The Creation of a Weight of Evidence Scoring Database for Risk Factors for Adverse Impacts to Birth Outcomes Using Expert Elicitation.

P.J.V. wrote the first draft of the manuscript. S.H. provided edits and wrote the results section. All coauthors reviewed the manuscript and provided editorial suggestions. T.C. created the online database.

The authors have nothing to report.

This study was supported by Corteva AgriSciences and SciPinion. All experts on the panel were blinded to the sponsor throughout the duration of the project. The only coauthors aware of funding were S.H., C.K. and T.C.

In the first round, members of the expert panel were instructed to identify the most common and important birth outcomes that are typically studied in epidemiology studies. This process identified 15 different birth outcomes of interest that were retained for subsequent weight of evidence assessments. These 15 outcomes, ranked in order of decreasing frequency, were: Fertility issues (e.g., failure to conceive) Mortality: miscarriage (pre ~20 weeks) Mortality: Neonatal death/stillbirth (post ~20 weeks) Preeclampsia Gestational diabetes Preterm birth Abnormal weight at gestational age (low or high) Abnormal length at gestational age Microcephaly Growth restriction (intra‐ or extra‐uterine) Low APGAR/ICU admission Congenital anomalies/major congenital malformations (MCM)/birth defects Delayed effects (neurodevelopmental) Delayed effects (immunological) Delayed effects (childhood cancer < 18 years of age) Fertility issues (e.g., failure to conceive) Mortality: miscarriage (pre ~20 weeks) Mortality: Neonatal death/stillbirth (post ~20 weeks) Preeclampsia Gestational diabetes Preterm birth Abnormal weight at gestational age (low or high) Abnormal length at gestational age Microcephaly Growth restriction (intra‐ or extra‐uterine) Low APGAR/ICU admission Congenital anomalies/major congenital malformations (MCM)/birth defects Delayed effects (neurodevelopmental) Delayed effects (immunological) Delayed effects (childhood cancer < 18 years of age) The expert panel identified 245 risk factors associated with the listing of 15 adverse birth outcomes. We grouped these 245 risk factors into 11 categories. We recognize there are some overlaps in these categories (e.g., air pollution is both an environmental exposure, and a place of residence risk factor), but we selected categories of risk factor that made most sense for generating summary weight of evidence scores. The number of risk factors identified and organized according to these categories are provided in Table  2 . Number of risk factors identified by category. Following the identification of the risk factor and birth outcome pairs, experts were asked to review the accompanying grids. They were asked to complete with a numerical weight of evidence rating (on a scale of “0” = no evidence to “5” = strong evidence; whole number entries only). They were explicitly told not to enter a value into every cell, but only those for which they thought they had appropriate expertise to do so. The expert elicitation scores were compiled and the average, standard deviation and number of scores per cell was tabulated in each of the 11 risk factor category tables. We focus our discussion below on the birth outcomes and risk factors that received higher scores and where at least five experts provided scores. We opted to use the mean ratings score as an indicator of strength of evidence. The weight of evidence scoring is summarized below (Tables  3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 ) in the form of a heat map where shades of orange are used to indicate increasing average scores (darkest shades associated with cells having an average score of 4.0–5.0, next darkest shade for cells with average scores of 3.0–3.99, and so on). The following sections provide brief synopses of the scores and highlight the risk factors‐birth outcome pairs that received the highest scores. Birth outcomes × female physiological factors. Note: Each cell contains the mean ± the standard deviation and number of experts who have scored that cell in parentheses. Birth outcomes × female medical factors. Note: Each cell contains the mean ± the standard deviation and number of experts who have scored that cell in parentheses. Birth outcomes × male physiological and medical factors. Note: Each cell contains the mean ± the standard deviation and number of experts who have scored that cell in parentheses. Birth outcomes × lifestyle factors. Note: Each cell contains the mean ± the standard deviation and number of experts who have scored that cell in parentheses. Birth outcomes × dietary and nutrition factors. Note: Each cell contains the mean ± the standard deviation and number of experts who have scored that cell in parentheses. Birth outcomes × residence factors. Note: Each cell contains the mean ± the standard deviation and number of experts who have scored that cell in parentheses. Birth outcomes × socioeconomic factors. Note: Each cell contains the mean ± the standard deviation and number of experts who have scored that cell in parentheses. Birth outcomes × occupational factors. Note: Each cell contains the mean ± the standard deviation and number of experts who have scored that cell in parentheses. Birth outcomes × environmental factors. Note: Each cell contains the mean ± the standard deviation and number of experts who have scored that cell in parentheses. Birth outcomes × medications/therapies. Note: Each cell contains the mean ± the standard deviation and number of experts who have scored that cell in parentheses. Birth outcomes × chemical exposures (environmental or occupational). Note: Each cell contains the mean ± the standard deviation and number of experts who have scored that cell in parentheses. Of the categories of risk factors, the experts' scoring indicated that the strongest evidence was for maternal physiological conditions (Table  3 ). Maternal age demonstrated the most consistent strong evidence as a risk factor across multiple outcomes, including fertility issues (4.85 ± 0.49 mean ± standard deviation), miscarriage (4.52 ± 0.68), neonatal death (4.22 ± 0.88), preeclampsia (4.18 ± 1.01), gestational diabetes (4.06 ± 1.06), preterm birth (4.10 ± 0.79), and congenital anomalies (4.40 ± 1.00). Multiple pregnancy showed strong evidence for numerous adverse outcomes, including preterm birth (4.70 ± 0.47), abnormal fetal weight (4.56 ± 0.89), preeclampsia (4.46 ± 0.78), miscarriage (4.38 ± 0.87), neonatal death (4.36 ± 0.74), and growth restriction (4.44 ± 0.86). Ovulatory function and oocyte quality demonstrated the strongest association with fertility issues (4.91 ± 0.30), indicating near‐unanimous expert agreement on its critical importance. Genetic factors showed particularly strong evidence for congenital anomalies (4.68 ± 0.58). Weight, height, and BMI showed strong evidence across multiple outcomes, including fertility issues (4.18 ± 1.29), preeclampsia (4.07 ± 0.88), gestational diabetes (4.18 ± 1.19), preterm birth (4.00 ± 0.97), and abnormal