Dementia diagnoses were made based on DSM-IV criteria, and further classified into AD and DLB. Mild dementia was defined as a Mini-Mental Status Examination (MMSE) score of ≥ 20 or a Clinical Dementia Rating (CDR) global score of = 1. Detailed medical information and comorbidities were gathered from medical records, supplemented by structured assessments. Dopamine transporter SPECT scans were available for most patients suspected of having DLB.

Pathological diagnosis was made on 56 subjects of the DemVest cohort, with an accuracy above 80% when compared to the clinical criteria (17).

In this study, the Global Leadership Initiative on Malnutrition (GLIM) Index served as the tool for evaluating nutritional status (18). Using a set of predefined criteria, two key factors, namely body mass index (BMI) and age, were employed to classify individuals into one of three distinct nutritional status categories: severe malnutrition, moderate malnutrition, or adequate nutritional status.

For severe malnutrition, individuals with a BMI less than 18.5 who were under 70 years old, as well as those with a BMI less than 20 who were 70 years or older, were identified as being severely malnourished. In the case of moderate malnutrition, individuals with a BMI less than 20 under the age of 70 and a BMI less than 22 for those 70 years or older were categorized as moderately malnourished. Participants who did not meet the criteria for either of these established categories were considered to have no signs of malnutrition.

The nutritional status of participants was assessed at the beginning of the study and annually over a period of 5 years of follow-up.

MRI scans were acquired at baseline using a standardized protocol on a 1.5-T Philips Intera scanner at Stavanger University Hospital. The 3D T1-weighted images were obtained with parameters including a repetition time (TR) of 10.0 ms, echo time (TE) of 4.6 ms, flip angle of 30°, number of excitations (NEX) of 2, acquisition matrix of 256 × 256, and voxel size of 1.0156 × 1.0156 × 1 mm3. Images were subjected to quality checks, and those with movement artifacts or poor quality were discarded. A standardized pre-processing approach was applied, involving movement correction and intensity normalization. Scans with poor quality were excluded, and the retained images were used for assessing TMT. (See Supplementary Material: figure 1)

For the measurement of the temporal muscle, on axial T1-weighted images, TMT was measured on the right and left sides separately in all patients, perpendicular to the long axis of the temporalis muscle, using the Sylvian fissure as a reference and the orbit roof. A mean TMT was calculated for each patient for further statistical analysis (Figure 1). Measurements were performed by one trained researcher, and we re-measured 20 random images to ensure good intra-rater reliability.

Measure of TMT in one of the participants. A representative case for the assessment of the TMT on MRI

a. Coronal view of the brain. The TMT was measured on the right and left sides separately in all patients, perpendicular to the long axis of the temporalis muscle. b. Axial view of the brain. The Sylvian fissure and orbit roof were used as points of reference for the measurement.

As potential confounders, demographic factors were considered such as gender and age. BMI was calculated using body weight and height with the formula BMI = weight (kg)/height (m)2. Comorbidities using the Cumulative Illness Rating Scale (CIRS) based on patient and informant reports and global cognitive performance using the Norwegian version of the MMSE (19). Instrumental and basic activities of daily living were evaluated with the Norwegian version of the Rapid Disability Rating Scale-2 (RDRS-2) (20). RDRS-2 was performed in person by the research nurses at the various outpatient clinics. Items were scored from 1 to 4 (Alone=1, with some help=2, with a lot of help, =3 and cannot perform=4) and then divided by the number of items (0 minimum and 4 maximum total score). The more functional decline is present, the higher the score becomes.

This study was approved by the regional ethics committee (approval code: 2010/633) and the Norwegian authorities for the collection of medical data. All data was handled and kept following national health and data privacy protocols. All participants signed an informed consent form before inclusion in the study.

A descriptive analysis was conducted, involving the estimation of percentages for categorical variables and means with standard deviations for quantitative variables. Most analyses were performed in the combined dementia group as well as AD and DLB separately. This descriptive analysis was also done for each study time point, at baseline and then yearly to the five-year follow-up, including the number of individuals for each study time point to describe attrition due to dropout, for both AD and DBL groups. (See Supplementary Material figure 1)

We also performed a correlation analysis using Spearman’s rho coefficient between the TMT and continuous variables (MMSE and RDRS) to evaluate any linear associations between these variables at baseline. To provide a visual representation of how TMT at baseline relates to the outcomes over time, scatter plots were generated using the predicted values or probability of the outcome for dichotomous variables (moderate and severe malnutrition) and the predicted values for the continuous variables, all adjusted for group, sex, and age.

For assessing associations between TMT and the clinical outcome measures, we conducted both unadjusted and adjusted linear and logistic regression models (adjusting for group, sex, and age). Similarly, we employed longitudinal outcome status analysis using unadjusted and adjusted (by group, sex, age, and time) linear or logistic mixed-effects models with random slopes to test associations between TMT at baseline and the outcomes over time. Age, sex, group, and time were included as fixed terms, while time was considered a random effect. Results were reported in terms of beta coefficients or odds ratios (ORs) along with their respective p-values.

Finally, a survival analysis using a Cox regression model was performed, both unadjusted and adjusted (by group, sex, and age), we considered CIRS and BMI as potential confounders but discarded after variable selection procedure and to avoid collinearity. To assess whether TMT was associated with mortality, the time-to-response variable was calculated from the date of inclusion in the study to the date of death. Participants who were still alive or for which the outcome of death was unknown were treated as censored. Adjusted relative risk of death and TMT stratified by group and sex were plotted. HR and their respective p-values were reported. Notably, none of the models were adjusted for BMI due to collinearity issues with TMT. We established statistical significance at p < 0.05 to evaluate the influence of covariates in the models or the hypothesis tests. All analyses were conducted using R Studio version 4.2.1.