Objective To compare predictive values of small-for-gestational-age (SGA) by different measures for secondhand smoke (SHS) exposure during pregnancy and to develop and validate a prediction model for SGA using SHS exposure along with sociodemographic and pregnancy factors.
Methods We compared the predictability of different measures of SHS exposure during pregnancy for SGA among 545 Chinese pregnant women, and then used the optimal SHS measure along with other clinically available factors to develop and validate a prediction model for SGA. We fit logistic regression models to predict SGA by single measures of SHS exposure (self-report, serum cotinine and CYP2A6*4) and different combinations (self-report+cotinine, cotinine+CYP2A6*4, self-report+CYP2A6*4 and self-report+cotinine+CYP2A6*4).
Results We found that self-reported SHS exposure alone predicted SGA (area under the receiver operating characteristic curve or area under the receiver operating curve (AUROC), 0.578) better than the other two single measures (cotinine, 0.547; CYP2A6*4, 0.529) or as accurately as combined SHS measures (0.545–0.584). The final prediction model that contained self-reported SHS exposure, prepregnancy body mass index, gestational weight gain velocity during the second and third trimesters, gestational diabetes, gestational hypertension and the third-trimester biparietal diameter Z-score could predict SGA fairly accurately (AUROC, 0.698).
Conclusions Self-reported SHS exposure at peribirth performs better in predicting SGA than a single measure of serum cotinine at the same time, although repeated biochemical cotinine assessments throughout pregnancy may be optimal. Our simple prediction model is fairly accurate and can be potentially used in routine prenatal care.
- Secondhand smoke
- Smoking Caused Disease
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Small-for-gestational-age (SGA) is a risk factor not only for neonatal morbidity and mortality,1 but also for long-term adverse outcomes in adolescence (ie, neurodevelopmental and behavioural disorders)1 ,2 and adulthood (ie, metabolic syndrome).3
Maternal secondhand smoke (SHS) exposure during pregnancy is a modifiable risk factor for SGA. Self-report and serum cotinine are two common methods for measuring SHS exposure during pregnancy.4 ,5 As an objective measure, serum cotinine from maternal active smoking is often believed to be more accurate in predicting adverse pregnancy outcomes (eg, SGA, preterm birth) and thus more valid to reflect true tobacco exposure than self-report.6 However, it is unknown if this conclusion can be generalised to maternal SHS exposure. Some evidence among non-pregnant populations suggests that recall bias by non-smokers who are exposed to SHS seems to be a matter of less concern than that by active smokers, given that smoking is a socially unfavourable behaviour,7 which may be more the case among pregnant women. In addition, our previous work suggests that serum cotinine concentration among pregnant women exposed to SHS can be largely affected by polymorphisms of the CYP2A6*4 gene (the key phase I enzyme for nicotine metabolism),8 which is consistent with previous studies on active smoking.9 Serum cotinine concentration is also substantially affected by the timing of the blood draw,10 as it usually increases first to a peak and then deceases rapidly (half-time, 16∼20 h) after smoking a cigarette or each exposure to SHS,11 which is an important issue particularly in clinical settings where the interval between SHS exposure and the blood draw is harder to control. So, it is necessary and valuable to compare different measures of SHS in terms of their predictive ability for SGA among non-smoking pregnant women, which has not been studied so far but could help to choose appropriate measures of SHS exposure for clinical purposes of predicting and treating SGA.
