Drs. Bostwick and Pankratz Reply

To the Editor: In our meta-analysis we reported that suicide risk in persons with affective disorders follows a hierarchy based on the intensity of the treatment settings. We found that lifetime suicide prevalence for patients who had ever been hospitalized for suicidality was the highest, at 8.6%, followed by 4.0% for those hospitalized without the specification of suicidality, 2.2% for mixed inpatient/outpatient populations, and less than 0.5% for the population that was not affectively ill.

We agree with Dr. Simon that his work with VonKorff (1998) further strengthens the primary findings of our meta-analysis. We cited their work in an attempt to credit them accordingly. We could not include their actual data, however, because the data did not conform to our inclusion criteria. We required a clearly defined cohort with affective disorder of whom at least 90% were followed up for an average of at least 2 years and for whom both deaths and suicides were specified. Of the hundreds of studies we reviewed, fewer than 50 met this rigorous standard. The study by Dr. Simon and VonKorff was not among them.

First, they included not only patients diagnosed with depression but also patients who had begun antidepressant treatment, even if only for an adjustment disorder. Second, their review of 35,546 subjects who were followed for 62,159 person-years resulted in an average follow-up of only 1.75 years. Third, their hierarchy differed from ours in that it included not only psychiatric inpatients and outpatients without the specification of suicidality but also any patient for whom an antidepressant was prescribed by any HMO provider. Last, we could not discern from their report the degree to which their cohort was successfully followed, as measured by the number with complete follow-up. Inclusion appeared to be based on continued HMO membership; the percentage of patients lost to follow-up because of lapsed HMO enrollment was not stated.

Drs. Kuo and Gallo report original data that exhibits hierarchical layering, which also supports our findings. As did Dr. Simon and VonKorff, they studied a population different from ours. The Baltimore ECA study (Eaton et al., 1997) drew on a community-based sample that was less severely ill than most of our cohorts. Their hierarchy distinguished between subjects with a history of suicide attempts, subjects with affective disorders, and subjects in the general population. In addition, the suicidality data they report in their letter is not present in the report on the Baltimore ECA study that they cite and therefore could not have been considered for inclusion in our meta-analysis.

We are gratified that the critiques from these authors essentially expand on and agree with the results of our meta-analysis. We recognize that our somewhat procrustean decision to apply rigorous criteria to a broad, nonstandardized, epidemiological literature on affective disorders and suicide necessitated excluding some important data. Our exclusions were not meant to detract from these authors’ studies. On the contrary, we acknowledge the importance of the work by Dr. Simon and VonKorff and by those involved in the Baltimore ECA study in helping us all better understand the risk of suicide among affectively ill patients.