Table 1.

List of covariates.

Covariate	Definition and Reference Levels	Details
Age	Mid-point of age range	MD prevalence and corresponding suicide risk are known to vary with age and by sex (Whiteford et al., 2015). Therefore, they are important sources of variation to consider in our analyses
Percent Female	Continuous covariate representing the proportion of females in study sample (Ranges from 0 to 1)
Follow-up time	Follow-up time in years	Suicide is a relatively rare outcome compared to other causes of death. Therefore, the duration of follow-up may impact the number of study-reported outcomes
Response rate	Proportion of sample remaining after loss to follow-up/dropout	The response rate provides important information about possible selection bias in the sample.
Disorder	GBD mental disorder categories (Reference: MDD)	These are the primary risk factors for suicide being assessed. MDD was chosen as the reference since it was the most commonly assessed disorder among selected studies.
Estimate adjustment	Indicator for whether or not effect size has been adjusted for potential confounders such as individual demographics, socioeconomic status, family psychiatric history, etc. Reference: Adjusted for potential confounders	Study-reported effect sizes may be adjusted for potential confounders that are known to influence the MD-suicide association. These may be different from (and typically lower) than unadjusted effect sizes. Therefore, our analyses examine variation in suicide risk by testing this methodological covariate.
Psychological Autopsy (PA) Method	Indicator for whether or not data was collected using psychological autopsy – which involves collecting data from all available sources such as family informants, medical records, and healthcare providers. (Reference: PA not used)	This covariate was assessed because the psychological autopsy method involves data collection from informants and therefore is susceptible to biases in measurement of psychopathology, event recall, choice of appropriate control groups, etc. (Brent, 1989).
Study design	Prospective (Reference) or retrospective design	We expected variation in study quality and effect sizes based on the choice of study design. Therefore, this covariate was included to examine if study design influenced pooled RRs (Brent, 1989).
Sampling type	Random/other methods (multistage, cluster sampling)	We tested this covariate because we expected studies using random sampling to have less biased samples than studies using other methods (Lester and Stack, 1989).
SDI	Sociodemographic Index value (SDI): A summarized metric of a location's socio-demographic development on a scale of 0 (lowest) to 1 (highest).	Higher SDI and HAQI are associated with better health outcomes and lower premature mortality. Therefore, these were tested in the model to see if they had an impact on suicide risk.
HAQI	Health Access Quality Index value: A summarized metric of healthcare access and quality on a scale of 0 (worst) to 100 (best). More details on the construction of the HAQI can be found in elsewhere (Barber et al., 2017).
High income locations	Locations that are classified as high-income as per World Bank income classification (Reference) vs other locations (Fantom and Serajuddin, 2016).	High-income locations are known to have better health outcomes and lower premature mortality than low and middle-income countries. Therefore, we tested this variable to see if it influenced resulting suicide risk estimates.