REPLY: We thank Mindel et al 1 for their interest in our recent study regarding amiodarone-associated optic neuropathy. Multivariable regression analysis and propensity score analysis are 2 of the most widely used methods for adjusting observed confounders. 2 As the application of propensity score analysis becomes increasingly popular, its limitations are also widely discussed. 3,4 In many cases, the results from multivariable regression and propensity score analysis are similar. 5 Moreover, both statistical methods have their own limitations and none of them solve the problem of unknown confounders. 2 The instrumental variable approach may be a better option, because it may help to adjust for unobserved confounders. 2 However, it is difficult to find an instrument closely related to the use of amiodarone, but not related to optic neuropathy. Therefore, we did not adopt this approach. We acknowledged this limitation of unobservable confounding in our article and chose Cox multivariable regression analysis as the analytic method for the current study.The second concern raised by Mindel et al may be a misunderstanding. The variables included in the main Cox regression model were age, sex, amiodarone use, and variables with a P value of <0.05 in the univariate analysis. A similar approach was adopted for our stratification analyses. In addition, we conducted the sensitivity analyses (Table 1; also Table 3 in the original article, available at www.aaojournal.org) by including all variables (age, sex, 7 comorbidities, and amiodarone use). The results remained robust. Obstructive sleep apnea was not included in the analysis because none of these patients with obstructive sleep apnea had the occurrence of optic neuropathy during the study period.In our study, 7764 patients in the control group, which accounted for 31.4% of all control subjects, had >1 comorbidity, compared with 3412 patients (55.3%) in amiodarone group. Of the 7764 control subjects with >1 comorbidity, 14 subjects (0.2%) had optic neuropathy, compared with 14 patients (0.4%) in the amiodarone group. After reconstructing the model by including age, sex, taking amiodarone or not, and having >1 comorbidity or not, patients with >1 comorbidity were significantly more likely to incur optic neuropathy (odds ratio [OR], 2.17; 95% confidence interval [CI], 1.17e4.03; P ¼ 0.01). The impact of amiodarone use on the optic neuropathy also remained robust in this model (OR, 2.20; 95% CI, 1.19e4.05; P ¼ 0.01).The final concern from Mindel et al is whether these patients had unilateral or bilateral ocular involvement. In this study, we retrieved our data from the Taiwan's National Health Insurance Research Database, which is secondary information and contains data for insurance, such as diagnostic code, examination code, medication, and date of visit. We used inclusion criteria with visual field and fundus examination, which were recognized from the examination codes in the database, to make our diagnosis more precise. However, this database did not provide access t...