Dreo Jurij, Jug Jan, Pavlovčič Tisa, Ogrin Ajda, Demšar Anita, Aljaž Barbara, Agatić Filip, Marusic Uros
BrainTrip, Ljubljana, Slovenia.
Faculty of Computer Science, University of Ljubljana, Ljubljana, Slovenia.
Dement Geriatr Cogn Disord. 2024;53(6):289-298. doi: 10.1159/000540225. Epub 2024 Jul 16.
Recent introductions of disease-modifying treatments for Alzheimer's disease have re-invigorated the cause of early dementia detection. Cognitive "paper and pencil" tests represent the bedrock of clinical assessment, because they are cheap, easy to perform, and do not require brain imaging or biological testing. Cognitive tests vary greatly in duration, complexity, sociolinguistic biases, probed cognitive domains, and their specificity and sensitivity of detecting cognitive impairment (CI). Consequently, an ecologically valid head-to-head comparison seems essential for evidence-based dementia screening.
We compared five tests: Montreal cognitive assessment (MoCA), Alzheimer's disease assessment scale-cognitive subscale (ADAS), Addenbrooke's cognitive examination (ACE-III), euro-coin handling test (Eurotest), and image identification test (Phototest) on a large sample of seniors (N = 456, 77.9 ± 8 years, 71% females). Their specificity and sensitivity were estimated in a novel way by contrasting each test's outcome to the majority outcome across the remaining tests (comparative specificity and sensitivity calculation [CSSC]). This obviates the need for an a priori gold standard such as a clinically clear-cut sample of dementia/MCI/controls. We posit that the CSSC results in a more ecologically valid estimation of clinical performance while precluding biases resulting from different dementia/MCI diagnostic criteria and the proficiency in detecting these conditions.
There exists a stark trade-off between behavioral test specificity and sensitivity. The test with the highest specificity had the lowest sensitivity, and vice versa. The comparative specificities and sensitivities were, respectively: Phototest (97%, 47%), Eurotest (94%, 55%), ADAS (90%, 68%), ACE-III (72%, 77%), MoCA (55%, 95%).
Assuming a CI prevalence of 10%, the shortest (∼3 min) and the simplest instrument, the Phototest, was shown to have the best overall performance (accuracy 92%, PPV 66%, NPV 94%).
近期用于治疗阿尔茨海默病的疾病修饰疗法的出现，为早期痴呆症检测事业注入了新的活力。认知“纸笔”测试是临床评估的基础，因为它们成本低廉、易于实施，且无需脑部成像或生物学检测。认知测试在持续时间、复杂性、社会语言偏见、所探究的认知领域以及检测认知障碍（CI）的特异性和敏感性方面差异很大。因此，进行生态有效且直接的比较对于基于证据的痴呆症筛查似乎至关重要。
我们在大量老年人样本（N = 456，77.9 ± 8岁，71%为女性）中比较了五项测试：蒙特利尔认知评估（MoCA）、阿尔茨海默病评估量表 - 认知子量表（ADAS）、Addenbrooke认知检查（ACE - III）、欧元硬币处理测试（Eurotest）和图像识别测试（Phototest）。通过将每项测试的结果与其余测试的多数结果进行对比（比较特异性和敏感性计算[CSSC]），以一种新颖的方式估计它们的特异性和敏感性。这避免了对诸如痴呆症/轻度认知障碍/对照的临床明确样本等先验金标准的需求。我们认为，CSSC能在排除因不同痴呆症/轻度认知障碍诊断标准以及检测这些病症的熟练程度所导致的偏差的同时，对临床性能进行更符合生态学的有效估计。
行为测试的特异性和敏感性之间存在明显的权衡。特异性最高的测试敏感性最低，反之亦然。比较特异性和敏感性分别为：Phototest（97%，47%）、Eurotest（94%，55%）、ADAS（90%，68%）、ACE - III（72%，77%）、MoCA（55%，95%）。
假设CI患病率为10%，最短（约3分钟）且最简单的工具Phototest被证明具有最佳的总体性能（准确率92%，阳性预测值66%，阴性预测值94%）。
