Association of maternal ethnicity and urbanicity on severe pediatric disease: a nationwide cohort study

We analyzed data from the Taiwan Birth Registry (TBR) 2004–2011, Taiwan Death Registry (TDR) 2004–2015, and National Health Insurance Research Database (NHIRD) 2004–2015. The nationality of the mother has been recorded in TBR since 2004. The NHIRD is the health insurance claims database of the Taiwan National Health Insurance (NHI) system. In 1995, Taiwan launched a single-payer NHI system, which has covered 99% of the population of the country since 2014 [14]. The databases contain registration and original claims data for reimbursement, including demographics characteristics, dates of clinical visits, medical diagnoses, medical expenditure, and details of prescriptions, examinations, and procedures. The TBR, TDR, and NHIRD are linked by a unique personal ID number, which is de-identified for research purposes.

A nationwide dynamic cohort, in which subjects are recruited at different times, was used. The advantages of such a cohort are that the number of subjects does not decline, or suffer from the problem of aging, over time [15]. A live birth during 2004–2011 was identified in the TBR and followed-up in the NHIRD or TDR until December 31, 2015 or the date of death, whichever occurred first (Fig. 1). Children of FBMs from developed countries such as Japan, Korea, South Africa, UK, and US, among others, were excluded, because they represented less than 0.5% of all births [7].

Mortality and severe diseases were studied. To avoid coding errors and misclassification bias in NHIRD, patients with severe diseases eligible to a catastrophic disease certificate were enrolled. In Taiwan, patients with severe or rare diseases (30 categories) are eligible to a catastrophic disease certificate (CDC). Patients with CDC can waive for the copayments of outpatient and inpatient. Issuance of the certificate requires a physician diagnosis of a catastrophic disease and a formal review by the NHI Administration (NHIA) with additional documentation including genetic studies, laboratory data, diagnostic images, or pathology reports. (see Additional file 1 for the 30 catastrophic diseases). To avoid spare data bias, only severe diseases with more than 100 incident cases during the follow-up period were included in this study [16].

Four groups were studied: urban TBMs, rural TBMs, urban FBMs, and rural FBMs. TBMs or FBMs was based on the maternal nationality from TBR. TBMs were the mothers whose nationality were Taiwan; whereas FBMs were the mothers whose nationality were foreign countries. The classification of urban or rural was based on the township in household registration of the mothers when they gave birth of the study children. In Liu’s study [17], they included population density (people/km²), population ratio of people with college or above educational levels, population ratio of elder people over 65 years old, population ratio of people of agriculture workers and the number of physicians per 100,000 people and used the cluster analysis with squared Euclidean distance and Wald’s minimum variance method, to study the urbanization stratification of varied township in Taiwan. They identified 7 clusters for the 359 townships of Taiwan: highly urbanized town, moderate urbanization town, emerging town, general township, aging town, agricultural town, remote towns and villages. In this study, we categorized the first 2 clusters as “urban” and the remaining 5 clusters as “rural”, proposed by Liu et al. [17] We classified health status at birth as small for gestational age (a birth weight below the 10th percentile), appropriate for gestational age (a birth weight between the 10th and 90th percentile), and large for gestational age (a birth weight above the 90th percentile).

Disease incidence rates were calculated by dividing the total number of children with severe diseases during follow-up by person-years at risk. The 95% CI of the disease incidence rate was computed based on the Poisson distribution because the n is large and the rate is very small here and the binomial distribution is very similar to Poisson distribution [18]. The chi-squared test, Fisher’s exact test, and analysis of variance (ANOVA) were used for univariable comparisons of differences among the 4 study groups, where appropriate. The log-rank test (univariable) and Cox proportional hazards model (multivariable) was used to examine differences among the 4 study groups and the reference group was rural TBMs. Other than the status of the study group, we also included the grouping of birthweight-gestational-age, sex, and maternal age in the Cox proportional hazard model, because we studied numerous health outcomes. When comparing urban and rural mothers, the reference group was rural mothers (rural TBMs and rural FBMs). When comparing TBMs and FBMs, the reference group was FBMs (urban and rural FBMs). SAS 9.4 was used for the statistical analysis. The significance level of the study was .05.

