| Received | : | Apr 30, 2020 |
| Accepted | : | Jun 05, 2020 |
| Published Online | : | Jun 10, 2020 |
| Journal | : | Annals of Epidemiology and Public health |
| Publisher | : | MedDocs Publishers LLC |
| Online edition | : | http://meddocsonline.org |
Cite this article: Jing G, Yang Y, Xie L, Zhang J, Bai Q, Pan L, Li Z. Analysis of Birth Defect Monitoring Data and Comparison of Prediction Models in Lanzhou, China. A Epidemiol Public Health. 2020; 3(1): 1018.
Background: To analyze the distribution of birth defects in Lanzhou, China from 2012 to 2018, to forecast the prevalence in the future, and to provide evidence for prevention of birth defects.
Materials and Methods: Monitoring datas for birth defects in Lanzhou from 2012 to 2018 were used to describe its epidemiological characteristics, such as the change of the prevalence and trend, urban-rural differences, the varieties and cis-position. Gray modeling, Curve estimation, Exponential smoothing, and ARIMA were used to predict the occurrence for the next year in the future.
Results: The average prevalence rate of total perinatal infants’ birth defects was 104.13 per 10000 perinatal infants in the past 7 years in Lanzhou, with an upward trend increasing by 7.74% on average each year. From 2012 to 2018, the average incidence of birth defects in urban and rural areas of Lanzhou was 105.61/10000 and 106.48/10000, respectively, and the overall trend was upward. The most common birth defects in Lanzhou from 2012 to 2018 were cleft lip ± cleft palate and Congenital heart disease. Exponential smoothing model can better fit the number of birth defects in Lanzhou. The predicted results of exponential smoothing model indicate that birth defects in Lanzhou in 2019 are lower than that in 2018.
Conclusions: The number of birth defects in Lanzhou from 2010 to 2018 is on the rise, and is still in a high incidence in 2019. Congenital heart disease and Polydactylism are the main types of birth defects, which should be strengthened in the future.
Keywords: Birth defects; Perinatal infants; Epidemiological characteristics; Model comparison; Prediction.
Birth Defects (BDs) consist of a group of conditions, including physical or biochemical abnormalities, present at birth, which may involve malformations, disruptions, deformations [1,2]. The causes of BDs are complex, which can result from genetic defects, non-genetic maternal environmental exposures, or a combination of genetic and non-genetic factors [3-5]. The risk for BDs affects everyone, irrespective of socioeconomic status, race-ethnicity or other demographic characteristics.
The World Health Organization (WHO) estimates that approximately 260,000 deaths (7% of all neonatal deaths) globally were caused by BDs in 2004. The prevalence rates of BDs is estimated reach to 4.7% in the developed countries, 5.6% in the middle-income countries, and 6.4% in the low-income countries [6,7]. As far as we know, China is the most populated developing country in the world, and the prevalence of BDs varies across different socioeconomic groups and areas [8]. China is also a middle-income country with the largest population in the world, which has 16 million births annually and is expected to have 0.9 million BDs each year [9]. According to the most recent surveillance data in 2011, BDs have become the second leading cause of infant deaths in China (the leading cause being premature/low birth weight), which plays a significant role in the main causes of spontaneous abortion, stillbirth, perinatal death, infant death and congenital disability [10]. BDs also affect the child’s and the family’s quality of life and place an intolerable economic burden to the family and the society, particularly in the setting of the opening of a two-child per family policy in 2015 [11].
Lanzhou is the capital and largest city of GanSu Province. It is located in the geometric center of the geographical map of the motherland and is also an important node city of the “Belt and Road”. Lanzhou governs three counties of Yuzhong County, Gaolan County, Yongdeng County, and five districts of Chengguan District, Qilihe District, Xigu District, Anning District, Honggu District [12]. Lanzhou is located in the northwestern part of China. It’s natural and environmental conditions are relatively vague, the level of economic development is backward, the people’s living standards are relatively low, and the maternal and child health care facilities are poor. There is still a large gap between the economically developed areas.
The objectives of this study are to estimate the prevalence and types of BDs in Lanzhou, to explore the prevalence of BDs through different predictive models to provide a basis for future prevention and prevention of BDs.
The surveying population was perinatal infants (including stillbirth, fetal death or live birth) born in the Maternal and Child Health Hospital of Lanzhou from 2012 to 2018. All surveillance data of BDs were collected from Maternal and Child Health Hospital of Lanzhou and confirmed between 28 weeks of gestation through 7 days after birth.
