Logistic Discriminant Analysis Research Articles

ScDLC: a deep learning framework to classify large sample single-cell RNA-seq data

BackgroundUsing single-cell RNA sequencing (scRNA-seq) data to diagnose disease is an effective technique in medical research. Several statistical methods have been developed for the classification of RNA sequencing (RNA-seq) data, including, for example, Poisson linear discriminant analysis (PLDA), negative binomial linear discriminant analysis (NBLDA), and zero-inflated Poisson logistic discriminant analysis (ZIPLDA). Nevertheless, few existing methods perform well for large sample scRNA-seq data, in particular when the distribution assumption is also violated.ResultsWe propose a deep learning classifier (scDLC) for large sample scRNA-seq data, based on the long short-term memory recurrent neural networks (LSTMs). Our new scDLC does not require a prior knowledge on the data distribution, but instead, it takes into account the dependency of the most outstanding feature genes in the LSTMs model. LSTMs is a special recurrent neural network, which can learn long-term dependencies of a sequence.ConclusionsSimulation studies show that our new scDLC performs consistently better than the existing methods in a wide range of settings with large sample sizes. Four real scRNA-seq datasets are also analyzed, and they coincide with the simulation results that our new scDLC always performs the best. The code named “scDLC” is publicly available at https://github.com/scDLC-code/code.

Soil Order-Land Use Index Using Field-Satellite Spectroradiometry in the Ecuadorian Andean Territory for Modeling Soil Quality

Land use conversion is the main cause for soil degradation, influencing the sustainability of agricultural activities in the Ecuadorian Andean region. The possibility to identify the quality based on the spectral properties allows remote sensing methods to offer an alternative form of monitoring the environment. This study used laboratory spectroscopy and multi-spectral images (Sentinel 2) with environmental covariates (physicochemical parameters) to find an affordable method that can be used to present spatial prediction models as a tool for the evaluation of the quality of Andean soils. The models were developed using statistical techniques of logistic regression and linear discriminant analysis to generate an index based on soil order and three indexes based on the combination of soil order and land use. This combined approach offers an effective method, relative to traditional laboratory methods, to derive estimates of the content and composition of soil constituents, such as electrical conductivity (CE), organic matter (OM), pH, and soil moisture (HU). For Mollisol index.3 with Páramo land use, a value of organic matter (OM) ≥8.6% was obtained, whereas for Mollisol index.4 with Shrub land use, OM was ≥6.1%. These results reveal good predictive (estimation) capabilities for these soil order–land use groups. This provides a new way to monitor soil quality using remote sensing techniques, opening promising prospects for operational applications in land use planning.

Imputation methods on retrospective breast cancer data in Tanzania: A comparative study

Background: Clinical datasets are at risk of having missing data for several reasons including patients’ failure to attend clinical measurements and measurement recorder’s defects. Missing data can significantly affect the analysis and results might be doubtful due to bias caused by omission incomplete records during analysis especially if a dataset is small. This study aims to compare several imputation methods in terms of efficiency in filling-in missing data so as to increase prediction and classification accuracy in breast cancer dataset. Methodology: Five imputation methods namely series mean, k-nearest neighbour, hot deck, predictive mean matching, expected maximisation via bootstrapping, and multiple imputation by chained equations were applied to replace the missing values to the real breast cancer dataset. The efficiency of imputation methods was compared by using the Root Mean Square Errors and Mean Absolute Errors to obtain a suitable complete dataset. Binary logistic regression and linear discrimination classifiers were applied to the imputed dataset to compare their efficacy on classification and discrimination. Results: The evaluation of imputation methods revealed that the predictive mean matching method was better off compared to other imputation methods. In addition, the binary logistic regression and linear discriminant analyses yield almost similar values on overall classification rates, sensitivity and specificity. Conclusion: The predictive mean matching imputation showed higher accuracy in estimating and replacing missing data values in a real breast cancer dataset under the study. It is a more effective and good approach to handle missing data. We recommend replacing missing data by using predictive mean matching since it is a plausible approach toward multiple imputations for numerical variables. It improves estimation and prediction accuracy over the use complete-case analysis especially when percentage of missing data is not very small.