fetal weight (4.06 ± 1.06). Several factors demonstrated moderate evidence (scores > 3.0) across multiple outcomes. Major trauma or injury during pregnancy showed strong evidence for miscarriage (4.31 ± 0.75) and preterm birth (4.30 ± 0.67), with moderate evidence for neonatal death (4.00 ± 0.77), all of which would be dependent on the extent of trauma and injury. Previous pregnancy and reproductive history showed consistently moderate‐to‐strong evidence across outcomes, including neonatal death (4.10 ± 1.10), preeclampsia (4.08 ± 1.08), gestational diabetes (4.00 ± 1.05), preterm birth (4.00 ± 1.11), and growth restriction (4.33 ± 0.87). Gestational weight gain showed strong evidence for abnormal fetal weight (4.23 ± 1.01) and moderate evidence across several outcomes, including preeclampsia (3.80 ± 0.79) and gestational diabetes (3.93 ± 0.92). Parity demonstrated moderate‐to‐strong evidence for multiple outcomes, including preeclampsia (4.07 ± 1.00) and neonatal death (4.00 ± 1.12). Fetal sex showed moderate evidence for several outcomes, particularly affecting fetal growth parameters, with stronger evidence for abnormal weight (3.88 ± 1.46) and length (4.00 ± 1.26) at gestational age. Low‐grade inflammation showed moderate evidence across several outcomes, including fertility issues (3.25 ± 2.22) and preeclampsia (3.67 ± 1.15). The interpregnancy interval showed moderate evidence only for preterm birth (3.58 ± 1.38). Several risk factors related to maternal medical conditions were identified as strong risk factors for impacts to birth outcomes (Table  4 ). The conditions showing the strongest evidence encompassed metabolic, infectious, and pregnancy‐specific complications. Preeclampsia demonstrated strong evidence for preterm birth (4.38 ± 0.74) and abnormal fetal weight (4.14 ± 1.21). Similarly, severe obesity showed strong evidence across multiple outcomes, including gestational diabetes (4.56 ± 1.01), abnormal fetal weight (4.11 ± 1.05), and preterm birth (4.08 ± 1.16). Infectious conditions demonstrated particularly strong evidence for certain outcomes. Viral infections showed strong evidence for congenital anomalies (4.73 ± 0.65), microcephaly (4.5 ± 1.07), and neurodevelopmental effects (4.0 ± 1.73). Sexually transmitted infections showed strong evidence for fertility issues (4.0 ± 1.55), miscarriage (4.2 ± 0.84), and congenital anomalies (4.0 ± 1.0). These patterns highlight the significant impact of maternal infections on fetal development. Pregnancy‐specific complications showed strong evidence for multiple adverse outcomes. Placental abruption demonstrated strong evidence for neonatal death (4.5 ± 0.58), preterm birth (4.25 ± 0.5), and abnormal fetal weight (4.0 ± 1.15). Excessive gestational weight gain showed strong evidence for gestational diabetes (4.38 ± 0.52) and abnormal fetal weight (4.5 ± 0.84), while low gestational weight gain showed strong evidence for abnormal fetal weight (4.0 ± 0.82). Several reproductive conditions showed strong evidence for specific outcomes. Polycystic ovarian syndrome showed strong evidence for fertility problems (4.17 ± 1.33). Multiple embryo transfer in IVF showed strong evidence for miscarriage (4.14 ± 0.9) and preterm birth (4.0 ± 1.15). Several less common conditions showed strong evidence for specific outcomes. Uncontrolled maternal phenylketonuria demonstrated strong evidence for congenital anomalies (4.29 ± 0.76) and neurodevelopmental effects (4.4 ± 0.89). Lack of antenatal care showed strong evidence for neonatal death (4.14 ± 0.69) and growth restriction (4.0 ± 0.0). Risk factors associated with physiological and medical conditions in males were considered to have high strength of evidence (Table  5 ). The experts indicated sperm/seminal plasma quality showed the strongest evidence for fertility issues (4.55 ± 0.82) and genetic factors demonstrated strong evidence for congenital anomalies (4.08 ± 1.44), with moderate evidence for fertility issues (3.64 ± 1.43). Paternal age showed moderate‐to‐strong evidence across multiple outcomes, with the strongest evidence for fertility issues (4.00 ± 1.07) and congenital anomalies (3.36 ± 1.50). The pattern suggests broader impacts of advanced paternal age on reproductive outcomes compared to other male factors. Several conditions showed moderate evidence (scores > 3.0) for specific outcomes: Cancer demonstrated moderate evidence for fertility issues (3.11 ± 1.76). Endocrine disorders showed moderate evidence for fertility issues (3.38 ± 1.41). Inflammatory conditions of the male reproductive tract showed moderate evidence for fertility issues (3.13 ± 1.73). Infections showed moderate evidence for fertility issues (3.11 ± 1.27). Cancer demonstrated moderate evidence for fertility issues (3.11 ± 1.76). Endocrine disorders showed moderate evidence for fertility issues (3.38 ± 1.41). Inflammatory conditions of the male reproductive tract showed moderate evidence for fertility issues (3.13 ± 1.73). Infections showed moderate evidence for fertility issues (3.11 ± 1.27). Some male factors were rated with low weight of evidence scores despite being highly prevalent. For example, anthropometric measures (weight, height, BMI) typically had lower scores with highest scores for fertility issues (2.86 ± 2.12). Similarly, metabolic disorders and microbial dysbiosis showed limited evidence for most outcomes. Some risk factors related to lifestyle were considered to have strong evidence by the experts (Table  6 ). The experts indicated that alcohol consumption showed strong evidence for impacting microcephaly (4.13 ± 1.46) and congenital anomalies (4.69 ± 0.60), with moderate‐to‐strong evidence across multiple outcomes including abnormal fetal weight (3.73 ± 1.27) and growth restriction (3.80 ± 1.03). Smoking/vaping showed strong evidence for abnormal fetal weight (4.00 ± 1.46) and moderate‐to‐strong evidence for growth restriction (3.88 ± 1.36). Illicit drug use demonstrated moderate‐to‐strong evidence across multiple outcomes, including fertility issues (3.67 ± 1.15), miscarriage (3.50 ± 1.07), neonatal death (3.63 ± 1.06), growth restriction (3.92 ± 0.79), and neurodevelopmental effects (3.67 ± 0.87), indicating broad impacts across the spectrum of pregnancy outcomes. Poor antenatal attendance showed moderate‐to‐strong evidence of association with several outcomes, including preterm birth (3.78 ± 0.83), abnormal fetal weight (3.43 ± 0.98), growth restriction (3.60 ± 0.89), and low APGAR score or ICU admission (3.50 ± 1.00). Some lifestyle factors showed moderate evidence of association with specific adverse outcomes. A sedentary lifestyle was moderately associated with gestational diabetes (3.40 ± 0.55), while poor oral health showed moderate evidence for preterm birth (3.00 ± 1.58). Marijuana use generally showed lower evidence across outcomes compared to other substances, with the strongest evidence for fertility issues (2.40 ± 1.95). Notable patterns emerged in the evidence distribution across different types of lifestyle factors. Substance use behaviors showed consistently