The purpose of cognitive assessment is to detect and measure impairment in any cognitive domain (attention, memory, language, executive function, praxis, visuospatial skills), using these findings in the diagnostic and treatment process. In some cases (for instance in consultations due to cognitive complaints, which are increasingly frequent1) this assessment is a fundamental part of neurological examination.2

Cognitive assessment may be informal: during medical history taking, the physician may detect disorientation, impaired language expression, marked memory problems,3 the “don’t know” sign,4 or the head turning sign,5 all of which suggest cognitive impairment (CI). However, assessment is usually performed with validated instruments for structured evaluation of the patient’s cognitive function. Like such other instruments as the ophthalmoscope or the reflex hammer, these cognitive instruments and tests must meet certain technical requirements, and the physician must be proficient in their use and the interpretation of results. In clinical practice, given the unavoidable limitations on consultation times, these instruments must necessarily be quick and simple to administer. Furthermore, these instruments should evaluate all the cognitive domains currently considered in the diagnostic criteria for different constructs of CI (attention, memory and learning, language, executive function, praxis, and visuospatial skills)6–8 and, needless to say, present good psychometric properties, like any other diagnostic tool.9 These multi-domain brief cognitive tests (BCT) should be validated specifically for the diagnosis of CI (and not only for dementia); otherwise, the tool would be unable to detect cognitive problems at a stage when the appropriate intervention may either revert the problem or delay or halt dementia progression.10 Normative data applicable to the population of interest should also be available for these BCTs, and these tools should ideally be available free of charge.11

Among the multi-domain BCTs validated in Spain for CI (Supplementary material, Table 1), only the MoCA12,13 and the ACE-III14,15 evaluate all cognitive domains. However, these tests require 10–15 and 15–20 minutes to administer, respectively; these administration times are excessive for general neurology consultations,16 and impracticable in primary care.17 Although the Mini-Mental State Examination,18 Eurotest,19 Fototest,20 Mini-Cog,21 and Clock Drawing Test22 present shorter administration times, they do not assess all cognitive domains. Furthermore, the Mini-Mental State Examination presents many other limitations that have led to a decline in its use in recent years.23

Physicians frequently administer several BCTs to evaluate more cognitive domains, increasing the sensitivity and diagnostic accuracy of these tests in cases of CI. The most widely known examples are the 7 Minute Screen, which combines an orientation task, a cued recall task, a semantic verbal fluency task, and the Clock Drawing Test24; and the Mini-Cog, which includes a delayed recall task and the Clock Drawing Test.21

The Fototest20 and the Mini-Cog21 are 2 very short tests (administration times of 3 and 2 minutes, respectively) that are widely used in Spain and have been validated specifically for CI. The former evaluates language, executive function, and episodic visual memory with free and cued recall, whereas the latter assesses attention, verbal memory, and visuospatial skills and visuoconstructive praxis. This study explores the hypothesis that the combination of the Fototest and the Mini-Cog, which together evaluate all the cognitive domains that should be included in a cognitive assessment, improves the diagnostic accuracy of either test alone while requiring very little time to administer (< 5 minutes), a particularly interesting feature in our setting.

Our objective was to evaluate the diagnostic accuracy and predictive validity of the combination of Fototest plus Mini-Cog in detecting CI in everyday clinical practice.

The study sample included 448 individuals with a mean age (SD) of 60.5 (17.7) years, with a slight predominance of women (53.8%) and individuals with more than primary education (52.2%); the prevalence of CI in this sample was 46.4%. The 2 randomly generated datasets (base and test, including 358 and 90 individuals, respectively) showed no significant differences in any of the sociodemographic variables studied, in the prevalence of CI, or in Fototest and Mini-Cog results.

The external sample, in contrast, presented a much higher prevalence of CI (77.0%), an older mean age (71.7 [10.0] years), a higher education level (60.7% of individuals had at least secondary education), and a lower prevalence of women (47.5%), although differences were not significant in the latter 2 variables. Fototest and Mini-Cog results were significantly lower in the external sample than in the study sample.

Table 1 summarises the sociodemographic characteristics, prevalence of CI, and Fototest and Mini-Cog results of the 2 samples and the 2 datasets.

The random decision tree model selected (among the 98 models generated by the platform) had an accuracy of 92.2% and a phi coefficient of 0.84 in the test sample, with a total of 7 levels (Fig. 1) (Supplementary Material, Table 3).

Figure 1.

Random decision tree model.

CI: cognitive impairment; FR: Fototest free recall score; LR: logistic regression; MC_rec: Mini-Cog recall score; no CI: no cognitive impairment; TR: Fototest total recall score.

(0.14MB).

Table 2 summarises the diagnostic parameters of the different models in the different samples and datasets. The diagnostic accuracy of the Fototest was moderate, but significantly higher than that of the Mini-Cog (0.93 ± 0.01 [AUC ± SE] vs 0.89 ± 0.02; t = 2.06; P = .04), while the diagnostic accuracy of the Comb-simple (0.95 ± 0.01) and Comb-LR models (0.98 ± 0.01) was significantly higher than that of either test alone (t > 4.00 and P < .0001 for all comparisons) (Fig. 2, Table 3). The Comb-DT model does not have an AUC, since it only includes one alternative for classification; however, its accuracy (95.2%) was significantly higher than that of the Mini-Cog (81.6%; χ2 = 32.24; P < .0001), the Fototest (84.9%; χ2 = 21.16; P < .001), and the Comb-simple (88.3%; χ2 = 11.2; P = .001) and Comb-LR models (91.6%; χ2 = 3.76; P = .05). Stepped models present the exact same diagnostic accuracy as the combined models they include, with the added benefit of significantly shorter administration times (197.3 [56.7] vs 233.9 [45.2]; t = 8.8; P < .0001).

