Secondary Logo

Journal Logo

Original Article: ENVIRONMENT

Air Pollution and Pulmonary Function in Asthmatic Children: Effects of Prenatal and Lifetime Exposures

Mortimer, Kathleena; Neugebauer, Romainb; Lurmann, Frederickc; Alcorn, Sianac; Balmes, Johnd,e; Tager, Iraa

Author Information
doi: 10.1097/EDE.0b013e31816a9dcb

ArticlePlus

Click on the links below to access all the ArticlePlus for this article.

Please note that ArticlePlus files may launch a viewer application outside of your web browser.

Numerous studies have demonstrated that short-term fluctuations in ambient air pollutants such as ozone (O3), particulates with median aerodynamic diameter <10 μm (PM10) and nitrogen dioxide (NO2) are associated with increased acute respiratory symptoms and decreases in pulmonary function in children, especially in asthmatics.1–3 Recent studies have reported negative effects of pollution on lung function in the first year of life4 and among college freshman.5 Given the heterogeneity of response to air pollution,6–8 it is important to identify subgroups of children with asthma who are susceptible to air pollution-related effects.

The issue of prenatal exposure has assumed increasing importance, since ambient air pollution exposures of pregnant women have been shown to lead to adverse pregnancy outcomes and to respiratory morbidity and mortality in the first year of life.9–12 Several of these studies found that exposures during the first trimester (a time of major lung and airway development) are associated with intrauterine growth retardation and preterm birth.13,14 Early-life factors have been shown to be robust predictors of future health outcomes in multiple conditions and have effects that extend into adulthood. Pulmonary function measures have been shown to be useful predictors of mortality.15 Consequently, the examination of the effects of early-life exposures on these outcomes may serve as useful early-warning signs of future morbidity and premature mortality.

This paper examines which prenatal and lifetime exposures to several ambient air pollutants were most predictive of current pulmonary function among a cohort of children aged 6–11 years with asthma.. We sought to identify characteristics of asthmatic children who are most responsive to these exposures.

METHODS

Data for this study were collected within the context of the Fresno Asthmatic Children's Environment Study, a longitudinal epidemiologic study of 315 asthmatic children in Fresno, California. The focus of this paper is a separately funded substudy of 232 members of the cohort for whom prenatal and chronic exposure could be obtained.

Children and their families were recruited through community-based advertisements, school nurses and local physicians. Eligibility criteria included physician-diagnosed asthma, age 6–11 years at the time of the baseline interview, and having a primary residence within 20 km of the California Air Resources Board's Fresno First St. air monitoring site. Only asthmatic children who had been symptomatic or used medication or had a physician visit for asthma in the 12 months previous to enrollment were eligible. The original protocol consisted of an in-person office visit (baseline), followed by a series of home surveys, telephone calls, and office visits.

An in-person, baseline questionnaire was administered in English or Spanish. Baseline data were used to assign each child a severity score based on the criteria of the Global Initiative for Asthma asthma-severity classification scheme.16 Spirometry was performed with a dry, rolling-seal spirometer (Spiroflow; P.K. Morgan Instruments, Andover, MA). Up to 8 attempts were allowed to obtain 3 acceptable tracings as described previously.17 Measures included forced vital capacity (FVC), forced expiratory volume in 1 second (FEV1), peak expiratory flow (PEF), forced expiratory flow between 25% and 75% of vital capacity (FEF25–75), and forced expiratory flow at 25% and 75% of vital capacity (FEF25 and FEF75, respectively). Also considered were the ratio of FEV1 to FVC (FEV1/FVC), and the ratio of FEF25–75 to FVC (FEF25–75/FVC). The FEF25–75/FVC ratio has the interpretation of the reciprocal of the time constant of the lung,18 similar to Meade's Vmax50/(VC × Pst(L)50) (ie, instantaneous flow at 50%, divided by vital capacity times elastic recoil pressure at 50% of vital capacity) and is reflective of intrinsic airway size.19 The mean of the first 3 (but not less than 2) acceptable tracings was calculated.