stronger evidence compared to behavioral factors like exercise or sleep patterns. For example, while alcohol and tobacco use showed strong evidence across multiple outcomes, factors like hot tub use or long periods of standing showed limited evidence. In the opinion of the experts consulted, some lifestyle factors had limited evidence. Exercise patterns (both excessive and insufficient) showed modest evidence across outcomes. Sleep disturbances and ability to make care decisions showed limited evidence for most outcomes. Among diet and nutrition factors, the experts indicated folic acid deficiency demonstrated the strongest overall evidence as a risk factor, particularly for congenital anomalies (4.72 ± 0.57) and impact on neurodevelopmental outcomes (3.57 ± 1.51) (Table  7 ). Similarly, low prenatal vitamin intake was rated as having strong evidence for congenital anomalies (4.00 ± 0.95) and moderate evidence across multiple outcomes including neonatal death (3.25 ± 1.26) and preeclampsia (3.33 ± 1.53). The experts indicated there was strong evidence for excess Vitamin A for specific outcomes, particularly microcephaly (4.00 ± 1.41) and congenital anomalies (4.00 ± 1.25). Iodine deficiency demonstrated strong evidence for neurodevelopmental effects (4.29 ± 1.11) and congenital anomalies (3.88 ± 0.83), consistent with its critical role in fetal brain development. Food insecurity was rated as having moderate‐to‐strong evidence across multiple outcomes, including preterm birth (3.60 ± 1.14), growth restriction (3.50 ± 1.05), and low APGAR/ICU admission (3.40 ± 1.14). Distance/access to fresh food, nutrition, and clean water showed similar patterns of moderate evidence across outcomes. Several vitamin deficiencies were rated as having moderate evidence for specific outcomes: Vitamin D deficiency was deemed to have moderate evidence for preeclampsia (3.50 ± 1.38) and gestational diabetes (3.40 ± 1.82). General vitamin deficiencies were scored as having moderate evidence for gestational diabetes (3.60 ± 1.14) and growth restriction (3.50 ± 1.29). Zinc deficiency was rated as having moderate evidence for congenital anomalies (3.71 ± 1.11) and abnormal fetal weight (3.00 ± 1.63). Calcium deficiency was rated as having moderate evidence for preeclampsia (4.00 ± 1.15). Vitamin D deficiency was deemed to have moderate evidence for preeclampsia (3.50 ± 1.38) and gestational diabetes (3.40 ± 1.82). General vitamin deficiencies were scored as having moderate evidence for gestational diabetes (3.60 ± 1.14) and growth restriction (3.50 ± 1.29). Zinc deficiency was rated as having moderate evidence for congenital anomalies (3.71 ± 1.11) and abnormal fetal weight (3.00 ± 1.63). Calcium deficiency was rated as having moderate evidence for preeclampsia (4.00 ± 1.15). Notable patterns emerged in the evidence distribution across different types of nutritional factors. Micronutrient deficiencies were rated with stronger evidence for developmental outcomes compared to macronutrient imbalances. For example, while folic acid and iodine deficiencies were rated as having strong evidence for neurodevelopmental outcomes, factors like protein intake or complex carbohydrates were scored as showing limited evidence across outcomes. Some nutritional factors were rated with limited evidence despite common concerns. The ratings for caffeine intake were stronger with modest evidence across outcomes, with the strongest evidence for growth restriction (3.00 ± 2.00). Similarly, probiotics/prebiotics were deemed to have consistently low evidence across all outcomes. The experts identified only 11 risk factors associated with residence (Table  8 ). The experts indicated there was the strongest evidence for distance/access to medical care as a risk factor for preterm birth (4.00 ± 1.00) and moderate‐to‐strong evidence as a risk factor for preeclampsia (3.83 ± 0.98), gestational diabetes (3.60 ± 0.89), and growth restriction (3.67 ± 1.21). Hazards in the home (lead paint, asbestos, mold) were rated with strong evidence for neurodevelopmental effects (4.00 ± 1.31) and moderate‐to‐strong evidence for congenital anomalies (3.88 ± 0.99). The experts felt high altitude had moderate evidence of an effect on abnormal weight at gestational age (3.43 ± 1.72) and growth restriction (3.25 ± 1.39), consistent with known effects of oxygen availability on fetal development. Access to insurance demonstrated moderate evidence across multiple outcomes, including preterm birth (3.43 ± 1.13), growth restriction (3.29 ± 1.50), and neurodevelopmental effects (3.20 ± 1.79). The expert panel identified 23 different risk factors for birth outcomes related to socioeconomic factors (Table  9 ). Experts indicated income inequality demonstrated consistently strong evidence for an association with preterm birth (4.67 ± 0.58), neonatal death (4.33 ± 1.15), low APGAR/ICU admission (4.33 ± 1.15), growth restriction (4.25 ± 0.96), and neurodevelopmental effects (4.33 ± 1.15). Housing instability was rated with a similar pattern of strong evidence across outcomes, including preterm birth (4.50 ± 0.58), neonatal death (4.33 ± 1.15), and growth restriction (4.33 ± 1.15). Maternal nativity/years lived in the US was rated with strong evidence for multiple outcomes, with ratings of 4.0 or higher for miscarriage, neonatal death, preeclampsia, gestational diabetes, preterm birth, and growth restriction. Employment status and ethnicity both demonstrated strong evidence for several outcomes. Employment status was rated with particularly strong evidence for preterm birth (4.33 ± 1.15) and abnormal fetal weight (4.00 ± 1.73), while ethnicity was scored with strong evidence for gestational diabetes (4.00 ± 1.15) and congenital anomalies (3.88 ± 1.13). Education level was rated with moderate‐to‐strong evidence across numerous outcomes, particularly for neurodevelopmental effects (3.86 ± 1.46), gestational diabetes (3.75 ± 1.50), and preterm birth (3.89 ± 1.27). Socioeconomic status demonstrated similar patterns, with strong evidence for preeclampsia (4.00 ± 1.00) and growth restriction (4.00 ± 1.07). Lack of antenatal care was rated with consistently moderate‐to‐strong evidence across outcomes, including gestational diabetes (3.83 ± 1.17), preterm birth (3.75 ± 0.71), and low APGAR/ICU admission (3.80 ± 0.84). Race was rated with moderate evidence across multiple outcomes, with the strongest evidence for preterm birth (3.73 ± 1.56). The expert panel identified 11 risk factors for birth outcomes that are related to occupational factors (Table  10 ). Experts indicated modest evidence for most occupational risk factors, with few showing strong evidence (mean score > 4.0) for adverse outcomes. Occupational stress demonstrated the strongest evidence across outcomes, with moderate‐to‐strong evidence for miscarriage (3.67 ± 1.03), preeclampsia (3.50 ± 1.00), preterm birth (3.50 ± 1.07), and neurodevelopmental effects (3.50 ± 1.29). Shift work emerged as a consistent risk factor, showing moderate evidence across several outcomes including miscarriage (3.13 ± 