Study	Reason for exclusion
Dobie 1992	ALLOCATION: Randomised PARTICIPANTS: Same setting and number of patients (N = 92) as in a series of papers where the most definitive trial is the one included in this review (Sullivan 1993). This originates an explicit replication bias.
Dobie 1993	ALLOCATION: Randomised PARTICIPANTS: Same setting and number of patients (N = 92) as in a series of papers where the most definitive trial is the one included in this review (Sullivan 1993). This originates an explicit replication bias.
Holgers 2011	ALLOCATION: Unclear (not declared) PARTICIPANTS: 75 patients meeting risk criteria for developing severe refractory tinnitus, without socially disabling hearing loss INTERVENTION: This study evaluated the same patient groups as the Zöger 2006 trial, which was already evaluated and excluded in the previous 2009 update of the present review. In the paper published in 2011 the follow‐up time was extended to 28 weeks and the authors focus specifically on evaluation of quality of life. Although this trial aimed to study sertraline, a benzodiazepine drug (oxazepam) was also administered to 9 out of the total 75 patients. Oxazepam was administered in the first 2 weeks to alleviate expected worsening of distress and to improve tolerability of the adverse side effects of sertraline (cited in the paper as "well known side effects of sertraline"). Being a benzodiazepine (a drug acting in the central nervous system) this may have an effect on tinnitus and represents a potential source of bias. OUTCOMES: High drop‐out rate with a relevant difference between the active drug group (40.5% drop‐outs after 28 weeks) and placebo group (10.5% in total after 28 weeks)
Jalali 2009	ALLOCATION: Cross‐over, randomised, triple‐blind, placebo‐controlled trial PARTICIPANTS: 36 patients between ages 21 and 65, with a complaint of non‐pulsatile tinnitus of more than 1 year duration. INTERVENTION: The study evaluated the effects of alprazolam on tinnitus. Although alprazolam is sometimes prescribed to treat depression, it is not actually classified as an antidepressive agent. Furthermore, patients in this trial were treated with an "active placebo" (chlorpheniramine maleate) to simulate the side effect of the active treatment (drowsiness). Although the authors say that it is "popularly believed that mild induced sedation (like that expected from chlorfeniramine) does not relieve tinnitus", this can originate strongly biased results (sleep disorders are one of most serious problems in tinnitus patients).
Katon 1993	ALLOCATION: Randomised PARTICIPANTS: Same setting and number of patients (N = 92) as in a series of papers where the most definitive trial is the one included in this review (Sullivan 1993). This originates an explicit replication bias.
Lopez‐Gonzalez 2007a	ALLOCATION: Prospective, randomised, single‐blind, placebo‐controlled study PARTICIPANTS: 150 patients with subjective tinnitus INTERVENTION: The interventions included a combination of drugs: although sulpiride is sometimes used for its antidepressive efficacy, hydroxyzine is not actually classified as an antidepressive agent
Lopez‐Gonzalez 2007b	ALLOCATION: Not randomised ("Un total de 100 pacientes diagnosticados de acùfenos se dividieron aleatoriamente a su llegada a la consulta en dos grupos de 50") PARTICIPANTS: 100 patients with tinnitus ("acufenos") INTERVENTION: The interventions included a combination of drugs: although sulpiride is sometimes used for its antidepressive efficacy, melatonin is not actually classified as an antidepressive agent
Roberts 2011	ALLOCATION: Randomised controlled trial PARTICIPANTS: 24 adult patients with tinnitus INTERVENTION: This trial is not placebo‐controlled, therefore it is excluded from the review. The trial compared vestipitant versus paroxetine in the treatment of tinnitus. There is no established evidence that the substance P antagonists/NK‐1 receptor antagonists, to which the drug vestipitant belongs, are effective antidepressants. Most of the drugs in the same class are (or have been tested as) antiemetics. The report also lacks a description of the dose (dosage) of the drugs investigated in the trial. There is only a general reference to "predose" or "postdose". This is an important criticism, because we cannot know the precise treatment and the trial cannot be 'replicated'.
Sullivan 1989	ALLOCATION: Not randomised
Sullivan 1994	ALLOCATION: Randomisation not explicit. There is no mention of randomisation or allocation method, although the context and the recruited patients were the same as Sullivan 1993. PARTICIPANTS: Same setting and number of patients (N = 92) as in a series of papers where the most definitive trial is the one included in this review (Sullivan 1993). This originates an explicit replication bias.
Zöger 2006	ALLOCATION: Randomised PARTICIPANTS: 76 patients with tinnitus were investigated with measures of tinnitus severity, depression and anxiety INTERVENTION: Although this trial aimed to study sertraline, a benzodiazepine drug (oxazepam) was also administered to 9 out of the total 76 patients OUTCOMES: High drop‐out rate and if oxazepam‐treated patients are excluded from the analysis, the percentage of patients lost from analysis becomes unacceptably high (31% in treatment group, 26% in placebo group)