We appreciate the comments from Drs Panei and Paola, hereafter referred to as PP, on our study published in BMJ.1 Our study (which was not sponsored by pharmaceutical industry) aimed to explore suicidal behavior as a potential adverse effect of ADHD medication. The main conclusion of our study was that no evidence was found for an increased rate of suicide-related events (suicide attempt/completed suicide) associated with the use of ADHD medication. We also observed a protective effect of psycho-stimulants on suicidal behavior. Issues concerning drug efficacy and other adverse effects of pharmacotherapy, though critical, are beyond the scope of our research question. We now take this opportunity to clarify some major points misunderstood by PP.
PP express strong suspicions against our observational (i.e. non-randomized) study design and statistical analysis. They state that “What we need are results from well-designed systematic reviews based on good quality clinical trials addressing the multiple social and familial factors responsible for ADHD”. We agree that randomized clinical trials are ideal for examining treatment efficacy and for establishing causality. However, clinical trials are often not practical for investigating such association for the following reasons. First, although it is possible to examine treatment-emergent suicidal ideation in clinical trials, any study participants with notable suicidal ideation should be provided acute intervention or be excluded from the trial due to ethical concerns. Accordingly, suicide attempt and completed suicide are unlikely to be observed thereafter. Second, since clinical trials are limited in sample size, duration of follow-up and patient compliance, they are not feasible for studying rare events as suicidal behavior. Thus, we find PP’s suggestion to discard observational studies unreasonably rigid.
PP state that “Second, the study by Chen et al. has major statistical limitations. … they used a complex Cox regression model splitting the population into thousands of groups using each patient as their own control”. PP gives three arguments as to why our “within-analysis” would be problematic. First, they argue that we are “presumably dangerously limiting [our] fragmented analysis to patients who will never attempt suicide…” This is mathematically, wrong. Rather, it is the other way around; the within-analysis is limited to those patients who attempt suicide at least once during follow-up. This is intuitively reasonable; we cannot compare the rate of suicide attempt during medicated and non-medicated periods in those patients who never make suicide attempts. Also, this is not a limitation, since those who never make suicide attempts (even if medicated) do obviously not need to worry about the elevated risk during medication. Second, PP argue that “In real life, because children differ in construct skills their decisions could reflect unpredictable cognitive and emotional patterns, determining complex interrelated variables that are difficult to analyze within each patient.” We interpret this as sentence as expressing a worry for unmeasured time-varying confounding. We agree with PP in this aspect, but note that the potential for unmeasured time-varying confounding can not be ruled out by any statistical method that we are aware of. We also note that we mentioned the potential for unmeasured time-varying confounding in the Discussion of our paper. Third, PP state that our design/analysis “fails to give readers the simple statistical measures, such as odds ratios (proportions), number needed to treat and number needed to harm”. We don’t agree that an odds ratio is a simpler or more transparent measure of association than a hazard ratio. But we do agree that number need to treat/harm is a more clinically useful measure than both hazard ratios and odds ratios. However, these measures cannot (as for as we know) be obtained from a within-analysis. They can be obtained from a more standard “between-analysis”, but such analysis would not automatically adjust for all (measured and unmeasured) time-stationary confounders, as the within-analysis does. Therefore, since we share PP’s concern about confounding, we chose to “sacrifice” the number needed to treat/harm, to be able to make an extensive confounding adjustment.
PP claim that there is a “the known association between medications prescribed for patients with ADHD and risk of concomitant suicidal behavior”. However, the references that PP give to support this claim 2 3 focused only on a positive relationship between use of atomoxetine and treatment-emergent suicidal ideation. The author of the meta-analysis did not draw any conclusion on suicidal behavior because only one suicide attempt and no completed suicide was observed in the clinical placebo-controlled trials which the meta-analysis was based on. Moreover, a recent update of this meta-analysis demonstrated no significant association between the use of atomoxetine and suicidal ideation, and still no more than 1 case of suicidal behavior was documented after 9 years of the FDA warning.4 Therefore, the “known association” for suicidal behavior referred to by PP was actually unknown, and even the associations for suicidal ideation varied from time to time. This is a good example of why clinical trials are not practical for examining potential treatment-emergent suicidal behavior.
PP also mentioned the negative attitude of parents of children with ADHD towards non-pharmacological interventions of ADHD in Italy. This might reflect inadequate instruction and counseling of parents on the management of the disorder. In addition, better evidence is needed for the efficacy of non-pharmacological interventions.5
We did observe a protective effect of psycho-stimulant on suicidal behavior, which makes a harmful effect of psycho-stimulants on suicidal behavior seem even less likely. A protective effect of psycho-stimulant is possible because dysfunction of impulse control resulted from imbalance of serotonergic system has been involved in the pathophysiology of both ADHD and suicidal behavior.6-8 Medications improving symptoms of impulse control might be beneficial for patients with ADHD who are also at increased risk of suicidal behavior, which is worthy of future investigation.