From 2004 through 2011, 1,603,794 infants were born. A total of 1,494,843 singleton live births were included after excluding stillbirths, non-singleton births, children with missing maternal ID numbers, premature newborns with a gestational age less than 24 weeks, implausible combinations of birth weight and gestational age, children of mothers from developed countries, and children with missing data on residence area. There were 682,982 (45.7%) singleton live births to urban TBMs, 662,818 (44.3%) to rural TBMs, 61,570 (4.1%) to urban FBMs, and 87,473 (5.9%) to rural FBMs. Regarding to nationality of FBMs, mainland China (52.94%) was the predominant country, followed by Vietnam (33.01%), Indonesia (7.29%), Philippines (2.24%), Thailand (1.94%), Cambodia (1.31%), Myanmar (1.28%) (Fig. 1).

The annual numbers of births from 2004 through 2011 declined in all 4 maternal groups, except for the sudden drop in 2010 which is Tiger year, a Chinese zodiac. In Chinese belief, people born in Tiger year tend to have an unlucky life [19]. The male:female ratio of 52:48 remained constant among the 4 maternal groups. The mean birth weight was lowest for rural TBMs, followed by urban TBMs, rural FBMs, and urban FBMs. The rate of low birth weight (< 2500 g), preterm rate (< 37 weeks) and percentage of SGA children was highest for rural TBMs, followed by urban TBMs, rural FBMs, and urban FBMs (Table 1).

Maternal age at delivery was mean age of 26.69 years for rural FBMs, 27.48 years for urban FBMs, 28.96 years for rural TBMs, and 30.79 years for urban TBMs. About 24.2% of TBMs were unemployed, whereas 82.1% of FBMs were unemployed. Similarly, most FBMs had an income less than p20 (NT$15840 per month) (Table 1).

Figure 2 ranks the incidence rates of 18 severe diseases for the 4 maternal groups. Congenital circulatory anomalies had the highest incidence rate, followed by all-cause mortality, psychoses with origin specific to childhood, congenital cleft palate and cleft lip, infantile cerebral palsy, vasculitis (in children, almost all Kawasaki disease), chromosomal anomalies, other congenital anomalies of the digestive system, congenital anomalies of the urinary system, long-term mechanical ventilation, type I diabetes mellitus, congenital hypothyroidism, major trauma/burn, lymphoid leukemia, congenital nervous anomalies, other and unspecified congenital anomalies, malignant neoplasm of the brain, and disorders of amino-acid transport and metabolism. The incidence rates of 8 severe diseases significantly differed among the 4 maternal groups, namely, congenital circulatory anomalies (p = .0002), all-cause mortality (p < .0001), psychoses (p < .0001), infantile cerebral palsy (p = .0081), vasculitis, including Kawasaki disease (p < .0001), congenital hypothyroidism (p = .0341), major trauma/burn (p < .0001), and congenital nervous anomalies (p = .0011). The category other and unspecified congenital anomalies was statistically significant (p = .0412) but was not considered because such events were combined from several congenital anomalies.

Figure 3 shows age-specific cumulative incidence rates and adjusted hazard ratios (HRs) for 8 severe diseases among children in the 4 maternal groups. Four major patterns were observed, as detailed below.

Incidence rates varied with age (Fig. 3). The cumulative incidence rates of congenital circulatory anomalies, congenital hypothyroidism, and congenital nervous anomalies increased quickly until age 1 year and increased more slowly until age 5 years. The cumulative incidences of infantile cerebral palsy and vasculitis rapidly increased until age 5 years and remained relatively constant thereafter. No child developed psychosis before age 2 years, but rates rose sharply, and differed in relation to urbanicity, after that. Incidence rates for major trauma/burn and all-cause mortality exhibited a similar pattern, except that rates rose slowly after age 18 months.

The longitudinal design allowed us to evaluate child health from birth until age 11 years. Severe diseases based on catastrophic diseases, which was carefully validated by physicians and reviewed by committee, minimized the problem of misclassification. However, children with mild physical diseases or mental disorders are not included. Hence, the incidence rates of the 18 severe diseases in this study tended to be lower than those reported previously. For instance, the incidence of congenital circulatory anomalies was 70.06 per 100,000 person-years in this study, but the incidence of congenital heart diseases in Taiwan was 13.08 per 1000 live births [39]. We anticipated that bigger effect of FBM/TBM or urban/rural on children’s mild health outcome. We did not examine how parental health status affects child health because maternal disease before or during pregnancy was not always recorded in the TBR. Future studies should review medical records to examine this issue. Furthermore, frequency of parental tobacco use and alcohol use disorders was not studied because lifestyle data were not available. Finally, the status of urban and rural was based on household registration of the mothers when they gave birth of the study children. We did not consider children who moved from urban to rural area or vise versa. Hence, for diseases with very different incidence rates between urban and rural area, such as congenital hypothyroidism, psychoses, etc., we suggest that the status of urban and rural should be treated as time-varying rather than static in future studies.