Implementing machine learning methods with complex survey data: Lessons learned on the impacts of accounting sampling weights in gradient boosting.

Despite the prominent use of complex survey data and the growing popularity of machine learning methods in epidemiologic research, few machine learning software implementations offer options for handling complex samples. A major challenge impeding the broader incorporation of machine learning into epidemiologic research is incomplete guidance for analyzing complex survey data, including the importance of sampling weights for valid prediction in target populations. Using data from 15, 820 participants in the 1988-1994 National Health and Nutrition Examination Survey cohort, we determined whether ignoring weights in gradient boosting models of all-cause mortality affected prediction, as measured by the F1 score and corresponding 95% confidence intervals. In simulations, we additionally assessed the impact of sample size, weight variability, predictor strength, and model dimensionality. In the National Health and Nutrition Examination Survey data, unweighted model performance was inflated compared to the weighted model (F1 score 81.9% [95% confidence interval: 81.2%, 82.7%] vs 77.4% [95% confidence interval: 76.1%, 78.6%]). However, the error was mitigated if the F1 score was subsequently recalculated with observed outcomes from the weighted dataset (F1: 77.0%; 95% confidence interval: 75.7%, 78.4%). In simulations, this finding held in the largest sample size (N = 10,000) under all analytic conditions assessed. For sample sizes <5,000, sampling weights had little impact in simulations that more closely resembled a simple random sample (low weight variability) or in models with strong predictors, but findings were inconsistent under other analytic scenarios. Failing to account for sampling weights in gradient boosting models may limit generalizability for data from complex surveys, dependent on sample size and other analytic properties. In the absence of software for configuring weighted algorithms, post-hoc re-calculations of unweighted model performance using weighted observed outcomes may more accurately reflect model prediction in target populations than ignoring weights entirely.

Endotoxin clustering with allergens in house dust and asthma outcomes in a U.S. national study.

BACKGROUND: Endotoxin is ubiquitous in the environment, but its clustering with indoor allergens is not well characterized. This study examined the clustering patterns of endotoxin with allergens in house dust and their association with asthma outcomes. METHODS: We analyzed data from 6963 participants of the 2005-2006 National Health and Nutrition Examination Survey. House dust sampled from bedroom floor and bedding was evaluated for endotoxin and allergens from fungi, cockroach, dog, cat, mites, and rodents. Two-step cluster analysis and logistic regressions were performed to identify the clustering patterns and their associations with current asthma and wheeze in the past 12 months, adjusting for covariates. RESULTS: Of the homes, 17.8% had low endotoxin and allergen levels in house dust (Cluster 1). High endotoxin level clustered with Alternaria and pet allergens in the homes of participants with a high socioeconomic status who own pets (Cluster 2) (48.9%). High endotoxin clustered with Aspergillus, dust mites, cockroach, and rodent allergens in the homes of participants with low socioeconomic status (Cluster 3) (33.3%). Compared to Cluster 1, Cluster 2 was associated with higher asthma prevalence (OR 1.42, 95% CI: 1.06-1.91) and wheeze (OR 1.32, 95% CI: 1.07-1.63). Cluster 3 was positively associated with wheeze only in participants sensitized to inhalant allergens (OR 1.42, 95% CI: 1.06-1.91) or exposed to tobacco smoke (OR 1.72, 95% CI: 1.15-2.60). CONCLUSIONS: The clustering of endotoxin with allergens in dust from homes with pets or of people with low socioeconomic status is associated with asthma and wheeze.

Association of urinary levels of bisphenols F and S used as bisphenol A substitutes with asthma and hay fever outcomes.

BACKGROUND: Bisphenols F (BPF) and S (BPS) are bisphenol A (BPA) analogs used as substitutes in consumer products. Despite previous reports of BPA's association with asthma, no studies have examined its structural analogs in relation to asthma and allergy outcomes. OBJECTIVE: To examine the association of urinary BPF, BPS, and BPA with asthma and hay fever in a US representative sample. METHODS: We analyzed data from 3,538 participants aged 12 years or older in the 2013-2016 National Health and Nutrition Examination Survey (NHANES). Children aged 6-11 years (N = 738), who did not have all covariate data available, were analyzed separately. Covariate-adjusted logistic regression was used to assess the association of the exposures with the outcomes. RESULTS: BPF, BPS, and BPA were detected in 57.1%, 88.4%, and 94.8% of the urine samples, respectively. Urinary BPF detection was positively associated with current asthma (odds ratio [OR]: 1.54, 95% confidence interval [CI]: 1.16-2.04) and hay fever (OR: 1.66, 95% CI: 1.12-2.46). Urinary BPS was associated with increased odds of current asthma in men (OR: 1.64, 95% CI: 1.13-2.40) and urinary BPA was associated with increased odds of asthma without hay fever in children aged 6-11 years (OR: 2.65, 95% CI: 1.05-6.68). CONCLUSION: Our nationally-representative findings document that BPF and BPS exposure is common in the US and that exposure to these BPA analogs is associated with asthma and/or hay fever. Our results suggest that BPF and BPS may not be safe alternatives to BPA; however, prospective studies should be conducted to confirm these results.

Synergistic Association of House Endotoxin Exposure and Ambient Air Pollution with Asthma Outcomes.

