Biostatistics Research

A Simulation Study Comparing Tree-Based Methods in Identifying Interactions of Continuous and Binary Variables for Prediction of Increased Risk of Disease

Sybil Prince Nelson — 2023-04-18

Tree-based methods are commonly used to create models that predict an output based on several input variables. Classification and Regression Trees (CARTs) is a popular algorithm that builds tree-like graphs for predicting continuous and categorical dependent variables, but it has been shown to be biased toward the inclusion of continuous variables. Conditional inference is a technique used to alleviate this bias. C.Logic is an alternative tree-based method that uses Boolean logic to create classification trees. Previous research has shown that C.Logic is superior to CART in identifying interactions that lead to an increased risk of disease. No comparison has been made between the C.Logic package and CART with conditional inference as found in a package called Party. In this paper, a simulation study is used to compare the capability of these two algorithms to identify interactions between continuous and binary variables. It is found that while both methods succeed in identifying correct interactions, C.Logic is more effective. The C.Logic algorithm does a better job of alleviating the bias toward continuous variables when attempting to identify interacting variables that lead to an increased risk of disease.

Spatial Characteristics of the Fungus Powdery Mildew (Erysiphe neolycopersici) on Tomatoes and its Spread in Industrial Greenhouses

Anastasia Sokolidi — 2023-01-12

In regions with cool temperate climates, tomatoes are grown on an industrial scale in large greenhouses. There the crops are susceptible to infection by powdery mildew, the fungus Erysiphe neolycopersici, which is introduced largely as fungal spores from outside the greenhouses and spread by wind within them. We have monitored the spread of the disease and mapped its distribution in four commercial greenhouses throughout the growing season to understand its aetiology. We modelled the patterns of infection geostatistically, each comprising a deterministic long-range trend plus a short-range spatially correlated random residual. We identified three main kinds of pattern; one consisted of a constant plus a spatially correlated residual, second comprised a linear trend throughout the greenhouse plus a correlated random residual, and third, the trend had the form of a bell akin to a Gaussian surface plus, again, a correlated random residual. Here, we show three examples of these distributions and the detail of their geostatistical analysis using both the traditional method of moments (MoM) estimation of variograms and residual maximum likelihood (REML) to separate the deterministic and random components. The analytical modelling is followed by ordinary punctual kriging in the first case, by universal kriging in the second, and by regression kriging in the third case to display the infection as isarithmic ("contour") maps. We interpret the first form of distribution as arising from numerous foci as spores landed on the leaves from various sources spread by air currents and the movement of workers along the paths through the greenhouse. In the second case, the disease seemed to have spread from an infection introduced through the main door in one corner of the greenhouse and spread from there by the workers and air currents. The third infection arose near the centre of the greenhouse by the main path and spread outwards from there. In all three examples, the main pathways seemed important routes along which the fungus spread.

Accuracy and Precision of Bone Scan, Magnetic Resonance Imaging (MRI) and Digital Radiography in Limb Salvage Surgery for Long Bone Tumors

Gomadam Kuppusamy Rangarajan — 2023-04-18

Introduction: The aim of this study is to evaluate the level of accuracy and precision of bone scan (BS), magnetic resonance imaging (MRI) and digital radiography (DR) to measure long bone tumors to design custom made prosthesis (CMP)/modular prosthesis (MP) in limb salvage surgery (LSS). Material and methods: There are two separate groups, one is phantom study and another one is patient’s study. Phantom study: done with Jaszack phantom for gamma camera (GC) and indigenous phantom for MRI and DR. Three independent imaging professionals (nuclear medicine physicians and radiologists) measured the distance between standardized, preselected points on the Jaszack phantom in the GC and indigenous phantom on the coronal and sagittal view of the MRI scan and in DR. The measured values were compared it with the known values for phantom measurement. Patient’s study: Patients with a malignant bone tumor of the lower/upper limbs enrolled from 2020-2021 at the institute were taken up for the retrospective study. Totally 36 patients were enrolled, 24 patients were male (Ages: 2 to 45 years) and 12 patients were female (Ages: 8 to 18 years). Three independent imaging professionals measured the patient’s long bone in the BS, MRI and DR and compared with histopathological specimen measurement after LSS. Statistical analysis: Descriptive statistics using appropriate measures of central tendency, dispersion, Karl-Pearson correlation coefficient and paired t-test were employed. Results: A near perfect positive correlation was evident between all three pairs of the BS, MRI scan and DR values and a positive agreement within 1 mm was around 95%. Conclusion: For the phantom study, we conclude that GC and MRI measurements are equal in physical measurements and multiplication correction factor (MCF) = 1. DR measurements were found to be near equal physical measurements and MCF = 0.9104 and three observer’s measurements values were also near normal.