Skin-prick allergen sensitivity tests to 14 antigens were performed with the MultiTest device (donated by Lincoln Laboratory, Decateur, IL), as described in eTable 1 (available with the online version of this article). Atopy was defined as one or more positive skin tests or a previous severe reaction to allergen skin testing.

Parents were asked to report the street address, city and state of all residences at which the mother lived during pregnancy or at which the child lived prior to enrollment in the study. Only addresses at which they lived for at least 3 months were recorded. Each address was geocoded with “EZ Locate Client” v.1.61, by Tele Atlas (www.geocode.com).

Pollutant concentrations were obtained from the Aerometric Information Retrieval System database supported by the US Environmental Protection Agency. Pollutants considered include O3, NO2, PM10, and carbon monoxide (CO). Pollutant metrics were determined by time-averaging the raw database concentrations. The estimates were mapped to the residences or ZIP code centroids based on inverse distance weighting of the monthly average concentrations from the air monitoring stations (up to 3) located closest to the residence location. Residences were assigned a quality code, based on the distance to the nearest monitor(s). A code of 1 meant that at least 1 station used in the spatial interpolation was located within 5 km of the point of interest (residence location). Similarly, code 2 was used when the closest station was located, relative to the point of interest, between 5 and 50 km for O3, NO2, and PM10, and between 5 and 25 km for CO. A code 3 was assigned when the closest station was located between 50 and 100 km for O3, NO2, and PM10, and between 25 and 50 km for CO.

Based on the age range of this cohort, pollutant levels for the years 1989–2000 were obtained. For all pollutants, means of the monthly 24-hour averages were created, and, for several pollutants metrics, monthly means of maximum values within a day or the monthly means of multihour intervals were created. These monthly values were averaged separately across several important developmental time-periods.

The study protocol was approved by the Committee for the Protection of Human Subjects of the University of California, at Berkeley. Written informed consent was obtained from parents/legal guardians prior to enrollment.

Statistical Methods

We ran age-adjusted models for each of the pollutant metrics of interest from a regulatory standpoint (24-hour average for NO2 and PM10 and 8-hour maximums for O3 and CO) (eTable 2). We developed a common set of plausible confounders or effect modifiers based on previous studies of pulmonary function and air pollution (Table 1). Their associations with pulmonary function values are presented in eTable 3. Low birthweight, premature birth, atopy and asthma diagnosis before age 2 were found not be on the causal pathway between early life exposures and pulmonary function at age 6–11 (data not shown); therefore, we considered them as potential confounders.

TABLE 1
TABLE 1:
Characteristics of the Fresno Asthmatic Children's Environment Study Subjects With Prenatal and Chronic Exposure Information Available (n = 232)

Some of the pollutants in the Fresno area are temporally correlated, as are pollutant metrics for different time periods (eg, lifetime vs. first 3 years of life). Strong correlation indicates there is a lack of information in the data to draw conclusions about the effect of individual pollutant metrics, which may lead to spurious findings. Therefore, rather than trying to model the effect of each pollutant metric on each outcome, we used methods to identify which pollutant metrics were most predictive of pulmonary function. In this exploratory analysis, a large series of metrics was considered simultaneously to identify variables that provided the best predictive model for each pulmonary function outcome. The Deletion/Substitution/Addition (DSA) routine20 uses the L2 loss function criterion (ie, minimizes the residues between observed and expected based on cross-validated model estimates). In this procedure, candidate predictors of the outcome, given all pollutant metrics and covariates (confounders and effect modifiers), are defined with polynomial generalized linear models generated with the algorithm (see below) under user-specified constraints. The space of candidate predictors is parameterized with 3 variables: the number of terms, the order of interactions, and sum of powers in each term. Models were restricted to size = 10, interaction = 2, and sum of powers = 3. To reduce the number of candidate confounders and modifiers (>100), a candidate-reduction step was implemented. Univariate regression analyses of the explanatory variables on each candidate variable were run and the P values that correspond with the test of the null hypothesis were used to rank the candidate variables. The model selection procedure relies on cross-validation and, thus, random splits of the data. For a given call, the data splits correspond to the aggregation of k independent 5-fold splits of the data. The value for k was set to 10. During this process, models that are influenced by outliers will not hold up in the cross-validation process and will not be chosen as the “best model.” They are not likely to be reported in any stable way over the 10 repetitions. Additional details about the Deletion/Substitution/Addition routine are presented in the appendix (available with the online version of this article).