Each delivery that was associated with a BD was reported using a registration card for BDs submitted by physicians in obstetrics and gynecology, pediatric or neonatal medicine through an online hospital-based survey, required by the Chinese government. Each case report card recorded basic maternal information, (including ethnicity, residence, family income, education, mother’s age, number of antenatal care visits, gestational age, number of reported abortions, pregnancy outcomes), birth information (gestational age, weight of birth), diagnoses of specific BDs, symptoms, early pregnancy disease(fever, virus infection, diabetes), medication use during early pregnancy, and family history. In addition to the case report card, an annual statement for each registered hospital was completed by professional physicians. Each annual statement contained 12 months of data such as the number of perinatal births, maternal age, residence, ethnicity, occupation, pregnancy history, gestational age of birth, gestational age, infant gender, number of BDs, and maternal illness.
Both case report cards and annual statement were reviewed and audited by maternal and child health hospitals and health administrative departments. Periodic quality control measures were in place at the monitored hospitals and occurred quarterly at the county-level and bi-annually at the city-level or provincelevel to assure reporting accuracy.
In this study, the Gray model, Curve fitting method, Exponential smoothing model and ARIMA model were used to dynamically predict the trend of BDs in Lanzhou, and the feasibility of four models in the prediction of BDs was discussed.
The GM is to transform the irregular raw data into one or more accumulated data processing methods, transform it into a more regular set of time series, and use discrete series to establish the dynamic model of the differential equation [13]. The GM prediction method is applicable to gray systems with partial information and partial information unknown. It contains both known information and unknown information (or some information is uncertain) [14]. In other words, the prediction of a gray process that changes in a certain azimuth while being timedependent.
The BDs data are considered as the original time series X=(𝑥0, 𝑥1, 𝑥2,…, 𝑥n),where n is the length of the time series. GM (1,1) is the main and basic model of grey predictions, that is to say, a single variable first order grey model, which is able to acquire high prediction accuracy despite requiring small sample size (but the sample size must be at least 4). The GM (1,1) model is suitable for sequences that show an obvious exponential pattern and can be used to describe monotonic changes. According to the data characteristics, this study will use the GM (1,1) model for prediction.
Finally, according to the prediction accuracy classification table, see Table 1, the test results is the prediction accuracy of the model [16].
Table 1
Table 1: Prediction accuracy classification table.
Curve fitting, also known as regression analysis, was used to find the “best fit” curve for a series of data points in this study. The curve fit often produces an equation that can be used to find points anywhere along the curve. Curve estimation were fitted to the data to quickly estimate regression statistics and produce related plots for 11 different models (linear, logarithmic, inverse, quadratic, cubic, power, compound, S-curve, logistic, growth and exponential models) [17]. F-test was used for selecting the best fitting curve for hypothesis testing. P-value <0.05 (two sided) was taken statistically significant. Similarly, R2 value > 0.80 was taken significant, where R2 is the correlation of the contribution of years (independent variables) in predicting BDs cases (dependent variables) [30]. In this investigation, x stands for the time (year) and y stands for the BDs cases.
For exponential smoothing method, also known as exponential weighted average method, the major advantage of this model is that it chooses the weighted average of each period as a decreasing exponential sequence, giving greater weight to the historical data closer to the forecast period. It might be suitable for the forecast of time-series data with trends and seasonality [18]. It’s consist of two models of seasonal and non-seasonal. The seasonal models mainly include simple seasonal, winteraddition, and winters multiplication model, and non-seasonal is made up primarily of holt-linear and brown-linear model. HoltWinter’s method is the extended form of the simple exponential smoothing that include the seasonality in the approach. In this study, SPSS 20.0 statistical software was used to model the data selection model by "Expert Modeler". In addition, Fitting indicators were analyzed and the optimal model in exponential smoothing model was selected to predict the BDs cases in 2019.
Autoregressive Integrated Moving Average (ARIMA) models, which take into account changing trends, periodic changes, and random disturbances in time series, are very useful in modeling the temporal dependence structure of a time series [19]. In epidemiology, ARIMA models have been successfully applied to predict the incidence of multiple diseases, such as BDs [20]. The Box-Jenkins was used to approach to ARIMA (p, d, q) modeling of time series (Box & Jenkins, 2010). This ARIMA model-building process is adopted to take advantage of associations in the sequentially lagged relationships that usually exist in periodically collected data [21]. The parameters selected for fitting ARIMA model are as follows: p, autoregressive order; d, difference; q, moving average order. The order of Moving Average (MA) and Autoregressive (AR) terms in the ARIMA model is determined by using the Auto Correlation Function (ACF) graph and Partial Auto Correlation Function (PACF) graph. The parameters of the model were estimated by the conditional least squares method. Diagnostic checking including residual analysis and the Akaike Information Criterion (AIC) was used to compare the goodnessof-fit among ARIMA models. The Ljung-Box [22] test was used to measure the ACF of the residuals.
