Inter-rater and test-retest reliability of quality assessments by novice student raters using the Jadad and Newcastle-Ottawa Scales

Introduction: Quality assessment of included studies is an important component of systematic reviews. Objective: The authors investigated inter-rater and test-retest reliability for quality assessments conducted by inexperienced student raters. Design: Student raters received a training session on quality assessment using the Jadad Scale for randomised controlled trials and the Newcastle-Ottawa Scale (NOS) for observational studies. Raters were randomly assigned into five pairs and they each independently rated the quality of 13-20 articles. These articles were drawn from a pool of 78 papers examining cognitive impairment following electroconvulsive therapy to treat major depressive disorder. The articles were randomly distributed to the raters. Two months later, each rater re-assessed the quality of half of their assigned articles. Setting: McMaster Integrative Neuroscience Discovery and Study Program. Participants: 10 students taking McMaster Integrative Neuroscience Discovery and Study Program courses. Main outcome measures: The authors measured inter-rater reliability using kappa and the intraclass correlation coefficient type 2,1 or ICC(2,1). The authors measured test-retest reliability using ICC (2,1). Results: Inter-rater reliability varied by scale question. For the six-item Jadad Scale, question-specific kappa s ranged from 0.13 (95% CI -0.11 to 0.37) to 0.56 (95% CI 0.29 to 0.83). The ranges were -0.14 (95% CI -0.28 to 0.00) to 0.39 (95% CI -0.02 to 0.81) for the NOS cohort and -0.20 (95% CI -0.49 to 0.09) to 1.00 (95% CI 1.00 to 1.00) for the NOS case-control. For overall scores on the six-item Jadad Scale, ICC(2,1)s for inter-rater and test-retest reliability (accounting for systematic differences between raters) were 0.32 (95% CI 0.08 to 0.52) and 0.55 (95% CI 0.41 to 0.67), respectively. Corresponding ICC(2,1) s for the NOS cohort were -0.19 (95% CI -0.67 to 0.35) and 0.62 (95% CI 0.25 to 0.83), and for the NOS case-control, the ICC(2,1) s were 0.46 (95% CI -0.13 to 0.92) and 0.83 (95% CI 0.48 to 0.95). Conclusions: Inter-rater reliability was generally poor to fair and test-retest reliability was fair to excellent. A pilot rating phase following rater training may be one way to improve agreement.