Standard errors were obtained for the semi-parametric model selected with this routine, using the GEE package in the software package R,21 the estimate of standard errors assumed an independence correlation structure. Note that this approach ignores the extra variability of our estimates introduced by the model-selection procedure, and thus possibly underestimates the variance.

Several sets of analyses were conducted using the Deletion/Substitution/Addition routine. First, for descriptive purposes, the list of candidate variables for each pulmonary function outcome model was restricted to the characteristics listed in Table 1 (ie, no pollutant metrics). Next, a more extensive analysis expanded the list of candidate variables to include pollutant-specific metrics for the entire pregnancy, each trimester, the first 3 years of life, the first 6 years of life, and the entire lifetime. Additional analyses were run that restricted the list of possible pollutant metrics to those that summarize the entire prenatal and entire lifetime intervals. In these last 2 sets of analyses, a single pollutant or combination of pollutants could have entered the best model.

The Deletion/Substitution/Addition algorithm is relatively new; therefore, for comparison purposes, the more traditional step-wise regression also was performed with SAS software (SAS Institute, Cary, NC). In this approach, variables are added 1 by 1 to the model; and the F statistic for a variable to be added must be significant (P < 0.05) at the user-specified entry level.

RESULTS

The characteristics of the subgroups included in this analysis are displayed in Table 1. A comparison of the pulmonary function values by characteristics is in eTable 3. The 6 children (3%) who were born in Mexico were not included in the prenatal analysis, due to our inability to obtain prenatal exposure information from Mexico. At the time of the baseline interview, only 20% of mothers were living in the home they had lived when pregnant with the enrolled child.

The distributions of pollutant metrics used for regulatory purposes are presented in Figure 1. As expected, the trimester-specific metrics are more variable than those for the entire prenatal or year-long metrics. Correlations between pairs of these pollutants are presented in Table 2. Only the entire pregnancy and lifetime measures are presented. In general, the strongest correlations are seen between NO2 and CO, which are both negatively correlated with O3.

FIGURE 1.
FIGURE 1.:
Distribution of pollutant metrics used for regulatory purposes. Timeframes: 1, entire prenatal period; 2, 1st trimester; 3, 2nd trimester; 4, 3rd trimester; 5, entire lifetime; 6, age 0–3 years; 7, age 0–6 years.
TABLE 2
TABLE 2:
Correlations Between Regulatory Pollutant Metrics for Prenatal and Lifetime Intervals

The results of the set of models that did not consider any pollutants in the candidate list are displayed in eTable 4. Table 3 displays the findings from the primary model selection process that included metrics for each trimester, the entire pregnancy, the first 3 and 6 years of life, and lifetime (birth through enrollment in the original study). In general, the effect sizes ranged from declines of 0% to 7.8% in the mean pulmonary function measure per interquartile-range (IQR) increase in the mean pollutant concentration. The list of variables selected for each model can be found in the table. As expected, height was selected in each model, with the exception of the ratio outcomes. For brevity, only the pollutant metrics selected are discussed below. The best model for FVC included a negative interaction between NO2 during the second trimester and African American race. The best model for FEV1 included a negative term for NO2 in the second trimester (cubed). The best model for PEF included a negative interaction between first trimester exposure to PM10 and mothers’ smoking status when pregnant. The best model for FEF25–75 included a positive (but not significant) effect of exposure to CO during the first trimester. For the FEV1/FVC ratio, a negative effect of CO during the first 3 years of life (squared) was chosen. The best model for FEF25–75/FVC identified a negative effect of CO during the first 6 years of life among children diagnosed with asthma before the age of 2 years. The best model for FEF25 included a negative interaction between exposure to PM10 during the second trimester and mother's smoking status during pregnancy, as well as a negative interaction between CO in the first 6 years of life and a diagnosis of asthma before the age of 2 years. No pollutant effect was identified for FEF75.

