• infarction
• stroke
• CT
• dementia
• diagnosis

Quis custodiet ipsos Custodes (who is to guard the guards themselves?).¹ What is the link between this rather pithy observation by a first century AD Roman poet and current research on diagnosing dementia? Read on!

The paper by Tian et al (this issue, pp 433–438)² explores the clinical utility of predictive testing of individuals with early cognitive problems. The hinterland between normal cognition and dementia is a nosological and terminological minefield. The problem is that before people develop full-blown dementia, they often pass through a stage of “pre-dementia”, referred to in Tian et al’s article as “questionable dementia”. Mild cognitive impairment, benign senescent forgetfulness, and age associated memory impairment are other, subtly different re-workings of the same phenomenon. In Tian et al’s paper, questionable dementia is defined as either the presence of subjective memory impairment affecting social or occupational functioning, but in the absence of impairment in other facets of cognition; or cognitive deficits in one or more domains without impairment of functioning. However, not everyone with the pre-dementia syndromes will go on to develop full-blown dementia, nor are all cases of dementia presaged by this phase.

Accurate early diagnosis of dementia is becoming increasingly important. In people with established dementia, early treatment with cholinesterase inhibitors may confer greater benefit than delayed treatment³ and provides the patient and family with a longer lucid time to make plans, do the undone, discuss treatments, and come to terms with the diagnosis.

The most useful clinical measure for patients is the positive predictive value (PPV). This answers the question “If I test positive, what is the likelihood that I will develop (or have) the condition the test is testing for?” The problem is that as the PPV rises, the negative predictive value (If the test is negative, what is the likelihood I will not have the disease) falls. In this case, Tian and co-workers found PPVs around 85% and negative predictive values around 45%. As they point out in their shrewd discussion, these figures are not helpful for most patients, who prefer to deal in certainties.

A significant issue with all research in this area is having a robust reference (gold) standard to diagnose dementia and evaluate the usefulness of your screening test. Most researchers rely on old favourites such as the DSM or NINCDS, although clinicians may be less rigid. When calculating parameters such as sensitivity and PPV, one has to regard these gold standards as inviolable. They are not. In fact, the sensitivity of antemortem diagnosis using these instruments in centres of excellence is at best around 90% when compared with the next hierarchy; histological examination.⁴ But who is to say that histology is always accurate (it cannot be, no test is). One can rapidly descend into a mire of iterative complexity and nosological uncertainty. The parallels between reference diagnostic standards and the Roman guard become obvious and we are left having to trust what we have, but feeling a little uneasy about it.