Statistical analysis

All questionnaire data were coded, checked in batches by two assessors, and entered into the study database in Research Electronic Data Capture (REDCap) software. The cleaned dataset was imported into SAS (version 9.4, SAS Institute, Inc. Cary, NC) or R (version 3.6.1) for subsequent analyses. To reiterate, in the present study, the T1 data and T2 data were analyzed separately as cross-sectional data for all regression analyses. Thus, our emphasis is on the findings at each time period: immediately after the disaster and approximately 1 year after. For those who participated at both time points, their answers at T1 were used for T1 regression analysis, and answers at T2 were used for the T2 regression analysis. First, descriptive statistics were computed for relevant categorical variables (frequency distributions) and continuous variables (e.g., mean, standard deviation, median, and quartiles). At each time period, unconditional logistic regression was used to regress each allergic symptom (yes/no) on each Harvey exposure (yes/no) while adjusting for age, sex, race/ethnicity, and education level. Results of the regression models were used to calculate corresponding odds ratios, 95% confidence intervals, and p-values. Linear regression was used to regress stress level (scored from 0 to 10) on Harvey exposures (yes/no), adjusting for age, sex, race/ethnicity, and education level. All regression analyses were performed separately for T1 data and T2 data. Considering that multiple testing may increase type-1 error rate, we corrected for multiple comparisons using the false discovery rate adjusted (FDR-adjusted) p-values; i.e., q-values. For the exploratory analysis of neighborhood-level socioeconomic disadvantage, first, at each time period (T1 and T2, separately), we tested for differences in relevant variables between low and high ADI groups using chi-square, T, or Wilcoxon rank-sum tests as appropriate. We then performed the earlier described unconditional logistic and linear regression analyses on the ADI-stratified data set. However, to avoid issues with model convergence, we did not adjust for the race/ethnicity and education in the ADI-stratified analyses. To supplement the foregoing regression analyses, we assessed whether there were significant differences in the health conditions reported at T1 versus T2 exclusively among the participants that provided data at both times (the paired-samples; N = 125). We used the McNemar test to assess the null hypothesis that the proportion of the “yes” response to a particular health condition at T1 = proportion of the “yes” response at T2. Notably, symptoms reported at both time points may have included symptoms that had already resolved. For stress, we used the paired-samples t-test to determine whether the mean difference between paired observations is different from zero.