TABLE 3
TABLE 3:
Best Predictive Model When Entire Prenatal, Trimester-Specific, First 3 and 6 Years, and Lifetime Metrics Were Considered

When only summary metrics for the entire pregnancy or lifetime were considered, the results were similar; effect sizes ranged from no effect to decreases of 15% in the mean pulmonary function measure per IQR increase in the pollutant concentration (eTable 4). Stepwise procedures (eg, PROC REG in SAS) do not consider all interactions and nonlinear terms unless specifically included in the model statement. Therefore, for the comparison of the model selection procedures, all interaction and nonlinear terms that were identified by the Deletion/Substitution/Addition procedure (ie, for FVC, NO2 during the second trimester × African American race) were calculated and included as candidates for that outcome in the more traditional step-wise approach (Table 4). For FVC, for example, both model selection results included the height and a negative interaction for NO2 during the second trimester and African American race. The stepwise model selection process also included negative terms for 2nd trimester exposure to CO and current inhaled steroid use, as well as a positive term for PM10 during the first 3 years of life.

TABLE 4
TABLE 4:
Model Selected by Traditional Stepwise Selection
TABLE
TABLE:
no caption available.

DISCUSSION

In multivariable analyses, negative effects on pulmonary function were found for exposure to PM10, NO2, and CO during key neonatal and early life developmental periods. Most of the effects were limited to subgroups of the population with characteristics typically associated with a greater prevalence or severity of asthma. Specifically, children who are African American, were diagnosed with asthma at an early age, exposed to maternal smoking during pregnancy or had a history of atopy or steroid use were the most responsive to the pollutants. The study findings highlight the importance of characterization of exposures during important developmental periods, such as specific trimesters or years of life, which may be useful to identify possible mechanisms by which pollutants exert their effect and to identify critical periods for intervention.

When the adjusted R-square values from the variables in the models chosen by the Deletion/Substitution/Addition procedure were compared with those from the traditional stepwise procedure, the results sere similar. More important, the overall conclusions are similar—prenatal and early-life exposure to CO and NO2 were negatively associated with lung function at age 6–11 years. However, our new procedure identified nonlinear effects for many of the associations. For several outcomes, multiple pollutants were included in the model selected. These findings highlight the need for flexible model-selection strategies (such as this one) that consider all relevant interactions and nonlinear terms and copollutants. Important interactions would have been missed and null effects may have been assumed when, in fact, key subgroups appear to be substantially more susceptible.

There are several other advantages to this newer model selection method. The Deletion/Substitution/Addition procedure uses cross-validation to identify the best model. SAS fits the model on the entire dataset and does not use the cross-validation; it is therefore more susceptible to over-fitting the model. The traditional stepwise selection tended to pick a larger model, and one with several pollutants. Often these pollutants had opposite signs, which may have counter intuitive interpretations due to the colinearity among the pollutants. In addition, the Deletion/Substitution/Addition approach can compare models based on different samples sizes, unlike methods such as the Akaike's Information Criteria, which can only compare across same sample size or nested models. These results lead us to feel that traditional step-wise method should not be considered the “gold standard” and that the cross-validation methods are superior.