1.55), preterm birth (3.22 ± 1.48), and fertility issues (2.86 ± 1.57). This pattern suggests broad impacts of circadian disruption on reproductive health. Heavy physical work showed similar patterns of moderate evidence, particularly for preterm birth (2.91 ± 1.22) and miscarriage (2.78 ± 1.09). Occupational radiation exposure rating included the strongest evidence for congenital anomalies (3.09 ± 1.58) and delayed effects like childhood cancer (2.56 ± 1.67), reflecting the known mutagenic properties of ionizing radiation. In general, evidence for immediate pregnancy outcomes was scored lower. The expert panel identified eight risk factors for birth outcomes related to environmental factors (Table  11 ). Experts indicated there was relatively moderate evidence for most environmental risk factors, with few factors being rated as having strong evidence (mean score > 4.0) for adverse outcomes. Trauma/domestic violence during pregnancy demonstrated the strongest evidence across outcomes, particularly for preterm birth (3.71 ± 0.76) and neonatal death (3.14 ± 0.69), suggesting significant impacts of acute physical and psychological stress on pregnancy. Secondhand smoke exposure was rated with moderate‐to‐strong evidence for several outcomes, including neonatal death (3.56 ± 1.24), congenital anomalies (3.25 ± 1.42), and miscarriage (2.90 ± 1.29). Air pollution emerged as a consistent moderate weight of evidence risk factor across multiple outcomes, with the strongest evidence for preterm birth (3.39 ± 1.33) and abnormal fetal weight (3.07 ± 1.44). The expert panel identified 32 medications that are risk factors for birth outcomes (Table  12 ). Several medications were found by the experts to have strong evidence (mean score > 4.0) for specific adverse birth outcomes, particularly regarding congenital anomalies and early pregnancy loss. The strongest evidence for an association was observed for well‐documented teratogens, with thalidomide showing the highest possible rating (5.0 ± 0.0) for both miscarriage risk and congenital anomalies. The experts indicated there was strong evidence for methotrexate as a risk factor for early pregnancy loss (4.67 ± 0.58) and congenital anomalies (4.5 ± 0.76). Isotretinoin (13‐cis‐retinoic acid) similarly demonstrated strong evidence as a risk factor for congenital anomalies (4.78 ± 0.44) and showed high ratings for both miscarriage (4.0 ± 1.0) and neonatal death (4.33 ± 0.58). Aminopterin demonstrated the maximum possible rating (5.0 ± 0.0) for both miscarriage risk and microcephaly, though these ratings were based on a smaller number of expert evaluations. Among anticonvulsant medications, several were rated as having strong evidence for congenital anomalies, including phenytoin (4.67 ± 0.50), valproic acid (4.5 ± 0.8), and anticonvulsants as a class (4.5 ± 0.71). Valproic acid additionally was similarly rated as providing strong evidence for miscarriage (4.0 ± 1.0) and with moderate evidence ratings for neurodevelopmental effects (4.25 ± 0.96), suggesting both immediate and longer‐term risks. Exposure to chemotherapy and/or radiotherapy was scored as having strong evidence for miscarriage (4.0 ± 0.63) and moderate evidence across multiple outcomes, including fertility issues (3.67 ± 1.86), neonatal death (3.67 ± 0.82), microcephaly (4.0 ± 1.0), and congenital anomalies (3.88 ± 0.64). Several medications were rated as providing moderate evidence (scores > 3.0) for specific outcomes while falling short of the strong evidence threshold. Angiotensin‐converting enzyme (ACE) inhibitors and angiotensin receptor blockers (ARBs) showed moderate evidence as risk factors for multiple outcomes, including neonatal death (3.67 ± 1.53 for both), preeclampsia (3.67 ± 1.15 for both), and growth restriction (3.75 ± 0.5 and 4.0 ± 1.0 respectively). Fertility‐related medications were rated with more targeted effects, with clomiphene citrate demonstrating strong evidence for fertility issues (4.0 ± 1.73) reflecting its specific mechanism of action and primary use in treating fertility problems. Among psychiatric medications, lithium and various anticonvulsants were deemed to provide moderate evidence for congenital anomalies (3.43 ± 1.27 and 4.5 ± 0.71 respectively) while having lower ratings for other outcomes. This suggests that their primary reproductive risks may be concentrated in early developmental periods rather than affecting later pregnancy outcomes. Notably, several commonly used medications were scored as providing consistently low evidence across outcomes. Antibiotics, paracetamol/acetaminophen, and lamotrigine generally received low ratings (most < 2.0) across all outcomes, suggesting relatively lower risk profiles. However, the number of experts rating these medications was often smaller than for well‐known teratogens, potentially reflecting less available evidence rather than proven safety. The expert panel identified 13 chemicals/chemical classes that might be risk factors for birth outcomes (Table  13 ). Experts indicated metals (lead, methylmercury, arsenic) demonstrated the strongest evidence (mean score > 4.0) for adverse birth outcomes, particularly for congenital anomalies (4.50 ± 0.71) and neurodevelopmental effects (4.36 ± 0.67). Ionizing radiation was indicated as having strong evidence for specific outcomes, including congenital anomalies (3.92 ± 1.44) and childhood cancer (3.73 ± 1.19), with moderate evidence for both early pregnancy loss (3.63 ± 1.06) and neonatal death (3.63 ± 1.06). A couple chemical classes (persistent organic chemicals and volatile organic chemicals) were scored as providing moderate evidence specifically for congenital anomalies (3.00 ± 1.41). All other chemicals and groups of chemicals were rated by the experts to have limited evidence as risk factors for any birth outcomes. We built a publicly available online database of the results of the experts' scores ( https://scipinion‐rfbo.onrender.com ). The database is fully searchable sortable and provides the following details by risk factor‐birth outcome pair: Count (number of experts who have scored the RF‐outcome pair) Count of experts who scored each value (e.g., 0, 1, 2, 3, 4, 5) Average of all scores Standard deviation of all scores Summary of the topic Citations for studies on that risk factor‐outcome pair Count (number of experts who have scored the RF‐outcome pair) Count of experts who scored each value (e.g., 0, 1, 2, 3, 4, 5) Average of all scores Standard deviation of all scores Summary of the topic Citations for studies on that risk factor‐outcome pair The database can be used to determine the risk factors and birth outcome pairs that have the highest and lowest expert scores, those that have the most scores (those with the highest degree of knowledge from the expert pool), those with the highest or lowest standard deviation. We foresee that the database, managed by SciPinion, will be a living database where additional experts can provide additional scores (although these will be coded as unblinded versus the original expert scores were all provided in a blinded manner), additional scientific manuscripts that inform the risk factor‐birth outcome pair, and provide edits to the text write up for each risk factor‐birth outcome pair in an ongoing and continuous manner. A rigorous protocol will be implemented to vet input as coming from experts with sufficient credentials and experience to assure a high quality of input is maintained. While the fine details of this protocol, and the administrative processes associated with the operations of this website are beyond the scope of this paper, they will be described on the website itself as all relevant associated details are finalized.