Accurate prediction or diagnosis of SGA is crucial in clinical decisions to treat SGA fetuses. A feasible, accurate and early enough prediction model can assist clinicians to initiate effective, timely interventions (eg, intravenous supply of amino acids,12 low-dose aspirin,13 pregnancy termination in late pregnancy14) for a suspected SGA fetus. Although valuable and useful, existing SGA prediction models have several important limitations. First, some of them miss some important predictors. For example, some SGA prediction models exclusively use ultrasound examination indicators such as estimated fetal weight and abdominal circumstance,15 without other routinely assessed risk factors such as prepregnancy body mass index (BMI) and gestational weight gain (GWG) that have good potential to improve the predictability.16 ,17 Second, few prediction models have been validated either internally or externally,18 which may limit their generalisability and thus weaken external researchers’ confidence of using them. Third, few studies have reported an optimal cut-off point for high predicted risk of SGA that warrants further clinical interventions. Finally, almost all of these models are under the assumption that the cost of a false negative is equal to that of a false positive. However, this assumption seems unrealistic because clinicians and patients usually believe a false negative SGA from the prediction model costs more than a false positive non-SGA, given the serious consequences of untreated SGA cases.
In this study, we compared predictive values of SGA by different measures for SHS exposure during pregnancy including self-report (time), serum cotinine (metabolite) and CYP2A6*4 (enzyme activity), and also developed and validated a prediction model for SGA using SHS exposure along with sociodemographic and pregnancy factors. We also provided the final prediction model with specific cut-off points by assigning equal or unequal weights to false positives and negatives.
We analysed two subsets of samples from a hospital-based case–control study (N=3566) conducted in two local Women and Children's Hospitals (Shenzhen and Foshan, Guangdong Province, southern China) from September 2009 to March 2011. This study was designed to investigate the gene-environmental factors for low-birth weight. Self-reported non-smoking pregnant women were recruited during their hospital visits for delivery. For each participating woman, only one pregnancy was recruited into the present study. Among the 3566 pregnant women, 545 women with complete data on self-reported time of SHS exposure during pregnancy, serum samples and CYP2A6*4 genotype (analytical sample 1) were selected to compare the predictive values of SGA by these three measures related to SHS exposure. Pregnant women in analytic sample 1 were older and more likely to live in an urban area, have a higher educational level and family monthly income and be less likely to be employed than those in the full sample. However, there were no differences in ethnicity, parity and GWG velocity during the second and third trimesters between the full sample and analytic sample 1 (see online supplementary table S1).
Ultrasound examination has become a part of routine prenatal care and some ultrasound measures, especially fetal biparietal diameter (BPD), have been widely used to predict or even diagnose SGA in China. Thus, of the same 3566 pregnant women, we included 1374 with complete data on ultrasound fetal BPD and other pregnant characteristics, sociodemographics and self-reported SHS exposure (analytical sample 2) to build and validate a prediction model for SGA.
This study was approved by the Institutional Review Board of School of Public Health, Sun Yat-sen University. All the participants signed informed consent and allowed their blood samples to be used for research purposes.
Trained medical students and physicians did face-to-face interviews with the participants using a structural questionnaire in the two hospitals after delivery. Participants self-reported the information on sociodemographic (ie, age, ethnicity, residential area, educational level, employment status and family monthly income), pregnancy (ie, prepregnancy weight, height and predelivery weight) and SHS-related characteristics (ie, exposure status and time).
We obtained participants’ other pregnancy characteristics (ie, parity, gestational hypertension, gestational diabetes and the offspring's gender, ultrasound fetal BPD and birth weight) from medical records. In our study, all diseases (eg, gestational hypertension and gestational diabetes) were diagnosed by experienced obstetricians after group discussion according to standard protocols.19 ,20 The diagnostic criteria for diseases were fairly consistent across the study periods. Obstetricians classified the diagnosed diseases according to the International Classification of Diseases, 10th edition (ICD-10). Several other obstetricians examined the reliability and integrity of medical records before they were delivered to medical record rooms. Librarians further checked the completeness of medical records periodically. Our well-trained research staff with a medical background abstracted final diagnoses from paper-based medical records. Although we did not find direct evidence on validity and reliability of abstracted data from medical records in Chinese hospitals, evidence from another country shows that the information extracted from medical records even by staff without clinical training was valid and reliable.21
We drew participants’ blood samples immediately (within 16–20 h, the half life of serum cotinine) after they were admitted into hospitals for the reason of labour. Two millilitres of collected blood samples were centrifuged at 3000 rpm/min to obtain serum to measure the concentration of cotinine. Another 2 mL of blood samples were used for genotyping CYP2A6*4. For the purpose of this study, we only included women whose blood was drawn within 20 h (ie, half lifetime of serum cotinine) after admission.