Finally, we agree with PP that well-designed research is needed to fully understand the benefits and risks associated with treatment for ADHD among both children and adults. Recommendations and guidelines for both pharmacological and non-pharmacological interventions should be evidence based.
References
1. Chen Q, Sjolander A, Runeson B, D'Onofrio BM, Lichtenstein P, Larsson H. Drug treatment for attention-deficit/hyperactivity disorder and suicidal behaviour: register based study. Bmj 2014;348:g3769.
2. Food and Drug Administration. Public health advisory: suicidal thinking in children and adolescents being treated with Strattera (atomoxetine), 2005.
3. Bangs ME, Tauscher-Wisniewski S, Polzer J, Zhang S, Acharya N, Desaiah D, et al. Meta-analysis of suicide-related behavior events in patients treated with atomoxetine. Journal of the American Academy of Child and Adolescent Psychiatry 2008;47:209-18.
4. Bangs ME, Wietecha LA, Wang S, Buchanan AS, Kelsey DK. Meta-Analysis of Suicide-Related Behavior or Ideation in Child, Adolescent, and Adult Patients Treated with Atomoxetine. Journal of child and adolescent psychopharmacology 2014.
5. Sonuga-Barke EJ, Brandeis D, Cortese S, Daley D, Ferrin M, Holtmann M, et al. Nonpharmacological interventions for ADHD: systematic review and meta-analyses of randomized controlled trials of dietary and psychological treatments. The American journal of psychiatry 2013;170:275-89.
6. Asberg M, Traskman L, Thoren P. 5-HIAA in the cerebrospinal fluid. A biochemical suicide predictor? Archives of general psychiatry 1976;33:1193-7.
7. Banaschewski T, Becker K, Scherag S, Franke B, Coghill D. Molecular genetics of attention-deficit/hyperactivity disorder: an overview. European child & adolescent psychiatry 2010;19:237-57.
8. Nordstrom P, Samuelsson M, Asberg M, Traskman-Bendz L, Aberg-Wistedt A, Nordin C, et al. CSF 5-HIAA predicts suicide risk after attempted suicide. Suicide & life-threatening behavior 1994;24:1-9.
Competing interests:
No competing interests
01 September 2014
Qi Chen
PhD student in Epidemiology
Arvid Sjölander, Associate Professor in Biostatistics; Henrik Larsson, Associate Professor in Epidemiology
Department of Medical Epidemiology and Biostatistics, Karolinska Institute
Rapid Response:
We appreciate the comments from Drs Panei and Paola, hereafter referred to as PP, on our study published in BMJ.1 Our study (which was not sponsored by pharmaceutical industry) aimed to explore suicidal behavior as a potential adverse effect of ADHD medication. The main conclusion of our study was that no evidence was found for an increased rate of suicide-related events (suicide attempt/completed suicide) associated with the use of ADHD medication. We also observed a protective effect of psycho-stimulants on suicidal behavior. Issues concerning drug efficacy and other adverse effects of pharmacotherapy, though critical, are beyond the scope of our research question. We now take this opportunity to clarify some major points misunderstood by PP.
PP express strong suspicions against our observational (i.e. non-randomized) study design and statistical analysis. They state that “What we need are results from well-designed systematic reviews based on good quality clinical trials addressing the multiple social and familial factors responsible for ADHD”. We agree that randomized clinical trials are ideal for examining treatment efficacy and for establishing causality. However, clinical trials are often not practical for investigating such association for the following reasons. First, although it is possible to examine treatment-emergent suicidal ideation in clinical trials, any study participants with notable suicidal ideation should be provided acute intervention or be excluded from the trial due to ethical concerns. Accordingly, suicide attempt and completed suicide are unlikely to be observed thereafter. Second, since clinical trials are limited in sample size, duration of follow-up and patient compliance, they are not feasible for studying rare events as suicidal behavior. Thus, we find PP’s suggestion to discard observational studies unreasonably rigid.
