Causal Inference
Correcting Exposure Misclassification via Control Variates: Effects of Opportunistic Infections on Lung Function in Children with Cystic Fibrosis Kevin Josey* Kevin Josey Josey Josey Josey Josey University of Colorado
Respiratory cultures from throat swabs or induced sputum guide treatment decisions for lung infections in cystic fibrosis (CF) patients, yet throat swabs exhibit imperfect sensitivity and specificity for detecting Pseudomonas aeruginosa and Staphylococcus aureus. We quantify the bias induced by this misclassification and develop a control variate-adjusted calibration weighting estimator to address multiple misclassified exposures. Our data comprise three disjoint subsamples: patients with only gold-standard sputum cultures; patients with only error-prone throat swabs; and a small validation sample with paired measurements. Using calibration weighting to treat misclassification as a missing data problem, we estimate causal effects in each subsample and demonstrate substantial attenuation bias in swab-based estimates. In our cohort of 550 CF patients ages 6-21, swab-based estimates fail to detect the significant negative effect of P. aeruginosa on lung function (-5.64% FEV, 95% CI: -9.75, -1.53) observed in sputum samples. The validation sample, consisting of 22 participants with paired measurements, reveals 73.7% sensitivity and 95.5% specificity for P. aeruginosa versus 95.8% sensitivity and 64.7% specificity for S. aureus. We show how the validation sample enables a control variate adjustment that integrates information from both estimators, reducing standard errors by 2-5% while preserving consistency. These findings suggest throat swabs may lead to under-treatment of P. aeruginosa infections and provide a methodological framework for causal inference with multiple misclassified exposures.
