Use of Electronic Medical Records to Estimate Changes in Pregnancy and Birth Rates During the COVID-19 Pandemic

Key Points Question Can electronic health care records be used to monitor and project changes in pregnancy and birth rates after the COVID-19 pandemic societal shutdown? Findings In this cohort study of pregnancies within a large US university health care system, a model using electronic medical records (used retrospectively from 2017 and modeled prospectively to 2021) projected an initial decline in births associated with the COVID-19 pandemic societal shutdown, predominantly related to fewer conceptions following the societal changes instituted to control COVID-19 spread. This decline was followed by a projected birth volume surge anticipated to occur in summer 2021. Meaning These findings suggest that electronic medical records can be used to model and project birth volume changes and demonstrate that the COVID-19 pandemic societal changes are associated with reproductive choices.


Introduction
The COVID-19 pandemic and associated societal measures to control the spread of the virus have brought about significant changes in almost every aspect of life in the US and globally. The economic and societal effects of the pandemic are vast, with consequences not only for large systems such as health care and education, but also for individuals and families. Previous large societal disruptions in the US, such as the 1918 H1N1 Influenza pandemic, 1 Great Depression (1929), 2 and Great Recession (2008) 3 have influenced population growth and fertility rates, but the exact effects of COVID-19 on fertility and birth rates are speculative to date. In the US, a multitude of factors could variably influence pregnancy rates. For example, economic concerns may lead people to postpone conception, whereas decreased access to contraceptive services could lead to increased unintended pregnancy rates. A 2020 Guttmacher Institute study 4 demonstrated that 40% of women reported changes in plans for childbearing due to the COVID-19 pandemic, 41% of women with children reported worry about being able to take care of their children, and 33% of women had to delay or cancel an appointment for reproductive health or contraception care. Additionally, the COVID-19 pandemic has highlighted longstanding economic and health disparities in the US, 4-6 yet how such disparities will influence fertility rates and obstetric complications are unknown at this time.
Anticipatory planning for birth rates is important for health care systems to appropriately anticipate increasing or decreasing staffing needs and patient volumes. Population size and population dynamics are of interest to economists to document size of economy and model working and/or aging populations. Often, the consequences of major societal events such as economic and natural disasters or infection pandemics are documented only after the fact or as decreasing birth rates are noted. At the University of Michigan Hospital, we use projection modeling and active management of estimated date of deliveries (EDD) to control obstetric birth volumes in our system and anticipate staffing needs for our birth center. In this analysis, we applied our EDD management modeling techniques to project anticipated delivery volumes after the COVID-19 pandemic and describe the projected decrease in birth rates from our center.

Methods
This observational cohort study including all pregnancy episodes within the health care system of the University of Michigan Hospital, a large US academic hospital, examined birth rates retrospectively from 2017 through current pregnancy episodes and modeled them prospectively to October 2021.
The primary exposure was the COVID-19 pandemic societal shutdown. The stay-at-home order in Michigan was placed on March 15, 2020. Our primary outcome of interest was the start of pregnancy episodes within our health care system, trends in the volume of pregnancies, and evaluation of potential explanations for these volume changes. This analysis was approved by the University of Michigan institutional review board, which exempted the study because it used deidentified data kept by the system for institutional quality improvement, and deemed that no direct informed consent from individual patients was required. Data reporting and analyses followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies.
Pregnancy episodes begin with a patient-initiated contact with our health care system to request pregnancy-related care, such as initial prenatal visit or ultrasound for pregnancy dating.
Pregnancy episodes remain open until 1 of the following events occurs: (1) delivery, (2) documented pregnancy loss (ie, miscarriage, ectopic, etc), or (3) no contact with the patient for 90 days after the EDD. The final EDD was determined by best obstetric estimate 7 based on last menstrual period (LMP) and ultrasound and is associated with the pregnancy episode in the medical record. We excluded any deleted episode that had been entered in error as noted by a comment entered in the electronic medical record (EMR).

