Safety From Observational Studies

SAFETY FROM OBSERVATIONAL STUDIES

To ascertain if SSRI treatment constitutes a risk for suicide in the general youth population (community treated individuals), many ecological studies on suicidality in relation to SSRI use have been performed. In these studies, temporal trends in completed suicides in relation to trends in SSRI utilization within specific countries were analyzed statistically. The findings, strengths and weaknesses of each of these studies will be reviewed.

Observational studies on the relationship between adult suicide and exposure to ATDs have been conducted on the Swedish population (Isacsson, Boethius and Bergman, 1992; Isacsson et al., 1997). These studies conclude that adult suicides are asso-ciated with little or no ATD use at the time of death and suggest that SSRIs protect against suicide. An alternative approach to the use of existing clin-ical data occurred in a British study that used a 2-year period of accident and emergency visit patient data to examine the relationship between deliber-ate self-harm and ATD class (Donovan et al., 2000). The authors found significantly more deliberate self-harm events following the prescription of an SSRI than for tricyclic antidepressants (TCAs) (p < 001). They infer that changing to an ATD class that is safer in over-dosage (SSRI rather than TCA) did not reduce the risk of morbidity from deliberate self-harm. The contrast between Swedish and British suicidality studies in treated adult populations points up the limits of generalizing from one country to another and from one approach to address drug safety to another. Neither study sheds light on the rela-tive risk for suicidality in SSRI community-treated persons.

In the United States where the recent concern has been on the risk for youth treated with SSRIs, admin-istrative reimbursement claims for prescription use in an insured population were collected for 1 month in 1989 and a corresponding 1 month period in 2001 (Olfson et al., 2003). ATD rates per 1000 patients receiving any medication (in contrast to enrolled youths) were calculated, stratified by sex and age group (10–14 years and 15–19 years) for counties with more than 100 prescriptions. To create data on suicides, Center for Disease Control and Prevention’s Compressed Mortality Files were extracted to produce suicide rates for each county by age group and sex. County-level suicide rates were converted to three-digit zip code region rates. Adjusted linear regression models were used to assess the association between the change in ATD medication (independent vari-able) and change in suicide rate (dependent variable) accounting for regional racial composition, median income and physicians per capita for two time points (1 month in 1990 and in 2000). The analysis was also presented stratified by sex, age group, median regional income and racial composition. A significant inverse relationship was observed between 1990 and 2000 for change in the regional rates of overall ATD medi-cation treatment and change in the regional suicide rates after adjusting for change in percentage white population, median income and number of physicians per capital. The attempt to assess the data in terms of TCA use and change in suicide rate was limited by the very low exposure to TCAs at each time point (1.2% and 0.8% or three youths in 1989 and six youths per region in 2000).

Several study limitations are prominent: The total sample size of exposed youths in 1 month window is very small to link with very rare events occurring at the population level, the assumption that prescriptions dispensed were consumed and the theoretical model suggesting that suicide and ATD use are strongly negatively correlated ignores many non-pharmacologic factors known to influence suicide rates, for example firearms reduction, non-pharmacologic therapies and the broad national trend for suicide reduction going back years before ATDs were being used. Also, it is difficult to interpret the data when the denominator is composed of youths treated with any medication because the period from 1990 to 2000 could have produced artifactual changes in medication-users based on health insurance cover-age plans. In addition, these patient-specific data are correlated with suicide events representing population data. The limitation of ecological data notwithstand-ing this study is being cited prominently to support the use of SSRIs in the treatment of pediatric depres-sion – a causal inference that is not justified by the data. A similar analysis focused on the association of prescriptions dispensed within a county and the suicide rate but found no association for total ATDs but a significantly higher rate for TCAs and lower rate for SSRIs (Gibbons et al., 2005). These data invite the alternative explanation that TCAs are more lethal than SSRIs when self-harm occurs. Observational studies based on patient-level rather than group-level data are generally more persuasive. Three examples are described below.

Medical examination and treatment reports dealing with suicidality have been obtained by Jick, Kaye and Jick (2004) from UK physicians reporting to the General Practice Research Database (GPRD). The GPRD physician office-based data were analyzed for the period from 1993 through 1999 and included treat-ment with TCAs as well as SSRIs in relation to suicide attempts and completed suicides. A matched case – control study was conducted, and the relative risk for newly diagnosed non-fatal suicidal behaviour was not different for amitriptyline, fluoxetine and paroxetine compared with the risk among dothepin users. New ATD use had to have occurred within 90 days before the index date for suicidal behaviour of the cases. The authors found no substantial difference in the effect of the four drugs on people aged 10–19 years. The Jick report lends little evidence to the risk for US treated youth because SSRI use is 3-fold higher for youth in the United States than in the United King-dom (Delate et al., 2004; Murray, de Vries and Wong, 2004). Greater use of drug combinations, higher dosing and greater duration of exposure are also likely to occur in US treatment patterns (Hunkeler et al., 2005). A separate analysis using GPRD data from the United Kingdom had similar findings to the Jick finding of no association except for an age-specific difference (Martinez et al., 2005). For current SSRI use in those 18 or younger, there was weak evidence for non-fatal self-harm relative to TCA use (p < 05). However, the possibility of channelling of SSRIs to patients at higher risk of suicidal behaviour cannot be ruled out.

A second patient-level study (Valuck et al., 2004) used US computerized commercial insurance claims data. The authors assessed suicide attempts from physician reports for the period from 1997 to 2003 for youths with a diagnosis of MDD, and the authors compared suicide attempt reports on those treated with ATDs for 2 months or less with depressed patients treated with ATDs for 6 months or more. The study was based on 138 physician-reported cases of suicide attempts in an adolescent popula-tion diagnosed with MDD ( n = 24 110) , thus yield-ing a suicide attempt rate of 0.57%. The Valuck analysis lacks credibility because outpatient physician visits are not likely to be coded for suicide attempts (E-codes) and ICD-9 diagnostic codes do not identify suicidality.

A third study utilized a health maintenance organi-zation (HMO) population in the northwest region of the country (Simon et al., 2006) and studied 82,285 ATD use episodes (n = 65 103 patients, mainly adults) with respect to the risk of suicide or serious suicide attempts (leading to hospitalization). The episodes included multiple events for an individual with more than one ‘new use’ episode (defined as no use in the past 6 months). During the 6 months following the start date of the dispensed prescription, the risk of serious suicide attempts was 1 in 1000 and the risk of suicide during acute-phase ATD treatment was 1 in 3000. The authors concluded that the data do not indicate a significant increase in risk of suicide or serious suicide attempt after starting treatment with newer ATDs. In a repeated-measures logistic regres-sion model with adjustment for age, sex and year of treatment, the risk of suicide death in the first month of treatment was not significantly higher than in subse-quent months (odd ratio = 1 2 95% CI = 0 5–2 9). The number of suicide deaths in adolescents (n= 3) was too small to support analysis of time trends. In agreement with the Jick findings, there was a higher risk of suicide attempt in the first week of ATD treat-ment than in subsequent weeks. The authors suggest that a causal model would require a randomized study and for single drug comparisons to placebo samples on the order of 300 000 to detect a 2-fold difference in risk of suicide death or serious suicide attempt during the first month of treatment. They recommend reliance on multiple data sources from both large observational studies and randomized trials.