Ordinal Trees and Random Forests: Score-Free Recursive Partitioning and Improved Ensembles

Existing ordinal trees and random forests typically use scores that are assigned to the ordered categories, which implies that a higher scale level is used. Versions of ordinal trees are proposed that take the scale level seriously and avoid the assignment of artificial scores. The construction principle is based on an investigation of the binary models that are implicitly used in parametric ordinal regression. These building blocks can be fitted by trees and combined in a similar way as in parametric models. The obtained trees use the ordinal scale level only. Since binary trees and random forests are constituent elements of the proposed trees, one can exploit the wide range of binary trees that have already been developed. A further topic is the potentially poor performance of random forests, which seems to have been neglected in the literature. Ensembles that include parametric models are proposed to obtain prediction methods that tend to perform well in a wide range of settings. The performance of the methods is evaluated empirically by using several data sets.

Measure of Departure from Conditional Symmetry Based on Cumulative Probabilities for Square Contingency Tables

For the analysis of square contingency tables with ordered categories, a measure was developed to represent the degree of departure from the conditional symmetry model in which there is an asymmetric structure of the cell probabilities with respect to the main diagonal of the table. The present paper proposes a novel measure for the departure from conditional symmetry based on the cumulative probabilities from the corners of the square table. In a given example, the proposed measure is applied to Japanese occupational status data, and the interpretation of the proposed measure is illustrated as the departure from a proportional structure of social mobility.

Clinical and echocardiographic risk score predicts need for hospitalization among patients with COVID-19

Background Age and medical co-morbidities are known predictors of disease severity in coronavirus disease-2019 (COVID-19). Whether baseline transthoracic echocardiographic (TTE) abnormalities could refine risk-stratification in this context remains unknown. Purpose To analyze performance of a risk score combining clinical and pre-morbid TTE features in predicting risk of hospitalization among patients with COVID-19. Methods Adult patients testing positive for COVID-19 between March 1st and October 31st, 2020 with pre-infection TTE (within 15–180 days) were selected. Those with severe valvular disease, acute cardiac events between TTE and COVID-19, or asymptomatic carriers of virus (on employment screening/nursing home placement) were excluded. Baseline demographic, clinical co-morbidities, and TTE findings were extracted from electronic health records and compared between groups stratified by hospital admission. Total sample was randomly split into training (≈70%) and validation (≈30%) sets. Age was transformed into ordered categories based on cubic spline regression. Regression model was developed on the training set. Variables found significant (at p<0.10) on univariate analysis were selected for multivariate analysis with hospital admission as outcome. β-coefficients were obtained from 5000 bootstrapped samples after forced entry of significant variables, and scores assigned using Schneeweiss's scoring system. Final risk score performance was compared between training/validation cohorts using receiver-operating curve (ROC) and calibration curve analyses. Results 192 patients were included, 83 (43.2%) were admitted. Clinical/TTE characteristics stratified by hospitalization are in Table 1. Moderate or worse pulmonary hypertension and left atrial enlargement were only TTE parameters with coefficients deserving a score (Table 1). The risk score had excellent discrimination in training and validation sets (figure 1 left panel; AUC 0.785 versus 0.836, p=0.452). Calibration curves showed strong linear correlation between predicted and observed probabilities of hospitalization in both training and validation sets (Figure 1, middle and right panels, respectively). ROC analysis revealed a score ≥7 as having best overall quality with sensitivity and specificity of 70–75% in both training and validation sets. A score ≥12 had 98% and 97% specificity and ≥14 had 100% specificity. Conclusion A combined clinical and echocardiographic risk score shows promise in predicting risk of hospitalization among patients with COVID-19, and hence help anticipate resource utilization. External validation and comparison against clinical risk score alone is worth further investigation. FUNDunding Acknowledgement Type of funding sources: None.

Value-Driven Multiple Criteria Sorting with Probabilistic Linguistic Information Considering Uncertain Assignment Examples

Multiple criteria sorting (MCS) dedicates to assigning alternatives to one of the predefined ordered categories according to their evaluation information on multiple criteria. The utility (value) function-based sorting is a popular MCS procedure, which requires decision-makers to express their preferences through assignment examples. By taking the assignment examples as reference alternatives, the additive value function, as the preferred model of a decision maker, can be built using the preference disaggregation technique. However, the existing literature hardly considered people’s hesitancy when determining assignment examples, and ignored applying linguistic evaluation information on qualitative criteria. To fill these research gaps, this study proposes a value-driven MCS procedure with probabilistic linguistic information considering uncertain assignment examples. Specifically, the probability linguistic term set, as a flexible information representation tool, is introduced to express the hesitancy of decision-makers regarding assignment examples and the performance of alternatives on qualitative criteria. Besides, to comprehensively reflect the preference of a decision-maker, a weighted additive value function is proposed based on the preference disaggregation technique to calculate the comprehensive scores of alternatives in which the weights are determined by the best-worst method. Finally, a case study on the sorting of down coats for sale demonstrates the applicability and superiority of our proposed method.