Forecasting model for disease risk period in chickpea x collar rot pathosystem

Collar rot caused by Sclerotium rolfsii Sacc. is one of the major biotic constraints of chickpea production worldwide. It is soil-borne fungi having wider host range and infection mainly occurs at the juvenile stage of crop growth resulting crop failure in no time. The pathogen is greatly influenced by soil temperature (ST) and soil moisture (SM) therefore, experiment formulated to develop a suitable forecasting model for its future use in computer simulation of plant disease prognostication by feeding only soil temperature and moisture data. The popular desi type chickpea variety Anuradha sown at different dates to get a range of soil temperature and soil moisture combination and its corresponding effect on disease incidence was recorded under natural epiphytotic conditions. The data obtained were analyzed using binary logistic regression and discriminant analysis to assess disease risk and non-risk period. The model developed was Y'= -73.9 + 1.251 SM + 0.017 ST. The outcome recorded, a unique statistically significant contribution of soil moisture (p value=0.029) on the establishment of the disease whereas, the effect of soil temperature was detected as statistically non-significant. The model developed and the correctness of the model determined to predict the disease severity with 80 % accuracy.

Age and sex estimation using fractal analysis in Brazilian adults: a discriminant analysis

This study assessed the accuracy of fractal analysis (FA) to estimate chronological age and sex in Brazilian adults for forensic investigations. The gender-balanced sample comprised lateral cephalometric radiographs of 120 individuals, stratified according to age (20-29, 30-39, 40-49, 50-59 years) and sex (female and male). A trained calibrated examiner measured the fractal dimension (FD) of the mandibular ramus and mandibular angle. Linear regression and multiple logistic discriminant analysis were carried out to explore the accuracy of FA. For all analyses, p-values < .05 indicated statistical significance. Overall, mean FD values were 1.49±0.10 for the mandibular ramus and 1.48±0.09 for mandibular angle. Results were more accurate in males than females for discriminating age and sex. The multiple discriminant analysis indicated that FA distinguished sex in 61.7% males and 58.3% females. In addition, the mean difference between actual and predicted value was 9.5 years and 10.1 years for men and women, respectively. Fractal analysis accurately identified sex- and age-related differences in the trabecular pattern of the mandible of Brazilian adults, confirming its utility for forensic investigations. Further studies investigating other populations are needed to assess the accuracy of FA for Forensic Dentistry.

Advancing Female Corporate Leadership: A Comparative Analysis of Female Versus Male Leadership Behaviour in India

This article investigates whether women managers possess any specific leadership attributes which will enable them to be promoted to leadership positions on merit grounds. It also examines whether women encounter barriers, which impede their progress to the top. Data for our study were collected from a sample of 372 managers in leadership positions across India, half of whom were women and the other half were men. The survey was done through the administration of a 21-item questionnaire developed by the authors. Through the application of decision tree analysis, logistic regression analysis and discriminant analysis, we arrived at findings that have reasonable predictive accuracy. Our results suggest that women managers in leadership positions are significantly more likely to get reticent team members to participate at meetings than their male counterparts. This is our principal finding. This is significant, given that contemporary leadership emphasizes inclusivity and distributed leadership. This is because, as our data shows, while women managers are as good as men, they additionally possess an attribute which makes them better suited than men as leaders. Another finding is that many women leaders differ from men in that they often struggle with work–life balance challenges. The suggestion is that companies could offer more options to women to achieve work–life balance. The third finding is that women leaders are perceived as more empathetic than men. The implication is that all of these constitute business arguments favouring organizational support for competent women managers advancing to leadership positions. Data are from India, focussing on Indian managerial leadership.

