Maternal and neonatal outcomes following preterm birth: a retrospective cohort study

Introduction

In the USA, preterm birth (PTB) accounted for 10.5% of live births in 20211 and an excess of $25.2 billion in healthcare costs (2016).2

There has been a steady increase annually in PTBs in the USA.3 Current trends indicate that the country is far from reaching its goal to reduce PTBs by 50%.4 Changes in the epidemiology of PTB require ongoing research to characterise the impacts resulting from this pregnancy outcome.

Given current trends in the USA, this study aimed to characterise the individual- and hospital-level burden associated with PTB. Outcomes were assessed by gestational age during the delivery hospitalisation and up to 30 days postdischarge using a hospital administrative database that is nationally representative of inpatient admissions in the USA.

Methods

This study used the Premier (PINC) AI Healthcare Database.5 Data from hospitals and healthcare systems were transferred, transformed and validated for completeness and accuracy by the data processor (PINC AI Applied Sciences).5 Approximately 25% of all inpatient admissions annually in the USA are represented.5 These data link to the PINC AI Applied Sciences’ Maternal Health Data, which contains detailed information from >450 hospital-based facilities that have opted to submit neonatal/infant encounter data for linkage.6 Both databases, accessible to the authors, are deidentified and compliant with the Health Insurance Portability and Accountability Act of 1996 (HIPAA) in accordance with the HIPAA Privacy Rule. This study was exempt from institutional review board approval, as it does not constitute human subjects research per 45 Code of Federal Regulations 46.102.

This study was non-interventional and retrospective by design. Inclusion/exclusion criteria were prespecified. Inclusion criteria for the primary study population were as follows: individuals 12–55 years old with an inpatient delivery at ≥23 weeks’ gestational age between 1 January 2016 and 30 September 2021. Linked neonatal records were available from 1 January 2019 to 30 September 2021, so the analysis of neonatal outcomes was restricted to this period. Prespecified exclusion criteria for the primary study population were as follows: individuals with a pregnancy that did not reach 23 weeks’ gestational age and pregnancies that were terminated,7 8 resulted in a stillbirth or were of a higher order multiple (≥3 fetuses). Deliveries with no or with conflicting gestational age diagnosis codes were also excluded. For the stillbirth exclusion criterion, the singleton pregnancy must have resulted in a stillbirth; in the case of twins, the pregnancy must have resulted in both twins stillborn. Only the first delivery was evaluated for individuals with repeat pregnancies in the data.

Additional exclusion criteria were applied to derive a subset of the primary study population; this subset was evaluated in sensitivity analyses and consisted of spontaneous PTBs only (rather than all eligible PTBs as in the primary study population). The following exclusions were applied for this subpopulation: evidence of a complicated pregnancy based on the presence of one or more of pre-eclampsia; eclampsia, hemolysis, elevated liver enzymes, and low platelets (HELLP) syndrome; fetal growth restriction; polyhydramnios; oligohydramnios; premature rupture of the amniotic membrane; intra-amniotic infection; placenta previa; placental abruption or pre-existing diabetes; any pregnancy with evidence of a neonatal, congenital or chromosomal condition; and any pregnancy with a non-reassuring fetal status code documented on the maternal record.

Mutually exclusive cohorts were formed based on delivery at <32 (very preterm), 32 to <37 (moderately preterm) or ≥37 (term) weeks’ gestational age. The index date was the date of first delivery; the index encounter was the delivery hospitalisation in which the index date occurred.

The baseline period was defined as the period between 20 weeks’ gestational age and the delivery hospitalisation for the maternal record. Maternal characteristics at baseline were reported overall and by cohort and consisted of age; race/ethnicity (Hispanic, non-Hispanic White, non-Hispanic Black, non-Hispanic Asian, other, unknown); smoking status during pregnancy (yes/no); payer status (commercial, Medicaid, Medicare, other); delivery year; hospital location (rural/urban); hospital region (Midwest, Northeast, South, West); plurality (singleton/twin); mode of delivery (vaginal/caesarean); the obstetric comorbidity score for severe maternal morbidity (SMM)9 and the obstetric comorbidity score for non-transfusion SMM.9

The obstetric comorbidity score is a validated measure whose total score is predictive of SMM.9 26 pre-existing and obstetric comorbidities plus maternal age are included, and it can be applied to any maternity discharge dataset using the International Classification of Diseases, Tenth Revision codes. Two total scores based on the inclusion or exclusion of blood transfusion-only cases are available.9 The linear combination of scores constitutes the total score for SMM and, separately, non-transfusion SMM; higher scores indicate greater comorbidity. The individual components and total scores are reported.

Prespecified data fields were used for all sociodemographic covariates. Diagnostic and procedural codes were used for sample selection, cohort identification and for all other covariates (plurality, vaginal/caesarean delivery, comorbidities). All codes used in this study are in online supplemental table S1. Covariates defined by diagnostic or procedural codes were assigned a ‘1’ if the code was documented and a ‘0’ if not. Covariates like race and ethnicity are assigned categories including ‘other’ and ‘unknown’ by the data processor prior to release; these data fields were analysed by their classification rather than as missing. Sample selection codes used for the subpopulation of spontaneous PTBs, examined in sensitivity analyses, are in online supplemental table S2.

