Presenile Dementia

For most diseases and injuries, processed data are modelled using standardised tools to generate estimates of each quantity of interest by age, sex, location, and year. There are three main standardised tools: Cause of Death Ensemble model (CODEm), spatiotemporal Gaussian process regression (ST-GPR), and DisMod-MR. Previous publications2 (link), 3 (link), 12 and the appendix provide more details on these general GBD methods. Briefly, CODEm is a highly systematised tool to analyse cause of death data using an ensemble of different modelling methods for rates or cause fractions with varying choices of covariates that perform best with out-of-sample predictive validity testing. DisMod-MR is a Bayesian meta-regression tool that allows evaluation of all available data on incidence, prevalence, remission, and mortality for a disease, enforcing consistency between epidemiological parameters. ST-GPR is a set of regression methods that borrow strength between locations and over time for single metrics of interest, such as risk factor exposure or mortality rates. In addition, for select diseases, particularly for rarer outcomes, alternative modelling strategies have been developed, which are described in appendix 1 (section 3.2).
In GBD 2019, we designated a set of standard locations that included all countries and territories as well as the subnational locations for Brazil, China, India, and the USA. Coefficients of covariates in the three main modelling tools were estimated for these standard locations only—ie, we ignored data from subnational locations other than for Brazil, China, India, and the USA (appendix 1 section 1.1). Using this set of standard locations will prevent changes in regression coefficients from one GBD cycle to the next that are solely due to the addition of new subnational units in the analysis that might have lower quality data or small populations (appendix 1 section 1.1). Changes to CODEm for GBD 2019 included the addition of count models to the model ensemble for rarer causes. We also modified DisMod-MR priors to effectively increase the out-of-sample coverage of uncertainty intervals (UIs) as assessed in simulation testing (appendix 1 section 4.5).
For the cause Alzheimer's disease and other dementias, we changed the method of addressing large variations between locations and over time in the assignment of dementia as the underlying cause of death. Based on a systematic review of published cohort studies, we estimated the relative risk of death in individuals with dementia. We identified the proportion of excess deaths in patients with dementia where dementia is the underlying cause of death as opposed to a correlated risk factor (appendix 1 section 2.6.2). We changed the strategy of modelling deaths for acute hepatitis A, B, C, and E from a natural history model relying on inpatient case fatality rates to CODEm models after predicting type-specific acute hepatitis deaths from vital registration data with specified hepatitis type.
DisMod-MR was used to estimate deaths from three outcomes (dementia, Parkinson's, and atrial fibrillation), and to determine the proportions of deaths by underlying aetiologies of cirrhosis, liver cancer, and chronic kidney disease deaths.

Observational studies (cross-sectional or longitudinal cohort studies) were included if they reported on community-dwelling older adults aged 60 years and above. This age cutoff point was selected because studies on frailty typically included participants aged 60 years and above (27 (link)). Cross-sectional, prospective cohort studies were included due to the small number of longitudinal studies that reported on the association between cognitive frailty and disability. Cognitive frailty was defined by the presence of frailty or prefrailty, and concurrent cognitive impairment was identified using validated physical frailty and cognitive assessments. A preliminary literature search has identified that most studies on cognitive frailty have slightly modified the definition of cognitive frailty by the consensus group, defining this condition with the presence of mild cognitive impairment instead of a CDR of 0.5 with the exclusion of concurrent Alzheimer’s disease or other dementias, and physical frailty using the modified Fried frailty phenotype (28 (link)). Thus, the utilization of CDR was not compulsory for study inclusion in this review if a validated cognitive assessment tool was reported. Studies must report the association between cognitive frailty and functional disability (ADL or IADL, mobility, physical function).
Studies were excluded if they included hospitalized or institutionalized older adults or those with neurological disorders or dementia. Conference abstracts, reviews, randomized controlled trials, protocols, and studies published in other languages besides English were excluded. Study titles and abstracts were screened based on the inclusion/exclusion criteria, and full texts of relevant studies were screened for eligibility. Data were extracted using a piloted data extraction form, including study and participant characteristics, frailty assessment and classification, and corresponding disability outcomes and measurement. Data extraction was conducted by 1 reviewer (K.F.T.) and checked by the second reviewer (S.W.H.L.), with discrepancies resolved by consensus.