Influence of weather factors, soil temperature and moisture on collar rot disease of betel vine (Piper betle L.) under coastal saline zone of West Bengal

A study was conducted to find out the influence of weather factors, soil temperature and soil moisture on the incidence of Sclerotium rolfsii Sacc. induced collar rot disease in betelvine (Piper betle L.), during 2016 to 2018. Fourteen soil and weather factors, taken from the agrometeorological observatory located at instructional farm of Ramkrishna Ashram Krishi Vigyan Kendra, Nimpith and recorded from a nearby betelvine boroj, were subjected to multiple regression, binary logistic regression and canonical discriminant analysis to develop a suitable disease forewarning model. The binary logistic model, Y(0/1) = 5.899 + 0.865 (Tmax) – 0.569 (SM) + 0.097 (BRHmin) was able to predict the disease risk with 78 per cent accuracy and correctly classified 94 per cent of cases during model validation in 2018. The weekly averages of maximum temperature (Tmax), soil moisture (SM) and minimum relative humidity inside the boroj (BRHmin) were found to be the most significant predictors of disease incidence, in this model. The soil moisture at 69 - 72 per cent of field capacity, minimum temperature of 25 - 27oC, maximum temperature of 33 - 36oC, average soil temperature of 28 - 30oC, minimum relative humidity of 60 - 72 per cent inside the boroj and maximum relative humidity of 83 - 89 per cent inside the boroj were found to be highly congenial for collar rot disease incidence in betelvine under coastal saline zone of West Bengal.

Predicting the outcomes of the Korean national accreditation system for higher education institutions: a method using disclosure data for outsiders

For outsiders to higher education institutions (HEIs) in South Korea, predicting the outcomes of the International Education Quality Assurance System (IEQAS)—a Korean institutional accreditation system for HEIs—is challenging. The annual IEQAS accreditation has been conducted behind closed doors; the assessment process is confidential, and there is little access to the data for the public and individuals. However, many stakeholders, such as overseas applicants to Korean HEIs, may want to predict whether particular universities will pass the upcoming IEQAS. Hence, we sought an alternative method for the outsiders to predict a binary result of the IEQAS accreditation by utilizing disclosure data that the Korean government has published. To best predict the outcomes, we mapped out a threefold discrete model combining logistic regression, discriminant analysis, and neural network. We collected the information disclosed by the Ministry of Education in 2019 on 138 Korean private HEIs and then analyzed the secondary public dataset in line with the discrete method that ensures generalizability. Results showed (i) three education investment factors, and one school operations factor appeared as key predictors among the tested indices; (ii) education cost per student within education investment proved to be the most crucial element; and (iii) while leveraging the disclosed data turned out to be reliable, neural network’s predictive accuracy was higher than those reported using logistic regression and discriminant analysis. By processing the publicly available disclosure data, our self-study model may effectively assist in predicting IEQAS outcomes, and it can also be used as a diagnostic, prior to accreditation, by local HEIs in other nations to check their preparedness and likelihood of success within similar contexts.

Non-invasive Urine Test for Molecular Classification of Clinical Significance in Newly Diagnosed Prostate Cancer Patients.

Objective: To avoid over-treatment of low-risk prostate cancer patients, it is important to identify clinically significant and insignificant cancer for treatment decision-making. However, no accurate test is currently available.Methods: To address this unmet medical need, we developed a novel gene classifier to distinguish clinically significant and insignificant cancer, which were classified based on the National Comprehensive Cancer Network risk stratification guidelines. A non-invasive urine test was developed using quantitative mRNA expression data of 24 genes in the classifier with an algorithm to stratify the clinical significance of the cancer. Two independent, multicenter, retrospective and prospective studies were conducted to assess the diagnostic performance of the 24-Gene Classifier and the current clinicopathological measures by univariate and multivariate logistic regression and discriminant analysis. In addition, assessments were performed in various Gleason grades/ISUP Grade Groups.Results: The results showed high diagnostic accuracy of the 24-Gene Classifier with an AUC of 0.917 (95% CI 0.892–0.942) in the retrospective cohort (n = 520), AUC of 0.959 (95% CI 0.935–0.983) in the prospective cohort (n = 207), and AUC of 0.930 (95% 0.912-CI 0.947) in the combination cohort (n = 727). Univariate and multivariate analysis showed that the 24-Gene Classifier was more accurate than cancer stage, Gleason score, and PSA, especially in the low/intermediate-grade/ISUP Grade Group 1–3 cancer subgroups.Conclusions: The 24-Gene Classifier urine test is an accurate and non-invasive liquid biopsy method for identifying clinically significant prostate cancer in newly diagnosed cancer patients. It has the potential to improve prostate cancer treatment decisions and active surveillance.