In this paper, we summarize our findings from a multistage expert elicitation to obtain weight of evidence scoring for risk factors suspected of increasing the risk of adverse birth outcomes. The obtained mean scores can be used as a tool to identify those risk factor and health outcome combinations for which additional research is needed, as well as combinations for which knowledge is firmly established. Weight of evidence scores were obtained for over a thousand combinations of risk factor and birth outcomes, providing a comprehensive summary tool for researchers in this area. Scores with narrow standard deviation also demonstrate relative expert consensus on the strength of evidence of a particular risk factor, while wider standard deviation may suggest less certainty among the experts included in this study. Risk factor‐birth outcome combinations with low scores represent those for which either minimal data exists or the weight of evidence for an association is weak. For those areas where minimal data exists, this might highlight a research need. Where sufficient data exists, but weak evidence is indicated, this may be an area of low priority for additional research. Likewise, those risk factor‐birth outcome combinations with the highest scores represent well‐known associations, although related research questions may remain unknown (e.g., biological mechanism, exposure windows, vulnerable subpopulations). We have purposively refrained from presenting summary risk estimates for these risk factor‐adverse outcome combinations as we feel these are best generated from rigorous systematic reviews that can incorporate aspects related to study quality and risk of bias assessments. An inspection of the weight of evidence scores across the 11 domains revealed that the weight of evidence scores were higher for female physiological factors. This represents in part the dominant roles maternal age, fetal sex, and reproductive history have across a wide range of adverse birth outcomes. In contrast, the summary scores for male physiological factors were much lower, indicating the experts felt there was less evidence for the effect of male physiological factors for most birth outcomes considered, apart from fertility issues (i.e., fecundability). The methodology to select experts to provide the weight of evidence scoring is a critical step in undertaking an expert elicitation. Our approach made use of objective criteria such as ranking experts based on years of experience and numbers of publications. Future efforts to recruit expertise across a number of predefined risk factors or health outcomes could consider stratified‐sampling expert recruitment, where a sufficient number of experts could be identified for each stratum. This could help mitigate overrepresentation of experts in some categories, which could in turn introduce some bias into the weight of evidence scores. While we did not adopt a stratified recruitment approach in this study, in our view, we don't feel that there was an overrepresentation in any of our adverse health outcomes and risk factors categories. Most of our identified experts had expertise across a wide number of risk factor groups and birth outcomes. An important strength of our expert elicitation was the high response rates that were achieved across the three rounds (100% in Round 1 and 93% in Round 2). However, the composition of our expert panel was mainly drawn from those based in North America and Oceana, with fewer experts from Europe and South America. Despite its diversity, it cannot provide full representation of research priorities. It is important to note that there would be underrepresentation of research themes that may be of greater importance in lower income countries. For example, neonatal mortality in lower income countries is driven largely by infections, asphyxia, and prematurity (Blencowe et al.  2013 ; Dhaded et al.  2015 ; Goldenberg and McClure  2015 ). The experts highlighted challenges in categorizing the risk factors across different domains. They pointed out these categories were often not mutually exclusive and that some risk factors could reasonably be expected to be included in multiple categories. For example, some potentially harmful exposures could occur in both the work (occupational) and residential settings. The categories were largely used for organizing purposes during the expert elicitation phase, and as such, can be adjusted over time as experts provide additional insights upon inspection of and working with the online database. The data collected from the expert elicitation formed the basis of the creation of an easy‐to‐use, menu‐driven system that makes the collected data available to public health professionals and the public at large. This information can be found at https://rfbo.scipinion.com . This includes a summary write up for each expert for each section, and summary scores (and number of expert ratings) for each cell. The database also allows for feedback (from vetted and qualified experts) to be provided, including the identification of newly published study(ies) that can inform the weight of evidence scores. Given the dynamic nature of studies on adverse birth outcomes, we hope to regularly revisit the weight of evidence scores annually. SciPinion will regularly update the website, and endeavor to conduct future surveys that focus on more specific risk factor and adverse health outcomes. The website will also highlight on an ongoing basis published systematic reviews and meta‐analyses that align with the risk factor and disease outcome combinations evaluated in this expert elicitation. The first set of scores were all provided in a blinded manner. No expert had knowledge of any other expert's scores so as not to bias their own scoring. As researchers wish to contribute to the scores after this publication, those scores will be differentiated as being non‐blinded. As time goes by, it will be interesting to see how the unblinded scores differ from blinded scores (if at all). We encourage researchers to help further populate references of studies that inform each risk factor‐birth outcome combination. This working database of scores and studies will be a valuable resource to the scientific, medical and research communities. Providing the expert elicitation data in a database format also allows for ongoing validation of the findings for this study. As new studies are