SPSS 22.0 was used to analyze the incidence of BDs in Lanzhou from 2012 to 2018, and different models were established to compare the incidence of BDs in Lanzhou in 2019. Using Excel 2013 compiled function to calculate the indicators of dynamic sequence of BDs in Lanzhou from 2012 to 2018. The dynamic series indexes mainly include absolute growth (cumulative absolute growth(CAG) and annual absolute growth(AAG), development speed (base ratio and chain ratio), growth rate (base ratio and chain ratio), average development rate and average growth rate. CAG= , AAG=, base base ratio development speed = chain ratio development speed= , base ratio growth rate = , and chain ratio development speed= , where x and xt denote observed and fitted values at time point t.
In addition, we use the Mean Absolute Percentage Error (MAPE), Error Square Sum (SSE) and the Average Absolute Relative Deviation (AARD) to evaluate the prediction accuracy of the four prediction models. MAPE= SSE= AARD= where x and xt denote observed and fitted values at time point t. A lower MAPE, SSE and AARD value indicates a better fit of the data. Finally, the fitted optimizing model was used for short-term forecasting of BDs cases for 2019 in Lanzhou. All analyses were performed using SPSS 22.0 with a significant level of P < 0.05.
A total of 346,729 perinatal infants were monitored from 2012 to 2018. Of the 346,729 perinatal infants, 3653 had BDs, with an average prevalence of 104.13 per 10,000 perinatal Infants (PIs). Trend analysis revealed that the annual prevalence rates of total BDs in the 7 years increased linearly (χ2 trend =49.753, P<0.001). The total prevalence rate increased by 7.74% annually from 2012 to 2018 on average (Table 2, Figure 1). In addition, the incidence of urban/rural BDs in Lanzhou is on the rise(χ2 urban trend =11.993, P<0.001; χ2 rural trend =44.029, P<0.001), and it’s higher in rural areas than in urban areas in 2018. (Table 2, Figure 2).
Table 2
Table 2: Incidence of Perinatal birth defects from 2012 to 2018 in Lanzhou (1/10000).
Figure 1: Incidence of Perinatal BDs from 2012 to 2018 in Lanzhou.
Figure 2: Incidence of Perinatal birth defects from 2012 to 2018 in Urban/Rural ares of Lanzhou.
Table 3 is the dynamic change data of BDs in Lanzhou from 2012 to 2018. As shown in the table 3, the number of cases of BDs in Lanzhou is on the rise as a whole. Perinatal BDs average speed of development is 107.74%, and the Perinatal BDs average speed of increment is 7.74%. Assuming that the number of BDs in Lanzhou will develop at the rate of 107.74% in 2019, it is estimated that the number of BDs in Lanzhou will reach 737 cases in 2019.
Table 3
Table 3: Dynamic changes in the number of cases of Perinatal birth defects in Lanzhou from 2012 to 2018.
The incidence and the sequence in the first five places were Congenital heart defects, Polydactylism, Cleft lip ± cleft palate, Limb shortening and Ankylodactylia (Table 4). The most common BDs in Lanzhou from 2012 to 2018 are Cleft lip ± cleft palate and Congenital heart disease. The incidence of cleft lip and palate decreased from 19.23/10,000 in 2012 to 5.49/10,000 in 2018, and the incidence of Congenital heart disease increased from 16.06/10,000 in 2012 to 46.87/10,000 in 2018. Moreover, the incidence of Polydactylism increased from 9.98/10,000 in 2012 to 19.77/10,000 in 2018. The change of other defect types is not obvious. (Table 5).
Table 4
Table 4: The occurrence and cis-position of all kinds of birth defects in Lanzhou from 2012 to 2018 (1/10,000).
Table 5
Table 5: Changes of birth defects in Lanzhou from 2012 to 2018 (1/ 10,000).
GM(1,1) Model
The results are as follows (Table 6). S0 =139.27, S1 =79.24, the means of the posterior error ratio c=S1/S0 ; little probability of error P=0.857. According to the judgment criteria of the GM(1,1) model prediction accuracy (Table 1), the posterior error ratio c grade is reluctantly qualified, little probability of error P grade is qualified, and the accuracy grade is the lower level of the two indicators. It is considered that the fitting accuracy of the model is in a barely qualified is y=1808.06e-0.18t-1402.06
Table 6
Table 6: Calculation Process of GM (1,1) Model for the Number of Birth Defects in Lanzhou.
