American Journal of Epidemiology

In this issue of the Journal, Snowden et al. (Am J Epidemiol. 2011;173(7):731–738) give a didactic explanation of G-computation as an approach for estimating the causal effect of a point exposure. The authors of the present commentary reinforce the idea that their use of G-computation is equivalent to a particular form of model-based standardization, whereby reference is made to the observed study population, a technique that epidemiologists have been applying for several decades. They comment on the use of standardized versus conditional effect measures and on the relative predominance of the inverse probability-of-treatment weighting approach as opposed to G-computation. They further propose a compromise approach, doubly robust standardization, that combines the benefits of both of these causal inference techniques and is not more difficult to implement.

The growing body of work in the epidemiology literature focused on G-computation includes theoretical explanations of the method but very few simulations or examples of application. The small number of G-computation analyses in the epidemiology literature relative to other causal inference approaches may be partially due to a lack of didactic explanations of the method targeted toward an epidemiology audience. The authors provide a step-by-step demonstration of G-computation that is intended to familiarize the reader with this procedure. The authors simulate a data set and then demonstrate both G-computation and traditional regression to draw connections and illustrate contrasts between their implementation and interpretation relative to the truth of the simulation protocol. A marginal structural model is used for effect estimation in the G-computation example. The authors conclude by answering a series of questions to emphasize the key characteristics of causal inference techniques and the G-computation procedure in particular.

Case-control studies have shown that regular use of nonsteroidal antiinflammatory drugs (NSAIDs) decreases bladder cancer risk, but few cohort studies have evaluated this association. The authors investigated NSAID use and bladder cancer in 3 large prospective studies (NIH-AARP Diet and Health Study; Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial; and U.S. Radiologic Technologists Study). Frequency of aspirin and nonaspirin NSAID use 1 year prior to baseline was ascertained using self-administered questionnaires. Study-specific hazard ratios and 95% confidence intervals were estimated using Cox regression models and were combined using a fixed-effects meta-analytic model. Data from all studies were aggregated, and aggregated hazard ratios were estimated. The analysis included 508,842 individuals, with 2,489 incident cases of bladder cancer. A reduction in risk was observed for individuals who reported regular use (>2 times/week) of nonaspirin NSAIDs compared with those who reported no use (hazard ratio (HR) = 0.92, 95% confidence interval (CI): 0.81, 1.04). The risk reduction was limited to nonsmokers (HR = 0.58, 95% CI: 0.41, 0.83) (Ptrend = 0.008) (Pinteraction = 0.02). No association was observed between regular aspirin use and bladder cancer risk (HR = 1.04, 95% CI: 0.94, 1.15). Results suggest that nonaspirin NSAIDs, but not aspirin, are associated with a reduction in risk of bladder cancer, particularly for nonsmokers.

To meet challenges arising from increasing rates of noncoverage in US landline-based telephone samples due to cell-phone-only households, the Behavioral Risk Factor Surveillance System (BRFSS) expanded a traditional landline-based random digit dialing survey to a dual-frame survey of landline and cell phone numbers. In 2008, a survey of adults with cell phones only was conducted in parallel with an ongoing landline-based health survey in 18 states. The authors used the optimal approach to allocate samples into landline and cell-phone-only strata and used a new approach to weighting state-level landline and cell phone samples. They developed logistic models for each of 16 health indicators to examine whether exclusion of adults with cell phones only affected estimates after adjustment for demographic characteristics. The extents of the potential biases in landline telephone surveys that exclude cell phones were estimated. Biases resulting from exclusion of adults with cell phones only from the landline-based survey were found for 9 out of the 16 health indicators. Because landline noncoverage rates for adults with cell phones only continue to increase, these biases are likely to increase. Use of a dual-frame survey of landline and cell phone numbers assisted the BRFSS efforts in obtaining valid, reliable, and representative data.

