Prenatal care and infant outcomes of teenage births: a Project WATCH study

One medical necessity that may mitigate some of these poor health profiles includes patient-physician interactions during prenatal care (PNC) visits. PNC is essential to the health of mother and baby [8‐14], as it gives clinicians the opportunity to advise young mothers and to prepare them for birth and motherhood. These interactions with health care providers can include exchanging pregnancy and birth information, facilitation of education, screening measures for abnormalities and complications for mother and baby, monitoring/continuous care, and preparation for childbirth and motherhood [12, 13]. PNC also provides the possibility of early detection and possible treatment of diseases, initiate timely intervention, promote overall wellness, and aid in facilitating informed birth choices [8, 14].

The most common and publicly well-validated indices for adequate PNC are the Kessner Index and the Adequacy of Prenatal Care Utilization (APNCU) index (or Kotelchuck Index). The Kessner Index classifies PNC into three categories: adequate, intermediate, and inadequate. For PNC to be considered adequate, initiation must begin in the first trimester and there must be 9 or more visits total for a pregnancy of 36 weeks or more [15]. The Adequacy of Prenatal Care Utilization (APNCU) index (or Kotelchuck Index) uses two parameters, including time of initiation and number of visits [16]. The APNCU index bases the number of visits off of the ACOG standards for uncomplicated healthy pregnancies that is used by practicing physicians as PNC standard of medical care [16]. This index categorizes care into four categories: inadequate, intermediate, adequate, and adequate plus. Adequate PNC is 80–109% of expected PNC visits, or 9 to 13 total visits [16‐18]. Both indices require initiation of care in the first trimester.

There are many adverse infant outcomes related to teenage births already known in the literature. Teen pregnancy has been linked with preterm birth and low birthweight babies [2, 6, 19]. While much literature has observed the association between poor infant outcomes and teenage births, the relationship is confounded by social and economic conditions [2]. Pregnant teenagers are more likely to be poverty-stricken, of minority racial or ethnic status, have less education, and be unmarried compared to their older peers [7, 20‐22].

While little is known about the relationship between PNC inadequacy and adverse infant outcomes of teenage births, one recent study showed that teenage mothers who received inadequate PNC (RR: 1.82 (95% CI: 1.39, 2.37)) and intermediate PNC (RR: 1.58 (95% CI: 1.83, 2.57)) were at increased risk of maternal morbidity when compared to teenage mothers who received adequate PNC [5]. The study also found that teenage mothers with maternal comorbidities who received inadequate PNC had 5 times increased risk of maternal mortality than those without maternal comorbidities and adequate PNC [5].

West Virginia (WV) is one of the poorest states economically and in terms of population health. According to the WV Behavioral Risk Factor Surveillance System (BRFSS), WV is ranked number 2 nationally for the highest prevalence of adults who report being in fair or poor health [23]. The prevalence of no healthcare coverage in the state is 14.9%, compared to only 10.1% nationally, and one fifth of adults do not have a personal healthcare provider (PCP) [23]. WV also has a very high rate of teenage births with 22.5 per 1,000 births being from a teenager aged 15–19 in 2020 [24] compared to the national average for 2020 of 15.3 per 1,000 [24].

In summary, there is a strong relationship between teen births and poor infant outcomes. Additionally, there is strong relationship between inadequate PNC and poor infant outcomes, while little is known about the relationships between teen births and inadequate PNC. To our knowledge no other study has examined infant outcomes of teen births directly in a statewide analysis. Our study aims to fill these gaps and examine the association of inadequate PNC and infant outcomes among teenage population giving birth in WV. We hypothesize that teenage mothers who do not receive inadequate PNC will have infants who are smaller for gestational age, lower APGAR scores, more likely to go to NICU, and are more likely to have longer hospital stays than infants of teenage mothers who receive adequate PNC. This is hypothesized because PNC is designed to monitor mother and fetus for possible complications that could arise over the course of the pregnancy. Without appropriate monitoring and evaluation complications that arise could go untreated leading to poorer infant outcomes.

The exposure variable for this study is inadequacy of PNC. Inadequacy of PNC is measured as a binary variable defined as inadequate care as < 10 PNC visits and adequate care a as ≥ 10 PNC visits. While there are many ways to define inadequate PNC, this method was chosen due to a previous study using this dataset by Umer et al. which determined by way of Receiver Operating Curve (ROC) analysis that 10 or 11 PNC visits optimized the sensitivity and specificity for increased risk of infant mortality, and also determined that the strength of the bivariate associations were stronger with < 10 PNC visits being the cut off for inadequate PNC [26]. This definition of inadequate PNC is consistent with both the Kessner and APNCU indices.

