In order to model (trends in) the number of births in the population, we retrieved all historical data of expected birth dates from 2014 to 2019 of Saltro, as registered during the screening around 11th week of pregnancy, and linked them to a dataset of Statistics Netherlands with live births from 2015‐ to 2017 as outcomes. To ensure a representative dataset, while also ensuring that data loss was minimized, we included only data from municipalities where Saltro had data on at least 55% of pregnancies. We aggregated the data to the number of births per week. We accounted for the uncertainty of the expected date of birth by using the national pregnancy length distribution (in weeks) of the Dutch national perinatal registry, Perined [8]. According to the Dutch legislation, this research did not fall under the Medical Research Involving Human Subjects Act (WMO). Only anonymized aggregated data were used for analysis. The research was conducted according to the declaration of Helsinki. Health care professionals provided verbal informed consent prior to the interviews.

To correctly forecast the number of births per week, we used the expected birth date as the base model. Consequently, we distributed the number of births for each week over the surrounding weeks according to the Perined distribution (Figure 1) and repeated this process 1000 times per week. This resulted in a calculated mean and SD per week. For a certain week, the number of expected date of births would actually be forecasted in the period between 17 weeks before and 5 weeks after the denoted week. When repeating this process for all weeks, the number in every week would become a sum of all these calculations.

To validate the modeling output on actual outcome data, we correlated the mean forecasted number of births for each week with the amount as reported by Statistics Netherlands. We combined the number per week to 2- and 3-weekly bins (ie, summing up the numbers for week 1 and 2, week 3 and 4, etc), and assessed whether this improved the correlation and thus predictive accuracy. Considering the potentially high variance of calculated number of births, binning the project number of births to 2- or 3-weekly bins may help smooth out large weekly differences. To compare the mean estimate and a 95% CI of our model to the actual number of births, we normalized the visualization of these two types of data, so that these two were comparable.

This work, and the development of the resulting tool, was performed using the Python programming language (version 3.8, Python Software Foundation) and the pandas (Version 1.1), numpy (version 1.19), and scipy (version 1.5) packages. We used the plotly package (version 4.14) for interactive visualization purposes.

After the model for the tool was validated, we presented our findings to stakeholders in the obstetrics field, including future end-users of the tool, such as managers and obstetric care providers (n=13) during the weekly consultation of the regional obstetric acute care (ROAZ midden Nederland). After an introduction explaining the modeling and the visualization, we sent the forecast to this status meeting on a 2-week basis.

In order to assess the added value of the forecasting tool in the obstetrics care, we invited regional stakeholders (eg, gynecologists, midwives, and managers) from the network regional consultation acute obstetric care (ROAZ Midden Nederland) for an evaluative stakeholder meeting 6 months after implementation of the tool. To this purpose, we created a topic list to investigate the experiences with staff planning and resource management after implementation of the tool as compared to before this implementation, and the experience with use of the forecasting tool (Table S1 in Multimedia Appendix 1). We then analyzed this discussion using Nvivo (version 12.6.1.970; QSR International), and extracted information about the aforementioned topics, as well as remarks on the tool, including possible areas for further research and improvements of the tool.

According to the Dutch legislation [9], this research did not fall under the Medical Research Involving Human Subjects Act (WMO). Only anonymized aggregated data were used for analysis. The research was conducted according to the Declaration of Helsinki. Health care professionals provided verbal informed consent prior to the interviews.

In total, Saltro screening data on 28,361 pregnancies were used from 23,778 patients with births per year ranging from 5052 in 2014 to 6047 in 2017 (mean 5681 per year). The Statistics Netherlands data from the included municipalities encompassed 35,394 births with births per year ranging from 8735 in 2016 to 9051 in 2014 (mean 8849 per year). The municipalities that were included in the development of the forecasting tool are given in Table 1, along with the number of pregnancies as reported by Saltro, and the actual number of births as reported by Statistics Netherlands.

The correlation between the modeling output and number of births as reported by Statistics Netherlands was 0.45 for weekly data. For the 2-weekly bins, we found that the correlation between the model output and the Statistics Netherlands increased to 0.61, and for 3-weekly bins, this was 0.67 (Figure 2; Figures S1 and S2 in Multimedia Appendix 2). All actual number of births were within the forecasted 95% CI, after normalization for differences in scale between modeling output and actual numbers (Figure 2).

The stakeholders chose a 2-weekly bin for forecast-width as this was considered a good trade-off between actionability with regards to planning (smaller bins mean higher granularity), but also having good predictive power. Additionally, the numbers were presented with an overall average over the preceding years (2015‐2020). From July 2021 onwards, we provided these forecasts every 4 weeks; in September, this interval was shortened to 2 weeks upon request. We performed forecasts over 13 weeks in advance, as we observed that forecasts using this time window were most stable in comparison with follow-up forecasts, because of incomplete data in the later weeks.