Exposure lagging and exposure-time window analysis are 2 widely used approaches to allow for induction and latency periods in analyses of exposure-disease associations. Exposure lagging implies a strong parametric assumption about the temporal evolution of the exposure-disease association. An exposure-time window analysis allows for a more flexible description of temporal variation in exposure effects but may result in unstable risk estimates that are sensitive to how windows are defined. The authors describe a hierarchical regression approach that combines time window analysis with a parametric latency model. They illustrate this approach using data from 2 occupational cohort studies: studies of lung cancer mortality among 1) asbestos textile workers and 2) uranium miners. For each cohort, an exposure-time window analysis was compared with a hierarchical regression analysis with shrinkage toward a simpler, second-stage parametric latency model. In each cohort analysis, there is substantial stability gained in time window-specific estimates of association by using a hierarchical regression approach. The proposed hierarchical regression model couples a time window analysis with a parametric latency model; this approach provides a way to stabilize risk estimates derived from a time window analysis and a way to reduce bias arising from misspecification of a parametric latency model.

Repeat-biomarker measurement error models accounting for systematic correlated within-person error can be used to estimate the correlation coefficient () and deattenuation factor (), used in measurement error correction. These models account for correlated errors in the food frequency questionnaire (FFQ) and the 24-hour diet recall and random within-person variation in the biomarkers. Failure to account for within-person variation in biomarkers can exaggerate correlated errors between FFQs and 24-hour diet recalls. For 2 validation studies, and were calculated for total energy and protein density. In the Automated Multiple-Pass Method Validation Study (n = 471), doubly labeled water (DLW) and urinary nitrogen (UN) were measured twice in 52 adults approximately 16 months apart (2002–2003), yielding intraclass correlation coefficients of 0.43 for energy (DLW) and 0.54 for protein density (UN/DLW). The deattenuated correlation coefficient for protein density was 0.51 for correlation between the FFQ and the 24-hour diet recall and 0.49 for correlation between the FFQ and the biomarker. Use of repeat-biomarker measurement error models resulted in a of 0.42. These models were similarly applied to the Observing Protein and Energy Nutrition Study (1999–2000). In conclusion, within-person variation in biomarkers can be substantial, and to adequately assess the impact of correlated subject-specific error, this variation should be assessed in validation studies of FFQs.

With advances in the effectiveness of treatment and disease management, the contribution of chronic comorbid diseases (comorbidities) found within the Charlson comorbidity index to mortality is likely to have changed since development of the index in 1984. The authors reevaluated the Charlson index and reassigned weights to each condition by identifying and following patients to observe mortality within 1 year after hospital discharge. They applied the updated index and weights to hospital discharge data from 6 countries and tested for their ability to predict in-hospital mortality. Compared with the original Charlson weights, weights generated from the Calgary, Alberta, Canada, data (2004) were 0 for 5 comorbidities, decreased for 3 comorbidities, increased for 4 comorbidities, and did not change for 5 comorbidities. The C statistics for discriminating in-hospital mortality between the new score generated from the 12 comorbidities and the Charlson score were 0.825 (new) and 0.808 (old), respectively, in Australian data (2008), 0.828 and 0.825 in Canadian data (2008), 0.878 and 0.882 in French data (2004), 0.727 and 0.723 in Japanese data (2008), 0.831 and 0.836 in New Zealand data (2008), and 0.869 and 0.876 in Swiss data (2008). The updated index of 12 comorbidities showed good-to-excellent discrimination in predicting in-hospital mortality in data from 6 countries and may be more appropriate for use with more recent administrative data.

Chronic diseases are now among the leading causes of morbidity and mortality in lower income countries. Although traditionally related to higher individual socioeconomic status (SES) in these contexts, the associations between SES and chronic disease may be actively changing. Furthermore, country-level contextual factors, such as economic development and income inequality, may influence the distribution of chronic disease by SES as well as how this distribution has changed over time. Using overweight status as a health indicator, the authors studied repeated cross-sectional data from women aged 18–49 years in 37 developing countries to assess within-country trends in overweight inequalities by SES between 1989 and 2007 (n = 405,550). Meta-regression was used to examine the associations between gross domestic product and disproportionate increases in overweight prevalence by SES, with additional testing for modification by country-level income inequality. In 27 of 37 countries, higher SES (vs. lower) was associated with higher gains in overweight prevalence; in the remaining 10 countries, lower SES (vs. higher) was associated with higher gains in overweight prevalence. Gross domestic product was positively related to faster increase in overweight prevalence among the lower wealth groups. Among countries with a higher gross domestic product, lower income inequality was associated with faster overweight growth among the poor.

