Maternal disease factors associated with neonatal jaundice: a case–control study

The National Health Insurance Service–National Sample Cohort (NHIS-NSC) established by the National Health Insurance Service in South Korea was used [16]. This database (DB) is a representative sample that randomly selected 1 million people, accounting for about 2.2% of the Korean population in 2002. Moreover, it contains sample data obtained from 2002 to 2013 [18]. In this study, the qualification DB and treatment DB of the NHIS-NSC were used. Five variables of the qualification DB (patient ID, sex, year, age, and income rank), five variables of statement data (patient ID, claim number, visit date, principal diagnosis, and additional diagnosis), and three variables of type of disease data (claim number, visit date, and diagnosis) in the treatment DB were utilized. Age is divided into 19 groups from 0 to 85 years old at 5-year intervals (age 0, 1–4, 5–9, …, and over 84). As the age of participants considered in this study was 15–49 years old, it was regrouped subsequently into three groups with ages 15–24, 25–34, and 35–49. The income rank is divided into 11 groups at deciles with medical aid beneficiaries, and it was regrouped into 5 groups at 20% intervals. Variables about diagnosis were distinguished using the Korean Standard Classification of Diseases, version 6 (KCD-6), which is the Korean modified version of the 10th revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10).

All methods were carried out in accordance with relevant guidelines and regulations.

The principal diagnosis and additional diagnosis of statement data were integrated into one diagnostic variable and were then merged with the type of disease data according to claim number. Based on the merged diagnosis data, details regarding delivery and age of the participants (15–49 years old) at the year of delivery date were extracted. Then, the pregnancy records of patients who had data about delivery were collected. In cases in which treatment for neonatal jaundice were provided before birth registration, when insurance claims were made by the mother, the diagnosis of neonatal jaundice is included in the mother’s record. Hence, these cases were included in the jaundice group. Cases with diagnostic codes correlated with neonatal jaundice within 4 weeks after the delivery date were included in the jaundice group. Meanwhile, the control group included cases in which the diagnosis of neonatal jaundice was not attached to the mother.

In this study, KCD-6 codes related to delivery [19‐21], pregnancy [19, 22, 23] and neonatal jaundice were selected to identify each event. Preterm delivery and multiple gestation were defined as at least one record of the related codes within 4 weeks before and after the delivery date.

As the NHIS-NSC included a sample established from the claims data, but not designed for the study, the diagnoses entered in the records did not always indicate new-onset diseases. The same diagnosis codes might have been recorded repeatedly. In such a case, it was counted as one. If the same person delivered several times, each delivery was considered an independent case. The minimum interval from delivery to diagnosis of the next pregnancy was 4 weeks. Considering that periviable birth is defined as delivery during at least 20 weeks of gestation [24], the minimum interval from the previous to the next delivery was 24 weeks. The maximum duration from the diagnosis of pregnancy to delivery was 44 weeks [25]. After cross-joining pregnancy and delivery records, the joint records were listed chronologically. The date of pregnancy diagnosis was defined as the visit date of the first pregnancy record among all pregnancy records. Cases with diagnostic codes related to abortion (O00-O08, pregnancy with abortive outcome) [22, 23, 26] or stillbirths (O36.4, maternal care for intrauterine death; Z37.1, single stillbirth; Z37.4, twins, both stillborn; and Z37.7, other multiple births, all stillborn) [19, 22, 26] up to 4 weeks after the delivery date were excluded. Deliveries assigned with codes including O82 and O84.2 (multiple delivery, all via cesarean section) were considered as cesarean section. Further, deliveries assigned with codes such as O80, O81, O83, and O84.0 (multiple delivery, all spontaneous) and O84.1 (multiple delivery, all using forceps and vacuum extractor) were considered as vaginal. If the two types of delivery were present, cesarean section (O82, O84.2) was prioritized. Cases in which the type of delivery was not identified were excluded.

In this study, two studies, study A and B, were conducted. One was about diseases during ANC (study A), and the second was about diseases 1 year before ANC (study B). In study A, claims data from each pregnancy diagnosis date to day 1 before the delivery date were extracted. Cases that had no record in the ANC period, other than diagnostic codes correlated with pregnancy, were excluded to identify possible maternal risk factors. In study B, claims data from 1 year before each pregnancy diagnosis date to day 1 before the pregnancy diagnosis date were extracted. In the analyses of both two studies, the only first three characters of the diagnosis codes were used.

The two-sided Fisher’s exact test with 95% CI for the categorical variables was performed. The t-test was used to assess continuous variables. To decrease the effect of confounding variables, the jaundice and control groups were matched at a ratio of 1:10 via propensity score matching (PSM) with nearest neighbor matching. Age at the time of delivery and income at the time of pregnancy diagnosis were considered covariates. MatchIt package [27] was used to perform PSM. The results obtained by repeating PSM 1,000 times by randomly shuffling the order of records were used for the analysis of matched samples. The average number of cases, odds ratio, and p-value were calculated only for significant findings from the 1,000 results obtained using PSM. If the odds ratio was infinite, it was excluded from the average. Diseases that have more than 900 significant results, with a mean odds ratio of > 1 and a mean p-value of < 0.05, were considered a risk factor. Moreover, those with a mean odds ratio of < 1 and a mean p-value of < 0.05 were considered a protective factor. Conditional logistic regression analyses, adjusted for preterm delivery, delivery mode, multiple gestation and ANC duration, were performed for the diseases which have more than 900 significant results in the univariable analyses. Survival package [28] was used to perform conditional logistic regression analyses.

