The goal of this study is to pilot a program that derives gestational age estimates from dried blood spot samples (heel-prick or cord blood) collected from a population-representative cohort in health and demographic surveillance sites in low-resource settings.

If successful, this will achieve the following objectives:

Primary objectives

Secondary objectives

At a population level, utilization of metabolic analysis for gestational age dating could support surveillance of the burden of preterm birth and research into the development of effective resource management strategies. We will seek to determine whether this is feasible in low-and middle- income settings.

This study was approved by the Ottawa Health Sciences Network Research Ethics Board (20180330-01H), Children’s Hospital Of Eastern Ontario Research Ethics Board (18/58X), the Stanford University School of Medicine Institutional Review Board (44656), the Kenya Medical Research Institute Scientific and Ethics Review Unit (SSC 2880), the University of Zambia Biomedical Research Ethics Committee (015-06-17), the Bangladesh Institute of Child Health (BICH-ERC-01-01-2019), and the Research Council of Zimbabwe (03744).

The Ottawa Hospital Research Institute (OHRI) has received funding from the Bill & Melinda Gates Foundation to cover project coordination costs in Ontario, Canada, including but not limited to sample analysis, gestational age estimation model execution and project management (OPP1184574). Stanford University received funding from the Bill & Melinda Gates Foundation to operationalize the project including site selection, protocol training, subcontracting to international sites, oversight and communication with international sites (OPP1182996).

This project is led by the OHRI with services provided by Newborn Screening Ontario (NSO), in partnership with the Department of Pediatrics, Stanford University and supported by the Bill & Melinda Gates Foundation. Sample collection sites include: Mirzapur, Bangladesh led by the Child Health Research Foundation; Kisumu, Kenya led by a team from the Kenya Medical Research Institute (KEMRI) and Center for Disease Control and Prevention (US CDC) collaboration; Lusaka, Zambia led by UNC-GPZ; and Harare, Zimbabwe led by researchers from University of Zimbabwe.

Investigators at Stanford University were responsible for identifying and training international partners in sample collection at facilities located in South Asia and sub-Saharan Africa. Identification of eligible site partners was based on capacity for sample collection from population representative cohorts and the ability to collect samples from both hospital and primary birthing facilities in order to evaluate whether metabolic gestational age dating approaches were valid estimates of population-level preterm birth rates.

We evaluated 19 sites for inclusion in the study. Sites were deemed acceptable if they met the following specific criteria: early pregnancy registration, ultrasound capacity, ability to obtain newborn heel-prick and cord blood samples, capacity to store and ship blood spot cards and cord blood samples, ability for patient follow up and demographic surveillance, and ability for follow-up testing for screen positive results for treatable metabolic diseases. Four sites were identified that broadly met our inclusion criteria. All sites were trained on cord blood and heel-prick derived dried blood spot collection.

Mizapur, Bangladesh. The Child Health Research Foundation is a demographic surveillance site (DSS) with births occurring at the Kumudini Hospital, local clinics, and at home. The DSS sites covers 374 km ², with a population of close to 300,000 people. One data collector covers approximately 400 households and visits every four months to collect demographic information. The preterm birth rate at this site is estimated to be 18–19% based on LMP. Half of the study population will be enrolled at the time of birth at the birthing center at the Kumudini Hospital, and half will be enrolled at 20 or before weeks’ gestation through the DSS to allow for early trimester ultrasound validation. Women enrolled at 20 weeks gestation will be transported to Kumudini Hospital for ultrasound examination. Heel prick samples will be collected for all infants, and cord blood samples will also be collected for infants born at Kumudini Hospital. An estimated 840 cord and 1455 heel prick samples will be collected from this site over an estimated 12-month period.

Kisumu, Kenya. The Kenya research center is located in Kisumu, at the KEMRI Center for Global Health Research. Field research sites are located in Siaya county where a maternal-infant surveillance platform hosting a prospective cohort of pregnant women and their infants is implemented from two community hospitals; Siaya County Referral Hospital (SCRH) and Bondo sub-County Hospital (BSCH). The SCRH is the main referral hospital for the county region. Bondo Hospital, located 23 km south, serves as a primary birthing center and enrollment site for the community. Prior studies have estimated the preterm birth rate to be 19% by a combination of LMP and ultrasound. Participants are enrolled at their first ANC visit or by research nurses or through trained community health workers (CHWs) who administer home-based urine pregnancy testing and refer pregnant women to the study clinics for further screening and enrollment. Eligible participants at this site include pregnant women 15–49 years of age, coming from a 10 km radius of the research facility, willing to deliver in the research hospital and not planning to relocate/migrate within one year of enrollment into the surveillance program. Women are typically enrolled prior to 20 weeks’ gestation at which time they undergo ultrasound, and are offered treatment for common illnesses including malaria, urinary tract infections and sexually transmitted infections. It is expected that 10% of babies will be born at home, with infants evaluated within 72 hours. Expected enrollment is approximately 780 cord and heel samples with paired ultrasounds for all women collected over one year.