Sulfur mustard, an agent used in chemical warfare, is an alkylating substance with carcinogenic potential. However, the precise long-term carcinogenic effects of mustard gas are unclear. Since 1952, the authors have conducted health surveys of former workers who were employed from 1929 to 1945 in a poisonous gas factory in Okuno-jima, Hiroshima, Japan. This prospective study was undertaken from 1952 to 2005 to examine the incidence of lung cancer among the workers who were exposed to mustard gas (n= 480), lewisite (n= 55), and/or diphenylcyanarsine (n = 178), as well as the incidence among unexposed workers (n = 969). The stochastic relation between exposure and lung cancer was explored on the basis of multistage carcinogenesis by using an accelerated hazard model with a transformed age scale. Mustard gas exposure was found to transform the age scale for developing lung cancer. One year of exposure in subjects ≤18 or >18 years old at first exposure shifted the age scale down by 4.9 years and 3.3 years, respectively. On the basis of the long-term follow-up of former workers in the poisonous gas factory, the authors concluded that sulfur mustard decreased the age at which people were at risk of developing lung cancer and that the effect declined with aging.

Higher egg and cholesterol intakes are associated with increased risk of type 2 diabetes mellitus. However, their association with gestational diabetes mellitus (GDM) has not been evaluated. The authors assessed such associations in both a prospective cohort study (1996&ndash;2008; 3,158 participants) and a case-control study (1998&ndash;2002; 185 cases, 411 controls). A food frequency questionnaire was used to assess maternal diet. Multivariable models were used to derive relative risks and 95% confidence intervals. Compared with no egg consumption, adjusted relative risks for GDM were 0.94, 1.01, 1.12, 1.54, and 2.52 for consumption of &le;1, 2&ndash;3, 4&ndash;6, 7&ndash;9, and &ge;10 eggs/week, respectively (P for trend = 0.008). Women with high egg consumption (&ge;7/week) had a 1.77-fold increased risk compared with women with lower consumption (95% confidence interval (CI): 1.19, 2.63). The relative risk for the highest quartile of cholesterol intake (&ge;294 mg/day) versus the lowest (<151 mg/day) was 2.35 (95% CI: 1.35, 4.09). In the case-control study, the adjusted odds ratio for consuming &ge;7 eggs/week versus <7 eggs/week was 2.65 (95% CI: 1.48, 4.72), and the odds of GDM increased with increasing cholesterol intake (P for trend = 0.021). In conclusion, high egg and cholesterol intakes before and during pregnancy are associated with increased risk of GDM.

Maternal angiogenic factors (placental growth factor, soluble fms-like tyrosine kinase 1 (Flt-1), and soluble endoglin) may be associated with fetal growth restriction, and the associations may differ according to stage of pregnancy. Among children born to pregnant women without preeclampsia in Norway between 1992 and 1994, 217 singletons with severe growth restriction (small for gestational age (SGA), <2.5th percentile) were compared with 378 singleton controls. For each angiogenic factor, SGA risk was related to concentrations in maternal serum collected in the first 2 trimesters, by using women with a serum concentration in the middle third at both samplings as reference. A low placental growth factor (lowest third) at both samplings was associated with high risk of SGA (odds ratio = 3.8, 95% confidence interval: 1.6, 8.8). An increase from the lowest to the highest third of soluble Flt-1 was associated with high SGA risk (odds ratio = 6.2, 95% confidence interval: 2.4, 16.1). Women with high soluble endoglin (highest third) at the second sampling had approximately a 3.5-fold increased risk of SGA. Low maternal soluble Flt-1 in early pregnancy followed by a strong subsequent increase in soluble Flt-1 and soluble endoglin was associated with a particularly high risk of severe fetal growth restriction.