We considered six sociodemographic variables that have been reported to be closely associated with SGA. They included family monthly income in Chinese yuan, maternal age in years, ethnicity (Han or minority), residential area (urban, suburban or rural), educational level (elementary school or illiteracy, high or middle school, college or higher) and employment status.
Pregnancy characteristics of our interest included maternal prepregnancy weight, height, predelivery weight, parity, gestational hypertension, gestational diabetes and ultrasound fetal BPD. Prepregnancy BMI (kg/m2) was defined as prepregnancy weight in kilograms divided by the square of height in metres. GWG (kg) was defined as the difference between predelivery and prepregnancy weights. To facilitate clinical use throughout pregnancy, we used GWG velocity instead of total GWG to predict SGA. A direct and accurate measure for GWG velocity was unavailable in this study, as we had only one weight measure right before delivery. However, since maternal weight usually does not change that much in the first trimester (gaining 1 kg on average) and also GWG is almost linear with gestational age in the second and third trimesters,22 we were able to estimate GWG velocity in the second and third trimesters as (predelivery weight—prepregnancy weight—1)/(gestational age—12). We recognised that this approach with a pseudo weight at the end of the first trimester could result in some errors, but these errors were outweighed by the benefits of obtaining an acceptable measure of GWG velocity in the second and third trimesters, a key factor for early prediction of SGA.
In the obstetric ultrasound examination, BPD measures the distance between the two sides of the fetus’s head. We obtained ultrasound fetal BPD from the second or third trimester ultrasound examination records. Using a reference population of normal Chinese fetuses of 14–40 gestation weeks,23 we calculated the gestational-age-specific BPD Z-score as (individual BPD–population mean)/(population SD).
SHS exposure measurements during pregnancy
The key exposure in the present study was SHS exposure, which contains nicotine and many other tobacco constituents such as polycyclic aromatic hydrocarbons. We used self-report and serum cotinine to measure SHS exposure during pregnancy. Given that all tobacco products contain nicotine and serum cotinine is the key metabolite of nicotine, we measured serum cotinine and used it as the proxy of nicotine which can reasonably reflect the level of SHS exposure. Pregnant women self-reported their SHS exposure status as well as daily time of exposure during pregnancy at home, in the workplace and in public places. Accordingly, we calculated the total daily time of SHS exposure by summing the daily times of exposure at these three locations. We measured serum cotinine by an ELISA kit (Immunalysis Corporation, Pomona, California, USA) with the limit of detection 1 ng/mL. The concentration of serum cotinine (nicotine metabolite) and other tobacco metabolites has been shown to be substantially influenced by the CYP2A6*4 gene polymorphism in our sample and others.9 ,10 Thus, we also considered CYP2A6*4 as an SHS-related variable along with serum cotinine to better reflect unmeasured internal tobacco exposure such as nicotine.
Midwifery nurses measured birth weight immediately after delivery. In most Chinese hospitals including the two participating hospitals in this study, last menstrual period (LMP) is the routine method to estimate gestational age. Obstetricians consider ultrasound to estimate gestational age only when pregnant women (usually a very small proportion, 0.9% in our sample) cannot recall their LMP. Therefore, in this study, we defined gestational age by combining LMP (gestational age=delivery date—LMP) and ultrasound estimate. Using a reference population of normal Cantonese newborns of 14–40 weeks of gestational duration,24 we calculated the gestational age-specific and sex-specific birth weight Z-score as (individual birth weight−population mean)/(population SD). Accordingly, SGA cases were defined as the gestational age-specific and sex-specific birth weight Z-score below −1.28 (equivalent to below the 10th centile in a hypothetical normal distribution) and Non-SGA controls were defined as the birth weight Z-score greater than −1.28.
