Disease-related determinants are associated with mortality in dementia due to Alzheimer’s disease

In this longitudinal study, we included 616 patients with a baseline diagnosis of dementia due to AD from a memory clinic-based cohort (the Amsterdam Dementia Cohort) who had a baseline visit between 2000 and 2014 [23]. Subjects were selected if a neuropsychological test battery was available at baseline, with a baseline Mini Mental State Examination (MMSE) score ≥ 16 and a minimum follow-up of 2 years. At baseline, patients received a standardized and multidisciplinary workup, including medical history; physical, neurological, and neuropsychological examinations; MRI; laboratory tests; and lumbar puncture for CSF measurements. Years of education and self-reported duration of complaints were recorded. For the assessment of activities of daily living, we used the Disability Assessment for Dementia (DAD) [24]. We included all data that were collected within 6 months of baseline diagnosis. Diagnoses were made in a multidisciplinary consensus meeting. Patients were diagnosed with probable AD using the criteria of the National Institute of Neurological and Communicative Disorders and Stroke/Alzheimer’s Disease and Related Disorders Association; all patients also met the core clinical criteria of the National Institute on Aging-Alzheimer’s Association for AD dementia [25, 26].

We recorded and defined the presence (yes/no) of hypertension (history of hypertension and/or use of antihypertensive drugs), hypercholesterolemia (history of hypercholesterolemia and/or use of cholesterol-lowering drugs), diabetes (history of diabetes mellitus and/or use of antidiabetic drugs), and cardiovascular disease (at least one of the following: history of coronary heart disease, heart failure, heart disease, peripheral vascular disease, stroke, and/or transient ischemic attack). Furthermore, we dichotomized smoking status (never smoked versus current or history of smoking) and counted the medications used per patient.

Cognitive function was assessed at baseline with a standardized test battery in which the MMSE was used for global cognitive functioning [27]. For memory, the Visual Association Test (VAT) and the Rey Auditory Verbal Learning Task (RAVLT) were included [28, 29]. To measure mental speed and attention, we used Trail Making Test A (TMT-A) and the forward condition of the digit span. Trail Making Test B (TMT-B) and the backward condition of the digit span were used for executive functioning [30, 31]. Language and executive functioning were tested by category fluency (animals) [32]. Missing data ranged from n = 19 (3%) (digit span forward) to n = 67 (11%) (RAVLT, delayed recall).

Subjects were scanned as part of clinical workup using a standardized protocol on a 1.0-, 1.5-, or 3.0-T system. All scans were visually rated by trained raters and subsequently evaluated in a consensus meeting with an experienced neuroradiologist [23]. Visual rating of medial temporal lobe atrophy (MTA) was performed using coronal T1-weighted images on a 5-point (0–4) scale from the average score of the left and right sides [33]. Global cortical atrophy (GCA) was assessed visually on axial fluid-attenuated inversion recovery (FLAIR) images (range of scores 0–3) [34]. Parietal atrophy was rated using T1-weighted and FLAIR weighted images viewed in sagittal, axial, and coronal planes by computing an average score of the left and right sides (range 0–3) [35]. WMH were rated on axial FLAIR images using a four-step scale (range 0–3) [36]. Lacunes were defined as deep lesions (3–15 mm) with CSF-like signals on all sequences and were dichotomized as present or absent. Microbleeds were defined as small, round foci of hypointense signal up to 10 mm in brain parenchyma on T2*-weighted gradient echo images. The total number of microbleeds was counted and divided into three categories: zero, one or two, and three or more microbleeds. MRI data were available for 485 (79%) subjects.

CSF analyses were performed at the Neurochemistry Laboratory at the Department of Clinical Chemistry of the VUmc. CSF was obtained by lumbar puncture between the L3-L4 or L4-L5 intervertebral space by using a 25-gauge needle and collected into polypropylene tubes. Within 2 h, the CSF was centrifuged at 1800 × g for 10 minutes at 4 °C, transferred to new polypropylene tubes, and stored at −20 °C until biomarker analysis (within 2 months). β-Amyloid 1–42 (Aβ₄₂), tau, and tau phosphorylated at threonine 181 (p-tau) were measured with commercially available enzyme-linked immunosorbent assays (Innotest; Fujirebio, Ghent, Belgium) [37]. CSF data were available for 466 (76%) subjects.

DNA was isolated from 10 ml of ethylenediaminetetraacetic acid blood. Apolipoprotein E (APOE) genotype was determined using the LightCycler APOE mutation detection method (Roche Diagnostics GmbH, Mannheim, Germany). According to APOE ε4 allele status, patients were dichotomized into carriers (hetero- and homozygous) and noncarriers. APOE status was available for 562 (91%) subjects.

For each patient, we obtained information on all-cause mortality (died yes/no with a date of death) from the Dutch municipal population register. This register was searched on 19 October 2016. Causes of death cannot be determined from this municipal registry. We defined follow-up duration as the time between the date of baseline AD diagnosis and the date of death or, if alive, between the date of baseline AD diagnosis and 19 October 2016.

To allow comparison of results on different tests within patients, all continuous variables were standardized to z-scores. All neuropsychological tests, except TMT-A and TMT-B, as well as CSF Aβ₄₂, were inverted by computing −1 × z-score, with the result being that a higher score implied more advanced disease. We used Cox proportional hazards models to assess associations between all baseline determinants and mortality, taking into account time to death, using the pooled results of the 15 imputations. Each measure was assessed unadjusted (model 1), adjusted for age and sex (model 2), and adjusted additionally for MMSE and duration of complaints as a proxy of disease severity (model 3). Effect modification, using interaction terms for each variable with *age and *sex, was not found. Subsequently, we aimed to select the optimal combination of determinants by constructing a multivariate model using forward selection. The model was built by assessing all variables and consecutively selecting the variable with the lowest p value in a stepwise manner until p was < 0.10. In case of several variables with the same lowest p value, we calculated the Wald statistics and selected the variable with the highest Wald value. Variables were added only when the overall model improved, as evaluated using the −2 log-likelihood ratio. In an additional set of analyses, we performed similar analyses based on nonimputed data, and the results were comparable (see Additional file 1: Table S1 and S2). Finally, we created Kaplan-Meier curves for each of the variables selected by forward selection. Because all variables except for sex were continuous values, we used tertiles for the survival curves. Data are represented as HRs with accompanying 95% CIs.