The authors assessed the relation between infant weight velocity and adult insulin resistance, specifically evaluating whether adult size and body fat distribution mediated the association. Data were from the Cebu Longitudinal Health and Nutrition Survey (Cebu, the Philippines), in which a birth cohort was followed to age 22 years (n = 1,409; 1983&ndash;2005). Insulin resistance was measured using homeostasis model assessment of insulin resistance (HOMA-IR). Weight velocity (g/month) from 0 to 4 months and from 0 to 24 months was assessed. The authors examined direct and total associations between early growth and adult HOMA-IR in linear regression models and used a nonparametric bootstrapping procedure to test indirect effects through adult body mass index (BMI; weight (kg)/height (m)2) and waist circumference. Infant weight velocity was positively associated with adult BMI and waist circumference, which positively predicted HOMA-IR. There were no total or direct effects of immediate postnatal weight velocity (0&ndash;4 months) on adult HOMA-IR, although indirect effects through BMI and waist circumference were significant. Weight velocity from 0 to 24 months positively predicted HOMA-IR among males only, while indirect effects were significant in both sexes. In a relatively lean sample of young adults from a population with rising rates of diabetes and cardiovascular disease, the authors found evidence for small indirect effects of infant weight velocity on adult insulin resistance mediated through adult BMI and waist circumference.

Short sleep duration is associated with incidence of overweight and obesity in preadolescent children. The authors performed regression analyses on data from the Quebec Longitudinal Study of Kindergarten Children (1986&ndash;1987), a prospective cohort study comprising 1,916 preadolescent children in Canada. The aim was to assess associations between time spent in bed and body mass index reported by mothers after adjusting for numerous confounding factors, such as pubertal status. Time-in-bed and body mass index trajectories were computed using a semiparametric model mixture. Time-in-bed trajectories were classified as short (15% of the preadolescents), 10.5-hour (68%), and 11-hour (17%) sleep-duration trajectories, decreasing over time. Body mass index trajectories were classified as normal weight (68% of the preadolescents), overweight (27%), and obese (5%). The short sleep trajectory was associated with an increased odds ratio of being in the overweight body mass index trajectory (odds ratio (OR) = 1.55, 95% confidence interval (CI): 1.39, 1.71) or in the obese body mass index trajectory (OR = 3.26, 95% CI: 3.20, 3.29) compared with the 11-hour trajectory. One hour less of sleep per night at 10 years of age was associated with an increased odds ratio of being overweight (OR = 1.51, 95% CI: 1.28, 1.76) or obese (OR = 2.07; 95% CI: 1.51, 2.84) at 13 years of age.

Susceptibility to the respiratory effects of air pollution varies between individuals. Although some evidence suggests higher susceptibility for subjects carrying variants of antioxidant genes, findings from gene-pollution interaction studies conflict in terms of the presence and direction of interactions. The authors conducted a systematic review on antioxidant gene-pollution interactions which included 15 studies, with 12 supporting the presence of interactions. For the glutathione S-transferase M1 gene (GSTM1) (n = 10 studies), only 1 study found interaction with the null genotype alone, although 5 observed interactions when GSTM1 was evaluated jointly with other genes (mainly NAD(P)H dehydrogenase [quinone] 1 (NQO1)). All studies on the glutathione S-transferase P1 (GSTP1) Ile105Val polymorphism (n = 11) provided some evidence of interaction, but findings conflicted in terms of risk allele. Results were negative for glutathione S-transferase T1 (GSTT1) (n = 3) and positive for heme oxygenase 1 (HMOX-1) (n = 2). Meta-analysis could not be performed because there were insufficient data available for any specific gene-pollutant-outcome combination. Overall the evidence supports the presence of gene-pollution interactions, although which pollutant interacts with which gene is unclear. However, issues regarding multiple testing, selective reporting, and publication bias raise the possibility of false-positive findings. Larger studies with greater accuracy of pollution assessment and improved quality of conduct and reporting are required.

