The statement made in the article, “alcohol abuse is associated with increased rates of physical activity in the general population, whereas dependence was not,” requires clarification. This statement reflects the results of an analysis (top left column of page 4) in which abuse status (yes vs. no) and dependence status (yes vs. no) were included as simultaneous predictors of meeting physical activity guidelines and abuse status and dependence status variables were overlapping and non-independent factors. That is, those positive for abuse could have alcohol abuse alone without concomitant dependence (pure abuse cases; n = 1654) or could have alcohol abuse with concomitant dependence (combined cases; n = 894), whereas those negative for abuse could have dependence without abuse (pure dependence; n = 522) or neither have abuse nor dependence (no alcohol disorder; n = 26,641). By including abuse and dependence status as simultaneous predictors, the overlap between abuse status and dependence status was statistically controlled for, hence removing confounding effects of cases with co-occurring abuse and dependence and allowing us to interpret the unique effects of abuse status and dependence status separately. Here, the results showed that, after accounting for this overlap, abuse status had a significant effect on rates of meeting physical activity guidelines (p < .0001), whereas dependence status did not (p = .52). In other words, when holding dependence status constant, abuse is related to rates of meeting physical activity guidelines; however, when holding abuse status constant, dependence is not related to rates of meeting physical activity guidelines.
To illustrate this effect in another way, we conducted post-hoc four-group pair-wise contrasts to isolate the unique effects of abuse status versus dependence status while holding the other diagnosis constant (results not presented in the final article). The results from this four-group analysis are consistent with the results described above and support the notion that variance in abuse status but not dependence status impact variation in physical activity outcomes. The rate of meeting physical activity guidelines for those with neither alcohol abuse nor dependence is 46.5%. The rate of meeting physical activity guidelines for those with alcohol abuse without dependence (pure abuse cases; 55.4%) is higher than those with no alcohol disorder diagnosis (pair-wise contrast: p < .0001). By contrast, the rate of meeting physical activity guidelines for those with alcohol dependence without concomitant abuse (pure dependence cases; 49.3%) is not different than those with no alcohol disorder diagnosis (pair-wise contrast: p = .44). The rate of meeting physical activity guidelines in those with co-occurring abuse and dependence (combined cases; 58.0%) is higher than those with no alcohol disorder diagnosis (pair-wise contrast: p < .0001) but not different than those with abuse alone (pair-wise contrast: p = .18), suggesting that the addition of dependence does not substantially raise rates of physical activity over and above the diagnosis of abuse. Similarly, the rate of meeting physical activity guidelines are lower among those with dependence without concomitant abuse (pure dependence cases; 49.3%) when compared to those who have dependence with co-occurring abuse (58.0%; pair-wise contrast: p < .01), suggesting that the addition of abuse does substantially add to likelihood of being physically active.
The sentence “As illustrated in Figure 1, rates of physical activity systematically increase from low to moderate symptom counts and subsequently decrease from moderate to high symptom counts” requires correction. A more accurate interpretation of the finding illustrated in that figure is that rates of meeting physical activity guidelines increase with symptom count faster at the lower end of the alcohol use disorder count and slows (and perhaps levels off) at the higher end of the alcohol use disorder count. Thus, after one gets to medium and then high levels of alcohol use disorder symptoms, the rates of meeting physical activity guidelines “flatten out” but do not “decrease” as originally stated.
We analyzed the data post-hoc using contrast statements to test for linear and quadratic relationships. PROC IML (SAS, 2008) was used to generate a coefficient matrix and the coefficients obtained can be added after the PROC GLM model statement. Testing the model in this fashion revealed a quadratic effect (p < .0001).
Consistent with this interpretation, we conducted a post-hoc analysis of the data by creating three groups (0–2 symptoms, 3–6 symptoms, and 7–11 symptoms). Here, we find significant differences between the “0 to 2” and “3 to 6” groups (p < .0001) as well as between the “0 to 2” and “7 to 11” groups (p < .05). The “0 to 2” group met physical activity requirements 46.4% of the time, the “3 to 6” group met them 57.1% of the time, and the “7 to 11” group met requirements 56.3% of the time. Specially, the “0 to 2 symptom” group exhibited the lowest frequency of meeting physical activity requirements in comparison to the middle- and high-symptom count groups. There were no differences between the “3 to 6” and “7 to 11” groups. Thus, quadratic analyses reflect that the speed with which rates of physical activity increase with symptom count is faster at the lower end of the alcohol use disorder count and slows (and perhaps levels off) at the higher end of the alcohol use disorder count.
The sentence on page 4 beginning “The quadratic term was. . .” should read “The quadratic term was significantly related to physical activity, OR (95% CI) = .97 (.94–.99), p < .05, as was related to disability, OR (95% CI) = .84 (.78–.92), p < .0001; non-alcohol mental disorder status, OR (95% CI) = .83 (.75–.93), p < .01; marital status, OR (95% CI) = .96 (.93–.99), p < .001; education, OR (95% CI) = 1.04 (.99–1.08), p = .07; age, OR (95% CI) =.98 (.98–.99), p < .0001; male sex, OR (95% CI) = 1.45 (1.37–1.54), p < .0001.”
Finally, several numbers were incorrectly reported in . should read as follows (changed values are bolded):
TABLE 1. Sample characteristics by alcohol use disorder (weighted).