Summary of main findings
FPs labeled 3.6 % of older persons with ‘probable dementia’, 8.2 % with ‘possible cognitive impairment or dementia’, 83.7 % ‘no signs of cognitive impairment’. They could not label the remaining 4.5 % of elderly. Persons labelled ‘possible cognitive impairment or dementia’ had more comorbidities and were more often iADL dependent than persons labelled ‘no signs of cognitive impairment’, irrespective of their age and CAMCOG score.
The CAMCOG did not confirm the presence of either dementia or aMCI in half of the persons in whom FPs considered cognitive impairment possibly present. The CAMCOG indicated some form of cognitive impairment in 12.5 % and suggested dementia in 10 % of persons in whom FPs were not aware of any such signs.
If we translate these figures to the average practice in this study with 3000 patients of whom 16.6 % (n = 498) are above the age of 65, FPs would suspect cognitive impairment in 41 persons. The CAMCOG would confirm this suspicion in 19, and would not confirm it in 22 persons. FPs would be unaware of cognitive impairment in 52 persons in which the CAMCOG would suggest such impairment.
There was substantial variation in the percentage of persons per cognitive category between FPs. Since the numbers were too small to provide estimates on PPV and NPV at practice level, it was not possible to determine the impact of case-mix differences on PPV and NPV variation between FPs. FPs who provided care for their population for less than three years could not classify the cognitive status of 35 to 40 % of their older patients.
Awareness of cognitive impairment was associated with ADL- and with iADL dependency; unawareness of cognitive impairment was associated with higher age.
Strengths and limitations
By using a nested design we guaranteed an accurate representation of the source population in our sample and facilitated correction for age and sex [
9,
10]. Furthermore, the method used appears to be appropriate to estimate FPs’ actual awareness or unawareness of dementia and aMCI, since not having to discuss their suspicion with the patient or document it in the medical records likely reduced barriers to identifying someone as possibly cognitively impaired.
A limitation of this study is the low response rate. We found that respondents were younger and more often male than non-respondents, although we could not demonstrate selective non-response related to cognitive status. Nevertheless, since age is strongly associated with cognitive impairment, correction for age may have reduced such bias if present. Previous research demonstrated a lower response rate among individuals with more cognitive impairment [
21]. If present, such a selection would result in underestimation of the PPV but also of the 1-NPV.
Another potential limitation was the time, on average almost nine months that elapsed between the index and reference test. Nevertheless, this delay did not seem to have impact on the estimate of the FPs’ accuracy in our longitudinal exploration.
The cognitive classification by the FPs had no absolute demarcation between categories and thus leaves room for different interpretations. This may have occurred if FPs doubted between ‘possible’ and ‘probable’ dementia. We cannot rule out that there were some false positives among elderly labeled with ‘probable dementia’.
The CAMCOG was validated in a population-based, purposive sample of older people with a relative overrepresentation of high age and cognitive impairment based on MMSE scores [
22]. As a result, the positive predictive value of the CAMCOG for dementia was likely lower in our study population. Results from another population based study support the hypothesis that the CAMCOG may overestimate the prevalence of dementia [
23,
24]. Since we used the CAMCOG as a reference standard, the potential effect on our results would be some underestimation of the PPV of FPs’ cognitive labels but an overestimation of the NPV of their labels. In addition, since the CAMCOG was only validated to measure amnestic MCI and not non-amnestic MCI, our results include only the amnestic MCI group. Further, we found substantial shifts in cognitive classification based on the CAMCOG between baseline and 1-year follow-up. This may have several causes: it may be a result of test-retest characteristics of the CAMCOG, the drop in the cut-off scores at the age of 75 (which caused a category shift in three respondents), the included population with predominantly earlier stages of cognitive impairment, and also treatment of specific conditions (e.g. heart failure) which may improve cognitive functioning in between tests may also explain improved CAMCOG scores [
22,
23]. However, since the total rate of CAMCOG scores suggestive of dementia and MCI remained stable we presume that these shifts had no substantial effects on our findings.
One should take into account the impact of the prevalence of a condition on the PPV and NPV. If, for example, the prevalence is very high, the PPV will become higher. Thus, our findings are prevalence specific and their external validity is limited to populations with a similar prevalence of cognitive impairment.
As we focused on persons without a firm diagnosis of dementia and did not verify patients labeled with ‘probable dementia’, we could not compute classical diagnostic accuracy measures (e.g. sensitivity and specificity), limiting the comparability of our results [
1,
6].
Comparison with existing literature
A new finding is that the likelihood of missing cognitive impairment was higher in older elderly. Particularly at earlier stages, FPs may incorrectly attribute signs of cognitive decline (e.g. dependency, forgetfulness) to normal ageing. Conversely, FPs may be preoccupied with the multiple chronic conditions of their older patient, and perhaps overlook diffuse cognitive impairments if present. Further, we confirmed the previously described association between awareness confirmed by a reference test and higher ADL dependency, showing that also at earlier stages, dependency is an important clue to cognitive impairment [
25].