Abstract 836: Immunologic markers and the progression of liver disease in HCV-positive individuals

Abstract Background: Host immune response and chronic inflammation associated with chronic hepatitis C virus (HCV) infection play a key role in the progression of liver disease from cirrhosis to hepatocellular carcinoma (HCC). We evaluated a large panel of circulating immunologic markers to better understand the natural history of HCV-related liver disease. Methods: We sampled 94 HCC, 68 cirrhotic and 100 non-cirrhotic controls from a Taiwanese prospective cohort study of chronically HCV-infected individuals. Multivariable polytomous logistic regression and canonical discriminant analysis (CDA) were used to compare baseline plasma levels for 102 markers in individuals who developed cirrhosis as compared to controls and those who developed HCC. P-values for individual markers were corrected for multiple testing using a false discovery rate of 10%; a p-value<0.05 was used to determine significance in CDA. Receiver operating characteristic curves were used to compare the predictive ability of marker groups from both analyses. Results: After adjusting for age, sex, years of follow-up, serum alanine aminotransferase level, alcohol and smoking, DEFA-1 (p-trend: <0.0001), ITGAM (p-trend: <0.0001), CCL18 (p-trend: 0.002) and ARTN (p-trend: 0.003) were all found to be positively associated with the development of cirrhosis when compared to non-cirrhotic controls; the same four markers were shown to be associated with HCC when compared to cirrhotic participants. Taken together, these four markers predicted cirrhosis and HCC development with 79% and 88% accuracy, respectively in adjusted models. CDA models isolated a twelve-marker signature that included DEFA1, CCL18, and ARTN, improving predictability of cirrhosis development to 91% accuracy in adjusted models. Using the same methods, a fourteen-marker signature that included DEFA-1, ITGAM, CCL18 and ARTN was found to predict HCC development with 97% accuracy when compared to cirrhotic individuals in adjusted models. Conclusion: To our knowledge, this is the first prospective study to assess immunologic markers in relation to liver disease progression in individuals chronically infected with HCV. While further validation is required, these findings highlight the importance of immunologic processes in HCV-related disease. Citation Format: Ilona Argirion, Ruth M. Pfeiffer, Zhiwei Liu, Allan Hildesheim, Ligia Pinto, Katherine A. McGlynn, Jessica L. Petrick, Tram Kim Lam, Thomas R. O'Brien, Kelly Yu, Chien-jen Chen, Hwai-I Yang, Mei-Hsuan Lee, Jill Koshiol. Immunologic markers and the progression of liver disease in HCV-positive individuals [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 836.

Selecting Classification Methods for Small Samples of Next-Generation Sequencing Data.

Next-generation sequencing has emerged as an essential technology for the quantitative analysis of gene expression. In medical research, RNA sequencing (RNA-seq) data are commonly used to identify which type of disease a patient has. Because of the discrete nature of RNA-seq data, the existing statistical methods that have been developed for microarray data cannot be directly applied to RNA-seq data. Existing statistical methods usually model RNA-seq data by a discrete distribution, such as the Poisson, the negative binomial, or the mixture distribution with a point mass at zero and a Poisson distribution to further allow for data with an excess of zeros. Consequently, analytic tools corresponding to the above three discrete distributions have been developed: Poisson linear discriminant analysis (PLDA), negative binomial linear discriminant analysis (NBLDA), and zero-inflated Poisson logistic discriminant analysis (ZIPLDA). However, it is unclear what the real distributions would be for these classifications when applied to a new and real dataset. Considering that count datasets are frequently characterized by excess zeros and overdispersion, this paper extends the existing distribution to a mixture distribution with a point mass at zero and a negative binomial distribution and proposes a zero-inflated negative binomial logistic discriminant analysis (ZINBLDA) for classification. More importantly, we compare the above four classification methods from the perspective of model parameters, as an understanding of parameters is necessary for selecting the optimal method for RNA-seq data. Furthermore, we determine that the above four methods could transform into each other in some cases. Using simulation studies, we compare and evaluate the performance of these classification methods in a wide range of settings, and we also present a decision tree model created to help us select the optimal classifier for a new RNA-seq dataset. The results of the two real datasets coincide with the theory and simulation analysis results. The methods used in this work are implemented in the open-scource R scripts, with a source code freely available at https://github.com/FocusPaka/ZINBLDA.