The authors investigated whether vitamin E intake was associated with amyotrophic lateral sclerosis (ALS) in the Nurses&rsquo; Health Study (1976&ndash;2004), the Health Professionals Follow-up Study (1986&ndash;2004), the Cancer Prevention Study II Nutrition Cohort (1992&ndash;2004), the Multiethnic Cohort Study (1993&ndash;2005), and the National Institutes of Health-AARP Diet and Health Study (1995&ndash;2005). ALS deaths were identified through the National Death Index. In the Nurses' Health Study and the Health Professionals Follow-up Study, confirmed nonfatal ALS cases were also included. Cohort-specific results were estimated using Cox proportional hazards models and pooled using random-effects models. Among 1,055,546 participants, 805 developed ALS. Overall, using vitamin E supplements was not associated with ALS. However, within cohorts with information on duration of vitamin E supplement use (231 cases), ALS rates declined with increasing years of use (P-trend = 0.01). Compared with nonusers, the multivariable-adjusted relative risk was 1.05 (95% confidence interval (CI): 0.60, 1.84) among users for &le;1 year (12 cases), 0.77 (95% CI: 0.33, 1.77) among users for 2&ndash;4 years (7 cases), and 0.64 (95% CI: 0.39, 1.04) among users for &ge;5 years (18 cases). For dietary vitamin E intake, the multivariable-adjusted relative risk comparing the highest quartile with the lowest was 0.79 (95% CI: 0.61, 1.03); an inverse dose-response was evident in women (P-trend = 0.002) but not in men (P-trend = 0.71). In this large, pooled prospective study, long-term vitamin E supplement use was associated with lower ALS rates. A possible protective effect of vitamin E deserves further consideration.

Confounding due to population stratification is a potential source of concern in population-based genetic association studies, particularly in recently admixed populations such as African Americans. Several methods have been developed to control for population stratification in the context of genome-wide association studies. Because these approaches require thousands of genotypes from genetic markers, they are not well suited to be used in genetic association analyses without genome-wide data. An alternative approach to control for population stratification is to estimate admixture proportions by using ancestral informative markers (AIMs). The authors evaluated whether a relatively small number of AIMs would be sufficient to estimate ancestral proportions in African Americans. They first estimated European admixture proportions in 1,757 subjects from the Black Women's Health Study (1995&ndash;2009) by genotyping an admixture panel of 1,373 AIMs; they then compared these results with those obtained using smaller sets of AIMs. The authors found that just 30 AIMs are needed to obtain very high correlation of estimates with the entire set (r = 0.89; P < 0.0001). A set of 200 AIMs gave an almost perfect correlation with the entire set (r = 0.98; P < 0.0001). These results show that a small number of AIMs are sufficiently precise to estimate European admixture in African Americans.

Few studies have examined the relation between usual physical activity level and rate of hip fracture in older men or applied semiparametric methods from the causal inference literature that estimate associations without assuming a particular parametric model. Using the Physical Activity Scale for the Elderly, the authors measured usual physical activity level at baseline (2000&ndash;2002) in 5,682 US men &ge;65 years of age who were enrolled in the Osteoporotic Fractures in Men Study. Physical activity levels were classified as low (bottom quartile of Physical Activity Scale for the Elderly score), moderate (middle quartiles), or high (top quartile). Hip fractures were confirmed by central review. Marginal associations between physical activity and hip fracture were estimated with 3 estimation methods: inverse probability-of-treatment weighting, G-computation, and doubly robust targeted maximum likelihood estimation. During 6.5 years of follow-up, 95 men (1.7%) experienced a hip fracture. The unadjusted risk of hip fracture was lower in men with a high physical activity level versus those with a low physical activity level (relative risk = 0.51, 95% confidence interval: 0.28, 0.92). In semiparametric analyses that controlled confounding, hip fracture risk was not lower with moderate (e.g., targeted maximum likelihood estimation relative risk = 0.92, 95% confidence interval: 0.62, 1.44) or high (e.g., targeted maximum likelihood estimation relative risk = 0.88, 95% confidence interval: 0.53, 2.03) physical activity relative to low. This study does not support a protective effect of usual physical activity on hip fracture in older men.

In time-to-event analyses, artificial censoring with correction for induced selection bias using inverse probability-of-censoring weights can be used to 1) examine the natural history of a disease after effective interventions are widely available, 2) correct bias due to noncompliance with fixed or dynamic treatment regimens, and 3) estimate survival in the presence of competing risks. Artificial censoring entails censoring participants when they meet a predefined study criterion, such as exposure to an intervention, failure to comply, or the occurrence of a competing outcome. Inverse probability-of-censoring weights use measured common predictors of the artificial censoring mechanism and the outcome of interest to determine what the survival experience of the artificially censored participants would be had they never been exposed to the intervention, complied with their treatment regimen, or not developed the competing outcome. Even if all common predictors are appropriately measured and taken into account, in the context of small sample size and strong selection bias, inverse probability-of-censoring weights could fail because of violations in assumptions necessary to correct selection bias. The authors used an example from the Multicenter AIDS Cohort Study, 1984&ndash;2008, regarding estimation of long-term acquired immunodeficiency syndrome-free survival to demonstrate the impact of violations in necessary assumptions. Approaches to improve correction methods are discussed.