The large number of exposures and modifiers examined may raise concerns about multiple testing. Given that this was the first study to examine this research question and is exploratory (given the small sample size), our approach erred on the side of too many, rather than too few, comparisons. For example, when our model-selection procedure is allowed to “choose” between a metric that covers the entire prenatal period and a metric that covers only the 2nd trimester, the cross-validation process revealed that the model with the 2nd trimester metric was superior. This more specific and biologically more plausible finding would not be revealed if only summary (or fewer) metrics were considered. The same argument applies to the inclusion of modifiers such as race. For FVC, the model with the interaction between African American race and NO2 was chosen over a model that also included main effects for each term, which suggests that the effect may be limited to, or at least substantially greater among, African American children.

Finally, the “state of the art” with respect to the research question warrants the detailed exploration undertaken. Current approaches to the multipollutant complexity of the ambient environment have been constrained by model-fitting choices. Moreover, they have required some “guessing” as to the proper form of the model without a thorough investigation of the strength of the guess. On theoretical grounds, the Deletion/Substitution/Addition method identified the “best fit” (ie, smallest cross-validated loss function) for the form and complexity. When combined with subject matter knowledge, this allows for the exploration of the sensitivity of results to a range of model specifications and presents results that are maximally robust with respect to these issues.

In contrast to previous findings,5 we did not find a negative effect of exposure to O3 among this cohort. As shown in Table 2, NO2 and O3 concentrations are negatively correlated (r = −0.4), while NO2 and CO concentrations are highly positively correlated (r = 0.7) in Fresno. NO2 and CO are markers for pollutants generated by urban combustion sources, primarily mobile sources in Fresno. Most of the residences are located in the urban and suburban areas that have elevated NO2 and moderate O3 concentrations. Few participants lived in the downwind areas (eg, near Parlier, CA) that have high O3 and low NO2 and CO concentrations. The absence of rural downwind exposures in our study population probably explains the lack of O3 association with the outcome. The NO2 and CO associations are stronger than the O3 association in this urban setting, where substantial scavenging of O3 by fresh NO emissions occurs.

Our data suggest that first and second trimester pollutant exposures (PM10 and NO2, specifically) had a negative effect on pulmonary function at age 6–11 years. An important consideration in the determination of the effects of various environmental exposures on respiratory health in children is the state of development of the lungs and the immune system at the time of exposure. The lungs begin to develop at 6 weeks of gestation and continue through distinct phases of progression. The airways and blood vessels are in place by 26 weeks’ gestation. The proliferation of alveoli occurs primarily from 26 weeks through birth, and continues during extrauterine life until approximately 8 years of age, although most alveoli are formed by age 2 years. In utero exposure to tobacco products has been shown to have effects on measures of airway function and not on lung volumes, which suggests that there is a vulnerable period with respect to respiratory tract development in the first and early second trimesters.22 Since tobacco smoke contains many compounds found in ambient air (NO2, CO, fine particulate, polycyclic aromatic hydrocarbon), it is reasonable to hypothesize that ambient pollutant exposures also alter the development of the fetal respiratory tract through similar mechanisms.

In utero exposure to tobacco smoke is known to be associated with low birth weight23 and with lower lung function later in childhood24,25 and the occurrence of asthma.26 In utero exposures to maternal smoking have been found to be associated with decreased pulmonary function in infants27,28 and children of school age, especially for small airway flows.25 Data from the NHANES3 study indicate that high cotinine levels in children aged 4–11 years were associated with 2% declines in FEV1 and 6% declines in FEF25–75.29 These effect sizes are quite similar to those seen for an IQR increase in the pollutants examined in this analysis.

There are limitations to this study and our analytic approach. Our findings are based solely on asthmatic children and may not apply to children without asthma. When exposure assignment is based on ambient central monitors, misclassification is always a concern. Most pollutants, however, do not vary a great deal within geographic areas of this size, and therefore, central monitors are likely to be a good representation of outdoor exposures. Children of this age do not attend schools that are far enough from their homes to warrant the use of a different monitor. It is possible that the observed significant negative health findings are confounded or due to chance. This is unlikely, given the wide range of confounders and modifiers considered, and the consistency with which these effects were identified across a range of outcomes, time periods and with extensive cross-validation and model selection strategies. The Deletion/Substitution/Addition approach automatically tests for interactions between all candidate covariates that could be problematic if biologically irrelevant covariates are identified. All of the interactions reported here, however, are biologically plausible, given an extensive literature on risk factors for asthma severity and susceptibility to environmental exposures.

