hemoglobin < 110 g/L, which were improved by iron supplementation after one month, they were considered to be IDA cases (Baker et al., 2010). The non-IDA anemia cases (2 cases) were excluded from this study.

| Statistical analysis
The data were double entered and verified by Epidata 3.0. After the data were cleaned, descriptive analysis was used to describe the distribution of the socio-demographics of infants and their parents.
Then, we generated different models using IBM SPSS Modeler 18.0, including neural network, Logistic regression, Bayes Net, and so on.
Finally, the multilayer perceptron model (MLP) of the neural network model was selected, with the highest accuracy rate. Boosting algorithm was used to enhance the accuracy of the model further. The outcome for the model building was IDA among recruited participants, and all the other variables were treated as exposure variables.
These variables include gestational age, birth weight, delivery mode, gravida, para, gender of infant, the number of months of exclusive breastfeeding, the number of months of complementary food (ironfortified rice flour) added, vitamin AD supplementation, vitamin D supplementation, Fe supplementation, Ca supplementation, physical development evaluation of infant, mother's reproductive age, mother's educational level, mother's years of living in Shanghai, whether mother had anemia, BMI < 18.5, BMI > 24, gestational diabetes, gestational hypertension, thyroid disease during pregnancy, weeks of iron supplementation during pregnancy, passive smoking during pregnancy of the mother, whether parents are engaged in medicalrelated career, and parents' annual income.
A ROC curve and the gain curve were obtained by using IBM SPSS Modeler 18.0, and the sensitivity and specificity of the predictive model were obtained by using the same software.

| RE SULTS
A total of 528 infants were recruited and included in this study.
Among these infants, 80 (15.2%) had lower hemoglobin values (<110 g/L) and were diagnosed as IDA. After using iron supplementary, their respective hemoglobin values increased.

| Demographic characteristics
The 528 infants were 187 ± 7 days old (between 173-230 days, one was less than 6 months, and eight were recruited in 7th month).

| Model selection and prediction model of iron deficiency anemia
After analyzing, the IBM SPSS Modeler presented ten different models, with an accuracy rate ranging between 76.5% and 97.3%.
The model, which had the highest accuracy rate (97.3%), was finally chosen (Model 1, Table 2). Twenty-six variables were selected for modeling based on previous studies. With the exception of the models we chose in this study, all the other models have accuracy rates of less than 87%, which are significantly lower than our current model.
The relative importance of these variables is shown in Figure 1.
We also calculated the weight of each included variable; the number of months of exclusive breastfeeding was the most important predictive variable, with a value of 0.197. Furthermore, mothers having anemia during pregnancy ranked second (0.173), while the third to fifth ranks was the number of months of adding iron-fortified rice flour (0.097), birth weight (0.094), and gestational age (0.063), respectively. All the other factors had values of 0.05 or below.

| Model verification
The model also gave us the results of the ROC curve (shown in Figure 2) and gained curve (shown in Figure 3). The results show