Lusaka, Zambia. Building on prior work of our pilot project nested in the ZAPPS-1 study, we partnered with UNC-GPZ to implement metabolic analysis for the participants of two existing study cohorts: the Zambian Preterm Birth Prevention Study (ZAPPS-2) and Improving Pregnancy Outcomes with Progesterone (IPOP) ^{8, 9} . This site has an estimated preterm birth rate of 15% by ultrasound ¹⁰ . Inclusion criteria for ZAPPS-2 and IPOP are women 18 years of age or older, residing in Lusaka with no plans to relocate. Women in the ZAPPS-2 cohort are HIV uninfected, whereas women enrolled in the IPOP study are HIV-1 positive and receiving antiretroviral therapy. HIV status and anti-retroviral therapy will be accounted for in the analysis phase. All enrollees are recruited prior to 24 weeks’ gestation in this cohort, and ultrasound occurs at the time of enrollment. Enrollment occurs over eight district clinics with delivery in the Lusaka University Hospital, creating a population representative sample for peri-urban Zambian women. The goal sample size for this site is estimated to be 600 cord and heel samples, all paired with ultrasound. Sample collection is expected to occur over one year.

Harare, Zimbabwe. The Zimbabwe site consists of two primary birthing clinics and the referral hospital with a large volume of community and hospital deliveries. Most pregnancies are registered after 24 weeks of gestation. Ultrasound is not routinely available unless complications are identified or by self-pay in a private clinic. Extensive laboratory capacity exists for follow-up testing of screen positive results; however, mass spectrometry is not available. The estimated preterm birth rate is 12% based on LMP. Half of participants in the Zimbabwe cohort will be enrolled at birth from the referral hospital and half will be enrolled during their first ANC visit at primary care clinics, at which time they will undergo ultrasound, yielding a total sample size of. Participants enrolled at primary care clinics will be transferred to the referral hospital if they go into preterm labour. The target sample size for this site is approximately 1000 participants (1000 heel and 500 cord samples) and we expect sample collection to occur over nine months.

Determining sample size requirements for an each site was guided by Steyerburg (2010) ¹¹ and based on typical birth rates at each site and number of ultrasounds that could be obtained. A pragmatic sample size of n=20 was planned for the smallest subgroup in which model performance metrics were to be calculated (ie.GA <34 weeks). This ensured sufficient sample size to calculate reasonably robust summary (i.e. mean and CI) of the reported performance metrics in each planned subgroup.

At all sites, parental informed consent will be sought prior to enrollment in the study, and consent forms will be available in local languages. All liveborn infants will be eligible for inclusion in this study. Women will be enrolled by research nurses during the first ANC visit or at delivery (depending on site enrollment criteria). Participants will be made aware that participation is voluntary, and the risks and benefits of enrollment will be discussed. Those choosing not to participate will continue to receive ANC and treatment according to local clinical standards. Participants will be assured that all reasonable measures will be taken to protect their personal information and that of their newborn. Participants enrolled at the Bangladesh, Zambia, and Kenya sites will also consent to storage and future use of their samples for perinatal research under authorization of presiding institutional committees.

Consenting parents will be informed that heel-prick collection may cause temporary discomfort to their newborn and will also be informed of the possibility that their child may be identified to be at risk of one or more treatable metabolic conditions. Participants will be informed that such incidental findings will be communicated to the study investigators and recommendations will be made to confirm diagnoses and help guide any necessary treatment of their infant. At the time of consent, participants will be informed of their right to request the full details of the screening results for their newborns.

Study personnel will collect, handle and transport heel-prick and cord dried blood spot samples from participants according to study-specific standard operating procedures ( Figure 1). Sample collection procedures have been published previously ¹² . In brief, cord blood will be collected within 30 minutes of delivery of the placenta into an uncoated, sterile syringe. Blood from the syringe (4–5 drops) will be applied to designated filter paper within pre-printed circles. Newborn heel-prick samples will be collected ideally between 24–72 hours after birth or prior to discharge if the newborn is released from hospital within 24 hours of delivery. The newborn’s heel will be warmed at the identified skin-puncture site to promote increased blood flow. The puncture site will be cleaned and air-dried. A sterile lancet or heel incision device will be used to puncture the lateral aspect of the plantar surface of the newborn’s heel. The first small drop of blood formed will be cleaned away and the formation of a second large drop of blood will be encouraged by intermittently applying gentle pressure to the newborn’s lower leg and heel. A filter card will be applied gently against the blood drop to soak through. Each preprinted circle on the filter card will be filled by subsequently formed drops of blood. Following heel-prick sample collection, the newborn’s foot will be elevated above the body and a sterile gauze pad or cotton swab will be pressed against the puncture site until the bleeding stops. No more than 500µL of blood of each sample type (heel prick or cord) will be required for this study. To reduce newborn discomfort during the heel prick sample collection, mothers will be encouraged to hold the baby close or breastfeed the baby before and during the procedure.