Statistical Analysis
We conducted descriptive data analyses using the pregnancy episode data from our health care institution's internal EMR system. The data included patient-specific demographic characteristics (eg, age and race), clinical characteristics (eg, parity), and socioeconomic indicators (eg, insurance status, zip code of residence). Race/ethnicity is patient-reported and documented in the EMR. Because this analysis consists of pregnancy episodes, no sex-specific analyses were performed. Baseline demographic characteristics in our obstetric population were analyzed in 2 ways: an annual comparison of pregnancy episodes between 2019 and 2020 and a comparison of new pregnancy episodes started before and after the COVID-19 mandated societal shutdown measures in March 2020. For the before and after comparisons, we defined time periods as January 1 through March 31, 2020, as the pre-societal shutdown period, and April 1 through June 30, 2020, as the post-societal shutdown period. Comparisons in demographic characteristics were compared with parametric or nonparametric tests as appropriate for continuous variables and with χ 2 tests for categorical variables.
We analyzed the volume of initiation of new pregnancy episodes in our system compared with prior years and explored the association between the COVID-19 lockdown measures on volume of pregnancy episodes and projected future births. We identified the number of pregnancy episodes that began each week between January 2017 and March 2021 and used an interrupted time series study to characterize the differences in pregnancy starts both before and after the initiation of the spring lockdown. A Poisson regression model was used, and Fourier terms were used to adjust for annual seasonal patterns in conception, as described by Bernal et al. 8 Because there is a delay between conception and the start of a pregnancy care episode of at least 2 weeks, we assigned the intervention binary variable of pre-and post-societal lockdown to begin March 29, 2020, 2 weeks after the start of the spring stay-at-home order, and last until June 14, 2020, 2 weeks after the end of the order. To examine whether prenatal care was being delayed during the COVID-19 pandemic, we examined the time between LMP and contact with the health care system in the time frame before the COVID-19 shutdown to the time period after the shutdown using both the Kolmogorov-Smirnov test and the Fleming-Harrington test for right-censored, 2-sampled time data. We projected future delivery volumes each month from October 2020 through February 2021 using EDDs of active pregnancy episodes with a growth rate applied, which represented the volume of patients who would deliver in an upcoming month who did not yet have a record in the EMR. This growth rate was based on data from 2019. We tested the accuracy of our birth volume projection models compared with actual birth volumes for December 2020 through February 2021.
We explored factors contributing to observed pregnancy volume changes including COVID-19 shutdown of in vitro fertilization (IVF) cycles and preterm birth rates. Rates of preterm births (ie, delivery <37 weeks' gestation) per week were explored and quantified using changepoint analysis to determine whether an abrupt change occurred during time-series data. Preterm birth rates before and after the identified changepoint were statistically compared using the Kolmogorov-Smirnov test.
Two-sided P < .05 were considered significant. Statistical analyses were performed with the R statistical programming language version 3.6.0 (R Project for Statistical Computing) and Microsoft Excel (Microsoft Corp). Missing data were handled in the following manner: missing race data were included in the other category, unknown Hispanic ethnicity was categorized as non-Hispanic, unknown insurance was categorized as non-Medicaid/Medicare, unknown zip codes were categorized as greater than 30 miles away, and unknown age was excluded from age analysis but included in pregnancy episode and birth projection analysis.   We used our institutional EDD capacity modeling to project future delivery volumes based on pregnancy episodes in our system (Figure 3). In this modeling, we captured known EDDs within our system and calculated anticipated delivery volumes based on those EDDs. As the time latency to  EDD increases, there are fewer EDDs in our system, which is expected because the early gestational ages of those pregnancies with EDDs that are more than 30 weeks away increase the proportion of EDDs that have not yet presented for care. Sixty days away from any given EDD, we are aware of 95%

Results
of patients who will deliver on a given day, and even 150 days (approximately 5 months) away from any given EDD, we have relatively complete information about projected deliveries (83% are known).
Based on our model, decreased delivery volume was expected from October 2020 through Given that pregnancy episodes are closed after delivery occurs, if there was a sizeable increase in preterm births, pregnancy episodes would fall out of the system and could contribute to the appearance of decreased pregnancy episode volumes. We found, on the contrary, that the overall preterm birth rates decreased after the onset of the COVID-19 pandemic (eFigure 2 in the Supplement). Changepoint analysis to detect abrupt changes in time-series data determined that July 2020 represented the time point after which preterm birth rates appeared to be different. Prior to July 2020, the rate was 13.3% (or 12.0 preterm births per week of 89.8 weekly births) vs 10.2% from July to December 2020 (9.1 preterm births of 89.7 weekly births) (P = .001).

