MMSE in primary care practice: why good tests can mislead in the wrong context.

The Mini-Mental State Examination (MMSE) is widely used in cognitive screening, including in primary care settings. This study evaluated the diagnostic performance of the MMSE using data from 390 community-dwelling older adults aged 65 to 98 years (M = 75.76, SD = 6.76). The MMSE's accuracy was assessed against clinical diagnoses and symptom severity levels based on the Global Deterioration Scale (GDS). Receiver operating characteristic (ROC) analysis, evaluating the ability to distinguish individuals with dementia from those without, provided an area under the curve (AUC) of .75 (95% CI: 0.67-0.84, p < .001). The optimal cutoff based on Youden Index was 21 and resulted in a sensitivity of .77 (95% CI [.728, .812]) and specificity of .65 (95% CI [.603, .697]), whereas the more conventional cutoff (24) showed lower sensitivity (.50; 95% CI [.450, .550]) but higher specificity (.82; 95% CI [.782, .858]). At the suggested cutoff, the MMSE identified all cases at the severe stage, 88% at the moderate stage, and 31% at the mild stage of dementias, as classified by the GDS. In contrast, the Quick Mild Cognitive Impairment screen (Qmci) identified nearly all cases across severity levels. Against previous dementia diagnoses, when employing a cutoff score of 24 the MMSE had a positive predictive value of .52 (95% CI [.395, .645]) and a negative predictive value of .81 (95% CI [.726, .894]), indicating modest diagnostic reliability in a primary care context. Similar results were obtained applying a cutoff score of 21. These findings highlight how base rates and test characteristics shape test accuracy and should guide decision-making. Overall, our findings highlight that the MMSE can produce a substantial number of false positives in contexts with a relatively low prevalence of dementia, such as primary care, challenging the common assumption of its low false-positive rate. More broadly, our study emphasizes the importance of considering the prevalence of the condition in a given context, as differences in prevalence can drastically affect the interpretation of results, particularly the positive predictive value, even when sensitivity and specificity remain unaffected.