Nutrition plays a critical role in maternal and fetal health; however, research on error in the measurement of energy intake during pregnancy is limited. The authors analyzed data on 998 women living in central North Carolina with singleton pregnancies during 2001&ndash;2005. Second-trimester diet was assessed by food frequency questionnaire. Estimated energy requirements were calculated using Institute of Medicine prediction equations, with adjustment for energy costs during the second trimester. Implausible values for daily energy intake were determined using confidence limits of agreement for energy intake/estimated energy requirements. Prevalences of low energy reporting (LER) and high energy reporting (HER) were 32.8% and 12.9%, respectively. In a multivariable analysis, pregravid body mass index was related to both LER and HER; LER was higher in both overweight (odds ratio = 1.96, 95% confidence interval: 1.26, 3.02; P = 0.031) and obese (odds ratio = 3.29, 95% confidence interval: 2.33, 4.65; P < 0.001) women than in normal-weight counterparts. Other predictors of LER included marriage and higher levels of physical activity. HER was higher among subjects who were underweight, African-American, and less educated and subjects who had higher depressive symptom scores. LER and HER are prevalent during pregnancy. Identifying their predictors may improve data collection and analytic methods for reducing systematic bias in the study of diet and reproductive outcomes.

Meta-analyses of genome-wide association studies are often based on imputed single nucleotide polymorphism (SNP) data, because component studies were genotyped using different platforms. One would like to include case-parent triad studies along with case-control studies in such meta-analyses. However, there are no published methods for estimating relative risks from imputed data for case-parent triad studies. The authors propose a method for estimating the relative risk for a variant SNP allele based on a log-additive model. Their simulations first confirm that the proposed method performs well with genotyped SNP data. As an empirical test of the method's behavior with imputed SNPs, the authors then apply it to chromosome 22 data from the Mexico City Childhood Asthma Study (1998&ndash;2003). For chromosome 22, the authors had data on 7,293 SNPs that were both genotyped and imputed using the software MACH, which relies on linkage disequilibrium with nearby SNPs. Correlation between estimated relative risks based on the actual genotypes and those based on the imputed genotypes was remarkably high (r2 = 0.95), validating this method of relative risk estimation for the case-parent study design. This method should be useful to investigators who wish to conduct meta-analyses using imputed SNP data from both case-parent triad and case-control studies.

Allergic conditions and biochemical measures are both used to characterize atopy. To assess questionnaires&rsquo; ability to predict biochemical measures of atopy, the authors used data on 5 allergic conditions (allergy, hay fever, eczema, rhinitis, and itchy rash) and serum-specific immunoglobulin E (IgE) levels from the 2005&ndash;2006 National Health and Nutrition Examination Survey. Atopy was defined as 1 or more positive specific IgEs (&ge;0.35 kU/L). Questionnaire responses were assessed for sensitivity, specificity, and positive and negative predictive values for atopy. In this population-based US sample, 44% of participants were specific IgE-positive and 53% reported at least 1 allergic condition. Discordance between atopy and allergic conditions was considerable; 37% of persons with atopy reported no allergic condition, and 48% of persons who reported an allergic condition were not atopic. Thus, no combination of self-reported allergic conditions achieved both high sensitivity and high specificity for IgE. The positive predictive value of reported allergic conditions for atopy ranged from 50% for eczema to 72% for hay fever, while the negative predictive value ranged from 57% for eczema to 65% for any condition. Given the high proportion of asymptomatic participants who were specific IgE-positive and persons who reported allergic conditions but were specific IgE-negative, it is unlikely that questionnaires will ever capture the same participants as those found to be atopic by biochemical measures.