PP state that “Second, the study by Chen et al. has major statistical limitations. … they used a complex Cox regression model splitting the population into thousands of groups using each patient as their own control”. PP gives three arguments as to why our “within-analysis” would be problematic. First, they argue that we are “presumably dangerously limiting [our] fragmented analysis to patients who will never attempt suicide…” This is mathematically, wrong. Rather, it is the other way around; the within-analysis is limited to those patients who attempt suicide at least once during follow-up. This is intuitively reasonable; we cannot compare the rate of suicide attempt during medicated and non-medicated periods in those patients who never make suicide attempts. Also, this is not a limitation, since those who never make suicide attempts (even if medicated) do obviously not need to worry about the elevated risk during medication. Second, PP argue that “In real life, because children differ in construct skills their decisions could reflect unpredictable cognitive and emotional patterns, determining complex interrelated variables that are difficult to analyze within each patient.” We interpret this as sentence as expressing a worry for unmeasured time-varying confounding. We agree with PP in this aspect, but note that the potential for unmeasured time-varying confounding can not be ruled out by any statistical method that we are aware of. We also note that we mentioned the potential for unmeasured time-varying confounding in the Discussion of our paper. Third, PP state that our design/analysis “fails to give readers the simple statistical measures, such as odds ratios (proportions), number needed to treat and number needed to harm”. We don’t agree that an odds ratio is a simpler or more transparent measure of association than a hazard ratio. But we do agree that number need to treat/harm is a more clinically useful measure than both hazard ratios and odds ratios. However, these measures cannot (as for as we know) be obtained from a within-analysis. They can be obtained from a more standard “between-analysis”, but such analysis would not automatically adjust for all (measured and unmeasured) time-stationary confounders, as the within-analysis does. Therefore, since we share PP’s concern about confounding, we chose to “sacrifice” the number needed to treat/harm, to be able to make an extensive confounding adjustment.
PP claim that there is a “the known association between medications prescribed for patients with ADHD and risk of concomitant suicidal behavior”. However, the references that PP give to support this claim 2 3 focused only on a positive relationship between use of atomoxetine and treatment-emergent suicidal ideation. The author of the meta-analysis did not draw any conclusion on suicidal behavior because only one suicide attempt and no completed suicide was observed in the clinical placebo-controlled trials which the meta-analysis was based on. Moreover, a recent update of this meta-analysis demonstrated no significant association between the use of atomoxetine and suicidal ideation, and still no more than 1 case of suicidal behavior was documented after 9 years of the FDA warning.4 Therefore, the “known association” for suicidal behavior referred to by PP was actually unknown, and even the associations for suicidal ideation varied from time to time. This is a good example of why clinical trials are not practical for examining potential treatment-emergent suicidal behavior.
PP also mentioned the negative attitude of parents of children with ADHD towards non-pharmacological interventions of ADHD in Italy. This might reflect inadequate instruction and counseling of parents on the management of the disorder. In addition, better evidence is needed for the efficacy of non-pharmacological interventions.5
We did observe a protective effect of psycho-stimulant on suicidal behavior, which makes a harmful effect of psycho-stimulants on suicidal behavior seem even less likely. A protective effect of psycho-stimulant is possible because dysfunction of impulse control resulted from imbalance of serotonergic system has been involved in the pathophysiology of both ADHD and suicidal behavior.6-8 Medications improving symptoms of impulse control might be beneficial for patients with ADHD who are also at increased risk of suicidal behavior, which is worthy of future investigation.
Finally, we agree with PP that well-designed research is needed to fully understand the benefits and risks associated with treatment for ADHD among both children and adults. Recommendations and guidelines for both pharmacological and non-pharmacological interventions should be evidence based.
References
1. Chen Q, Sjolander A, Runeson B, D'Onofrio BM, Lichtenstein P, Larsson H. Drug treatment for attention-deficit/hyperactivity disorder and suicidal behaviour: register based study. Bmj 2014;348:g3769.
2. Food and Drug Administration. Public health advisory: suicidal thinking in children and adolescents being treated with Strattera (atomoxetine), 2005.
3. Bangs ME, Tauscher-Wisniewski S, Polzer J, Zhang S, Acharya N, Desaiah D, et al. Meta-analysis of suicide-related behavior events in patients treated with atomoxetine. Journal of the American Academy of Child and Adolescent Psychiatry 2008;47:209-18.
4. Bangs ME, Wietecha LA, Wang S, Buchanan AS, Kelsey DK. Meta-Analysis of Suicide-Related Behavior or Ideation in Child, Adolescent, and Adult Patients Treated with Atomoxetine. Journal of child and adolescent psychopharmacology 2014.
5. Sonuga-Barke EJ, Brandeis D, Cortese S, Daley D, Ferrin M, Holtmann M, et al. Nonpharmacological interventions for ADHD: systematic review and meta-analyses of randomized controlled trials of dietary and psychological treatments. The American journal of psychiatry 2013;170:275-89.
6. Asberg M, Traskman L, Thoren P. 5-HIAA in the cerebrospinal fluid. A biochemical suicide predictor? Archives of general psychiatry 1976;33:1193-7.
7. Banaschewski T, Becker K, Scherag S, Franke B, Coghill D. Molecular genetics of attention-deficit/hyperactivity disorder: an overview. European child & adolescent psychiatry 2010;19:237-57.
8. Nordstrom P, Samuelsson M, Asberg M, Traskman-Bendz L, Aberg-Wistedt A, Nordin C, et al. CSF 5-HIAA predicts suicide risk after attempted suicide. Suicide & life-threatening behavior 1994;24:1-9.
Competing interests: No competing interests