Author Impact Factor: A Framework for Evaluating Authorship and Scientific Contribution

Brian C. Drolet — 2021-11-24

In the decade since Hirsch defined h, there has been widespread acceptance of the h-index as a bibliometric indicator. Although the h-index has been validated in numerous applications and settings, the bibliometric has some important limitations. Most importantly, the h-index does not account for authors' individual contributions to manuscripts within their h-defining body of work. Since each author makes a variable contribution to a piece of scientific work, an author-adjusted index would more fairly reflect scholarly productivity. We propose the author impact factor (AIF), which accounts for authorship position and number of co-authors, to adjust the h-index and more fairly account for contributions to the body of scientific work. The AIF is calculated from the h-index and an author's proportional contribution (α) to each h-defining manuscript. The α is based on authorship position and the number of co-authors. Using the golden ratio (φ), the calculation of α for each h-index defining manuscript is simple and axiomatic. To demonstrate the utility of this index, we calculated the AIF for a sample of high-impact scientists. The results show that the AIF maintains all the benefits of the h-index while adjusting the bibliometric for author-specific factors. Therefore, AIF more accurately reflects total "research output" and can be used to better compare authors' scientific contributions.

Nonoptimal Follow-up Times Make It Difficult to Detect the Epidemiological Inverse Relationship Between 25-Hydroxyvitamin D and Lung Cancer

Ari Voutilainen — 2021-09-06

Background: Previous studies have reported controversial conclusions regarding the association between circulating 25-hydroxyvitamin D (25-HVD) and lung cancer risk. Objectives: To test the hypothesis that the controversial conclusions can be due to different follow-up times (FUT) and because of interpreting findings predominantly based on statistical significance. Methods: The Kuopio Ischaemic Heart Disease Risk Factor Study provided data. We used the Cox regression to study the association between 25-HVD and lung cancer risk in 2578 middle-aged Finnish men. Out of them, 808 were free of cancer and willing to participate in follow-up examinations 11 years after baseline. We repeated all analyses for them. Results: Higher circulating 25-HVD predicted lower lung cancer risk over the entire follow-up period of 33 years. The hazard ratio (HR) of the highest vs. the lowest 25-HVD tertile adjusted for age, smoking, alcohol consumption, body weight status, inflammatory status, physical activity, and diet was lowest, 0.39 (95% CI: 0.17-0.87), when the FUT was 15 years. The HR was statistically significant (p < 0.05) only when the FUT was 15-17 years. In the sub-cohort, Cohen's d denoted a large or medium effect during the first 11 years. Conclusions: The optimal FUT in this prospective cohort study investigating the association between circulating 25-HVD and lung cancer risk in a middle-aged male population, 50-55 years, was 15 years. When the population aged for 11 years, shorter FUTs became more pertinent. In general, interpreting results of prospective cohort studies with respect to FUTs and effect sizes may lead to more precise conclusions. Moreover, researchers should consider FUTs when they combine studies meta-analytically.

Biological Network Mining

Zongliang Yue — 2023-04-10

In this survey, we explore the latest methods and trends in constructing and mining biological networks. We delve into cutting-edge techniques such as weighted gene co-expression network analysis (WGCNA), step-level differential response (SLDR), Biomedical Entity Expansion, Ranking and Explorations (BEERE), Weighted In-Network Node Expansion and Ranking (WINNER), and Weighted In-Path Edge Ranking (WIPER) from the Bioinformatics community, as well as breakthroughs in graph mining methods like parallel subgraph mining systems, temporal graph algorithms, and deep learning. To ensure a solid foundation, we provide an introductory-level overview of six well-established network types in systems biology. In addition, we offer a concise and accessible overview of strategies for network construction, including gene co-expression networks (GCNs), gene regulatory networks (GRNs), and literature-mined biomedical networks. We explain biological network mining in interdisciplinary domains, catering to both biomedical researchers and data mining experts. Our goal is to provide a comprehensive guide that doesn't require a significant time investment. We believe that these current trends will help readers become familiar with the topic and the practical applications of these tools in real-world studies.