Table 7
Table 7: The Main parameters of birth defects curve fitting.
Table 7 is the model summary and hypothesis test results of 11 models of curve fitting. This study is mainly based on the statistical significance of model hypothesis test. The more R2 and variance F are used as the optimal model criteria. A higer R2 and lower F value are indicate a better model fit of the data. The fitting results show that the determination coefficient R2 is 0.849 and F is 11.249 after quadratic function fitting, and the model has statistical significance. Therefore, the quadratic function model can better reflect the trend of BDs in Lanzhou from 2012 to 2018. The fitting model equation is y=310.714+41.952t+2.167t2 (Figure 3).
Figure 3: Quadratic Function Fitting Curve of birth defects Cases in Lanzhou City from 2012 to 2018.
Figure 4: Time Series Diagram of birth defects in Lanzhou from 2012 to 2017.
The expert modeler is used to fit the model, and the exponential smoothing model is used to construct the model. The optimal model selection, parameter estimation and prediction are carried out. The statistical results of each model are shown in Table 8. By comparing the statistical results of three seasonal fitting methods of exponential smoothing model, and combining with the selection principle of evaluation index, the winters additive model of exponential smoothing model is finally determined as the best model. The stable R2 , R2 , RMSE and MAPE of the model are the best values in the three models, and the Ljung-Box Q statistic of the model is 0.086, with no significant difference (P>0.05), indicating that the residual is a white noise sequence, suggesting that the Winters additive model has better fitting effect (Figure 5).
Table 8
Table 8: Multiple linear regression analysis between the clinical index and the total COHIP-SF 19 score.
Figure 5: Prediction of the number of birth defects in Lanzhou.
The ARIMA model is suitable for BDs from 2012 to 2018 and is tested by predicting the cases of BDs in 2019. Different ARIMA models were tested to determine the best fitting model. Based on the distribution characteristics, we conducted six models, ARIMA(1,0,1)(1,0,1)12, ARIMA (1,0,1) (1,1,1)12, ARIMA (2,0,1) (1,0,1)12, ARIMA(1,0,1) (2,0,1)12, ARIMA(2,0,1) (2,0,1) 12, and ARIMA(2,0,1)(2,1,1)12. Of all the models tested, the ARIMA(2,0,1)(2,0,1)12 model was the best fit for the data (Table 9).
Table 9
Table 9: Comparisons of tested models.
According to the size of MAE, SSE and AARD of GM(1,1), Curve fitting, Exponential smoothing model and ARIMA, it can be concluded that Winters addition method in exponential smoothing model can better fit the number of BDs in Lanzhou (Table 10). The predicted results of exponential smoothing model indicate that BDs in Lanzhou in 2019 are lower than that in 2018 (Figure 6).
Table 10
Table 10: Comparing results of model fitting accuracy.
Figure 6: the Prediction of BDs by different models.
BDs are a global problem, but their impact on infant and childhood death and disability is particularly severe in low and middle income countries. Serious BDs can be lethal. For those who survive early childhood with BDs, they can experience lifelong mental, physical, auditory, and visual disabilities that exact harsh human and economic tolls on them, their families, and their communities. Up to 70% of BDs could either be prevented, or with proper care, cured or ameliorated [23,24]. As the largest developing country in the world, a hospital-based surveillance system for monitoring BDs was established in China in 1986 [25].
Many hospitals and institutions have generated a large number of birth defect files and publications. There are large regional differences in birth defect prevalence. Our study performed a comprehensive analysis of differences in the perinatal infant prevalence of BDs in Lanzhou, China over a 7 year time period. The average prevalence rate of total perinatal infants’ BDs was 104.13 per 10000 PIs in the past 7 years in Lanzhou, with an upward trend increasing by 7.74% on average each year. From 2012 to 2018, the average incidence of BDs in urban and rural areas of Lanzhou was 105.61/10000 and 106.48/10000, respectively, and the overall trend was upward. The upward trend in BDs epidemiology is potentially explained by several factors. Firstly, the liberalization of the second-child policy, the increase in the proportion of older pregnant women in Lanzhou, resulting in an increase in the incidence of BDs related; Secondly, Lanzhou has implemented various maternal subsidies projects and improved prenatal screening and prenatal diagnosis technology in grass-roots hospitals, making the detection of perinatal BDs increased.