A Machine Learning decision-making tool for extubation in Intensive Care Unit patients

Background and Objective: To increase the success rate of invasive mechanical ventilation weaning in critically ill patients using Machine Learning models capable of accurately predicting the outcome of programmed extubations. Methods: The study population was adult patients admitted to the Intensive Care Unit. Target events were programmed extubations, both successful and failed. The working dataset is assembled by combining heterogeneous data including time series from Clinical Information Systems, patient demographics, medical records and respiratory event logs. Three classification learners have been compared: Logistic Discriminant Analysis, Gradient Boosting Method and Support Vector Machines. Standard methodologies have been used for preprocessing, hyperparameter tuning and resampling. Results: The Support Vector Machine classifier is found to correctly predict the outcome of an extubation with a 94.6% accuracy. Contrary to current decision-making criteria for extubation based on Spontaneous Breathing Trials, the classifier predictors only require monitor data, medical entry records and patient demographics. Conclusions: Machine Learning-based tools have been found to accurately predict the extubation outcome in critical patients with invasive mechanical ventilation. The use of this important predictive capability to assess the extubation decision could potentially reduce the rate of extubation failure, currently at 9%. With about 40% of critically ill patients eventually receiving invasive mechanical ventilation during their stay and given the serious potential complications associated to reintubation, the excellent predictive ability of the model presented here suggests that Machine Learning techniques could significantly improve the clinical outcomes of critical patients.

THEFT DETECTION IN POWER UTILITIES USING ENSEMBLE OF CHAID DECISION TREE ALGORITHM

Theft of electricity and fraud in energy consumption billing are the primary concerns for Distribution System Operators . Because of those illegal activities, it is believed that billions of dollars are wasted each year. DSOs around the world continue to use conventional time consuming and inefficient methods for non-technical loss detection, particularly in underdeveloped countries . This research work attempts to solve the problems as mentioned above by designing an effective model for detecting electricity theft to classify fraudster customers in a power delivery system. The key motivation for this current study is to support the DSOs in their fight against the stealing of electricity. Initially, the proposed method uses the monthly energy customer consumption data obtained from Multan Electric Power Company (MEPCO) Pakistan to segregate fraudsters and honest customers. The Bagged Chi - square Automatic Interaction Detection (CHAID) based decision tree algorithm is then used to classify the honest and fraudster consumers.Furthermore, based on the mentioned metrics, the performance superiority of the Bagged CHAID-based NTL detection method is validated by comparing its efficacy with that of few well-known state-of-the-art machine learning algorithms such as Artificial Neural Network (ANN), Support Vector Machine (SVM),) Logistic Regression (LR), Discriminant Analysis and Bayesian Network (BN).

Cricket Player Selection using Machine Learning

Cricket has always been a popular game since its invention in the world. Moreover, it became a religion in India. The selection committees like BCCI,PCB,ACB etc. pick the players based on their previous performances in domestic cricket tournaments like IPL,Ranji Trophy, Syed Mushtaq Ali Trophy etc. by committee decisions but there is no application for selection process till now. To develop an application we need player performance analysis and assessment. This paper suggests an important approach for Selecting Cricket players by Evaluating his Statistics and Provides a comparative look at machine learning techniques in cricket player selection. In this paper a model for Bowlers and Batsmen Separately was proposed which was implemented using Random Forest, Ada Boost, Support Vector Machines(SVM), Light GBM, Cat Boost, Logistic Regression Linear Discriminant Analysis(LDA), Voting Classifier, Naïve Bayes. The findings obtained by the suggested methodology in this paper are the same as in the Cricket board selected team players.

