Lyketsos 2003.

Methods	Randomised, placebo‐controlled, parallel, 12‐week, flexible clinical trial after a 1‐week, single‐blind placebo phase.
Participants	Country: USA Setting: 44 outpatients Inclusion criteria: NINCDS‐ADRDA for probable Alzheimer's disease, and DSM‐IV for major depressive episode, MMSE =>10 Exclusion criteria: unstable medical condition, lifetime diagnosis of schizophrenia, bipolar, pre‐AD anxiety disorder, current substance misuse, acutely suicidal or requiring hospital admission
Interventions	1. Sertraline (25 mg titrated to 150 mg/day, mean peak doses 113 mg) 2. Placebo
Outcomes	The principal outcome measure was response to treatment. Cornell Scale for Depression in Dementia (CSDD) Hamilton Depression Rating Scale Activities of daily living subscale of the Psychogeriatric Dependency Rating Scales The Neuropsychiatric Inventory (NPI) MMSE
Notes	Good, well‐designed study. Attempted to address the diagnostic difficulties, e.g. possible overlap of symptoms of Alzheimer's disease and major depression, if DSM‐IV criterion 2 were to count toward the diagnosis of depression of major depressive episode it had to be clearly because of loss of pleasure (anhedonia) and not entirely because of loss of interest. Similarly, DSM criterion 8 had to be attributed to indecisiveness and not entirely due to difficulty concentrating. Supported by NIMH, Bethesda, Md. grant 1R01‐MH56511. Dr Lyketsos has been or is a consultant and advisor for and has received or receives research support from several companies, as declared in the paper.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"patients were assigned randomly using a random numbers generating computer program, in blocks of 6, without stratification, to sertraline or to placebo"
Allocation concealment (selection bias)	Low risk	The research pharmacist implemented random allocation and masked treatment assignment was communicated by telephone to study staff.
Blinding (performance bias and detection bias) All outcomes	Unclear risk	"Identical appearing pills." Response to treatment was rated by 2 study psychiatrists (PVR or MS), who reviewed patients' scores on baseline and follow‐up depression rating scales after the study was completed. Rated each patient on 3‐point global scale as a non‐responder, partial responder or full responder. These psychiatrists were blind to medication or placebo and did not personally manage any patient in the study. No algorithm was used, instead they used their best clinical judgement. If there were any disagreements, the 2 raters met to provide a mutually agreeable rating.
Incomplete outcome data (attrition bias) All outcomes	Low risk	50 patients signed consent, 3 did not meet criteria, therefore 47 entered the wash‐in phase. 3 patients withdrew prior to randomisation (1 hospitalised for depression, 1 due to delirium due to urinary tract infection, and 1 had a cardiac arrest). 24 patients were randomised to receive sertraline and 20 placebo. 3 patients withdrew before completion in the sertraline arm (adverse events: 1, lack of efficacy: 2) and 5 in the placebo arm (death: 1, lack of efficacy: 3, withdrawal of consent: 1). These attrition figures are acceptable.
Selective reporting (reporting bias)	Low risk	The prespecified outcome measures were all reported. However, the adverse event data could not be included in our review, because they were presented for every 3‐week period and the numbers could not be combined.
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	"Identical appearing pills."
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	Response to treatment was rated by 2 study psychiatrists (PVR or MS), who reviewed patients' scores on baseline and follow‐up depression rating scales after the study was completed. Rated each patient on 3‐point global scale as a non‐responder, partial responder or full responder. These psychiatrists were blind to medication or placebo and did not personally manage any patient in the study. No algorithm was used, instead they used their best clinical judgement. If there were any disagreements, the 2 raters met to provide a mutually agreeable rating.