In total, 4 stakeholders attended the evaluative stakeholder meeting, including 3 stakeholders that worked in hospitals and 1 stakeholder that worked as community midwife.

In this study, we developed, validated, and implemented a tool to forecast and visualize trends of births in the greater Utrecht region, using routinely measured and collected calculated date of births by a primary care laboratory, to support management of the regional acute obstetric care, including staff planning. We developed a forecasting model and found that a 2- and 3-weekly binned forecasting model provided good predictive power when it comes to forecasting trends within birth numbers. Subsequently, we presented our findings to stakeholders in obstetrics and then implemented a tool to visualize our forecasts using a 2-weekly binned forecast. Although 3-weekly binned forecasts showed slightly higher predictive power, the choice for 2-weekly binned forecast was made, as this provided a more granular view of obstetric care demand. After 6 months, using an evaluative stakeholder meeting, we found that, although our tool was not used in day-to-day staff planning but rather on a (global) operational level, we were able to add value to the obstetrics field by adding a sense of urgency, resulting in improved regional collaborations and operational actions and improving QoC in obstetrics.

QoC is multifaceted. According to stakeholders, the use of our tool in daily practice enabled the focus on timeliness. Simultaneously, the tool supported collaborations in obstetric care, leading to more efficient and available obstetric care. A well-coordinated acute obstetric care chain is important, as impaired communication between health care providers is detrimental to patient safety and continuity of care [10].

We implemented the forecasting tool during the COVID-19 pandemic. This crisis sparked the demand for more insights into obstetric health care demand, especially since during the pandemic, births varied more than before the pandemic. Examples include a birth wave after lockdown and fewer preterm births during the first lockdown [11,12]. Similar to other literature, we found a decrease in birth numbers in 2020 and an increase in birth numbers in 2021. Additionally, the surge of COVID-19 required more flexible resource management, as resource requirements changed or were scarce [13]. Considering that our data are able to capture the true birth rate in the general population opens up additional angles for research. Further study could focus on the decline in birth rate as seen in Nordic countries and Germany after 2021 [14,15], by studying the trend in expected births in our data. The increase in care demand and need for flexible resource management due to the pandemic fueled the development and implementation of the tool, as regional resources had to be more efficiently allocated due to high demands. Stakeholders reported that our tool was helpful in achieving these aims, by offering concrete data that heightened the sense of urgency within the obstetric care.

The academic literature and practice alike highlight the challenges of forecasting acute care demand, as it is often driven by unforeseen events or factors that fall outside the scope of routine data collection [1,2]. Indeed, we observed in our data that it is very hard to forecast the weekly number of births, considering their weekly variance. However, widening the scope of the forecasting interval and retrieving information from these data, such as trends, are of added value to the obstetrics domain according to stakeholders in obstetric care.

Our research should be understood in the light of the following limitations. We have included only limited data, originating from municipalities for which data on at least 55% of pregnancies was available. The extent to which our study therefore suffers from selection bias remains unclear. As the obstetrics care in the Netherlands and the reporting of calculated date of births is standardized, we believe bias is limited. However, future studies could focus on combining data sources for the estimated dates of birth, in order to increase the data coverage, and to mitigate reliance on a single data source. Additionally, our dataset is relatively small. With larger numbers, the use of statistical models such as Autoregressive Integrated Moving Average, or machine learning models may be beneficial [16]. Considering we had data on the actual number of births for 3 years, the use of such models was currently infeasible. To the best of our knowledge, no other forecasting model for the number of births is currently used in the Netherlands.

As stakeholders have stated in the stakeholder meeting, there are several directions for further research. First of all, data should be shared between stakeholders to improve regional (acute) obstetric care. This could help further improve forecasts, as this can help identify subgroups of pregnancies, so that these groups and their health care demand can be further monitored. Additionally, data sharing could lead to further standardization of data collection, making data collection and research to improve obstetric care more feasible and impactful.

Implementation of a forecasting tool for number of births based on available data across the health care system during the COVID-19 pandemic led to increased communication, awareness, and collaboration between obstetric health care providers, showing that combining several sources of data within the routine acute care can add value to the routine acute care process. Providing visualizations of forecasted births created a sense of urgency about the high care demand situation and the impact of this demand on quality of care. Our tool thereby facilitated improved collaborations and cohesion between health care providers in the Utrecht region and aided further initiatives that can increase efficient use of resources in response to high care demand beyond pandemic situations. Further research should aim at improving regional obstetric acute care by fostering data sharing in order to improve health care demand forecasts.