Table 2.
Best-fitting Model of Predictive Associations Between Type and Number of 12-Month DSM-IV/CIDI Disorders and 12-Month SED Among 2647 Respondents With 1 or More 12-Month Disordersa
| Disorder | OR (95% CI) | PARPb |
|---|---|---|
| Mood disorders | ||
| Major depressive episode/dysthymia | 4.5* (3.8–5.5) | 30.0 |
| Bipolar disorderc | 0.9 (0.7–1.3) | 2.2 |
| Any mood disorder | … | 31.4 |
| Anxiety disorders | ||
| Agoraphobiad | 0.8 (0.4–1.3) | 1.5 |
| Generalized anxiety disorder | 1.6* (1.2–2.3) | −0.6 |
| Social phobia | 1.3* (1.0–1.7) | 2.2 |
| Specific phobia | 1.6* (1.3–2.0) | 9.8 |
| Panic disordere | 3.9* (2.6–5.9) | 2.3 |
| Posttraumatic stress disorder | 2.2* (1.3–3.6) | 2.9 |
| Separation anxiety disorderf | 1.0 | −2.2 |
| Any anxiety disorder | … | 10.9 |
| Behavior disorders | ||
| Attention-deficit/hyperactivity disorder | 3.1* (2.3–4.2) | 8.4 |
| Oppositional-defiant disorder | 2.5* (1.9–3.4) | 15.6 |
| Conduct disorder | 18.4* (13.6–25.0) | 35.9 |
| Eating disordersg | 1.0 (0.7–1.5) | −1.3 |
| Any behavior disorder | … | 54.5 |
| Substance disorders | ||
| Alcohol abuseh | 1.3 (0.9–1.8) | −1.3 |
| Drug abuseh | 1.3* (1.1–1.6) | 2.8 |
| Any substance disorders | … | 1.9 |
| No. of distress/behavior disorderi | ||
| Exactly 2 disorders | 0.8 (0.6–1.2) | … |
| Exactly 3 disorders | 1.0 (0.5–1.9) | … |
| ≥4 Disorders | 6.1* (1.5–24.7) | … |
Abbreviations: CIDI, Composite International Diagnostic Interview; OR, odds ratio; PARP, population-attributable risk proportions; SED, serious emotional disturbance.
Based on a model that includes predictors for 14 of the 15 types of disorders (compared with the omitted category of separation anxiety disorder); counts of 2, 3, and 4 or more distress or behavior disorders; and controls for independent sociodemographic variables (ie, age at interview, sex, and race/ethnicity). See Table 2 in the companion article2 in this issue for information on 12-month prevalence of disorders. By independent sociodemographic variables we mean the subset of sociodemographic variables that could not plausibly be caused by the respondent’s SED. This was the best-fitting model from a wide variety of models compared using the Bayesian information criterion as the measure of model fit. Model selection was not sensitive to the criterion used because other criteria examined also selected this as the best-fitting model. The 14 disorder-specific ORs differ significantly among themselves (F13,29 = 52.9; P < .001), whereas the 3 number-of-disorders coefficients are significant as a set (F3,39 = 3.4; P = .03). Asterisks indicate OR is significant using 2-sided tests with an α level of .05.
The PARP is the proportion of cases of SED predicted under the model to be due to each of the disorders considered here. See the “Analysis Methods” subsection in the “Methods” section for a description of the simulation method used to calculate PARP. The PARP estimates do not sum to 100% because PARP is calculated by deleting 1 disorder or class of disorders from the model at a time, resulting in the effects of comorbidity being included in each of the estimates.
Bipolar disorder includes bipolar I, bipolar II, and subthreshold bipolar disorder.
With or without a history of panic disorder.
With or without agoraphobia.
Separation anxiety disorder is the contrast category, with an implicit OR of 1.0.
Eating disorders include anorexia nervosa, bulimia nervosa, and binge-eating behavior.
With or without dependence.
No predictor variable for having exactly 1 disorder was included in the model because this would have been collinear with the coefficients associated with the individual disorders. The disorder-specific coefficients consequently can be interpreted as effects of pure disorders (ie, disorders that occur to respondents who have exactly 1 disorder), whereas the coefficients associated with numbers of disorders can be interpreted as deviations from the ORs associated with the products of these disorder-specific effects among respondents with more than 1 disorder.