The Applicability of Item Response Theory Based Statistics to Detect Differential Item Functioning in Polytomous Tests

The study used statistical procedures based on Item Response Theory to detect Differential Item Functioning (DIF) in polytomous tests. These were with a view to improving the quality of test items construction. The sample consisted of an intact class of 513 Part 3 undergraduate students who registered for the course EDU 304: Tests and Measurement at Sule Lamido University during 2017/2018 Second Semester. A self-developed polytomous research instrument was used to collect data. Data collected were analysed using Generalized Mantel Haenszel, Simultaneous Item Bias Test, and Logistic Discriminant Function Analysis. The results showed that there was no significant relationship between the proportions of test items that function differentially in the polytomous test when the different statistical methods are used. Further, the three parametric and non-parametric methods complement each other in their ability to detect DIF in the polytomous test format as all of them have capacity to detect DIF but perform differently. The study concluded that there was a high degree of correspondence between the three procedures in their ability to detect DIF in polytomous tests. It was recommended that test experts and developers should consider using procedure based on Item Response Theory in DIF detection.

Comparison of multinomial logistic discriminant analysis (mlgda) and classification and regression tree (cart) performance in classifying the impact of working children

Working children have some impacts that befall them. They are hampered by access to education, exploited in working hours and even experience the two impacts at the same time. This research focus on applying and comparing the classification performance of the impact of working children by multinomial logistic discriminant analysis (MLgDA) and classification and regression tree (CART). MLgDA is useful for dealing with cases that do not meet the linearity assumption and suitable for data containing numerical and categorical predictors. CART is not affected by outliers, collinearity and heteroscedasticity among predictors. The data sourced from National Labor Force Survey 2018. All predictors are categorical and characteristics between regions are different. More than half of working children in Indonesia face some impacts, reaching 54.31%. The best classification performance is produced by MLgDA, with an accuracy of 78.73%. MLgDA is superior to CART in almost all measures.

On symmetric semiparametric two‐sample problem

We consider a two-sample problem where data come from symmetric distributions. Usual two-sample data with only magnitudes recorded, arising from case-control studies or logistic discriminant analyses, may constitute a symmetric two-sample problem. We propose a semiparametric model such that, in addition to symmetry, the log ratio of two unknown density functions is modeled in a known parametric form. The new semiparametric model, tailor-made for symmetric two-sample data, can also be viewed as a biased sampling model subject to symmetric constraint. A maximum empirical likelihood estimation approach is adopted to estimate the unknown model parameters, and the corresponding profile empirical likelihood ratio test is utilized to perform hypothesis testing regarding the two population distributions. Symmetry, however, comes with irregularity. It is shown that, under the null hypothesis of equal symmetric distributions, the maximum empirical likelihood estimator has degenerate Fisher information, and the test statistic has a mixture of χ2 -type asymptotic distribution. Extensive simulation studies have been conducted to demonstrate promising statistical powers under correct and misspecified models. We apply the proposed methods to two realexamples.

Epidemiological pathology of A\u03b2 deposition in the ageing brain in CFAS: addition of multiple A\u03b2-derived measures does not improve dementia assessment using logistic regression and machine learning approaches

