Participants and study design
Patients with MCI [1] were consecutively recruited from the Centre for Neurodegenerative Disorders and the Centre for Alzheimer Disease, University of Brescia, Brescia, Italy. Demographic characteristics, family history, and clinical features were carefully recorded. All patients considered in the present study underwent a standardized neuropsychological evaluation; brain magnetic resonance imaging; at least one diagnostic marker of brain amyloidosis, i.e., cerebrospinal fluid Aβ1-42 dosage and/or amyloid positron emission tomography scan; and TMS intracortical connectivity measures, as described below.
Patients’ data were then anonymized, randomized, and presented to two neurologists, one with long-lasting experience in a tertiary dementia care center (AP, rater 1) and one with 5-year experience in a secondary referral center for the diagnosis and the cure of dementia (MSC, rater 2), in three consecutive steps. In 50% of the cases (arm 1), the two raters were made aware of the following: step 1—demographic characteristics, family history, clinical and neuropsychological assessment, and structural imaging data (henceforth defined as “clinical work-up”); step 2—amyloid marker data; and step 3—TMS intracortical connectivity measures.
In the other 50% of the cases (arm 2), the two raters were made aware of the following: step 1—clinical work-up, step 2—TMS intracortical connectivity measures, and step 3—amyloid marker data (see Fig. 1, study design).
On the basis of the data obtained at each of the three steps, the two neurologists were asked to (a) formulate their etiological diagnosis (MCI-AD, MCI-FTD, MCI-DLB, or MCI-other), (b) to rate their diagnostic confidence (DC) that cognitive impairment was due to AD on a structured scale ranging from 0 to 100% (DC-AD, 0–100%), (c) to rate their confidence that cognitive impairment was due to FTD on a structured scale ranging from 0 to 100% (DC-FTD, 0–100%), (d) to rate their confidence that cognitive impairment was due to DLB on a structured scale ranging from 0 to 100% (DC-DLB, 0–100%),and (e) to rate their confidence that cognitive impairment was due to other neurodegenerative causes on a structured scale ranging from 0 to 100% (DC-other, 0–100%). Thus, the highest DC supported the formulated diagnosis. Any change in diagnosis or DC in the subsequent steps could only be attributed to knowing such results.
Moreover, a “gold standard” diagnosis (i.e., MCI-AD, MCI-FTD, MCI-DLB, or MCI-other) was provided by the dementia experts (AB, AA, and BB), who had the subjects in charge and who had complete access to all available information, such as the clinical work-up, amyloid markers, TMS intracortical connectivity measures, and follow-up evaluations.
Clinical work-up
The set of mandatory information for each recruited subject, which were presented to the two neurologists during the clinical work-up evaluation, included the demographic characteristics (age, sex, family history, past medical history, and comorbidities), the conventional structural brain imaging findings, and the results of the neuropsychological assessment, including global cognitive functions, long-term memory, executive functions, and language and visual spatial abilities, as previously reported [20]. Mini-Mental State Examination and Clinical Dementia Rating scales were considered to test global cognitive functions [21, 22]. The Basic and Instrumental Activities of Daily Living [23, 24], Neuropsychiatric Inventory [25], and Geriatric Depression Scale [26] were also considered.
All the above data were provided to the two raters in step 1.
Amyloid markers
We considered cerebrospinal Aβ1-42 analysis or amyloid positron emission tomography imaging as markers of amyloidosis. Lumbar puncture was carried out in the outpatient clinic according to standard procedures, and cerebrospinal fluid analysis was performed using an ELISA assay (INNOTEST, Innogenetics, Ghent, Belgium) [27]. According to our internal cutoff scores, a cerebrospinal fluid AD-like profile was defined as cerebrospinal fluid Aβ1-42 ≤ 650 pg/mL (along with cerebrospinal fluid total Tau ≥ 400 pg/mL).
Amyloid positron emission tomography imaging was acquired using 370 MBq (10 mCi) of 18F-florbetapir or 18F-flutemetamol, and visual readings were performed by a nuclear medicine physician who was blinded to the patients’ diagnosis, following the procedures provided by the ligand manufacturer, as previously reported [9].
Cerebrospinal fluid Aβ1-42 dosage (along with Tau) and/or amyloid positron emission tomography results (“positive” vs. “negative”) were provided to the two raters in either step 2 or step 3, according to randomization.