These findings are based on observational data, and, as such, are subject to bias due to observed and unobserved confounding. Consequently, the effects observed cannot necessarily be interpreted as causal. However, given the broad range and parameterizations of confounders and modifiers that were considered in the model selection, it is likely that most key confounders have been addressed.

These findings are important for several reasons. There is an expanding literature regarding the harmful effects of prenatal exposures on children's health. Our data provide support for the concept that effects of prenatal exposures can be seen among school-aged children. As has been shown in the literature regarding passive smoke exposure, it is difficult to disentangle the effect of prenatal and postnatal exposures. Our models suggest that each time period of exposure may contribute independently to different dimensions of school-age children's pulmonary function. For 4 of the 8 pulmonary-function measures, prenatal exposures were more influential than early-lifetime metrics, while, in contrast, the ratio measures (FEV1/FVC and FEF25–75/FVC) were most influenced by postnatal exposures. This may reflect the fact that airway development and final alveolarization occur at different stages and, therefore, exposure during these phases affects different dimensions of lung health. The effect sizes for prenatal exposures are comparable to those for tobacco exposure during pregnancy. The large number of pregnant women and young children exposed to these pollutants, however, may increase the magnitude of the resulting public health burden. If these effects remain in adulthood, children exposed in early-life may be at greater risk for serious and debilitating respiratory disease.

We identified clinically meaningful health effects of prenatal and early life pollutant exposures. The Deletion/Substitution/Addition procedure identified interactions, nonlinear effects, and copollutant effects that were not as readily identified with a more traditional analytic approach. These findings suggest that prenatal and early-life exposures to NO2, CO, and PM10 adversely affect the pulmonary function of asthmatic children through preadolescence, and that the timing of the exposure may be more important than overall dose. Most effects were limited to subgroups of the population with characteristics typically associated with asthma severity. Evaluation of population subgroups at greater risk for adverse respiratory responses to air pollution is important for the estimation of the public health burden imposed by air pollution. Identification of susceptible groups may provide useful scientific information about the mechanisms for air pollution-related health effects.

ACKNOWLEDGMENTS

We recognize the important work of Lucas Carlton for geocoding the addresses. We also acknowledge the contributions of the entire FACES staff.