In contrast to previous findings, living alone was not significantly associated with unawareness of present cognitive impairment in our study [
26]. This may reflect that relatives usually wait to ‘help’ FPs become aware until cognitive impairment reaches a more advanced stage. Neither did we find the association between detection and contact frequency that was demonstrated previously [
7]. The milder cognitive impairment and thus more subtle symptoms and possibly less variation in contact frequency of the studied group might explain this.
Interpretation
We chose the combined reference test outcome ‘dementia or aMCI’ for the primary comparison with the index test because the identification of dementia starts with the recognition of cognitive decline. However, the clinically most relevant reference test for any index test would - completely accurately -indicate those patients who would benefit from treatment of the target condition. In the case of dementia this is not straightforward. There is no curative treatment and pharmacological treatment has only temporary and small effects on cognition and sometimes ADL (and not on other outcomes, like global impression) and is often accompanied by side-effects if tolerated at all [
27,
28]. Still, there are reasons to establish this diagnosis: most people would like to know about it, support can be offered and care arranged, the diagnosis may influence medical decisions (e.g. the use of medication adherence aids) and it may affect the extent to which patients can be engaged in such decisions [
29‐
32]. To interpret the different reference test outcomes and to link these outcomes to clinical actions, we would suggest that further diagnostic evaluation should be considered in all persons with a CAMCOG score below the cut-off for dementia, even more so when this is accompanied by iADL dependency. When the CAMCOG score suggests only aMCI, the benefit of further diagnostic evaluation is, due to its variable course, not straightforward in the primary care setting [
33]. Still, there are indications that persons with MCI are more vulnerable and FPs may consider a more pro-active approach towards these patients necessary [
34‐
37].
That a substantial proportion of persons in whom FPs’ suspected cognitive impairment scored well on the CAMCOG suggests that FPs find it hard to distinguish between earlier stages of cognitive impairment and normal ageing. This finding contributes to our understanding of un-established dementia diagnoses in primary care. Unsureness about whether cognitive deficits are at all present may increase the barrier to raising the issue during regular consultations with the patient or family. In contrast, although a decreased level of functioning helps FPs to become aware of cognitive impairments, the presence of such functional decline may also be erroneously interpreted as a sign of cognitive deterioration.
That a substantial proportion of persons in whom FPs’ had not noticed signs of cognitive impairment had a CAMCOG score suggestive of aMCI or dementia suggests that complete unawareness of cognitive impairment is a major explanatory factor for missed diagnoses of dementia in primary care. However, this should be seen in context: awareness is not only a result of how capable FPs are of recognizing cognitive impairment, but also of whether or not patients and their relatives bring it to bear (e.g. consultation frequency, discussing versus masking memory impairment) and of how the healthcare system is organized (e.g. time available per patient contact).
Still, FPs’ awareness of cognitive impairment is likely influenced by their own knowledge about symptoms and signs of dementia and by their attitude regarding the value of such awareness. This attitude shows quite some variation: some FPs value establishment of a diagnosis as a very important part of their function towards the patient and family, whereas others feel that provision of good care does not require a diagnosis, or that a diagnosis may even be harmful to the patient [
25,
38‐
40].
Implications for practice and future research
A vast body of evidence shows that, although patients’ receptiveness to discuss this topic will vary, the large majority will appreciate an open attitude to disclose suspicions of dementia [
26]. Relatives, although not unequivocally in favour of disclosure towards the patient, do usually value clarity about the presence of dementia as well [
41]. This suggests that FPs may be less hesitant to discuss suspected cognitive impairment.
At present, most guidelines on the diagnosis and management of dementia do not provide guidance for practitioners on how to diagnose, inform about and manage cognitive problems other than dementia, like MCI [
42,
43]. We assume that if such guidance would be available, this would decrease barriers for FPs to discuss suspicions of cognitive impairment.
Like with other missed diagnoses, not considering the diagnosis at all may be an important cause of FPs’ limited awareness of cognitive impairment. In this respect, it would be interesting to explore how FPs awareness would develop if they would be primed to the possibility of cognitive impairment, for example by performing the classification procedure every year. As a results FPs may also become more alert to other signs of dementia like mistakes with repeat medication, gait disturbances, weight loss or apathy [
44‐
46].
Awareness may also be improved by measures not directly involving the FP. Other PCP co-workers may be stimulated to communicate about signs of cognitive impairment. Public campaigns aimed to reduce fear and stigma and to inform about available services may help patients or their relatives seek professional help sooner [
47‐
49].