Results with a lower bound of > 1 or an upper bound of < 1 and a p-value of < 0.05 were considered significant. igraph package [29] was used to make a network image for the identified risk/protective factors. R (version 3.6.2) [30] was used in all analyses.

Figure 1 shows the flowchart of the case selection process. 555,474 of 560,645 women, included in NHIS-NSC from 2002 to 2013, had significant claims data about diagnoses. 67,967 of those 555,474 women had claims data related to diagnosis of delivery and were 15 to 49 years old at the time of the delivery. 65,442 participants of them, who had delivery records, had claims data related to the diagnosis of pregnancy. With pairing the delivery and pregnancy records, 91,477 delivery cases (64,723 women), satisfied with the time intervals which were defined as inclusion criteria in this study, were identified. 116 cases of 91,477 delivery cases were excluded as the delivery modes were not identified, and 131 cases of abortion or stillbirth were excluded subsequently. The participants with several delivery cases were included in each step unless all cases were excluded. Among 91,230 delivery cases, 5,111 cases incomplete on qualification DB around the gestation period and 7,800 cases that had no diagnosis records except pregnancy or delivery during ANC were excluded to consist of the sample for study A. For study B, 12,691 incomplete cases and 4,418 cases with no diagnosis records during ANC were excluded.

The sample in study A included 78,319 cases (n = 57,718). Among them, 366 cases (n = 364) were included in the jaundice group and 77,953 cases (n = 57,517) in the control group. The sample in study B had 74,121 cases (n = 54,787). Among them, 298 cases (n = 296) were included in the jaundice group and 73,823 cases (n = 54,620) in the control group. The n value indicated the number of mothers, not delivery cases. If a mother has delivered several times, it can be included in both the jaundice and control groups. Thus, the total number of patients in the jaundice and control groups did not correspond to the total population. The jaundice group accounted for 0.47% (366 in 78,319 cases) and 0.40% (298 in 74,121 cases) of all delivery cases in studies A and B, respectively. There was a significant difference in terms of income at the time of pregnancy diagnosis, multiple gestation, and ANC duration between the two groups in study A, but not in study B. However, the Cochran–Armitage trend test result (chi-square test for trend in proportion) for income was significant in studies A and B (p-value = 0.002 and 0.010, respectively). There was a significant difference in the mode of delivery between the two groups in studies A and B (Table 1).

The matched sample for study A had 4,026 cases (jaundice group: 366, control group: 3,660), and that for study B had 3,278 cases (jaundice group: 298, control group: 2,980). All the 1,000 matched samples significantly differed in terms of the type of delivery in both two studies (Table 2). There was also a significant difference in ANC duration in all the 1,000 matched samples of study A.

Tables 3 and 4 show the unadjusted odds ratios for neonatal jaundice according to disease that showed significant results in unmatched samples for studies A and B, respectively.

In study A, obstetrical tetanus (A34) had the largest OR (212.33, 95% CI: 2.71–14,121.54). Fever of other and unknown origin (R50) had the lowest OR (0.23, 95% CI: 0.03–0.84).

In study B, polyarteritis nodosa and related conditions (M30) had the largest OR (95% CI: 6.35-infinite). However, there was only case in the jaundice group. Pain associated with micturition (R30) had the lowest OR (0, 95% CI: 0–0.92).

Acute bronchitis (J20), vasomotor and allergic rhinitis (J30), gastritis and duodenitis (K29), dyspepsia (K30), and alopecia areata (L63) showed significance in the unmatched samples of the two studies. Among them, the OR of alopecia areata (L63) was > 1.

For diseases that showed significance more than 900 times in 1000 times of PSM, the average number of cases and average odds ratio are depicted in Fig. 2 (network image) and Table 5. Adjusted ORs were calculated for the diseases which have more than 900 significant results in the univariable analyses, as the primary outcome of this study.

In study A, among the probable risk factors, the disease with the highest OR was other diseases of digestive system (K92; adjusted OR: 14.12, 95% CI: 2.70–82.26), which was present in 0.20% of all matched cases, and the incidence of leiomyoma of the uterus was the highest (D25; adjusted OR: 3.22, 95% CI: 1.59–6.52), accounting for 1.14% of all matched cases. Among the probable protective factors, gastritis and duodenitis had the lowest OR (K29; adjusted OR: 0.39, 95% CI: 0.22–0.69), accounting for 8.43% of all matched cases, and the incidence of vasomotor and allergic rhinitis was the highest (J30; adjusted OR: 0.58, 95% CI: 0.37–0.92), which accounted for 9.46% of all matched cases (Fig. 2a).

In study B, the possible risk factor was salpingitis and oophoritis (N70; adjusted OR: 3.33, 95% CI: 1.59–6.94), which accounted for 1.26% of all matched cases. Among the probable protective factors, heartburn had lowest OR (R12; adjusted OR: 0.29, 95% CI: 0.12–0.71), which accounted for 5.29% of all cases, and the incidence of gastritis and duodenitis (K29; adjusted OR: 0.75, 95% CI: 0.58–0.95) was the highest, which accounted for 44.15% of all cases (Fig. 2b).

There was no common risk factor in both studies. However, the common protective factors included vasomotor and allergic rhinitis (J30) and gastritis and duodenitis (K29).