released or additional researchers review the database, updates can be made as needed to adjust the relevance of certain risk factors and their impact on the 11 birth outcome categories. While the expert elicitation was able to identify numerous risk factors and provide weight of evidence scores, additional work is needed to identify clusters of related risk factors. Our expert elicitation approach did not endeavor to determine the extent to which factors were connected or interrelated with each other. For example, there are disparities in ambient concentrations in air pollution by race and income (Jbaily et al.  2022 ; Hajat et al.  2015 ), and many lifestyle behaviors are interrelated and also related to multiple health conditions that are in turn risk factors for adverse birth outcomes. In our view, additional synthesis that assesses how strongly related the risk factors would provide further insight on the necessity to control for these factors as confounders or evaluate as effect modifiers. Despite this deficiency in the expert elicitation herein, the listing of the risk factors provides some guidance to researchers about what core data should be collected in their studies. We recognize that the compiled weight of evidence scores cannot be directly used to determine what risk factors are confounders or effect modifiers in an individual study. Confounding occurs when the true effect of the exposure on the outcome is distorted by some other factor, leading to a biased effect measure (Rothman et al.  2008 ). In contrast, effect modification describes the phenomenon where the association between two variables is different based on the level of a third factor (i.e., the effect modifier). Complicating matters further, a variable can function as a confounder in some contexts and as an effect modifier in others. Additionally, a risk factor may act as a confounder in one study population but not in another, depending on how its prevalence and characteristics vary across groups. Moreover, as noted by VanderWeele, the impacts of confounding and effect modification are relative to what other variables are being conditioned upon (Vander Weele  2012 ). In practice, confounding is identified by assessing the extent to which a measure of association changes following adjustment for the confounder. Consequently, the database we have created cannot be used to determine whether a factor is a confounder or an effect modifier in any single study. Rather, our database provides a list of variables to consider collecting at the design stage of a study. Thereafter, other epidemiological methods such as directed acyclic graphs, evaluation of change in risk estimates, as well as etiological understanding of disease processes could be followed to assess the role of these factors when evaluating causal relationships. Relatedly, while our expert elicitation does identify relevant risk factors that may be confounders, it is impractical for us to provide guidance on how these risk factors should be modeled. It is our view that approaches to control for the identified risk factors evaluate empirically which methods best account for the confounding influence of these other risk factors. That said, collecting more detailed information about a putative confounder will provide more flexibility to modeling possible confounding. For example, it is preferred to collect data on specific age, rather than age group. It is our hope that the measures created from our expert elicitation are used by the broader research communities, including those undertaking primary research, and those who communicate the findings from this research to various stakeholders, including the general population. With that in mind, we see several possible ways our website can be used; these could, for example, include: Prioritization of evidence: The scores provided by the experts can provide some guidance as to which risk factors have the strongest evidence for specific outcomes, while also highlighting gaps. Hypothesis generation: The database can be used as a tool to promote new studies and do so in a way that encourages interdisciplinary work as the data do show connections across risk factors such as environmental, social, and behavioral determinants of health. Methodological guidance: Our risk factor database can serve as a tool for researchers in that it can assist them to identify relevant risk factors as candidate covariates for adjustment, mediator, or effect modification analysis. Grant applications and study design: Our website can help justify research priorities given the transparent evidence‐weighting scores that are provided. Policy and guideline development: Our findings can provide an evidence‐informed input into guideline committees (e.g., WHO, CDC, health ministries). Prioritization of evidence: The scores provided by the experts can provide some guidance as to which risk factors have the strongest evidence for specific outcomes, while also highlighting gaps. Hypothesis generation: The database can be used as a tool to promote new studies and do so in a way that encourages interdisciplinary work as the data do show connections across risk factors such as environmental, social, and behavioral determinants of health. Methodological guidance: Our risk factor database can serve as a tool for researchers in that it can assist them to identify relevant risk factors as candidate covariates for adjustment, mediator, or effect modification analysis. Grant applications and study design: Our website can help justify research priorities given the transparent evidence‐weighting scores that are provided. Policy and guideline development: Our findings can provide an evidence‐informed input into guideline committees (e.g., WHO, CDC, health ministries). Expert elicitation represents a viable approach to synthesize the epidemiological evidence for risk factors of adverse birth outcomes. While our elicitation identified a large number of health outcomes and associated risk factors, the weight of evidence ratings narrowed these combinations to clearly identify those for which there is strong evidence, which could indicate areas for which more research should be prioritized or may reflect areas where there is substantial consensus and not necessarily a need for further research. Risk factor‐birth outcome combinations with lower weight of evidence scores represent either those areas that have insufficient data or sufficient data and a weak association, the former being a research priority. Interpretated as such, we also would like to highlight that these scoring combinations can serve as a guide to prioritize future systematic reviews and meta‐analysis on specific risk factors. Lastly, the science on birth outcomes and associated risk factors is constantly evolving and building to our knowledge base. With this in mind, we hope that the online database we have created provides a valuable resource for keeping investigators updated and informed, guiding research priorities, and supporting the design and conduct of future studies.