Labels with unique site-specific de-identified participant identifiers (ID) will be affixed to each filter card at the time of collection. Samples will be air-dried for 3–4 hours or overnight after which they will be transported to a designated secure study location (hospital office or laboratory) and stored in a room temperature setting prior to shipping. Samples will be shipped at room temperature from the collection sites to NSO every seven days by a preferred courier service. We have previously found samples to be stable for up to two months at room temperature in climate-controlled conditions ¹³ . Appropriate shipping standards will be communicated to the collection sites to minimize the risk of compromised sample integrity during shipping.

An electronic shipment manifest containing clinical and demographic information essential for clinical interpretation of newborn metabolic profiles will be relayed to NSO and OHRI ahead of shipping and included as a hard copy in the shipment ( Table 1). Upon receipt of the samples at NSO, the sample manifest will be cross-referenced against the physical sample cards. Any discrepancies will be resolved through follow-up with the collection site. Secondary accession numbers will be applied for use by internal NSO systems.

NSO is Ontario’s provincial newborn screening program, located at the Children’s Hospital of Eastern Ontario Research Institute (CHEO RI). All samples collected will be directed to NSO for laboratory analysis. NSO is equipped with sufficient personnel and laboratory resources to accommodate the inclusion of international clinical and research samples into the organization's standard workflow. The organization is currently providing newborn screening services to remote communities in other Canadian provinces/territories, and has piloted international screening services ¹⁴ .

NSO will refer to organization and international standards to guide quality management of samples ^{15, 16} . Each sample received will be reviewed for specimen quality and quantity. Newborn screening test calculations assume that the blood is evenly distributed within the circle and completely saturates both sides of the filter paper, and so a satisfactory newborn screening specimen will have blood fully soaked through to the back of the filter paper. Only satisfactory samples will be used for study analysis. Samples can be deemed unsatisfactory for several reasons as outlined in Table 2. If the sample quality is unsatisfactory due to insufficient quantity of blood, the sample will be excluded from analysis or undergo partial analysis.

NSO will generate Quality Reports summarizing the quality of samples received in each shipment. These reports will be sent electronically to the collection site and used to address issues of sample collection, handling and storage prior to shipment. The Report will summarize the Batch ID for the batch of samples received, specimen details and the quality of the samples, as outlined in Table 2.

Standard practice at NSO is to screen each sample for metabolites indicative of risk for approximately 30 treatable conditions. Full-panel analyses as per standard NSO practice will be executed on samples when sample quality is satisfactory.

Quarterly Reports summarizing study progress will be generated by NSO and electronically shared with the study sites, Stanford University and OHRI. Quarterly Reports will include a summary of the analytical results collected up to the end of that quarter, and summarize the total number of samples received, sample collection data (time of collection), newborn characteristic data (gestational age, sex, multiple birth etc.) and the number and distribution of samples excluded from analysis based on ‘unsatisfactory’ quality.

The conditions screened as part of the NSO screening panel are rare. It is possible that an infant will be identified to be at high risk for one of these screened conditions through sample analysis. Such risk poses clinical and ethical considerations for gathering newborn screening data for research purposes. While this study involves newborn screening, this study is not being undertaken as a newborn screening initiative and is designed explicitly as a non-interventional study.

Real time notification of incidental findings will be made for conditions that can be feasibly treated at the collection sites. Sample integrity, potential for confirmatory testing and potential for intervention were all taken into consideration when determining the feasible management of incidental findings, and our approach and rationale have been described in detail elsewhere ^{6, 12, 17} . Based on the above considerations, three high-priority conditions have been identified for real-time reporting: congenital hypothyroidism (CH), hemoglobinopathies, and medium-chain acyl-CoA dehydrogenase deficiency (MCADD).

In brief, in the event that an infant screens positive for a priority condition, second-tier analysis of samples, including repeat confirmatory testing, CFTR mutation analyses, TBX1 and purine profile assays, will be executed as per a modified workflow based on standard NSO practice, provided sufficient sample is available from the original blood spot card. Electronically encrypted Alert Reports will be provided as soon as analysis is complete, and the screening results have been interpreted. The Alert Report will indicate the disease for which the infant is at high risk, the flagged analyte, the analyte level, and the reference ranges and thresholds for the flagged analyte. To ensure that communication of clinically relevant findings between investigators is successful, the collection site will confirm receipt of all Alert Reports within 24 hours. The report will recommend that positive screen results be communicated to the participant’s family, and that appropriate follow-up testing be conducted along with any necessary medical intervention, implemented in accordance with local treatment guidelines. The collection site will report back to the lead site with a completed Alert Report identifying confirmatory testing and medical follow-up.

