Smoking behavior is associated with suicidality in individuals with psychosis and bipolar disorder: a systematic quantitative review and meta-analysis

Abstract

Smoking behavior has been well-established to be more prevalent in individuals with psychosis and bipolar disorder compared to the general population. However, reports about higher suicide attempt prevalence of smoking compared to non-smoking patients suggest that smoking behavior may contribute to identifying at-risk groups of patients in a comparatively easy manner. In the present systematic quantitative review, we provide meta-analytical evidence on the smoking and suicide attempt link in 22 studies (k = 27 independent samples; N = 11,452) of patients with psychosis and bipolar disorder. We observed a small meaningful effect of smoking on suicide attempts (OR = 1.70; 95% CI [1.48; 1.95]), indicating that smokers have 1.70 the odds of having reported a suicide attempt compared to non-smokers. This effect generalized across diagnosis type (i.e., schizophrenia vs. bipolar spectrum disorder), sample type (i.e., in-vs. outpatients), and participant sex. However, the observed summary effect appeared somewhat inflated due to publication process-related mechanisms, showing some evidence for effect-inflating publication bias and a decline effect. In all, the presently observed smoking and suicide attempt link appears to be small but meaningful and robust, thus suggesting smoking status represents a useful variable for the identification of at-risk populations for suicide attempts.

Keywords: bipolar disorder; meta-analysis; psychosis; quantitative review; smoking; suicide.

Figure 1

PRISMA-flow chart for study inclusion.

Figure 2

Forest plot of all included studies according to study precision. Effect sizes are provided in log odds ratios with 95% confidence intervals (CI). Symbol size varies according to study precision, with larger squares and shorter whiskers indicating higher study precision.

Figure 3

Effects of study publication year on log odds ratios. Symbol size varies according to study precision, with larger bubbles indicating higher study precision. Solid line represents the linear regression; dashed lines represent 95% confidence bands.

Figure 4

Cumulative forest plot according to study publication year. Effect sizes are provided in log odds ratios with 95% confidence intervals (CI).

Figure 5

Bubble plot for effects of women percentage within samples on log odds ratios. Symbol size varies according to study precision, with larger bubbles indicating higher study precision. Solid line represents the linear regression; dashed lines represent 95% confidence bands.

Figure 6

Contour-enhanced funnel plot. Solid dots indicate observed effect sizes; circles indicate effect sizes that were imputed symmetrically to the adjusted effect according to the Trim-and-Fill method. The non-0 vertical lines indicate the observed (dotted) and Trim-and-Fill-based adjusted (dashed) effect.

Figure 7

p-value distributions of significant values according to p-curve. The observed p-curve includes 21 statistically significant (p < 0.05) results, of which 21 are p < 0.025. There were 6 additional results entered but excluded from p-curve because they were p > 0.05.