Rationale: House endotoxin and ambient air pollution are risk factors for asthma; however, the effects of their coexposure on asthma are not well characterized.Objectives: To examine potential synergistic associations of coexposure to house dust endotoxin and ambient air pollutants with asthma outcomes.Methods: We analyzed data of 6,488 participants in the National Health and Nutrition Examination Survey 2005-2006. Dust from bedding and bedroom floor was analyzed for endotoxin content. The Community Multiscale Air Quality Modeling System (CMAQ) and Downscaler Model data were used to determine annual average particulate matter ≤2.5 µm in aerodynamic diameter (PM2.5), ozone (O3), and nitrogen dioxide (NO2) exposures at participants' residential locations. The associations of the coexposures with asthma outcomes were assessed and tested for synergistic interaction.Measurements and Main Results: In adjusted analysis, PM2.5 (CMAQ) (odds ratio [OR], 1.12; 95% confidence interval [CI], 1.07-1.18), O3 (Downscaler Model) (OR, 1.07; 95% CI, 1.02-1.13), and log10 NO2 (CMAQ) (OR, 3.15; 95% CI, 1.33-7.45) were positively associated with emergency room visits for asthma in the past 12 months. Coexposure to elevated concentrations of house dust endotoxin and PM2.5 (CMAQ) was synergistically associated with the outcome, increasing the odds by fivefold (OR, 5.01; 95% CI, 2.54-9.87). A synergistic association was also found for coexposure to higher concentrations of endotoxin and NO2 in children (OR, 3.45; 95% CI, 1.65-7.18).Conclusions: Coexposure to elevated concentrations of residential endotoxin and ambient PM2.5 in all participants and NO2 in children is synergistically associated with increased emergency room visits for asthma. Therefore, decreasing exposure to both endotoxin and air pollution may help reduce asthma morbidity.

Endotoxin predictors and associated respiratory outcomes differ with climate regions in the U.S.

RATIONALE: Although endotoxin is a recognized cause of environmental lung disease, how its relationship with respiratory outcomes varies with climate is unknown. OBJECTIVE: To examine the endotoxin predictors as well as endotoxin association with asthma, wheeze, and sensitization to inhalant allergens in various US climate regions. METHODS: We analyzed data on 6963 participants in the National Health and Nutrition Examination Survey. Endotoxin measurements of house dust from bedroom floor and bedding were performed at the University of Iowa. Linear and logistic regression analyses were used to identify endotoxin predictors and assess endotoxin association with health outcomes. RESULTS: The overall median house dust endotoxin was 16.2 EU/mg; it was higher in mixed-dry/hot-dry regions (19.7 EU/mg) and lower in mixed-humid/marine areas (14.8 EU/mg). Endotoxin predictors and endotoxin association with health outcomes significantly differed across climate regions. In subarctic/very cold/cold regions, log10-endotoxin was significantly associated with higher prevalence of wheeze outcomes (OR:1.48, 95% CI:1.19-1.85 for any wheeze, OR:1.48, 95% CI:1.22-1.80 for exercise-induced wheeze, OR:1.50, 95% CI:1.13-1.98 for prescription medication for wheeze, and OR:1.95, 95% CI:1.50-2.54 for doctor/ER visit for wheeze). In hot-humid regions, log10-endotoxin was positively associated with any wheeze (OR:1.66, 95% CI:1.04-2.65) and current asthma (OR:1.56, 95% CI:1.11-2.18), but negatively with sensitization to any inhalant allergens (OR:0.83, 95% CI:0.74-0.92). CONCLUSION: Endotoxin predictors and endotoxin association with asthma and wheeze differ across U.S. climate regions. Endotoxin is associated positively with wheeze or asthma in cold and hot-humid regions, but negatively with sensitization to inhalant allergens in hot-humid climates.

Sensitization and Exposure to Pets: The Effect on Asthma Morbidity in the US Population.

BACKGROUND: Although pets are found in more than 50% of US homes, the effect of pet allergen exposure on asthma morbidity in the US population is not well documented. OBJECTIVE: To determine the effect of dog and cat allergen exposures on asthma morbidity in the US population. METHODS: The National Health and Nutrition Examination Survey is a representative sample of civilian US population. Data on asthma, dog and cat allergen levels in bedroom dust, as well as specific IgE to dog and cat were analyzed for all participants 6 years or older. RESULTS: Pets are common in the United States, with more that 50% of households having a dog or a cat or both. The prevalence of allergic sensitization in the National Health and Nutrition Examination Survey population was similar for dog and cat, with both being approximately 12%. Among those who were sensitized, exposure to elevated levels of pet allergens was associated with an increased prevalence of asthma and asthma attacks. Indeed, 44.2% of the asthma attacks were attributable to exposure to high levels of dog allergen in the bedroom among patients with asthma sensitive to dog and 30.3% were attributable to cat allergen exposure among the comparable cat-sensitive and exposed group. Projecting these results to the US population indicates more than 1 million increased asthma attacks each year for the dog-sensitive and exposed group and more than 500,000 increased asthma attacks for the cat-sensitive and exposed population of patients with asthma. CONCLUSIONS: Exposure to elevated levels of dog and cat allergens among those sensitized individuals with asthma is associated with excess asthma attacks. Reducing pet allergen exposures has the potential for a significant decrease in asthma morbidity.