| D ISCUSS I ON
Knowing the variables that can predict the risk of having IDA accurately is essential for preventing IDA. We built a neural network model with high accuracy for predicting the risk of having IDA among infants at our center. The results of our study indicate that the number of months of exclusive breastfeeding, whether mothers had anemia during pregnancy, and the number of months of adding iron-fortified rice flour rank in the top three standings. These findings can help develop practical intervention strategies.
The results of this study show that the number of months of exclusive breastfeeding is the most important predictive variable. The more extended the exclusive breastfeeding, the more likely the infant has IDA. This result is similar to a large number of previous studies (Burden et al., 2007;Capozzi, Russo, Bertocco, Ferrara, & Ferrara, 2010;Monterrosa et al., 2008). The normal infant can access sufficient iron from breast milk until the infant has doubled his or her birth weight, which occurs at about 4-6 months of age in a term, normal birth weight infant (Domellof, 2007).
Although the iron absorption rate of breast milk is five times that of formula milk, the amount of iron in breast milk is still insufficient for infants (Li, 2011 which is at least 50% (WHO, 2017). We should recommend exclusive breastfeeding according to WHO, but providing additional iron supply to exclusively breastfeeding infants proves essential and integral in preventing IDA.
We also found that maternal anemia during pregnancy is the second crucial predictive variable of IDA. The most recent estimates about anemia showed that the global prevalence of anemia, in 2011, among women of reproductive age was 29% (WHO, 2016b). Although the anemia rate during pregnancy in our study (15.5%) was much lower than this prevalence, the anemia rate among pregnant women in Shanghai is still high. WHO aimed to achieve a 50% reduction of anemia in women of reproductive age by 2025 (WHO, 2017); further reduction of anemia is a significant challenge. For infants, iron that is obtained from a mother during the fetal stage can generally meet their needs between 4 and 5 months of age. But if the mother has anemia during pregnancy, maternal iron storage decreases, transferrin receptor compensatory conversely increases, and the placental iron uptake capacity decreases. This reduces fetal iron gain among infants (Pasricha, Drakesmith, Black, Hipgrave, & Biggs, 2013). For such infants, the iron obtained from the mother is depleted more quickly, and the infant is more prone to anemia. Providing further iron supply to pregnant women is considered an essential prevention strategy.
The months of adding iron-fortified rice flour ranked third among all the chosen variables. To meet the needs of the infants, adding complementary food timely and effectively is vital to prevent IDA (Domellof, 2007;Jonsdottir et al., 2012). The American Academy of Pediatrics (AAP) recommends exclusive breastfeeding infants should get a daily supplement of 1 mg/kg iron from 4 months old until the infant can intake adequate iron-bearing foods to reduce iron deficiency (Baker et al., 2010). But some studies support that iron supplements are beneficial in iron-deficient children, but there is a risk of incurring adverse effects in those who are iron sufficient (Domellof, 2010). In China, most of the parents are generally not accepting of iron supplements, so iron-fortified rice flour can be used as a means of supplementing iron for infants of 4 months old and above (Jonsdottir et al., 2012;Qasem, Fenton, & Friel, 2015). Chinese Dietary Guidlines, (2016) point out that the first complementary foods for infants should be ironrich foods, such as iron-fortified rice flour, minced meat, and so on.
Birth weight and gestational week were two other significant predictors. Previous studies have indicated that the last three months of pregnancy are a critical period of fetal intrauterine iron transport (Li, 2011). Iron is actively transported to the fetus F I G U R E 1 The ranking of different predictive variables in the final model in Shanghai, China, 2015China, -2017 F I G U R E 2 ROC curve of the selected prediction model for predicting iron deficiency anemia among infants in Shanghai, China, 2015China, -2017 through the placenta during pregnancy, and the fetus maintains a high intrauterine hemoglobin level. So, the fetus can adapt to the intrauterine environment, which relatively lacks oxygen. The larger the gestational week, the more iron the fetus gets. And there is about 75 mg iron per kilogram, so the more significant the body weight, the higher the iron content (Dewey & Chaparro, 2007). In other words, the larger the birth weight and gestational weeks, the higher the iron content, and the less likely the infant will have IDA.
The prediction model can be used to predict whether the infant will have IDA, based on the situation of the infants and their parents.

| CON CLUS ION
The prediction model established in this study showed that exclusive breastfeeding, maternal anemia during pregnancy, and nontimely supplementation of complementary food were the top three risk factors for IDA in infants. WHO recommends exclusive breastfeeding before six months old, but iron supplementation should be added. Following strategies can be implemented to prevent IDA in infants: (a) We should enhance parents' acceptance of iron supplementation by health education; (b) we should continue to improve iron supplementation during pregnancy, even 78.8% participants had taken iron during pregnancy; (c) we should supervise and urge pregnant women with anemia to get treatment as soon as possible, to improve their iron nutrition; (d) for nonexclusive breastfeeding infants, we suggest they take complementary food timely and effectively to prevent IDA.

ACK N OWLED G M ENTS
The author thanks all the participants who participated in this study.

F I G U R E 3
Gain curve of the selected prediction model for predicting iron deficiency anemia among infants in Shanghai, China, 2015China, -2017 CO N FLI C T O F I NTE R E S T None declared.

E TH I C A L A PPROVA L
Ethical approval was obtained from the ethics review committees at the Fenglin Community Health Service Center (Shanghai, China) prior to the survey launch. Informed consent was obtained from all the participants by e-signing the inform consent form online.