All experts who took part in this project were made aware of the project objectives and disclosures and signed a contract indicating their agreement to these parameters. This included that the experts would be blinded to the sponsor, the sponsor would be blinded to the experts, and the experts would be blinded to each other during the panel deliberations. All experts agreed to have their names disclosed in the manuscript, and being named as a coauthor was decided by each expert upon review of the first draft of the manuscript.

Over the past few decades, there have been many improvements in maternal and fetal health, and the associated birth outcomes in middle‐ and higher‐income countries (Goldenberg et al.  2018 ), although it has been noted that there have been inequalities in this progress globally (Global Burden of Disease Child Adolescent Health Collaboration  2017 ). These improvements in maternal and fetal health have been attributed to a variety of factors ranging from advances in diagnostic imaging and screening (Rubesova and Barth  2014 ) maternal health care (Khorrami et al.  2019 ), medical treatments (Schuler et al.  2022 ), and a better understanding of the etiology for a wide range of adverse birth outcomes (Strong et al.  2021 ). The latter has been shaped by findings from a large epidemiological literature that has incorporated a diversity of study designs, study populations, and methods to characterize exposures across a range of jurisdictions. As noted in an earlier review by Savitz et al. ( 2006 ), epidemiological investigations of pregnancy outcomes are distinct from studies of other health outcomes for several reasons. These include the closer temporal proximity of determinants and outcomes, non‐random allocation of risk factors, heterogeneity of adverse outcomes, and the importance of sociodemographic and racial disparities in outcomes. While there have been improvements in pregnancy outcomes, preterm birth and its complications are leading causes of mortality among children under 5 years of age (Liang et al.  2024 ). We also note that recent data suggest that even in high income countries like the United States, there has been a regression for some pregnancy‐related outcomes including increased rates of maternal mortality (Joseph et al.  2024 ; Hoyert  2022 ). The prevalence of low birth weight remains high at approximately 15%–20%, or nearly 20 million births worldwide each year (World Health Organization  2014 ; Okwaraji et al.  2024 ). Of course, other types of adverse birth outcomes, such as congenital anomalies, stillbirth, small for gestational age, and infant mortality, also contribute significantly to population health burdens. These outcomes not only impact infant health but also have serious implications for maternal health. The effects are especially pronounced in developing regions like Sub‐Saharan Africa, where nearly two‐thirds of global maternal deaths occur (UNFPA, World Health Organization, UNICEF, World Bank Group  2019 ). For all these reasons, research on adverse birth outcomes continues to evolve and is ongoing in a multifactorial and multi‐jurisdictional fashion. Given the diversity of risk factors across a range of specific birth outcomes, there have been numerous published reviews and meta‐analyses that have sought to synthesize some of this evidence. These reviews nearly always focus on summarizing the literature and the associations between a single etiologic risk factor and a specific outcome, or in some cases a limited number of closely related outcomes. However, there is no comprehensive assemblage of these risk factors across a comprehensive series of birth outcomes. Although these types of systematic reviews provide valuable insight, an evaluation of the relevance of all possible risk factors through systematic review would be highly challenging. In such situations, decision analysis represents a methodology to apply scientific theoretical approaches to address multi‐criteria problems. In our case, our aim was to synthesize the perceived strength of evidence for an association between risk factors, identified to be relevant by experts, and a diverse range of adverse birth outcomes. To do so, we opted to use a risk‐based decision analysis through expert elicitation. Expert elicitation is an established approach for incorporating quantitative or qualitative input from a group of experts about unknown quantities or parameters (Colson and Cooke  2018 ). While informal solicitation of expert knowledge often informs policy decision‐making, structured expert elicitation uses a formalized methodology to develop a probabilistic distribution of the state of knowledge to aid decision analysis (Bojke et al.  2022 ). The use of expert elicitation has informed a number of complex health‐related issues from climate change (Morgan et al.  2001 ), air quality (Roman et al.  2012 ; Knol et al.  2009 ), pesticides (von Krayer Krauss et al.  2004 ) as well as health care decision‐making (Bojke et al.  2022 ). The US Environmental Protection Agency has incorporated expert elicitation into processes related to addressing uncertainties in environmental sciences (US Environmental Protection Agency  2011 ). Given the complexities of prioritizing research activities on the study of birth outcomes, soliciting judgments from experts can help better understand uncertainty in a way that facilitates decision‐making. As described by Knol et al. ( 2010 ), expert elicitation represents a means that is structured and transparent and useful to address these uncertainties. Moreover, as Colson and Cooke ( 2018 ) draws attention to, it is an approach that is particularly valuable when “problems are too urgent, or stakes are too high to postpone measures until more complete knowledge is available.” While a number of reviews exist for establishing best practices in expert elicitation, including Morgan and Henrion ( 1990 ), Cooke ( 1991 ), and O'Hagan et al. ( 2006 ), there is no formally recognized and established methodology for structuring an expert elicitation process (Bojke et al.  2022 ). Often, expert elicitations require a built‐for‐purpose protocol, although existing protocols can provide guidance (Soares et al.  2024 ). More recently, Bojke et al. ( 2022 ) and Soares et al. ( 2024 ) have worked to establish a set of principles for expert elicitation in health care decision‐making. Where possible, the insights and best practices recognized through the literature for expert elicitation were incorporated into the methodological approach for this paper. With this background, SciPinion initiated this project whose objectives were to identify an expert elicited contemporary listing of risk factors most relevant to the study of birth outcomes and to provide a relative scoring of these risk factors. This objective was motivated by the goal of providing insight on priorities for research and how best to account for the role of numerous risk factors in epidemiological studies of birth outcomes. The input provided by this expert elicitation culminated in the creation of a publicly available weight of evidence database ( https://scipinion‐rfbo.onrender.com ) which is described herein, which will allow for ongoing development and validation of the judgments provided by experts.