Maternal outcomes examined were postpartum complications defined by a composite of sepsis, shock, acute renal failure, cardiac event, thromboembolic event, acute respiratory distress syndrome or haemorrhage; intensive care unit (ICU) admission; death; index encounter length of stay (LOS) and hospital cost of care and readmission within 30 days postdischarge from the index encounter. Neonatal outcomes consisted of morbidity defined by a composite of respiratory distress syndrome, bronchopulmonary dysplasia, haemorrhage, periventricular leukomalacia, retinopathy of prematurity, sepsis, meningitis or necrotising enterocolitis; neonatal death; neonatal ICU (NICU) admission; NICU LOS and hospital cost of care and readmission within 30 days postdischarge from the index encounter. For composite outcomes, occurrence was defined by documentation of any one of the components.

Delivery costs and LOS were assessed during the index encounter; all other outcomes were assessed during the index encounter and any readmission(s) within 30 days of discharge. Costs were inflated to 2022USD and reflect total costs to treat the individual during the encounter (they are not total charges for billed items during the encounter). Codes for the study outcomes are in online supplemental table S3.

Covariates were assessed descriptively overall and by cohort. Means and SDs were examined for continuous covariates; frequencies and percentages were examined for categorical covariates. Means were compared using the Mann-Whitney U test; percentage distributions were compared using Fisher’s exact test.

An assessment of model fit supported the use of regression models with logit link and binomial distribution for dichotomous outcomes, log link with negative binomial distribution for count outcomes and log link with gamma distribution for cost outcomes. Two-part generalised linear models were used for conditional outcomes. Bivariable and multivariable regression analyses were conducted.

The following covariates were specified a priori for all multivariable regression models: maternal race/ethnicity, payer status, smoking status, hospital location, region, obstetric comorbidity score for SMM, plurality, mode of delivery, and delivery year. Maternal age was included in the obstetric comorbidity score and not as a standalone covariate. The obstetric comorbidity score for non-transfusion SMM is a subset of the obstetric comorbidity score for SMM; consequently, only the latter was included.

Sensitivity analyses were conducted whereby all statistical analyses were replicated for the subpopulation of spontaneous PTBs.

Adjustment for multiple comparisons using the Bonferroni method was performed. A 2-sided α-level=0.0025 (ie, 0.05/20 endpoints) indicated statistical significance in regression analyses. SAS V.9.4 (SAS Institute, Cary, North Carolina) was used.

Results

A total of 4 303 772 maternal records were included, of which 635 569 (14.8%) were linkable to neonatal records (online supplemental table S4). There were 65 531 individuals who delivered very preterm, 359 272 who delivered moderately preterm and 3 878 969 who delivered at term. Among the neonates included, 10 105 were born very preterm, 60 857 were born moderately preterm and 564 607 were born at term.

Maternal characteristics are shown in table 1. Most (88.1%) births occurred in an urban setting, 45.9% were in the South, 98.3% were singletons and 32.9% were caesarean deliveries. Comorbidity scores indicated significantly higher comorbidity per person in the PTB cohorts relative to the term cohort (table 1); the frequency of each component was higher among the PTB cohorts than in the term cohort (online supplemental table S5). Characteristics among all eligible births and among spontaneous PTBs were consistent (online supplemental tables S6 and S7).

Relative to term delivery, PTB was independently associated with worse outcomes (table 2). PTB had significantly higher per-person hospital resource use and costs relative to birth at term (table 3).

Figure 1 shows that the frequency of postpartum complications followed a decreasing trend from 23 weeks’ gestational age until week 39. At this point, the frequency of complications steadily increased; a non-linear trend in the outcome across weeks at term was observed. Figure 2 illustrates trends in the frequency of individual morbidities comprising each of the maternal and neonatal composite outcomes, respectively. Postpartum haemorrhage was the most frequent maternal morbidity across weeks’ gestational age (figure 2A), and respiratory distress syndrome was the most frequent neonatal morbidity across weeks’ gestational age (figure 2B).

Spontaneous PTB was independently associated with worse outcomes relative to birth at term (online supplemental table S8). Sensitivity analyses also showed significantly higher per-person hospital resource use and costs associated with spontaneous PTB (online supplemental table S9).

Discussion

In this study, PTB was independently associated with worse maternal and neonatal outcomes during the delivery hospitalisation and up to 30 days postdischarge, as well as notably higher hospital-based resource use and costs per person compared with birth at term. Results were consistent with those reported among the spontaneous PTB sample. The most pronounced burden was consistently observed among the very PTB cohort. The aOR for maternal postpartum complications among the moderately PTB cohort in the main analysis is interpreted in the context of the totality of the results inclusive of the risk per 10 000 deliveries reported as well as maternal ICU admission and resource use, which can reflect severity of the maternal condition and the non-linear trend in this outcome across weeks gestational age at term. Collectively, the results demonstrate significant maternal-, neonatal- and hospital-level burden associated with overall and spontaneous PTB in the USA.

