In this 13-year longitudinal study of 247 patients with mild-to-moderate AD, we found that increased cerebral inflammation was the only AD-related pathology that was an independent risk factor for death caused specifically by dementia (Table 3). Other significant risk factors were low cognitive ability, frontal atrophy, and medial temporal lobe atrophy. Tau, amyloid, or cerebrovascular pathology were not significant variables in the Cox regression model for death caused by dementia.
To the best of our knowledge, the present study is the first to show that cerebral inflammation is independently associated with early death in AD. Cerebral inflammation, as measured by CSF sVCAM-1 level, was a risk factor for early death in all mortality groups (Table 3). Previous studies reported that neuroinflammation is involved in the pathogenesis of AD [18, 30–33]. It has been suggested that chronic neuroinflammation represents a risk factor for AD in the elderly through an acceleration of senescence in microglia, with a reduction in their neuroprotective function leading to the formation of senile plaques and neurofibrillary tangles . A recent neuropathologic study found that cerebral inflammation measured as glial response is an essential mechanism that determines which of the patients with a high load of tau and amyloid pathology were demented or nondemented . Thus, inflammation might be a key component in the toxicity of β-amyloid, as many individuals have a high burden of amyloid and tau pathology without any clinical symptoms . Furthermore, Forassi et al.  showed that systemic inflammation is a strong independent risk factor for death in an elderly population.
The other significant risk factors identified (lower cognitive status, frontal atrophy, and medial temporal lobe atrophy) should probably be interpreted as downstream biomarkers, which reflect the disease duration and disease stage. Other studies have also found that low cognitive function and atrophy at baseline are significant risk factors for mortality [36–39]. Frontal atrophy was a stronger predictor than medial temporal atrophy (Table 3), possibly because medial temporal atrophy is an earlier step in the disease process, whereas frontal atrophy occurs later, as an indicator of end-stage dementia . Another explanation could be that AD patients with frontal lobe degeneration constitute a subset of AD with a more rapid disease progression. This assumption is supported by previous results, that revealed that tests of frontal cognitive function, but not those of parietal and temporal function, independently predicted a more rapid AD progression .
The results of the subanalyses showed that AD-related factors, but not concomitant diseases or medications, caused early death in AD (Table 4). This was further supported by the fact that neither older age nor male sex predicted specific-cause death (Table 4), even though they are regarded as traditional predictors of death among elderly individuals.
Neuroleptic/sedative drugs were associated with early death in the all-cause model, which was in line with previous studies that showed that psychotic symptoms and/or neuroleptic drugs increase mortality rates [4, 41, 42]. The mean survival time of 6.4 years found here was in agreement with a large review of 42 studies on AD mortality . Large differences are found in the prevalence of death causes, but overall, pneumonia and dementia have been reported most commonly [3, 11, 13, 44], which differs somewhat from the results of the present study (Figure 1). We believe this can be explained by the manner in which the cause of death was entered on the death certificate, rather than by an actual difference in death causes. If no other specific diagnosis was apparent at the time of death, the attending physician most likely regarded dementia as the cause of death.
One of the strengths of this study was that it constitutes the world’s longest longitudinal study with CSF biomarkers of AD, which ensures a high diagnostic accuracy and a low number of censored cases (patients still alive at follow-up). We also examined predictors of specific-cause death due to dementia, and not only all-cause death, which is essential when trying to map key mechanisms of mortality in AD. Other advantages of our study included the multivariate design of the analysis, which allowed us to examine independent effects of the most common pathologies and hallmarks of AD, and the inclusion of patients from a routine clinical setting.
We chose to measure cerebral inflammation by using sVCAM-1 levels in the CSF, based on previous validating studies [45–50]. sVCAM-1 is a cell-surface molecule that regulates leukocyte migration and is activated by proinflammatory factors such as cytokines and interleukins . Moreover, it has been associated with glial activation (expressed on astrocytes) . Numerous studies have associated increased levels of sVCAM-1 with several neuroinflammatory conditions, and it has been significantly correlated with inflammatory cerebral lesions assessed by using contrast MRI [45–50]. sVCAM-1 is also a well-proven marker of endothelial dysfunction and has been associated with atherosclerosis  and small-vessel disease . However, based on our results, we believe that sVCAM-1 is related more directly to neuroinflammation, rather than to cerebrovascular pathology, as sVCAM-1, not vascular lesions and BBB integrity, was an independent significant variable in the main analysis performed to predict death caused by dementia (Table 3). Vascular lesions and BBB integrity were, however, separately significant when examining death also caused by cerebrovascular disorders, which validates them as adequate markers of vascular pathology in this data set.
To warrant our conclusions, the results of this exploratory study should be tested in larger longitudinal cohorts including additional factors that are associated with shorter survival (for example, diabetes) and using other markers of inflammation such as IL-1 on either the protein or mRNA level.