The main outcome variables of interest were length of infant hospital stay (LOS), small for gestational age (SGA), infant stay in the Neonatal Intensive Care Unit (NICU) and APGAR score. APGAR score is a method for assessing an infant after birth. Elements in the APGAR score assessment include; color of infant, heart rate, reflexes, muscle tone, and respiration rate [27]. These variables of interest are based on previous literature [2, 6, 19] and information that is collected and available within the dataset. Infant stay in NICU, APGAR score and SGA were analyzed as binary variables. Infant stay in NICU was binary (yes vs no) and captures if the infant was admitted to the NICU, including those transferred to a NICU at a different hospital. For this study 5-min APGAR score that ranges from 0 – 10 was recoded into low APGAR score being less than 7 and normal being 7 or greater this cutoff value is based on literature [27]. Using data on birthweight (grams) and gestational age (weeks), gestational age categories were computed. Small for gestational age (SGA) was defined as infants born with a birth weight below the 10th percentile, appropriate for gestational age (AGA) defined as infants born with a birth weight between the 10th percentile and the 90th percentile, and large for gestational age (LGA) was defined as infants born with a birth weight above the 90th percentile. These cut-off values were based on recommendations by the World Health Organization’s (WHO) [28]. AGA and LGA were combined in this study due to both groups having very similar PNC rates. LOS was analyzed as a continuous count variable of days in the hospital, where infant discharge date was subtracted for date of birth. In this dataset the discharge date is captured on final hospital discharge so even if an infant was transferred to a different unit or hospital, the discharge date captured is the final discharge date.

Socio-demographic and confounding variables were controlled for in this analysis, including maternal race, education, parity, insurance payment method, smoking status, diabetes, substance use, and maternal age. Project WATCH collects data on race as a categorical variable with categories white, Black, Asian, Hispanic, Multiracial, and other. For this study, race was dichotomized into white vs not white. Maternal age was gathered as a continuous data ranging from 11 to 51 years old. Data was subset to only include teen births aged 19 and under. Maternal education was collected as a continuous variable and was recategorized for the purpose of demographics into 8^th grade or less, 9^th grade, 10^th grade, 11the grade 12^th grade, and some college. Parity included number of previous pregnancies and was collected as a continuous variable. Parity was then categorized as 0, 1, 2, and 3 or more for demographic purpose but was then recoded into 0 and 1 or more for simplicity in the model. For this study, insurance status was originally collected as private insurance, WV Medicaid, self-pay, other, and unknown was reclassified into private insurance vs other. Smoking was collected as nicotine use during pregnancies and was based on self-report data. Smoking status was categorized as yes vs no. Substance use data includes opioids, sedatives/hypnotics, cannabinoids, alcohol, stimulants, phencyclidine- PCP, gabapentin, and antidepressants and was based on either self-report, prenatal records, and drug test of birthing person, data categorized as yes vs no. Diabetes was collected as type I, type II, gestational diabetes, or none and was recategorized to any diabetes and no diabetes. Maternal age and education were highly correlated (r = 0.60, < 0.0001), so only maternal age was included in the models.

All statistical analysis was conducted in SAS version 9.4 (SAS Institute, Cary NC). Missing data was treated using pairwise deletion. Basic descriptive statistics were performed on all variables. Frequencies and percentages were calculated for adequacy of PNC groups in teen births categorical demographic characteristics, and covariates for the full sample, and then stratified by adequate/inadequate PNC. Chi-square tests were performed with accompanying p-values presented to determine the significance of the associations between covariates and PNC. Means and standard deviations were calculated for continuous variables. Bivariate associations for continuous variables included t-tests or Mann–Whitney tests to evaluate their relationship with PNC. Logistic regression analysis was used to examine the bivariate relationship between PNC and each categorical health outcome, small for gestational age (SGA), NICU, and low APGAR score, with covariates. Covariates were binary coded for the logistic regression analysis, including payment method (WV Medicaid vs other), Race (white vs not white), Parity (0 vs 1 or more), smoking (no vs yes), substance use (no vs yes), diabetes (no vs yes), while maternal age was kept continuous. Results are presented as odds ratio (OR) and adjusted OR (aOR) along with 95% confidence intervals (CI). Kaplan–Meier curves were used to determine if the probabilities of LOS differed between PNC groups, and a Weibull model was used to perform a survival analysis to determine the bivariate relationship between PNC groups and the continuous count outcome, infant length of stay (LOS). A Weibull model was selected due to ties, appropriate model shape, and having the lowest AIC value [29]. Adjusted and unadjusted hazard ratios (HR) and 95% confidence intervals (CI) along with p-values and regression coefficients were calculated for having inadequate PNC; this type of model was selected due to high number of ties (e.g., 2 day stays) among a large proportion of the sample. All covariates were adjusted for in the final model. Since prematurity is in the causal pathway to LOS, a post hoc sensitivity analysis was conducted to determine that appropriateness of the model by stratifying by term and pre-term births, and comparing the resulting HRs.