Step 1: comparison of different SHS exposure measures (analytic sample 1)
The area under the receiver operating curve (AUROC) generated by logistic regression models was used to compare the discrimination accuracy of different SHS exposure measures (self-reported time, serum cotinine level, CYP2A6*4, self-report+cotinine, cotinine+CYP2A6*4, self-report+CYP2A6*4 and self-report+cotinine+CYP2A6*4) in predicting SGA alone or along with those sociodemographic and pregnancy characteristics with a p value less than 0.2 in univariate analyses. For the self-reported time of SHS exposure or serum cotinine concentration, we used the Youden Index (sensitivity+specificity—1) within the receiver operating characteristic (ROC) curve to identify the optimal cut-off point to define the binary self-reported SHS exposure and serum cotinine for predicting SGA.
Step 2: building prediction models for SGA (analytic sample 2)
We first randomly divided analytic sample 2 into training (80%, N=1100) and validation data sets (20%, N=274). In the training data set, we used stepwise logistic regression to select significant predictors into the final prediction model for SGA. We then used the fixed coefficients in the final model derived from the training data set to calculate the expected risk of SGA of each newborn in the validation data set. The expected risk of SGA was then used as a predictor to build the ROC curve for observed SGA cases in the validation data set. We evaluated the performance of the final prediction model by calculating the sensitivity, specificity, negative predictive value (NPV) and positive predictive value (PPV) at different probability cut-off points (0.16–0.66) in the training and validation data sets. We also evaluated NPV and PPV by applying the prevalence of SGA (6.9%) among the general Cantonese population.25 In order to facilitate clinical use for different purposes (classification vs prediction), we used three criteria to select optimal cut-off points for SGA probability: (1) to maximise the sum of sensitivity and specificity; (2) to maximise the sum of PPV and NPV. (3) to maximise the sum of PPV and 2×NPV. The difference between the latter two criteria was that we assigned equal (1:1 in criterion 2) or unequal (2:1 in criterion 3) weights to false negatives (1-NPV) and false positives (1-PPV). We considered a prediction model with an AUROC of 0.70 or higher as clinically useful.26 The model calibration (the extent of matching between predicted and observed risk of SGA) was assessed by the Hosmer-Lemeshow (H-L) test.
We used the multiple imputation approach (Markov chain Monte Carlo, five imputations) to impute the missing data on sociodemographic and pregnancy characteristics except ultrasound fetal BPD. All data analyses were performed in SAS V.9.2 (SAS Institute, Cary, North Carolina, USA).
Table 1 shows sociodemographic, pregnancy and SHS-related characteristics of the two analytic samples.
Comparison of different SHS measurements in predicting SGA
Table 2 shows predictive values for SGA by different measures of SHS exposure during pregnancy including each of self-reported daily SHS exposure time (min/day), serum cotinine level (ng/mL) and CYP2A6*4 (*1/*1, *1/*4 and *4/*4), as well as their pairwise and all combinations. The AUROC of self-reported daily SHS exposure time alone was 0.578 (95% CI 0.526 to 0.631). Using it as the reference group, serum cotinine (0.547, 95% CI 0.492 to 0.602) or CYP2A6*4 (0.529, 95% CI 0.473 to 0.584) alone, the combination of self-report and serum cotinine and the combination of serum cotinine and CYP2A6*4 achieved a lower AUROC, although the differences were not statistically different. The combination of self-report and CYP2A6*4 (0.580, 95% CI 0.524 to 0.637), and the combination of self-report, serum cotinine and CYP2A6*4 (0.584, 95% CI 0.528 to 0.640), achieved a similar AUROC with self-report alone. According to the ROC analysis, the optimal cut-off point of self-report daily SHS exposure time for predicting SGA was 0.145 min/day (sensitivity=0.519, specificity=0.615, NPV=0.795, PPV=0.307). To facilitate practical use in clinical settings, we decided to use 0 min/day (very close to 0.145 min/day) as the conventional cut-off point, which had almost the same predictive values with the optimal cut-off point. The optimal cut-off point of serum cotinine for predicting SGA was 1.910 ng/mL (sensitivity=0.720, specificity=0.390, PPV=0.807, NPV=0.279). Accordingly, we set the conventional cut-off point as 2 ng/mL, which had similar predictive values (sensitivity=0.674, specificity=0.405, PPV=0.790, NPV=0.272) with the optimal cut-off point.