PTB in the USA is a public health concern. Rates remain staggeringly high and may reflect a lack of prioritisation of maternal, neonatal and infant health in the USA.10 To date, this study is one of the largest population-based studies of its kind. It is the only study to address the research question of interest using the PINC AI Healthcare Database and linked Maternal Health Data. The estimates reported have good external validity as evidenced through the comparability of key characteristics between the overall maternal study sample and national birth statistics from 2016 to 2021 reported by the Centers for Disease Control and Prevention.11 Given the known changes in the epidemiology of PTB, including increased incidence, geographic variability and socioeconomic disparity,3 10 12 this study warrants an urgent need to advance the healthcare landscape, improve outcomes for mothers and their newborn babies and inform the prioritisation of strategies intended to reduce the frequency of PTB.

The results for postpartum complications in the main analysis must be considered in context. The magnitude of association at 32 to <37 weeks’ gestational age is small despite being statistically significant. Importantly, the magnitude of association reflects the non-linear trend in the frequency of complications across weeks’ gestational age at term; this is consistent with mounting evidence reporting improved outcomes with birth at 39 weeks.13–16 Hence, these results, in the context of all evidence from this study, including the findings for spontaneous PTBs, indicate that maternal outcomes are worse among individuals delivering preterm compared with at term.

The data from this study have important implications for both clinical practice and policy. For example, prenatal education on risk factors for PTB and its epidemiology is warranted for all individuals who are pregnant. In this study, the prevalence of comorbid conditions and other risk factors like smoking, substance use and mental health disorders was higher among individuals who delivered preterm; the highest and lowest estimates were consistently observed in the very preterm and term cohorts, respectively. Many risk factors like smoking are modifiable; thus, transparent and continual counselling, education and conversations between individuals who are pregnant and their healthcare providers should be standard practice. To alleviate the onus that falls on individuals who are pregnant, healthcare providers should proactively engage with their patients to inform them of the risks and outcomes associated with PTB. Knowledge of available resources and support networks before, during and after pregnancy is necessary and can be achieved through regular communication with healthcare providers.

The data also highlight the need for policy makers to advance the healthcare landscape. Key policy actions are outlined in the March of Dimes Report Card and include but are not limited to the expansion of benefits, education, parental support networks and funding for research on predictive, diagnostic and therapeutic innovations to prevent PTB and/or improve PTB care strategies.10 Solutions for optimising the allocation of medical resources to mitigate the economic burden of PTB are essential. Data from this study, and future insights using the PINC AI Healthcare Database, serve as important resources to inform policy.

The PINC AI Healthcare database is a nationally representative and externally valid hospital-based database which limits selection bias. Nonetheless, mention of study limitations is warranted; these include known limitations consistent with those for secondary healthcare data sources, including potential coding errors and incomplete or missing data. The study design and methodology precluded an assessment of the multiple causes of PTB and causality in general, which would require cause-specific adjustment for relevant clinical characteristics to the extent they are observable in the data source. Although the sensitivity analyses focused on a subsample consisting of spontaneous PTBs, future research is warranted to better understand the causes and mechanisms of PTB in relation to outcomes. This study enabled an assessment of statistical associations only; future studies may consider incorporating other methods, prespecified a priori, that better support causal assessments in observational studies, such as inverse probability of treatment weighting. Still, these methods rely on observed (measured) confounders. Thus, unmeasured confounding and therefore residual confounding may still be present; as a result, the ability to draw causal conclusions is hampered. Additionally, baseline data prior to delivery hospitalisation may be limited in this study. Most covariates, however, were able to be measured during the delivery hospitalisation. In this study, 14.8% of maternal records were linked to a neonatal record; although this represents a fraction of the overall sample, the size of the linked data renders meaningful evaluations to be conducted. This study could not distinguish first-ever births, as births outside the database network were not captured. This is not considered a major limitation, however, given the comparability in the distribution of key characteristics between the study sample and birth estimates reported by the Centers for Disease Control and Prevention over a comparable period.11 Finalised hospital payment amounts were not available in the data and may differ from the costs reported, which represent hospital costs incurred during the hospital encounter. Study results are conservative, as non-medical and/or longer-term maternal and neonatal outcomes and costs were not examined (due to limited observation in the hospital-based dataset). Given the potential life-long impact of complications from PTB, additional analyses are needed to incorporate non-medical and/or longer-term outcomes and costs; coupled with the outcomes and costs reported in this study, PTB lifetime costs can then be re-estimated. The present study examined composite measures of maternal and neonatal complications; as such, further research into the relationship of PTB and specific individual morbidities, as appropriate, would be worthwhile to pursue in future research. Study replication on a regular basis is necessary to assess societal progress towards the goal achievement of reducing PTB; this is feasible given the rate at which data are accrued and updated by the data processor. The generalisability of the study findings may be limited by study inclusion and exclusion criteria and the limited proportion of neonatal records that were able to be linked.

Conclusion

This study found a significant individual- and hospital-level burden associated with PTB and spontaneous PTB in the USA. It provides robust and modern-day population-based evidence signalling a clear need for prioritisation of maternal and neonatal health in the USA.