Measure of departure from average marginal homogeneity for the analysis of collapsed ordinal square contingency tables

Summary In clinical research, collected data are often classified into ordered categories using a set threshold to evaluate efficacy and safety of treatment. Data can be summarized as a shift table, which displays the change in the frequency of subjects across specified categories from the baseline to post-baseline. Although ordered categories are sometimes recombined into three categories, the combined patterns vary. To consider various collapsed patterns comprehensively, this paper proposes a new measure that represents the degree of departure from average marginal homogeneity, and can distinguish between two kinds of marginal inhomogeneity. Additionally, applications of the proposed measure to clinical data are discussed.

Methodological approaches for analysing data from therapeutic efficacy studies

Several anti-malarial drugs have been evaluated in randomized clinical trials to treat acute uncomplicated Plasmodium falciparum malaria. The outcome of anti-malarial drug efficacy studies is classified into one of four possible outcomes defined by the World Health Organization: adequate clinical and parasitological response, late parasitological failure, late clinical failure, early treatment failure. These four ordered categories are ordinal data, which are reduced to either a binary outcome (i.e., treatment success and treatment failure) to calculate the proportions of treatment failure or to time-to-event outcome for Kaplan–Meier survival analysis. The arbitrary transition from 4-level ordered categories to 2-level type categories results in a loss of statistical power. In the opinion of the authors, this outcome can be considered as ordinal at a fixed endpoint or at longitudinal endpoints. Alternative statistical methods can be applied to 4-level ordinal categories of therapeutic response to optimize data exploitation. Furthermore, network meta-analysis is useful not only for direct comparison of drugs which were evaluated together in a randomized design, but also for indirect comparison of different artemisinin-based combinations across different clinical studies using a common drug comparator, with the aim to determine the ranking order of drug efficacy. Previous works conducted in Cameroonian children served as data source to illustrate the feasibility of these novel statistical approaches. Data analysis based on ordinal end-point may be helpful to gain further insight into anti-malarial drug efficacy.

Utilizing the Logistic Regression Model in Analyzing the Categorical Data of Economic Effects

The categorical data has a significant role in representing statistical binary variables, and they are analyzed by means of grouping the response variable into ordered categories. Thereby, the dependent variable becomes of type binary qualitative variable. The data related to the financial position of world countries is classified within the categorical data. This work is to study the economic effects of an individual's different factors on determining the richness or poorness levels of a selected population of countries. Moreover, a logistic regression model is to be created to estimate these levels. As a sample of research, the categorical data relevant to the financial status of 20 Arabic countries were drawn from the website of the World Bank, WB. In addition, for comparison purpose, another similar set of categorical data was generated by MATLAB too. The paper has been based on two hypotheses, first is the well-known regression models, like the ordinary least squares or maximum likelihood, are not accurate in case of binary qualitative variables. Second, is utilizing the logistic regression model as an alternative model to achieve the paper goal. The paper results, for both WB data and MATLAB data, have successfully proved the ability of the logistic regression model in manipulating the categorical data and predicting the coefficients of the corresponding regression models.

Robust Ordinal Regression: User Credit Grading with Triplet Loss-Based Sampling

With the development of social media sites, user credit grading, which served as an important and fashionable problem, has attracted substantial attention from a slew of developers and operators of mobile applications. In particular, multi-grades of user credit aimed to achieve (1) anomaly detection and risk early warning and (2) personalized information and service recommendation for privileged users. The above two goals still remained as up-to-date challenges. To these ends, in this article, we propose a novel regression-based method. Technically speaking, we define three natural ordered categories including BlockList , GeneralList , and AllowList according to users’ registration and behavior information, which preserve both the global hierarchical relationship of user credit and the local coincident features of users, and hence formulate user credit grading as the ordinal regression problem. Our method is inspired by KDLOR ( kernel discriminant learning for ordinal regression ), which is an effective and efficient model to solve ordinal regression by mapping high-dimension samples to the discriminant region with supervised conditions. However, the performance of KDLOR is fragile to the extreme imbalanced distribution of users. To address this problem, we propose a robust sampling model to balance distribution and avoid overfit or underfit learning, which induces the triplet metric constraint to obtain hard negative samples that well represent the latent ordered class information. A step further, another salient problem lies in ambiguous samples that are noises or located in the classification boundary to impede optimized mapping and embedding. To this problem, we improve sampling by identifying and evading noises in triplets to obtain hard negative samples to enhance robustness and effectiveness for ordinal regression. We organized training and testing datasets for user credit grading by selecting limited items from real-life huge tables of users in the mobile application, which are used in similar problems; moreover, we theoretically and empirically demonstrate the advantages of the proposed model over established datasets.