Upon completion of sample collection, sites will provide OHRI and Stanford with a detailed data collection form with maternal and neonatal covariates required for model validation and analysis ( Table 3). The data will be paired with matching newborn screening data provided by NSO into a single, combined dataset to be used for analysis. Once the primary analytic goals of the project have been met, these data will be stripped of patient identifiers and made available as a shared dataset by site and shared with the appropriate participating institutions in the pursuit of open collaborative research. A common data use agreement will be generated to govern the sharing and use of the data.

The research team at the Ottawa Hospital Research Institute (OHRI) will lead primary data analyses for the generation of gestational age, preterm birth and SGA estimates, and all regulatory activities required at OHRI and CHEO RI, including but not limited to research contracts, data/material sharing agreements and ethics approvals. Each site will be provided with copies of their own cohort data for use in future analyses in consultation with the overall group.

Model-based gestational age estimation and classification of preterm birth, SGA/AGA/LGA using clinical covariates and newborn screening metabolites. Newborn gestational age will be estimated from models derived using multivariable regression coupled with elastic net regularization and including the following covariates:

Classification of preterm birth, as well as SGA/AGA/LGA is conducted using an iterative multitask classification machine learning algorithm that simultaneously derives classification models for both preterm birth status and SGA/AGA/LGA and uses the prediction of one to inform the prediction of the other. These models include birthweight, sex, multiple birth status and newborn screening analytes as predictors, similar to Model 3 above. The model classifying preterm birth status will also use the prediction of SGA/AGA/LGA from the other model, and vice versa, improving the accuracy of both predictions, and will also employ an elastic net or lasso regularization approach similar to gestational age estimation models. Gestational age size classification (SGA/AGA/LGA) will be based on the INTERGROWTH-21 standards ¹⁸ .

Models were trained and internally validated in independent training and validation/test cohorts of infants from Ontario, Canada. These pretrained models can then be applied to the data for infants from the external cohorts to produce gestational age estimates. In order to evaluate the accuracy of gestational age estimates calculated from the models, the model estimate will be compared to the ultrasound reference gestational age for each infant, and the residual error calculated (model gestational age minus reference gestational age). These residuals can be squared, absolute value taken, summed and averaged to calculate different metrics including: mean square error (MSE); standard error of estimation, also known as root mean square error (RMSE); and mean absolute error (MAE), which is the average of the absolute value of the residual across all subjects (or subsets of subjects). Additionally, we will calculate the proportion of model-derived estimates that fall within ± 1 and ± 2 weeks of reference ultrasound gestational age for those samples with gestational age available. MAE will be the main performance metric we will use to evaluate model accuracy, but multiple metrics will be calculated and reported to facilitate comparisons to other models developed by our group and others. Model-derived frequency of preterm birth will be compared to the preterm birth frequency based on reference ultrasound gestational age by dichotomizing the model and reference gestational age values, where preterm is defined as gestational age <37 weeks and term is defined as gestational age ≥37 weeks.

For classification models predicting preterm birth status and SGA/AGA/LGA, we will report area under the ROC curve, as well as sensitivity/specificity and positive predictive values. We will assess model calibration by constructing and visually inspecting calibration plots of and through calculation of the calibration slope and intercept for a regression of model gestational age vs. reference gestational age, and for predicted vs actual probabilities of preterm birth and SGA/AGA/LGA for classification models.

Assessment of other gestational age measures. We will collect additional assessments of gestational age, including anterior lens vascularity, Ballard score, anthropometric measures, LMP and first trimester ultrasound to compare the accuracy of gestational age estimates with our algorithm.

Knowledge Translation will be facilitated through the relationships between researchers, funding organization, and the communities involved in the project. Our findings will be disseminated through publication in peer-reviewed Open Access journals and presentations at academic conferences. We will promote awareness of our findings through press release within our institutions, locally and abroad. The data is collectively owned by OHRI, CHEO and Stanford. Third party de-identified data sharing is possible with the expressed written agreement of all applicable parties and with the development of the appropriate Data Sharing Agreements.

All ethics approvals and contracts are in place between all partners. NSO began receiving samples from Zambia, Bangladesh and Kenya in April 2019 and sample collection is nearing completion at these sites. For Kenya and Bangladesh we have not observed the same hesitancy to perform heel prick blood sampling as in our previous studies ⁶ . Sample collection in Zimbabwe was planned to start in March 2020 but study initiation was delayed due to the COVID-19 global pandemic.