For all births to teens in WV during the study period, compared to teenagers who received adequate PNC, the odds of an infant admitted to NICU was significantly increased when inadequate PNC was received during the pregnancy (Table 2; aOR: 1.84, CI:(1.41, 2.42), p < 0.0001). The odds of infants having a low 5- minute APGAR score when teenagers received inadequate PNC were significantly increased (Table 2; aOR: 3.26, CI:(2.03,5.22), p < 0.0001). Increased odds were found for infants being SGA when born to teens who received inadequate PNC compared to those who received adequate PNC, those results were not statistically significant (Table 2; aOR: 1.08, CI:(0.78, 1.50), p = 0.6302).

An exploration of significant confounders also revealed interesting information about teenage births and PNC. Individuals who had diabetes (Type I, II or gestational) were almost 2.5 times greater odds (aOR: 2.46, CI:(1.57,3.85), p < 0.0001) to have an infant admitted to the NICU. It was also found that teenagers who smoke were at twice the odds (aOR: 1.98, CI:(1.52,2.58), p < 0.0001) and teenagers that used substances were 1.5 times greater odds (aOR: 1.55, CI:(1.14,2.11), p = 0.0048) to have an infant classified as SGA.

Results of the Kaplan–Meier analysis show statistically different probabilities for LOS between teens who received inadequate and adequate PNC -2log(LR) test (χ2 = 58.72, p =  < 0.0001) LOS being longer for infants of teens who received inadequate PNC (mean stay: 4.37 days (SD = 14.74) compared to 2.95 days (SD = 7.15) Shown in Fig. 1.

The Weibull estimates, HRs, and their corresponding 95% CI and p-values are presented in Table 3. The results of the survival analysis show that infants of teenagers that received inadequate PNC had longer LOS compared to infants of teenagers that received adequate PNC (Est. = -0.33, HR: 0.72, CI:(0.65,0.81), p < 0.0001). The analysis also found that infants of non-white teenagers (Est. = -0.4, HR: 0.67, CI:(0.58,0.77), p < 0.0001) and infants of younger teenagers (Est. = -0, HR: 0.91, CI:(0.88,0.91), p < 0.0001) also had longer median LOS compared to their counterparts.

There are limitations to this analysis, one of the most pertinent is the lack of information on a potential confounder of financial support within the household. There is a known association of poverty and teen pregnancy, and poverty and inadequate PNC [7, 20‐22]; not including this known confounder might mean the implications of the study are indirectly due to SES rather than directly inadequate PNC. However, the inclusion of medical insurance status in the model may help to mitigate some of this potential bias. Similarly, there are many unknown health conditions of both the infant and mother that could have an association with infant outcomes in which this dataset does not account for, one in particular being maternal hypertension, while maternal hypertension is known to be a common morbidity in pregnancy and childbirth it is not collected in this dataset in the future we hope to be able to explore the association of inadequate PNC and infant outcomes while including maternal hypertension as a covariate. We were limited to de-identified data, because of this we are unable to account for births to multiples (twins, triplets, etc.) in this analysis. The lack of an ethnicity variable creates a limitation, Hispanic is considered a race in the race variable which doesn’t permit us to analysis ethnicity directly. While the state is largely non-Hispanic future work should include analyzing a state with more ethnic diversity. This is also a cross-sectional analysis; therefore, causal inferences cannot be drawn as temporality could not be assessed. The results of this study may not be generalizable to populations outside of WV. While not generalizable, the results of the study could potentially lead further research on a wider scale to determine association between LOS, infant size, and low APGAR score and PNC among teenage pregnancies other states or on a national level. Finally, the definition of PNC was previously defined in the Project WATCH dataset and could not be adjusted to fit other indexes of PNC [15, 46]. Despite these limitations, this study demonstrates the potential harm inadequate PNC can have on infant outcomes such as LOS, SGA, and low APGAR scores among teenage pregnancies in the state of WV.