Several study limitations should be noted. First, the analyses used retrospective AOO reports to reconstruct temporal order. Differential recall error could bias results. Second, we excluded respondents without a history of abuse from a diagnosis of dependence, leading to a restriction in the coverage of dependence to those with socially maladaptive or hazardous use. This restriction is likely to be small, though (Degenhardt et al. 2008
; Degenhardt et al. 2007b
). Third, the only AOO information recorded was age of onset of first symptom of abuse (alcohol, illegal drugs) or dependence (nicotine). To the extent that mental disorders that begin subsequent to these first symptoms predict subsequent progression to dependence, we will under-estimate the overall predictive effects of these disorders in our analysis. Fourth, as we focused only on time-lagged predictive associations, we also under-estimated the predictive effects of mental disorders on subsequent onset of substance dependence in the year of onset of the mental disorders.
The simulated PARP estimates are broadly consistent with those in the one earlier simulation study of this type ever undertaken, in a series of cross-national WHO surveys (Kessler et al. 2001
). These earlier estimates, though, focused exclusively on active disorders and considered narrower range of externalizing disorders than the NCS-R.
It is important to recognize that the PARP and NNT estimates are based on two unrealistic assumptions: that the observed associations between mental disorders and later substance dependence are entirely due to causal effects of mental disorders; and that it would be possible to prevent or cure 100% of these mental disorders with treatment. The first assumption is unrealistic in that mental and substance dependence are almost certainly influenced by common causes (Glantz et al. 2005
; Krueger et al. 2007
). The second assumption is unrealistic because no intervention for mental disorders approaches 100% effectiveness (Connor et al. 2006
; Gilchrist & Arnold, 2008
; Nelson, 2008
; Sartorius et al. 2007
We might have attempted to estimate the latter effect, as we have information on age of first seeking treatment for mental disorders. However, cases that seek treatment are typically more severe than those that do not, often leading treatment to be associated with increased rather than decreased risk of subsequent persistence, severity, and onset of secondary disorders. Because of this problem, we made no attempt to estimate the extent to which treatment of mental disorders predicted subsequent risk of substance dependence.
In light of the above considerations, the actual effects of real treatment of primary mental disorders would probably be smaller than the upper bound estimates reported here. For example, if only half the mental disorders treated were cured (a reasonable upper bound based on the results of published treatment effectiveness studies) and if only half the predictive effects of mental disorders on later substance dependence are causal, then the actual PARP associated with real-world interventions might be no more than 25% as large as the PARP estimates reported here and the NNT would be four times as large as the NNT estimates reported here. NNT is the critical statistic here, as cost-effectiveness is judged in terms of costs per effectively treated case. Based on reasonable best-case assumptions, NNT would be in the range 76-177 for anxiety-mood disorders and 40-47 for externalizing disorders. Numbers these large are well outside the range considered cost-effective to prevent a single case of substance dependence. In light of this fact, even though we found that mental disorders significantly predict subsequent substance dependence, we cannot conclude that prevention or early successful treatment of mental disorders would have a cost-effective impact in preventing subsequent substance dependence in the general population.
At the same time, the NCS-R data show clearly that people with mental disorders have a meaningfully elevated risk of substance dependence. This means that information about mental disorders might be useful as part of a risk formula to target preventive interventions even if the focus of the interventions was on risk factors other than on the mental disorders themselves. Externalizing disorders might be especially important risk markers in this regard (Glantz et al. 2005
; Hicks et al. 2004
; Verona & Sachs-Ericsson, 2005
), as they are the most strongly predictive of later substance dependence and the only class of disorders for which no difference was found in the magnitude of survival coefficients associated with active and remitted disorders.
It is also noteworthy that the effects of mental disorder treatment interventions in preventing onset of secondary substance dependence, although too small to provide a primary
justification for these interventions, might be considered important secondary outcomes in evaluating such interventions. Follow-up over a period of several years or longer might be needed to detect these effects, so the addition of a long-term follow-up component to experimental interventions to treat mental disorders could be valuable in documenting secondary benefits such as this (Kessler et al. 2008
). Furthermore, even if interventions to treat mental disorders would not completely avert cases of substance dependence, they might mitigate the severity, course, or collateral problems associated with substance dependence and, in particular, cases of comorbidity.