Providing evidence of construct validity for the risk of bias indicators in clinical trials on Alzheimer's Disease and other dementia

• Grant number: 16/22586-5
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: February 01, 2017
• End date: January 31, 2018
• Field of knowledge: Health Sciences - Medicine - Psychiatry
• Principal Investigator: Hugo Cogo Moreira
• Grantee: Filipe Nishiyama
• Host Institution: Escola Paulista de Medicina (EPM). Universidade Federal de São Paulo (UNIFESP). Campus São Paulo. São Paulo , SP, Brazil

Abstract

Different systematic reviews have been conducted about de effectiveness and efficiency of interventions on many typology of Alzheimer 's disease (AD) and other dementia; these typologies vary from drugs intervention, like ibuprofen, cannabis and thiamine, to those based on vitamin B6 and B12. On the Cochrane database, a group devoted to provide evidence though systematic reviews, we can find, nowadays, 25 systematic reviews about the treatment of AD and other dementia, totalizing 95 clinical trials. Since the clinical trials may contain different natures of bias (e.g., selection bias [by random sequence generation or allocation concealment], performance bias, detection bias based by patient-reported outcomes, detection bias by all-cause mortality, attrition bias - both short and long term and reporting bias), the Cochrane systematic reviews have the practice to judge the risk of bias trough the items above described, classifying them as Likert scale as following: low, unclear and high risk. There are other ways to judge the risk of bias/clinical trials quality on clinical trials systematic reviews (JADAD, for example). Until now, both mentioned tools do not have studies regarding their construct validity. Construct validity is one type of validity (like convergent validity, divergent validity, criteria validity) that aims to judge how an item of an inventory/scale/battery are assessing a non-directly observed phenomenon (in our case, risk of bias). In other words, a construct validity provide evidence both empiric and theoretical about a tool/questionnaire/inventory. The statistical technique used for this kind of validity is called confirmatory factor analysis. This project aims to provide construct validity evidence to the set of seven indicators proposed by Cochrane to judge the risk of bias. For doing so, we will use the seven Cochrane indicators as observed variables (also called indicators) and underlying them, we will have a factor called risk of bias. Such analytical typology is generally used to judge features/areas/psychic factors and one of the innovations of this project is to use such technique to judge a non-human factor, but a feature found on clinical trials - the risk of bias.We will use Mplus (a specific software to work on latent variables) and the following adequacy index to judge the underlying model of the Cochrane judging items of risk of bias: Dz, Comparative Fit Indice (CFI), Tucker-Lewis index (TLI), Root Mean Square Error of Approximation (RMSEA) e Weighted Root Mean Square Residual (WRMR). (AU)