Authors: David S. Miller, XIngmei Wang, Alan Kott
Published: 2020-12-07
DOI: 10.1002/alz.045353
Source: Full article
AbstractBackgroundAudio recording of eCOA MMSE administration enables identification of both administration and scoring errors. Either can result in incorrect MMSE scores, potentially leading to inclusion of inappropriate subjects and/or biased assessment of change after randomization. Our current analysis explores the effect of protocol design, MMSE scores close to inclusion criteria and visit type on MMSE scoring and/or administration errors.MethodData were pooled from 2 types of early dementia protocols – Type A where MMSE inclusion criteria need only be met at screening, and type B where they needed to be met at both screening and baseline. All assessments were audio recorded and reviewed by independent clinicians for the presence of administration and scoring errors. Operationally, subjects were defined as close to inclusion criteria if their MMSE scores were within 2 points of specified thresholds. Logistic regression with 3‐way interaction was used to explore the hypotheses.Result11,391 MMSE assessments were reviewed, and 2,706 (23.8%) had either an administration and/or scoring error. In both protocol types, administration errors were significantly less likely at baseline than screening, regardless of MMSE score proximity to inclusion threshold. In type A protocols, administration errors were more likely if scores were close to inclusion criteria. For type B protocols this was true only at screening. Scoring errors were significantly less likely at baseline, if MMSE scores were close to inclusion (both protocol types) or not (type A). Scoring errors were significantly more likely at screening but not baseline when scores were close to inclusion (only type A). No direct effect of protocol design was observed on the likelihood of either scoring or administration errors.ConclusionOur data indicate that scoring and administration errors occur more frequently at screening and in subjects with scores close to inclusion criteria. Despite some differences, the effect of protocol design seems less likely to result in errors to the same extent as the other 2 parameters. Given that data coming from subjects close to inclusion criteria at screening are at the highest risk of having a scoring or administration error indicates that data may be manipulated to meet inclusion criteria.