Table 3 shows that the changes in AUROC for SGA by adding different SHS-related measures to the regression model already with sociodemographic and pregnancy characteristics. Overall, the pattern of changes in AUROC was similar to that in table 2 mentioned above. This pattern of changes in AUROC is consistent for continuous and binary measures of self-report and serum cotinine. In addition, adding self-reported time as a binary variable (0 vs >0 min/day) resulted in an increase in AUROC to a similar extent (0.678, 95% CI 0.626 to 0.731 vs 0.684, 95% CI 0.631 to 0.737) as adding self-reported time as a continuous variable (min/day). Therefore, we decided to use the binary self-reported SHS exposure time (0 vs >0 min/day) as the final indicator in the SGA prediction model as below.
SGA prediction model
Using the training data set (N=1100) of analytic sample 2, we derived the final prediction model for SGA as “Logit (Probability of SGA)=2.16+0.46×binary self-reported SHS exposure time (0 vs >0 min/days)−0.15×prepregnancy BMI (kg/m2)−1.75×GWG velocity in the second and third trimesters (kg/week)−1.62×gestational diabetes (Yes/No)+1.08×gestational hypertension (Yes/No)−0.36×third trimester BPD Z-score”. Table 4 was the ORs of SGA for binary SHS exposure (0 vs >0 min/days), prepregnancy BMI (per unit), GWG velocity in the second and third trimesters (kg/week), gestational diabetes, gestational hypertension and third trimester BPD Z-score (per unit), respectively. The AUROC of the final prediction model was 0.698 (95% CI 0.660 to 0.737, figure 1) and the H-L test indicated no significant differences (p=0.782) between observed and predicted risks of SGA.
In the validation data set, the final model yielded an AUROC of 0.672 (95% CI 0.588 to 0.755) and the H-L test indicated no significant differences (p=0.820) between observed and predicted risks of SGA. Table 5 shows the predictive values of the final prediction model under different cut-off points of SGA probability (range, 0.16–0.66) in the training data set, the validation data set and the general Cantonese population with an SGA prevalence of 6.9%. The cut-off point of SGA probability that maximised the sum of sensitivity and specificity was 0.22 (Sensitivity=0.621, Specificity=0.685, PPV=0.335, NPV=0.876). The cut-off point of SGA probability that maximised the sum of PPV and NPV was 0.48 (Sensitivity=0.063, Specificity=0.986, PPV=0.542, NPV=0.805) and this point also maximised the sum of PPV and 2×NPV. Compared to the training data set, the cut-off point of 0.22 in the validation data set yielded lower sensitivity (0.571), specificity (0.681) and PPV (0.281) but slightly higher NPV (0.879); also, the cut-off point of 0.48 yielded lower sensitivity (0.020) and PPV (0.333), but higher specificity (0.991) and NPV (0.823). In the general Cantonese population, PPV was 0.111 and NPV was 0.961 at the cut-off point of 0.22, while NPV was 0.225 and NPV was 0.934 at the cut-off point of 0.48.
