A comparison of five clock scoring methods using ROC (receiver operating characteristic) curve analysis

Objective To compare the accuracy of five clock scoring methods for detecting dementia in English-speaking patients. Design A prospective cohort study. Setting A general geriatric outpatient clinic in southwest Sydney, Australia. Participants A total of 127 consecutive new referrals to the clinic, of mean age 78.2 years. Measurements The clock drawing test was conducted at the beginning of each clinic appointment by a blinded observer. Each patient was then assessed by a geriatrician, who collected demographic data, administered the modified Barthel index, the geriatric depression scale, and the Folstein Mini-Mental State Examination, and categorised each patient as demented or not demented, according to DSM-4 criteria. Each clock was scored according to the methods of Mendez, Shulman, Sunderland, Watson and Wolf-Klein, and evaluated for reliability, and predictive accuracy, using receiver operating characteristic (ROC) curve analysis. Results The area under the ROC curve was largest for the Shulman (0.79, 95% CI 0.70-0.85) and Mendez (0.78, 95% CI 0.70-0.85) methods. Both predicted dementia more accurately than the Sunderland (area = 0.71) and Watson (area = 0.65) methods (p < 0.05). The inter-rater (0.81-0.93) and intra-rater (0.87-0.96) correlation coefficients were high for all five methods. Conclusions While substantial differences among the clock scoring methods were evident in our sample, the accuracy of each was modest at best. Unless further studies in relevant settings suggest otherwise, we caution on the use of clock drawing alone to screen for dementia. Copyright (C) 2001 John Wiley & Sons, Ltd.