The authors declare no conflicts of interest.

Recognizing the complexities and diversity of adverse birth health outcomes and their risk factors, we sought expert elicitation through the recruitment of an interdisciplinary group of experts representative of fields related to birth outcomes. These included individuals with expertise in epidemiology, birth outcomes/defects, obstetrics and gynecology, reproductive and developmental effects, and pediatrics. As wide a net as possible was cast to engage a diversity of experts. A total of over 32,342 emails were sent to potential candidates based on SciPinion's database of experts as well as a targeted PubMed search for researchers who have published in the fields listed above. A total of 358 experts applied to our recruitment. Of these, 24 were rejected for not providing sufficient information to allow for their expertise to be vetted. The remaining 334 experts were rated (scored) by SciPinion based on objective measures of expertise, including: years of experience, number of publications, number of first and last author publications, and scores on key words related to the topic. This study did not include stratified‐sampling expert recruitment, which may have increased control over stratum bias. However, we do not believe any expert stratum is over or under‐represented in the expert pool, except perhaps years of experience since our selection algorithm weights experience and number of publications as a positive factor. Based on a ranking of scores, 30 experts were selected for participation in the SciPi (collection of SCIentific oPInions). Two experts dropped out after Round 1, and therefore, our analysis is based on ratings provided by 28 experts. The expert panel participated in the elicitation between April and July 2024. A listing of the panel of experts who participated is provided in Table  1 . Listing of experts who participated in the expert elicitation panel. Conflict of interest was not explicitly controlled for in the study. All experts were blinded to the sponsor of the study to eliminate this potential source of influence. The sponsor was blinded to the panel participants until submission of the manuscript to the journal. All experts were blinded to each other during their involvement in this panel. Following the selection of experts to participate, the experts were asked to provide input at three time points. The methods used at each interval are described below. In this round, the experts were asked to identify the birth outcomes and risk factors they were aware of, and those they deemed most important. The experts were encouraged to identify risk factors for adverse impacts to birth outcomes across a range of risk factor categories, including, but not limited to: physiological, lifestyle, residency, socioeconomic, occupational, and environmental. Experts were also asked to provide citations in support of the outcomes and risk factors they identified. Implicit in these instructions was the expectation that experts would not identify risk factors for which there was considerable evidence that they were unrelated to the risk of adverse outcomes. Such an approach would allow for weight of evidence scoring for risk factors for which our group of experts identified as relevant. This approach would cast a wide net and allow for weight of evidence ratings for risk factors for which varying levels of evidence were available thereby maximizing dissemination of information for future use of the publicly available database that we created. The compilation of risk factors was organized by the research team and sent back to the panel for review and finalization. Before the second round, we compiled and summarized data collected in Round 1. These summaries were made available for all experts to review. The experts were then prompted whether they would like revisions on the identified series of risk factors and health outcomes, allowing for correction of potential inaccuracies. This resulted in some minor changes to the listings. Following this step, we finalized the list of risk factors and adverse health outcomes to include in a weight of evidence assessment. The identified birth outcomes and risk factors were entered into a spreadsheet and organized into 11 different risk factor categories (each organized into a separate tab in Excel). In total, there were 15 health outcomes (columns) in each of the 11 tables, and 143 risk factors (rows) combined across the 11 tables. This yielded a total of 2145 cells for which ratings were possible. Experts where then charged with scoring combinations of outcomes and risk factors. Each expert was asked to provide a weight of evidence score for each cell, representing a specific risk factor and health outcome, for which they had working knowledge of the literature. The experts scored this cell with a value ranging from 0 to 5 where 0 represented “no evidence” and 5 represented “strong evidence” of an association between the specified risk factor and corresponding health outcome. Following this scoring, the weight of evidence scores from the 28 experts were aggregated across all risk factors and disease outcome combinations. We retained descriptive statistics for each cell that included the number of provided scores, the mean score, and the standard deviation of these scores. To summarize the relative ratings across all cells for this manuscript, heat maps were generated. Additional details and resources are available in the online database. A database was created to house the results of the expert elicitation scoring. The database was created using and is housed at https://scipinion‐rfbo.onrender.com . This database contains a brief overview of the relevance of each of the 11 domains of risk factor for each health outcome. It also contains summaries of the weight of evidence scores provided through the expert elicitation.