Transcranial magnetic stimulation intracortical connectivity measures
TMS protocols were carried out as previously published [11]. We considered SICI [28] and ICF [29], which predominantly reflect GABAAergic and glutamatergic neurotransmission, respectively [15], and SAI [30], which primarily reflects cholinergic transmission [15].
Briefly, SICI, ICF, and SAI were studied using a paired-pulse technique, employing a conditioning-test design. For all paradigms, the test stimulus was adjusted to evoke a motor evoked potential (MEP) of approximately 1 mv amplitude in the right first dorsal interosseous muscle.
For SICI and ICF, the conditioning stimulus was adjusted at 70% of the resting motor threshold (RMT), employing multiple interstimulus intervals (ISIs), including 1, 2, 3, and 5 ms for SICI and 7, 10, and 15 ms for ICF [11, 28, 29]. SAI was evaluated employing a conditioning stimulus of single pulses (200 μs) of electrical stimulation delivered to right median nerve at the wrist, using a bipolar electrode with the cathode positioned proximally, at an intensity sufficient to evoke a visible twitch of the thenar muscles [11, 30]. Different ISIs were implemented (− 4, 0, + 4, + 8 ms), which were fixed relative to the N20 component latency of the somatosensory evoked potential of the median nerve.
For each ISI and for each protocol, 10 different paired conditioning-target stimuli and 14 control target stimuli were delivered in all participants in a pseudo-randomized sequence, with an intertrial interval of 5 s (± 10%). Stimulation protocols were conducted in a randomized order. All of the participants were capable of following instructions and reaching complete muscle relaxation; if, however, the data was corrupted by patient movement, the protocol was restarted and the initial recording was rejected.
The operators who performed TMS (VC and VD) were blinded to the subjects’ amyloid marker status and clinical or neuropsychological evaluation. Mean SICI-ICF (1, 2, 3 ms/7, 10, 15 ms) and mean SAI (0, +4 ms), as well as SICI-ICF/SAI ratio, were calculated, as previously reported [11]. SICI-ICF/SAI ratio was provided to the two raters, and they considered the previous published cutoff value of 0.98 [11] in either step 2 or step 3, according to randomization.
Statistical analysis
Sociodemographic characteristics of the patients as well as descriptive features of the DCs were provided through mean, standard deviation, 95% confidence intervals (95% CI), and median values.
Considering the experimental design (with repeated measures within arms, raters, and assessments and, thus, with variance structure dependence), and taking into consideration the diagnostic confidence distributions (skewed and with a positive mass at zero) of the four outcomes (DC-AD, DC-FTD, DC-DLB, DC-other), generalized estimating equation models with Tweedie distribution and log link-function were adopted to assess the association of the three factors: arms (arm 1 [clinical work-up➔amyloid markers➔TMS], arm2 [clinical work-up➔TMS➔amyloid markers]), raters (rater1, rater2), and single assessments (clinical work-up, TMS, amyloid markers) with DC. A first evaluation of the four DCs data with respect to arms, raters, and assessments was provided regardless of the diagnosis, by performing three generalized estimating equation models with DC as dependent variable and each of the three factors, and their triple interaction, as independent factors. Subsequently, a detailed evaluation of the additional contribution of the assessments (clinical work-up, clinical work-up PLUS either TMS or amyloid markers, and clinical work-up PLUS both markers) in explaining the DC variability was performed for each of the four diagnoses (MCI-AD, MCI-FTD, MCI-DLB, or MCI-other).
Finally, the association of DC of each of the five sections (independent variables) with the “gold-standard” diagnosis (i.e., MCI-AD, MCI-FTD, MCI-DLB, and MCI-other as, in turn, dependent variables) was evaluated through logistic regression models. Performance of each assessment section in predicting the “gold-standard” diagnosis was evaluated through receiver operating characteristic (ROC) curves, and the corresponding area under the curve (AUC) values, applied on predictive probability scores obtained by the logistic models. High values of AUC (greater than 0.8) indicate good performance of independent variables in predicting the diagnosis. Comparison of AUC was performed by the DeLong test.
Statistical significance was assumed at p < 0.05. Data analyses were carried out by “mclust” and “InformationValue” packages of R statistical software (URL http://www.R-project.org/) and IBM SPSS Statistics for Windows, version 21.0, Armonk, NY: IBM Corp.