The effects of celebrity suicide on copycat suicide attempt: a multi-center observational study

Abstract

Background

The effect of celebrity suicides on copycat suicide attempts is not well known. Our objective was to determine the association between celebrity suicide and copycat suicide attempts.

Methods

We conducted a retrospective multicenter observational time series analysis. Celebrity suicides were selected by an operational definition via three nationwide television news internet sites from January 2005 to December 2008. The reference week was defined as the week preceding date of suicide notification to the public. Then two pre-event weeks and four post-event weeks were analyzed for suicide attempts. We derived a prediction model for suicide attempt visits for each ED for these seven observational weeks using a General Additive Model with data from the National Emergency Department Information System (NEDIS) database. We calculated the mean excess visit (EV = observed visit − expected visit) and mean excess visit ratio (EVR = EV/expected visit). We tested the mean EV and EVR between reference weeks versus the observational weeks using independent t test and repeated measures ANOVA.

Results

Five celebrity suicides occurred during the study period. Total number of ED visits was 5,453,441 in the 85 EDs over the 4-year period, and suicide attempt or self-injury occurred in 27,605. The mean excess visit for each observational interval per ED was less than 0.1 during pre-event periods but increased to 0.695 in the second post-event week. EVs were significantly higher in the first to the third post-event weeks (p = 0.02, p < 0.01, p = 0.03, respectively) compared to reference week. The mean EVRs were significantly higher (=0.215) in the second post-week intervals compared with the reference week (p = 0.03). Mean EVs and mean EVRs showed significant increase in the post event period compared with the observational period (p = 0.001 in EV, p = 0.021 in EVR).

Conclusion

From a prediction model using a 4-year nationwide ED database, ED visits for suicide attempts or self injury increased following the announcements of celebrity suicides.

Appendix: General additive model

The methods described here represent a generalization of multiple regressions (which is a special case of general linear models). Specifically, in linear regression, a linear least-squares fit is computed for a set of predictor or X variables, to predict a dependent Y variable. The well known linear regression equation with m predictors, to predict a dependent variable Y, can be stated as:

$$ Y = b_{0} + b_{1} *X_{1} + \cdots + b_{\text{m}} *X_{\text{m}} $$

where Y is predicted values of the dependent variable, X ₁ through X _m values for the predictor variables, b ₀, and b ₁ through b _m the regression coefficients estimated by multiple regression.

A generalization of the multiple regression model would be to maintain the additive nature of the model, but to replace the simple terms of the linear equation b _i *X _i with f _i (X _i ) where f _i is a non-parametric function of the predictor X _i . In other words, instead of a single coefficient for each variable (additive term) in the model, in additive models an unspecified (non-parametric) function is estimated for each predictor, to achieve the best prediction of the dependent variable values.

The generalized linear model instead of general linear model is better when the dependent variable is non-normal and the dependent variable values are predicted from a linear combination of predictor variables via a link function. To illustrate, in the general linear model a response variable Y is linearly associated with values on the X variables while the relationship in the generalized linear model is assumed to be

$$ Y = g(b_{0} + b_{1} *X_{1} + \cdots + b_{\text{m}} *X_{\text{m}} ) $$

where g(…) is a function. Formally, the inverse function of g(…), say gi(…), is called the link function; so that:

$$ {\text{gi(mu}}Y )= b_{0} + b_{1} *X_{1} + \cdots + b_{\text{m}} *X_{\text{m}} $$

where mu-Y stands for the expected value of Y.

Generalized Additive Models allows you to choose from a wide variety of distributions for the dependent variable, and link functions for the effects of the predictor variables on the dependent. Normal, gamma, and Poisson distributions are follows:

$$ {\text{Log link:}}F(z) = \log (Z) $$

We can summarize the final model, combining the notion of additive models with generalized linear models, to derive the notion of generalized additive models, as:

$$ {\text{gi}}({\text{mu}}Y) = S_{i} (f_{i} (X_{i} )) $$

In other words, the purpose of generalized additive models is to maximize the quality of prediction of a dependent variable Y from various distributions, by estimating unspecific (non-parametric) functions of the predictor variables which are “connected” to the dependent variable via a link function.

The final model was as follows (allowing all EDs a specific prediction model):

$$ {\text{log[}}E(W) ]= S (x_{i} )+ D ( {\text{season)}} $$

E(W) is expected number, x _i week (k = 4, degree of freedom), S smoothing of splines, D is the dummy variable for adjustment.

The S (cubic spline scatter plot smoother) is a special smoothing technique for 2-dimension scatter plots, which generally produces a smooth generalization of the relationship between the two variables in the scatter plot. The cubic spline smoother is often used in generalized additive models, to estimate the unspecific (non-parametric) function of the predictor variables that best predicts the (transformed) dependent variable values [6].