Predictive values for SGA by different SHS exposure measures
We found that self-reported SHS exposure time during pregnancy alone could predict SGA better than once measure of serum cotinine before delivery and/or CYP2A6*4. Binary self-reported SHS exposure time during pregnancy (0 vs >0 min/day) predicted SGA as well as the more detailed SHS exposure time did. This seemingly surprising finding may be explained by the fact that (1) like most of the previous studies,27 ,28 blood samples used for measuring cotinine were collected only once before delivery in this study and thus might not accurately reflect SHS exposure during the entire pregnancy, which could theoretically compromise the validity of serum cotinine as an objective measure for SHS exposure. Although we are not sure, we suspect that repeated measures of serum cotinine during pregnancy, for example, the first, second and third trimesters, may improve the predictability of serum cotinine for SGA; (2) compared to the non-pregnant population, pregnant women are usually more health-conscientious. They may pay more attention to the smoking environment and are thus more likely to accurately recall their SHS exposure; (3) as the key metabolite of nicotine, serum cotinine is only a marker for nicotine and cannot represent other harmful constituents in tobacco smoke that may also impact fetal growth. Moreover but less relevantly, it should be noted that our finding is different from some evidence suggesting that serum cotinine is more reliable than self-report in reflecting active smoking.6 This may be explained by the differential levels of recall bias by pregnant women regarding SHS exposure versus active smoking. Compared with SHS exposure, active smoking during pregnancy was more socially undesirable or even unacceptable, and is thus more subject to being under-reported. Taken together, from a measurement standpoint, the optimal approach of SHS measurement might be repeated biochemical cotinine assessments throughout pregnancy; however, this is not often practical. Therefore, in the absence of this optimal situation, self-reported SHS measure at peribirth performs better than a single cotinine assessment in predicting SGA at the same time.
Different from our hypothesis, we found that adding CYP2A6*4 could not improve the predictability of serum cotinine. One possible explanation is that as the downstream reaction product of nicotine, the concentration of serum cotinine may already reflect the contribution of the CYP2A6 enzyme to the metabolic conversion from nicotine to cotinine. Therefore, adding the CYP2A6*4 genotype (mostly impacting the nicotine metabolism) to the model already with serum cotinine may not contribute substantially new information for predicting SGA.
Compared to the self-report, measuring serum cotinine and/or genotype CYP2A6*4 is more costly and time-consuming. For example, in this study our estimated total cost, including the blood draw, transportation, processing, lab test and personnel, was ¥31 (equivalent to $5.1) for serum cotinine per sample by the ELISA method. A more accurate approach such as gas chromatography for measuring serum cotinine could cost more. The corresponding cost was ¥27 (equivalent to $4.4) for genotyping CYP2A6*4 per sample by the PCR method. However, the estimated total cost for self-report was only ¥14.3 (equivalent to $2.3), including recruitment, interview, questionnaire and compensation for subjects. With a slightly higher predictability and half or lower cost, self-reported SHS exposure during pregnancy seems to be more cost-effective to predict SGA than serum cotinine with or without CYP2A6*4. In addition, our finding suggested that the binary measure (0 vs >0 min/day) of self-reported SHS exposure predicted SGA similarly to detailed self-reported SHS exposure time (continuous). We are not sure about the reason(s) for this similarity, but it does not appear to be explained by a very low threshold (eg, close to zero) of SHS exposure time that is sufficient for causing SGA, as our supplemental analysis suggested a dose–response association between self-reported SHS exposure time and risk of SGA (adjusted OR=1.50, 95% CI 0.92 to 2.47 for 1–10 min/day and 1.94, 95% CI 1.21 to 3.11 for 11+min/day, compared to 0 min/day).
Prediction models for SGA
The early prediction, diagnosis and treatment of SGA are important for reducing SGA-related morbidity and mortality, which can be facilitated by an accurate, simple and economic prediction model for SGA. The AUROC of our final prediction model was very close to our preset cut-off level (0.7) of clinically use. Our model is relatively simple and contained only six economic and clinically available variables including binary SHS exposure, prepregnancy BMI, GWG velocity in the second and third trimesters, gestational diabetes, gestational hypertension and third trimester BPD Z-score. These variables can be obtained with small extra efforts in routine prenatal care: binary SHS exposure and prepregnancy BMI by asking pregnant women, gestational diabetes, gestational hypertension and GWG from medical records, and third trimester fetal BPD from routine prenatal ultrasound examination. In summary, acceptable accuracy and simplicity allow our prediction model to be potentially used in routine prenatal care.
