4.3. Statistical methods

In the original study I, a logistic regression analysis was used to examine an association of hospital-treated atopic disorder to depression after adjusting for sex (male/female), father’s social class in 1966 (classes I–II/classes III–IV/farmers) (Rantakallio 1979), and the dwelling place in 1980 (urban/rural) (Rantakallio et al. 1992).

In the original study II, the proportions of prick test positive cases were compared, separately for males and females, with those of prick test negative ones. Unadjusted odds ratios (OR) and their 95% confidence interval (CI) were calculated for those who were atopic to estimate the increased risk of being depressed sometime in their lifetime. Logistic regression analysis was used to examine the association between atopy and depression, adjusting for confounding variables presented in Table 1. In clinical practice, the presence of atopy requires both positive skin prick tests and clinical symptoms. Therefore, a logistic regression analysis was used to investigate the association between allergic disorders ascertained by skin prick test responses and depression adjusting for confounding variables.

In the original study III, the continuation ratio model was used to estimate the association between depression and allergic symptoms verified by skin prick test responses. This model partitions the analysis of the outcome variable into three different logistic regression models (doctor-diagnosed depression in three different HSCL-25 depression subscale cut-off points (> 1.54, > 1.74 and > 2.00)) for dichotomous responses (Agresti 1990). Non-depressive subjects in the outcome variable consisted of those cohort members who had neither self-reported doctor-diagnosed lifetime depression nor depressive symptoms in the HSCL-25 depression subscale (mean score = 1.0). The final models used odds ratios (OR) and 95% confidence intervals (95%CI) after controlling for confounders presented in Table 1.

In the original study IV, associations of independent variables (cohort member’s atopy verified by positive skin prick test response, parental atopy and sibling atopy) and potential confounding factors with self-reported and hospital-treated depression of a cohort member were analyzed separately for both genders with the Chi-Squared Automatic Interaction Detector (CHAID) technique (Moles & Bedi 1997). CHAID divides a population into a series of groups, based on their ability to predict a dependent variable (i.e., depression in this study). The result is a tree diagram. At each stage in the generation of the tree, multiple chi-square-tests are performed in order to select the current most significant predictor of the dependent variable. When there are no more statistically significant predictors (i.e., p-value < 0.05), the generation of the tree stops. Due to the small number of hospital-treated depressions among the cohort members, Fisher’s Exact test was also used alongside with the chi-square test. In addition, multivariate logistic regression analysis was used to examine the association between exposure variables (cohort member’s atopy, parental atopy and siblings’ atopy) and self-reported depression of a cohort member when potential/possible confounding factors (presented in Table 1) were adjusted. In order to estimate whether an individual atopic disorder in combination with maternal, paternal or a sibling’s atopy were risks with regard to a proband’s depression, odds ratios with 95% confidence intervals were calculated for male and female cohort members, characterized by several combinations of family atopy variables. A group with no own/family member’s atopy was used as a reference group in the logistic regression analyses.

Statistical analyses were performed using the SPSS version 9.0 for Windows in the original publication I (SPSS Inc. 2001), and by using the SAS software (version 8e) for Windows in the original publications II–IV (SAS Institute Inc. 1999)