China is a country with high incidence of BDs. The incidence of BDs is increasing year by year. In 2010, the incidence rate is 149.9/10000, rising to 157.0/10000 in 2014 and 175.5/10000 in 2016 [26]. The average incidence of BDs in Lanzhou (104.68/10000) is lower than the national level. Considering the reasons, it may be that with the promulgation of the National Comprehensive Prevention and Control Program of BDs [27], the grass-roots maternal and child health institutions in Lanzhou can actively respond to the policy call, conscientiously implement the prevention and control program of BDs, take certain measures in the prevention and control of BDs, and reduce the incidence of BDs.
From 2012 to 2018, the occurrence categories of BDs in Lanzhou were Congenital heart disease, Polydactylism, Cleft lip ± cleft palate, Limb shortening and Ankylodactylia. According to WHO reports, the most common serious congenital anomalies in the world are Congenital heart disease, Neural tube defects and Down's syndrome (anomalies). The most common types in the United States are Down's syndrome, Cleft lip, and Congenital heart disease [28]. Muscle and urinary system defects are the most common in Russia [29]. By above knowable, Congenital heart disease is a BD worthy of attention at present, which is in a high incidence in Lanzhou. The results of cis-positions of BDs from 2012 to 2018 show that before 2015, BDs were mainly Congenital heart disease and Cleft lip and palate. After 2015, Congenital heart disease and Polydactylism were the main types of BDs. In recent years, the incidence of Congenital heart disease has increased, on the one hand, due to the development of medical and health level, the continuous improvement of prenatal screening and diagnosis technology, such as the popularization and wide application of four-dimensional color Doppler ultrasound, the detection rate of Congenital heart disease has increased. On the other hand, it is not excluded that the incidence of Congenital heart disease increases due to the increase of risk factors related to Congenital heart disease. These results suggest that the prevention and treatment of Congenital heart disease is not optimistic.
The early recognition of distributions is significantly important for BDs disease control and prevention. The effectiveness of statistical models in forecasting future BDs incidence has been proved useful [30]. The surveillance system is a good way to collect and analyze BDs disease data. With high quality surveillance data, BDs disease may be accurately detected and forecasted. The selections of the BDs forecasting techniques is very important. In the present study, we conducted a comparative study of four typical forecasting methods in the forecasting of the BDs diseases, namely four methods GM(1,1), Curve fitting, exponential smoothing and ARIMA model methods. We have also compared the differences among these methods in both observed value and predicted value. By comparing the results of model evaluation index (SSE, MAPE, AARD), it is found that the order of the four models is: exponential smoothing, ARIMA, Curve fitting and GM(1,1).
The four models selected in this study have different prediction theories with different prediction ideas. The GM(1,1) prediction model needs less data and is suitable for the data with good smoothness of the original data. Moreover, the prediction is more accurate, the accuracy is higher and the calculation is simple. However, because the GM(1,1) model relies on historical data and does not take into account the relationship between various factors, the medium-term and long-term prediction error is large [31]. The Curve fitting model is easy to calculate and can save time for short-term prediction, but it is not suitable for medium-term and long-term prediction because of large errors in data prediction with large fluctuations [32]. Exponential smoothing model has been widely used in the prediction of many diseases. Its main idea is to obtain the predicted value by weighting the actual data. According to the characteristics of the near-time data, the predicted value is given more weight, while the far-time data is given less weight, which makes the predicted value less affected by the long-term data, and more affected by the near-term data [33]. ARIMA model is more suitable for time-varying series. It uses the characteristics of data stationarity to predict the future. It is suitable for short-term and medium-term prediction. It has poor long-term prediction effect [34,35]. Moreover, the sequence must satisfy stationarity and non-white noise. Otherwise, it is not appropriate to establish ARIMA model.
This study chooses the best model from the above models to forecast the BDs in Lanzhou in 2019. The prediction results show that the occurrence of BDs in Lanzhou in 2019 is close to that in 2018. It suggests that strengthening the prevention and control of BDs is an important public health problem in Lanzhou in the future years.
In summary, the occurrence of BDs in Lanzhou has shown an upward trend in recent years, and the types of defects are Congenital heart disease and Polydactylism. In addition, the predicted results show that the number of BDs in 2019 is close to that in 2018. Therefore, in the future, Lanzhou should increase the prevention and control of BDs.
The Science and Technology Planning Project of Lanzhou. Project number: 2018-3-8.
Data available on request from the authors. The data used in this study come from the Maternal and Child Health Hospital of Lanzhou, which does not support sharing because of the hospital itself.
We always work towards offering the best to you. For any queries, please feel free to get in touch with us. Also you may post your valuable feedback after reading our journals, ebooks and after visiting our conferences.
