Table 3. Risk of bias table of included non- or quasi-randomised studies, using Risk Of Bias In Non-randomized Studies–of Interventions (ROBINS-I) tool [18].

Non-Randomised Study	Domains of Analysis
	Confounding	Participant Selection	Classification of Interventions	Unintended deviation from intervention	Missing outcome data	Outcome measurement	Selective reporting	Overall Rating
Itil, Shrivastava [24]	Serious risk	Low risk	Low risk	Not enough information	Serious risk	Serious risk	Serious risk	Serious risk
	Psychiatric comorbidities not controlled for; also, no baseline demographic or clinical characteristics discussed, with only mention of similarity between the two groups in baseline CGI	Selection of subjects with at least mild-moderate depression (HAMD> = 15) unlikely to introduce significant bias given this is the population most commonly prescribed antidepressants	Intervention well defined and not determined retrospectively	Adherence issues not stated or discussed	Proportion of missing participants greater for fluvoxamine group than other groups and mITT excluded almost 50% of sample	Despite subject blinding, significantly greater proportion of side effects in active groups due to rapid uptitration likely allowed inference of active treatment	No pre-specified plan of analysis stated, and selective reporting with extensive detailing of some measures in table and brief mention of other scores in body
Van Ameringen, Mancini [25]a	Critical risks	Serious risk	Low risk	Not enough information	Low risk	Critical risk	Low risk	Critical risk
	Lacks control group to control for time and engagement effect, especially when target sample of socially phobic subjects were reviewed regularly; lax exclusion criteria	Inclusion of only generalised social phobia assumes no difference with performance subtype; recruitment from referrals to anxiety disorders clinic likely biases sample towards those with less severe social phobia who are more motivated and/or less impaired to tolerate outside scrutiny of their social phobia that was not factored in analysis	Intervention well defined and not determined retrospectively	Adherence issues not stated or discussed	No missing data	Awareness of intervention makes self-report vulnerable to bias, particularly in sample of socially phobic patients who by definition fears perceived criticism	Analysis occurred according to pre-specified plan, all FQ:Agoraphobia scores reported
Mancini and Ameringen [26]a								Critical risk
Finkel, Richter [30]	Serious risk	Serious risk	Low risk	Not enough information	Serious risk	Low risk	Critical risk	Critical risk
	Inclusion of those with mild cognitive impairment by setting MMSE threshold for inclusion at 24+ and not factoring impact of MMSE scores in analysis	Intentional exclusion of those with treatment-resistant depression introduced bias	Intervention well defined and not determined retrospectively	Even though study assessed adherence by pill count and serum levels, serum levels and pill counts of sertraline not reported or discussed, and criteria for exclusion from study based on pill count was lax	Proportion of participants missing from analysis much higher in nortriptyline group, and only those who completed study was analysed	Subjects blinded, and unlikely to have inferred treatment status as both groups were antidepressants	Pre-specified plan contradictory: stating “for all continuous measures… mean score and mean change score from baseline were computed”, but later stated secondary outcome measures (which include continuous measures) were to only include “changes from baseline”; for POMS, only those who completed study were analysed (vs pre-specified ITT), interaction with time not factored
Ozdemir, Boysan [31]	Critical risk	Serious risk	Low risk	Not enough information	Low risk	Serious risk	Low risk	Critical risk
	Psychiatric comorbidities or concurrent medications not controlled for	Inclusion of only inpatient population biases sample towards those with severe MDD and complex comorbidities not factored in analysis	Intervention well defined and not determined retrospectively	Adherence issues not stated or discussed	No missing data	Subjects not blinded, and voluntary status of admission not stated, so highly vulnerable to biases in self-reported measures	Trial analysed in accordance with pre-specified plan, with all POMS scores reported

^aRisk of bias assessment results identical.