REFERENCES

1.Delfino R, Quintana P, Floro J, et al. Association of FEV1 in asthmatic children with personal and microenvironmental exposure to airborne particulate matter. Environ Health Perspect. 2004;112:932–941.
2.Delfino RJ, Zeiger RS, Seltzer JM, et al. Association of asthma symptoms with peak particulate air pollution and effect modification by anti-inflammatory medication use. Environ Health Perspect. 2002;110:A607–A617.
3.Mortimer KM, Neas LM, Dockery DW, et al. The effect of summer ozone on inner city children with asthma. Eur Respir J. 2002;19:699–705.
4.Lacasana M, Esplugues A, Ballester F. Exposure to ambient air pollution and prenatal and early childhood health effects. Eur J Epidemiol. 2005;20:183–199.
5.Tager I, Balmes J, Lurmann F, et al. Chronic exposure to ambient ozone and lung function in young adults. Epidemiology. 2005;16:751–759.
6.Mortimer KM, Tager IB, Dockery DW, et al. The effect of ozone on inner-city children with asthma: identification of susceptible subgroups. Am J Respir Crit Care Med. 2000;162:1838–1845.
7.Ross M, Persky V, Scheff P, et al. Effect of ozone and aeroallergens on the respiratory health of asthmatics. Arch Environ Health. 2002;57:568–578.
8.Jalaludin B, Chey T, O'Toole B, et al. Acute effects of low levels of ambient ozone on peak expiratory flow rate in a cohort of Australian children. Int J Epidemiol. 2000;29:549–557.
9.Ritz B, Yu F. The effect of ambient carbon monoxide on low birth weight among children born in Southern California between 1989 and 1993. Environ Health Perspect. 1999;107:17–25.
10.Ritz B, Yu F, Chapa G, et al. Effect of air pollution on preterm birth among children born in Southern California between 1989 and 1993. Epidemiology. 2000;11:502–511.
11.Oliveira Z, Bettiol H, Gutierrez M, et al. Factors associated with infant and adolescent mortality. Braz J Med Biol Res. 2007;40:1245–1255.
12.Bell M, Ebisu K, Belanger K. Ambient air pollution and low birth weight in Connecticut and Massachusetts. Environ Health Perspect. 2007;115:1118–1124.
13.Shaheen S, Sterne J, Montgomery S, et al. Birth weight, body mass index and asthma in young adults. Thorax. 1999;54:396–402.
14.Seidman D, Laor A, Gale R, et al. Asthma in adult life—is there an association with birth weight? Arch Dis Child. 1991;66:584–587.
15.Sorlie P, Kannel W, O'Connor G. Mortality associated with respiratory function and symptoms in advanced age. The Framingham Study. Am Rev Respir Dis. 1989;140:379–384.
16.Health NIO. NHLBI/WHO Workshop Report: Global Strategy for Asthma Management and Prevention. Bethesda, MD: U.S. Department of Health and Human Services National Heart, Lung, and Blood Institute; 2002. Publication No. 02-5075.
17.Mortimer KM, Fallot A, Balmes JR, et al. Evaluating the use of a portable spirometer a study of pediatric asthma. Chest. 2003;123:1899–1907.
18.Tager I, Weiss S, Munoz A, et al. Determinants of response to eucapneic hyperventilation with cold air in a population-based study. Am Rev Respir Dis. 1986;134:502–508.
19.Mead J. Dysanapsis in normal lungs assessed by the relationship between maximal flow, static recoil, and vital capacity. Am Rev Respir Dis. 1980;121:339–342.
20.Sinisi S, van der Laan M. The deletion/substitution/addition algorithm in loss function based estimation: applications in genomics. J Stat Methods Mol Biol. 2004;3.
21.R Project for Statistical Computing. Available at: http://www.r-project.org/.
22.Hanrahan J, Tager I, Segal M, et al. The effect of maternal smoking during pregnancy on early infant lung function. Am Rev Respir Dis. 1992;145:1229–1235.
23.Ward C, Lewis S, Coleman T. Prevalence of maternal smoking and environmental tobacco smoke exposure during pregnancy and impact on birth weight: retrospective study using Millennium Cohort. BMC Public Health. 2007;7:81.
24.Tepper R, Williams-Nkomo T, Martinez T, et al. Parental smoking and airway reactivity in healthy infants. Am J Respir Crit Care Med. 2005;171:78–82.
25.Gilliland F, Berhane K, McConnell R, et al. Maternal smoking during pregnancy, environmental tobacco smoke exposure and childhood lung function. Thorax. 2000;55:271–276.
26.London S, Gauderman J, Avol E, et al. Family history and the risk of early-onset persistent, early-onset transient, and late-onset asthma. Epidemiology. 2001;12:577–583.
27.Moshammer H, Hoek G, Luttmann-Gibson H, et al. Parental smoking and lung function in children: an international study. Am J Respir Crit Care Med. 2006;173:1255–1263.
28.Tager I, Weiss S, Munoz A, et al. Longitudinal study of the effects of maternal smoking on pulmonary function in children. N Engl J Med. 1983;309:699–703.
29.Mannino D, Moorman J, Kingsley B, et al. Health effects related to environmental tobacco smoke exposure in children in the United States: data from the Third National Health and Nutrition Examination Survey. Arch Pediatr Adolesc Med. 2001;155:36–41.

Supplemental Digital Content

© 2008 Lippincott Williams & Wilkins, Inc.