Figure 2.

Diagnostic accuracy of the Fototest and Mini-Cog, used alone or in combination.

Comb-LR: logistic regression combined model; Comb-simple: simple combined model.

(0.12MB).

The validation of the models in the test dataset largely replicates the predictions of the models generated in the base dataset, except for the models using random decision trees, which present a slightly lower diagnostic accuracy, as may be expected due to the overadjustment effect. The models also showed a very high diagnostic accuracy in the external sample, with accuracy values ranging from 86.9% to 91.8% (except for the Mini-Cog, at 78.7%), although the difference between the external sample and the base dataset was not statistically significant (81.6%; χ2 = 0.29; P = .59).

The Supplementary Material (Tables 4–11) presents the contingency tables and classical measures of diagnostic accuracy (sensitivity, specificity, predictive values, likelihood ratios) for each model and in each sample for the selected cut-off points and criteria.

The results of this prospective study, which was conducted under conditions of routine clinical practice, clearly show that the combined use of the Fototest and Mini-Cog surpasses the diagnostic accuracy of either instrument alone. These results are consistent, as they were replicated in 2 independent samples: one with the same sociodemographic and clinical characteristics as the sample used to develop the models, and another sample that was completely different in terms of sociodemographic and clinical characteristics (drawn from a cognitive-behavioural neurology consultation, older age, higher prevalence of CI, and different geographical location). The greater diagnostic accuracy of the combined models is probably due to greater comprehensiveness of the cognitive assessment, as these models evaluate more cognitive domains. This, in turn, facilitates the identification of different profiles of CI, which could enhance diagnostic studies. However, our study did not address this issue, which should be explored in future research.

Our results are also consistent with and support the available data from Spain regarding the diagnostic accuracy of the instruments used. The only previous study of Mini-Cog in Spain reported an AUC of 0.88 (SE: 0.01),29 which is practically identical to that found in our study (0.89 [0.02]); one study reconstructed the Mini-Cog score from the Clock-Drawing Test and the Mini–Mental State Examination, so we do not consider it comparable.28 Similarly, the diagnostic accuracy of the Fototest in the present study (0.93 [0.01]) is very similar to that reported in previous studies in Spain (0.86 [0.02],48 0.93 [0.02],49 0.95 [0.01]20) and even in Argentina (0.93 [0.03]50).

The complex models (Comb-LR and Comb-DT) provide higher diagnostic accuracy than the Comb-simple model, probably due to the weighting of diagnostic contributions from each test and the inclusion of sociodemographic variables (sex, age, education level) in these models. These variables can be especially relevant, particularly in the case of the Mini-Cog.29 These complex models have the disadvantage that they involve a computational process that may be difficult to apply during consultations. The simple model, in contrast, is more practical for use in the clinical setting, as it simply requires adding the scores from both instruments. However, one of the authors (RVC) developed a program to facilitate the application of these instruments, incorporating a calculator that performs these computations automatically, providing a report that includes all results (Supplementary Material, Fig. 1). This application, Predi-Cog51 (https://www.hipocampo.org/Predi-Cog.asp), is available online, free of cost, at www.hipocampo.org.52

Based on our results, we recommend applying the tests in a stepped manner (Fig. 3), which allows direct classification of a substantial percentage of individuals, without compromising diagnostic accuracy, using only the Fototest (base dataset, 36.6%; test dataset, 37.8%; external sample, 42.6%), saving time (197 [56.7] vs 233.9 [45.2] seconds).

Figure 3.

Recommended strategy for combined use of the Fototest and Mini-Cog.

CI: cognitive impairment; No CI: no cognitive impairment.

(0.15MB).

Our study has some limitations, such as the exclusion of individuals without at least primary education, which limits the possibility of extrapolating our results to the population with lower education levels. In any case, our previous data already recommended avoiding the use of Mini-Cog in this population.28,29 Another Fototest weakness is that only 63.9% of the individuals in the study sample underwent formal cognitive assessment. However, we only excluded individuals without cognitive complaints and with Fototest scores above the 10th percentile, which makes it unlikely that these individuals would have CI, thus minimising the risk of partial verification bias.53 Lastly, while the index and reference assessments were performed by different professionals who were blinded to other results, and diagnosis was based on reference test results, the professional who issued the diagnosis was not blinded to the results of the BCTs. Our study also has several strengths, such as its naturalistic, pragmatic nature, the large sample size, and, most importantly, the double validation of the results in 2 independent samples with very different characteristics.

In conclusion, the combined use of the Fototest and Mini-Cog, 2 very brief tests that are simple to administer and that evaluate all the cognitive functions that should be addressed in a brief cognitive assessment, enhances the already high diagnostic accuracy of both instruments. Stepped use of these tests is more efficient than their combined use, since it achieves a significant decrease in mean administration time (< 200 seconds) while maintaining accuracy.

The data that support the findings of this study are available upon reasonable request.

C. Carnero Pardo is the creator of the Fototest. Under the terms of its Creative Commons licence, the Fototest may be used and distributed for non-commercial purposes provided that it is not modified and its authorship is explicitly acknowledged.