Sensitivity analyses for data missing at random versus missing not at random using latent growth modelling: a practical guide for randomised controlled trials

• DOI: 10.1186/s12874-022-01727-1
• Publication Date: 2022-09-24T09:02:34Z
• Authors: Andreas Staudt; Jennis Freyer-Adam; Till Ittermann; Christian Meyer; Gallus Bischof; Ulrich John; Sophie Baumann
• Keywords: Adult; Male; Medicine (General); Growth curve model; Participant attrition; Dropout; Research; MNAR; 01 natural sciences; MAR; R5-920; Data Interpretation, Statistical; Humans; Female; 0101 mathematics; Randomized Controlled Trials as Topic

Abstract Background Missing data are ubiquitous in randomised controlled trials. Although sensitivity analyses for different missing mechanisms (missing at random vs. not random) widely recommended, they rarely conducted practice. The aim of the present study was to demonstrate assumptions regarding mechanism trials using latent growth modelling (LGM). Methods Data from a brief alcohol intervention trial used. sample included 1646 adults (56% female; mean age = 31.0 years) general population who had received up three individualized feedback letters or assessment-only. Follow-up interviews were after 12 and 36 months via telephone. main outcome analysis change use over time. A three-step LGM approach First, evidence about process that generated accumulated by analysing extent values both conditions, patterns, baseline variables predicted participation two follow-up assessments logistic regression. Second, models calculated analyse effects These assumed applied full-information maximum likelihood estimation. Third, findings safeguarded incorporating model components account possibility random. For purpose, Diggle-Kenward selection, Wu-Carroll shared parameter pattern mixture implemented. Results true generating remained unknown, unequivocal: control group reduced their time, but no significant differences emerged. There clear efficacy, neither be nor those Conclusion illustrated allows assessment how sensitive conclusions efficacy an mechanism. researchers familiar with LGM, it is valuable statistical supplement safeguard against nonignorable missingness. Trial registration PRINT prospectively registered German Clinical Trials Register (DRKS00014274, date registration: 12th March 2018).