We formulated the prediction of dementia as a binary classification problem (dementia, control); therefore, evaluation metrics, such as accuracy, F1-score, balanced accuracy, precision, specificity and sensitivity, were used to measure the performance of the subsets of features. The following evaluation metrics were used:

True positives (TP): number of dementia cases that were correctly classified.

False positives (FP): number of healthy subjects incorrectly classified as dementia cases.

True negatives (TN): number of healthy subjects correctly classified.

False negatives (FN): number of dementia cases incorrectly classified as healthy subjects.

Accuracy (%): the proportion of correct classifications among total classifications:

$$\text{Accuracy}=\frac{\text{TP}+\text{TN}}{n}$$
where n is the number of total classifications per test.

Sensitivity (%): The proportion of correctly classified dementia cases.

$$\text{Sensitivity}=\frac{\text{TP}}{\text{TP}+\text{FN}}$$

Specificity (%): The proportion of correctly classified healthy subjects.

$$\text{Specificity}=\frac{\text{TN}}{\text{TN}+\text{FP}}$$

Precision: The proportion of subjects classified as dementia cases who have dementia.

$$\text{Precision}=\frac{\text{TP}}{\text{TP}+\text{FP}}$$

F1-score (F-measure) (%): Harmonic mean of precision and sensitivity.

$$\text{F}1=2\times \frac{\text{Sensitivity}\times \text{Precision}}{\text{Sensitivity}+\text{Precision}}=\frac{\text{TP}}{\text{TP}+\left(\text{FP},+,\text{FN}\right)/2}$$

We attempted the classification of dementia status in 146 samples after removing missing values from the 177 that were used in the feature selection process. The 146 samples had a slight class imbalance, with 89 demented versus 57 non-demented patients. Before training our models, we randomly selected 57 patients from the demented group using the sample() function from the random module in Python3. Then, the rows were shuffled using sklearn.utils version 0.22.2.post1. As a result, 114 samples were utilized after balancing the class label. The 32 samples were held out for final assessment. The hippocampal tau stage feature, which had 50% missing values, was dropped during the training process. Age and brain weight were removed before training the models, ending up with 22 features and 114 samples for classification. The dataset was split into a training set of 70% (80 samples) and a testing set of 30% (34 samples).
Seven classification algorithms were trained to classify individuals’ dementia status from the 22 top-ranked features. Scikit-learn version 0.22.2.post1 was used to implement and train the ML classifiers, and then measure their classification performance. Logistic regression was implemented using the sklearn.linear_model package where penalty was set to 12, the regularization parameter C was set to 1, the maximum number of iterations taken for the solvers to converge was set to 2000, and other parameters were set to default values. A decision tree classifier was implemented using the sklearn.tree package. K-nearest neighbors classifier was implemented using the sklearn.neighbors with the number of neighbors set to 5, the function “uniform weights” used for prediction, the “Minkowski” distance metric utilized for the tree, and with other parameters were set to default values. The linear discriminant analysis classifier was implemented using the sklearn.discriminant_analysis package with singular value decomposition for solver hyperparameter and other parameters were set to default values. The Gaussian naïve Bayes classifier was implemented using sklearn.naive_bayes. The support vector machine with a radial basis function kernel (SVM-RBF) was implemented using sklearn.svm with the regularization parameter C set to 1, the kernel coefficient gamma = “scale” and other parameters were set to default values. The support vector machine with a linear kernel (SVM-LINEAR) was implemented using the sklearn.svm package with regularization parameter C set to 1, with a “linear” kernel, gamma coefficient “scale” and other parameters were set to default. The sklearn.metrics package was used to report classification performance. Training and performance evaluation were performed 500 times, from which the average performance measure was calculated as overall performance. Accuracy, balanced accuracy, F1-score, precision, sensitivity and specificity utilizing regression plots were measures used for performance. ML models and feature selection libraries were built using Python 3.7.3.