Methods Needed to Measure Predictive Accuracy: A Study of Diabetic Patients

Diabetes is one of the leading causes of morbidity and mortality and it can result in several complications such as kidney failure, heart failure, stroke, and blindness making it a major medical and public health concern in the United States. Statistical methods are important to detect risk factors and identify the best sampling plan to determine predictive bounds for diabetic patients' data. The main objective of this paper is to identify the best fit bootstrapping sampling method and to draw the predictive bound considering diabetes patient data. A random sample was used from the National Health and Nutritional Examination Survey (NHANES) for this study. We found that there were significant relationships between age, marital status, and race/ethnicity with diabetes status (p<0.001) and no relationship was observed between gender and diabetes status. We ran the logistic regression to identify the risk factors from the data. We identified that the significant risk factors are age (p<0.001), total protein (p<0.001), fast food (p<0.0339), and direct HDL (p<0.001). This study provides evidence that the parametric bootstrapping method is the best fit method compared with other methods to estimate the predictive error bounds. These findings will be of great significance for identifying the best sampling methods, which can increase the statistical accuracy of laboratory clinical research of diabetes. This will also allow for the determination of precise risk factors that will best represent the data by detecting mild and extreme outliers from disease observations. Therefore, these results will be useful for researchers and clinicians to select the best sampling methods to study diabetes and other diseases in order to maximize the accuracy of their results. This article is part of a Special Issue entitled: Oxidative Stress and Mitochondrial Quality in Diabetes/Obesity and Critical Illness Spectrum of Diseases - edited by P. Hemachandra Reddy.