First, retrospectively self-reported maternal prepregnancy weight, predelivery weight and height were subject to recall bias. Second, the generalisability of our prediction model for SGA may be limited as the participants were recruited from two local hospitals. In particular, our prediction model may perform well only among a high-risk population due to oversampling SGA newborns in this case–control study. Although we observed acceptable PPV and NPV after applying the average risk of SGA among the general Cantonese pregnant women, further research is much needed to evaluate its external validity in an independent sample. Third, selection bias was very likely to exist as most sociodemographic, pregnancy and SHS-related characteristics significantly differed among the full sample and the two analytic samples (see online supplementary table S1). We suspect that most of these differences might be explained by the much higher percentage of urban residents in analytic sample 1 (87.7%) compared to that in analytic sample 2 (31.0%). Most Chinese pregnant women who live in urban areas are voluntarily admitted to hospital 2–3 days for labour before the due date, whereas most pregnant women who live in rural areas are admitted to hospital only when parturition starts. As a result, urban pregnant women were more likely to be included in analytic sample 1 which needs blood samples to measure serum cotinine and genotype CYP2A6*4. Fourth, owing to the relatively low limit of detection, the serum cotinine ELISA kit might not be sensitive to distinguish low-level SHS exposure from non-exposure. Fifth, our results might be biased by the fact that pregnant women reported their SHS exposure status after knowing their babies’ birth outcomes. However, most Chinese women delivering SGA babies are not aware of this birth outcome, because SGA is not included in the routine communication between healthcare providers and the postpartum woman and her family members. Therefore, it is reasonable to assume that in our sample the SGA status should not substantially bias postpartum women’s recall of SHS exposure during pregnancy. Sixth, we obtained BPD information from ultrasound data, which was subject to measurement error. Seventh, as mentioned above, our estimated GWG velocity in the second and third trimesters was also subject to measurement errors due to the use of pseudo weight at the end of the first trimester. Finally, our once measure of serum cotinine shortly before delivery might only reflect SHS exposure during late pregnancy but not earlier pregnancy.
What this paper adds
We compared different measures of secondhand smoke (SHS) in terms of their predictive ability for small-for-gestational-age (SGA) among non-smoking pregnant women, which have not been studied so far but could help to choose appropriate measures of SHS exposure for clinical purposes of predicting and treating SGA.
Self-reported SHS exposure at peribirth performs better in predicting SGA than a single measure of serum cotinine at the same time, although repeated biochemical cotinine assessments throughout pregnancy may be the optimal measure.
Using self-reported SHS exposure measure and other clinically available factors, we established a new SGA prediction model which is fairly accurate and can be potentially used in routine prenatal care.
The authors thank Dr Rachel F Tyndale from the University of Toronto for sharing her methods of genotyping CYP2A6.
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Files in this Data Supplement:
- Data supplement 1 - Online supplement
CX and XW contributed equally to this work.
Contributors CX contributed to hypothesis generation, study design, data collection, data analysis, result interpretation and drafting of the manuscript. WC and XW contributed to hypothesis generation, study design, data analysis plan, result interpretation and revision of the manuscript. ZN, PD, TL, YH, JL, SY, XG and DJ contributed to the study design, field survey, blood sample collection and laboratory tests.
Funding This work was supported by the National Natural Science Foundation of China (grant numbers 30872164, 81172758), the Guangdong Population and Family Plan Committee (grant number 200835) and the International Program of Project 985, Sun Yat-Sen University (awarded to CX).
Competing interests None.
Patient consent Obtained.
Ethics approval Institutional Review Board of School of Public Health, Sun Yat-sen University.
Provenance and peer review Not commissioned; externally peer reviewed.