Single-cell DNA targeted sequencing (scDNA-seq) to test therapeutic vulnerabilities in urothelial cancer (UC) patient-derived organoids (PDO).

464 Background: Genomic alterations in FGFR3, PIK3CA, and CDKN2A are common actionable targets in urothelial cancer (UC). We aimed to determine the efficacy of alpelisib, a PIK3CA inhibitor, and abemaciclib, a CDK4/6 inhibitor in bladder cancer patient-derived organoid using single-cell targeted DNA sequencing. Methods: We established a patient-derived UC organoid (PDO) harboring the FGFR3 mutation (p.Y375C), the PIK3CA (p. E452K) mutation, and CDKN2A deletion, which we characterized using whole-genome sequencing. We generated dose-response curves of alpelisib, abemaciclib, and erdafitinib (an FGFR3 inhibitor) in PDO cells to determine the IC50 concentrations. The scDNA-seq (Tapsteri) platform was used to measure changes in the variant allele frequencies (VAF) and clonal fractions post-treatment. The Chi-square test for trend was used to test for linear trend across ordered categories. Results: scDNA-seq was performed after treating PDO cells with 3uM erdafitinib or DMSO. A total of 7000 single cells were obtained (4179 cells treated with erdafitinib vs. 2821 cells treated with DMSO). After removing variants mutated in <50% of cells, we identified 94 clonal variants. As expected, cells harboring FGFR3 Y375C have significantly decreased post erdafitinib treatment compared to DMSO-treated cells (66.05% vs. 82.36%, p<0.0001). We identified mutations in two genes ( RAB31 and SMAD4) that were associated with clonal expansion following FGFR3 inhibition (12% vs. 53% and 13% vs. 26%, respectively). We treated PDO cells with 3uM abemaciclib. We identified three genes harboring SNVs ( RB1, GNAQ, and SMAD4). The SNVs harboring cells were significantly decreased after abemaciclib compared to DMSO (p<0.0001). Then, we treated the cells with 1uM alpelisib for 72 hours alone or combined with abemaciclib (0.1uM). We identified that 100% of cells harbored the PIK3CA mutation E452K at pre-treatment, which limited our ability to detect significant changes in VAF post-treatment. Instead, we analyzed the effect on the FGFR3 Y375C clone. Using trend analysis, there was a significant reduction of FGFR3-mutant cells observed across the three conditions, abemaciclib + alpelisib vs. alpelisib alone vs. DMSO (74% vs. 85% vs. 92%, trend test p<0.0001), suggesting in vitro efficacy of alpelisib alone and significant synergism with the addition of abemaciclib. Conclusions: This study established the feasibility of using scDNA-seq as a promising tool to study the clonal evolution patterns in patient-derived UC organoids. Combined pharmacologic inhibition of CDK4/6 and PIK3CA showed more in vitro sensitivity than PIK3CA inhibition alone.

Association Between Infection Control Resource Use and the Number of Penalties Under Medicare’s Hospital-Acquired Condition Reduction Program

Background: The Hospital-Acquired Condition Reduction Program (HACRP) is a pay-for-performance Medicare program that promotes reducing patient harm, particularly healthcare-associated infections (HAIs). We examined the association between infection-control–related activities and the number of penalties a hospital received between fiscal years 2015 and 2018. Methods: We used logistic regression with ordered categories to assess infection control resource use and the number of penalties, an ordered categorical dependent variable with 5 categories ranging from 0 to 4, as of 2018. Data sources included National Healthcare Safety Network, American Hospital Association Annual Survey, Medicare Impact and Cost Report files, and Data.Medicare.gov. We excluded hospitals lacking data to calculate any HACRP score or component score for HAI and hospitals missing observations for model variables (301 hospitals). We assessed the following model variables: teaching hospital status, infection preventionists (IP) per 1,000 beds, surveillance hours per week per bed, other infection control activities per week per bed, nurse-to-bed ratio, housekeeping expenditure per 10,000 beds, nursing position vacancies per bed, bed size, electronic health record (EHR) implementation, number of skilled nursing beds, rural or urban location, and Medicare patient case-mix (cmi_quartiles). Results: In our model, negative logit model point estimates indicated that increased values of the variable are associated with a lower odds of having a higher number of penalties. The final data set consisted of 3,004 US hospitals. Lower penalties were significantly associated with higher IP-to-bed ratio. Although the point estimates were <1, an association between lower penalties and higher nurse-to-bed ratios or electronic health records was not demonstrated (Table 1). Conclusions: Our results suggest that after controlling for selected hospital structural factors, incremental resources related to infection control have a protective association with HCARP penalties.Funding: NoneDisclosures: None