Discussion
In this cohort study, we demonstrated EMR data on pregnancy episode volumes and projected birth volumes could be monitored and projected with accuracy without waiting for changes in birth volume to signal decreasing (or increasing) birth rates after major societal events. We documented an anticipated decline in births after the COVID-19 pandemic, starting in November of 2020 and persisting until spring of 2021, after which is projected a rebound in anticipated births that may exceed anticipated birth volume based on annual trajectories derived from the prior 5 years of institutional data. In our institution, we modeled pregnancy episode volume to surveil for capacity constraints and projected periods of high and low delivery volume for internal planning reasons.
These same modeling techniques can be applied to estimate impacts on anticipated birth rates within a hospital or health care system, or for local/state epidemiologic surveillance. Our data suggest that the anticipated decrease in the birth rate may be best explained by lower conception rates in the weeks and months immediately following the March 2020 COVID-19 pandemic major societal shutdown. Additionally, we found that preterm birth rates may have decreased after the COVID-19 pandemic shutdown.
Pandemics and other major societal events alter population dynamics by both changing fertility rates and changing aging and death rates. [1][2][3][9][10][11] Changing birth rates in other societal crises have been linked retrospectively to changes in economic conditions, morbidity and mortality rates among reproductive age populations, and other destabilizing societal conditions (eg, separations caused by war deployments, access to health care/contraception). Often, changes in birth rates are recorded as birth rates change, not modeled prospectively to anticipate these changes and plan accordingly. 1 How the COVID-19 pandemic may affect birth rates has been speculated in the lay press 12-17 but has not been fully documented, even as the societal impacts of the pandemic have persisted for longer than a year.
Population dynamics are of interest for governments, businesses, and economists because fluctuations in young and aging, workforce, and school-aged populations are critical variables in the ability to plan appropriately for social well-being, to make investments, and to anticipate economic patterns. 9,11,18,19 In fact, encouragement of childbearing is the focus of recent government and societal policies, such as 12-month paid parental leave and other financial bonuses for childbearing in countries concerned with declining fertility rates. 18

Strengths and Limitations
Strengths of our study include the ability to use novel modeling techniques to project birth rate volumes with relative certainty prospectively instead of waiting until decreased birth rates are observed. Additionally, by layering on demographic characteristics, such as maternal age and race/ ethnicity, assisted reproduction, and preterm birth rates, we can estimate the effects of the COVID-19 pandemic on birth demographic characteristics prospectively. We could apply changing demographic patterns within our models to anticipate fertility rates in specific patient populations, plan for highand low-risk obstetric volumes, and plan for neonatal intensive care unit and pediatric subspecialty needs.
This study had several limitations that should be considered. First, our data are based on a single tertiary care academic center that serves both a local population as well as transported referral populations from throughout the state of Michigan. Thus, our findings may or may not be generalizable to other centers or other regions of the country. Second, we found no significant maternal demographic differences from pre-to post-COVID-19 pandemic time frames. This may be because it is too early to detect those nuanced changes or because of the specific demographic composition of our obstetric population. Third, our projections require that the pregnancy be known to the health care system. Thus, we fail to capture early pregnancy losses, terminations, or pregnancies that might be ongoing but not presenting for prenatal care.

Conclusions
In this cohort study, we documented decreased birth rates following the COVID-19 pandemic societal changes, followed by a projected birth volume surge, suggesting that major societal changes may factor into reproductive choices. We demonstrated the use of EMR modeling to project birth rates and investigate in real-time the underlying reasons for changes in observed birth rates. These projection modeling techniques can be used in partnerships between hospitals and governmental or societal organizations to minimize the detrimental effects of the COVID-19 pandemic on society.