Aβ-amyloid deposition is a key feature of Alzheimer’s disease, but Consortium to Establish a Registry for Alzheimer's Disease (CERAD) assessment, based on neuritic plaque density, shows a limited relationships to dementia. Thal phase is based on a neuroanatomical hierarchy of Aβ-deposition, and in combination with Braak neurofibrillary tangle staging also allows derivation of primary age-related tauopathy (PART). We sought to determine whether Thal Aβ phase predicts dementia better than CERAD in a population-representative cohort (n = 186) derived from the Cognitive Function and Ageing Study (CFAS). Cerebral amyloid angiopathy (CAA) was quantitied as the number of neuroanatomical areas involved and cases meeting criteria for PART were defined to determine if they are a distinct pathological group within the ageing population. Agreement with the Thal scheme was excellent. In univariate analysis Thal phase performed less well as a predictor of dementia than CERAD, Braak or CAA. Logistic regression, decision tree and linear discriminant analysis were performed for multivariable analysis, with similar results. Thal phase did not provide a better explanation of dementia than CERAD, and there was no additional benefit to including more than one assessment of Aβ in the model. Number of areas involved by CAA was highly correlated with assessment based on a severity score (p < 0.001). The presence of capillary involvement (CAA type I) was associated with higher Thal phase and Braak stage (p < 0.001). CAA was not associated with microinfarcts (p = 0.1). Cases satisfying pathological criteria for PART were present at a frequency of 10.2% but were not older and did not have a higher likelihood of dementia than a comparison group of individuals with similar Braak stage but with more Aβ. They also did not have higher hippocampal-tau stage, although PART was weakly associated with increased presence of thorn-shaped astrocytes (p = 0.048), suggesting common age-related mechanisms. Thal phase is highly applicable in a population-representative setting and allows definition of pathological subgroups, such as PART. Thal phase, plaque density, and extent and type of CAA measure different aspects of Aβ pathology, but addition of more than one Aβ measure does not improve dementia prediction, probably because these variables are highly correlated. Machine learning predictions reveal the importance of combining neuropathological measurements for the assessment of dementia.

A Log-Linear Modeling Approach for Differential Item Functioning Detection in Polytomously Scored Items.

A log-linear model (LLM) is a well-known statistical method to examine the relationship among categorical variables. This study investigated the performance of LLM in detecting differential item functioning (DIF) for polytomously scored items via simulations where various sample sizes, ability mean differences (impact), and DIF types were manipulated. Also, the performance of LLM was compared with that of other observed score-based DIF methods, namely ordinal logistic regression, logistic discriminant function analysis, Mantel, and generalized Mantel-Haenszel, regarding their Type I error (rejection rates) and power (DIF detection rates). For the observed score matching stratification in LLM, 5 and 10 strata were used. Overall, generalized Mantel-Haenszel and LLM with 10 strata showed better performance than other methods, whereas ordinal logistic regression and Mantel showed poor performance in detecting balanced DIF where the DIF direction is opposite in the two pairs of categories and partial DIF where DIF exists only in some of the categories.

Statistical Methodology for Multiclass Classifications: Applications to Dementia.

Alzheimer's disease (AD) is a common, devastating disease which carries a heavy economic burden. Accelerated efforts to identify presymptomatic stages of AD and biomarkers to classify the disease are urgent needs. Currently, no biomarkers can perfectly discriminate individuals into multiple disease categories of AD (no cognitive impairment, mild cognitive impairment, and dementia). Although many biomarkers for diagnosis and their various features are being studied, we lack advanced statistical methods which can fully utilize biomarkers to classify AD accurately, thereby facilitating evaluation of putative markers both alone and in combination. In this paper, we propose two approaches: 1) a forward addition procedure in which we adapt an additive logistic regression model to the setting for disease with ordered multiple categories. Using this approach, we select and combine multiple cross-sectional biomarkers to improve diagnostic accuracy, and 2) a method by extending the Neyman-Pearson Lemma to the ordered three disease categories to construct optimal cutoff points to distinguish multiple disease categories. We evaluate the robustness of the proposed model using a simulation study. Then we apply these two methods to data from the Religious Orders Study to examine the feasibility of combining biomarkers, and compare the diagnostic accuracy between the proposed methods and existing methods including model-based methods (ordinal logistic regression and quadratic discriminant analysis), a tree-based method CART, and the Youden index method. The two proposed methods facilitate evaluations of biomarkers for conditions with graded, rather than binary, classifications. The evaluation of the performance of different approaches provides guidance of how to choose approaches to address research questions.