Comparison of multiple linear regression and machine learning methods in predicting cognitive function in older Chinese type 2 diabetes patients.

INTRODUCTION: The prevalence of type 2 diabetes (T2D) has increased dramatically in recent decades, and there are increasing indications that dementia is related to T2D. Previous attempts to analyze such relationships principally relied on traditional multiple linear regression (MLR). However, recently developed machine learning methods (Mach-L) outperform MLR in capturing non-linear relationships. The present study applied four different Mach-L methods to analyze the relationships between risk factors and cognitive function in older T2D patients, seeking to compare the accuracy between MLR and Mach-L in predicting cognitive function and to rank the importance of risks factors for impaired cognitive function in T2D. METHODS: We recruited older T2D between 60-95 years old without other major comorbidities. Demographic factors and biochemistry data were used as independent variables and cognitive function assessment (CFA) was conducted using the Montreal Cognitive Assessment as an independent variable. In addition to traditional MLR, we applied random forest (RF), stochastic gradient boosting (SGB), Naïve Byer's classifier (NB) and eXtreme gradient boosting (XGBoost). RESULTS: Totally, the test cohort consisted of 197 T2D (98 men and 99 women). Results showed that all ML methods outperformed MLR, with symmetric mean absolute percentage errors for MLR, RF, SGB, NB and XGBoost respectively of 0.61, 0.599, 0.606, 0.599 and 0.2139. Education level, age, frailty score, fasting plasma glucose and body mass index were identified as key factors in descending order of importance. CONCLUSION: In conclusion, our study demonstrated that RF, SGB, NB and XGBoost are more accurate than MLR for predicting CFA score, and identify education level, age, frailty score, fasting plasma glucose, body fat and body mass index as important risk factors in an older Chinese T2D cohort.

Urokinase Plasminogen Activator Deficiency Aggravates Cationic Bovine Serum Albumin-Induced Membranous Nephropathy Through T Helper Cell Type 2-Prone Immune Response in Mice.

Urokinase plasminogen activator (uPA) is a crucial activator of the fibrinolytic system that modulates tissue remodeling, cancer progression, and inflammation. However, its role in membranous nephropathy (MN) remains unclear. To clarify this issue, an established BALB/c mouse model mimicking human MN induced by cationic bovine serum albumin (cBSA), with a T helper cell type 2-prone genetic background, was used. To induce MN, cBSA was injected into Plau knockout (Plau-/-) and wild-type (WT) mice. The blood and urine samples were collected to measure biochemical parameters, such as serum concentrations of immunoglobulin (Ig)G1 and IgG2a, using enzyme-linked immunoassay. The kidneys were histologically examined for the presence of glomerular polyanions, reactive oxygen species (ROS), and apoptosis, and transmission electron microscopy was used to examine subepithelial deposits. Lymphocyte subsets were determined using flow cytometry. Four weeks post-cBSA administration, Plau-/- mice exhibited a significantly higher urine protein-to-creatine ratio, hypoalbuminemia, and hypercholesterolemia than WT mice. Histologically, compared to WT mice, Plau-/- mice showed more severe glomerular basement thickening, mesangial expansion, IgG granular deposition, intensified podocyte effacement, irregular thickening of glomerular basement membrane and subepithelial deposits, and abolishment of the glycocalyx. Moreover, increased renal ROS levels and apoptosis were observed in Plau-/- mice with MN. B-lymphocyte subsets and the IgG1-to-IgG2a ratio were significantly higher in Plau-/- mice after MN induction. Thus, uPA deficiency induces a T helper cell type 2-dominant immune response, leading to increased subepithelial deposits, ROS levels, and apoptosis in the kidneys, subsequently exacerbating MN progression in mice. This study provides a novel insight into the role of uPA in MN progression.

Visfatin and Retinol Binding Protein-4 in Young-Onset Type 2 Diabetes Mellitus.

Background and Objectives: The prevalence of type 2 diabetes mellitus in adolescents has increased rapidly in recent decades. However, the role of adipokines on pathophysiology in young-onset type 2 diabetes mellitus (YDM) is not clear. In this article, we explored the relationships between the adipokines (visfatin and retinol binding protein 4 (RBP4)) and metabolic syndrome (MetS) components in both YDM and late-onset type 2 diabetes mellitus (ODM). Materials and Methods: There were 36 patients with YDM (23.6 ± 4.8 years) and 36 patients with ODM (54.3 ± 10.1 years) enrolled. Visfatin, RBP4, and MetS components were measured. The relationships between visfatin, RBP4 and MetS components were assessed in YDM and ODM. Results: The visfatin, but not the RPB4 level, was significantly higher in YDM than in ODM. After adjusting for age and body mass index, visfatin was not related to any MetS components except that there was a negative correlation with fasting plasma glucose (FPG). As for RPB4, triglyceride was found to be positively and FPG negatively related to RBP4 in YDM. However, in ODM, the only positive relationship that existed was between RBP4 and diastolic blood pressure. Conclusions: In conclusion, both visfatin and RBP4 had certain roles in diabetes and MetS although their relationships were different in YDM and ODM. Further studies are needed to explore their physiological and pathological effects in glucose metabolism.

Predictors of abnormality in thallium myocardial perfusion scans for type 2 diabetes.

Type 2 diabetes mellitus (T2DM) increases coronary artery disease (CAD) risk. In this study, we used T2DM clinical variables to predict abnormality in thallium-201 myocardial perfusion scans (Th-201 scans). These clinical variables were summed stress score (SSS), summed rest score, and summed difference score (SDS), with data obtained from 368 male and 428 female participants with T2DM. Multiple linear regression results were as follows. In male participants, body mass index (BMI) and creatinine (Cr) were associated with SSS (ß = 0.224, p < 0.001; ß = 0.140, p = 0.022, respectively), and only BMI was associated with SDS (ß = 0.174, p = 0.004). In female participants, BMI and high-density lipoprotein cholesterol level were associated with SSS (ß = 0.240, p < 0.001; ß = - 0.120, p = 0.048, respectively), and only BMI was correlated with SDS (ß = 0.123, p = 0.031). Our multivariate logistic regression indicated that in male and female participants, BMI was the only independent indicator of high SSS (SSS ≥ 9). In this study, we demonstrated that male patients have a higher SSS and SDS than female patients do in Th-201 scans for T2DM in a Chinese population. For male and female patients, BMI was the strongest predictor of abnormality in Th-201 scans. Our results can help clinicians identify patients with T2DM at high risk of CAD.

Reduced Expression of Urokinase Plasminogen Activator in Brown Adipose Tissue of Obese Mouse Models.

In recent decades, the obesity epidemic has resulted in morbidity and mortality rates increasing globally. In this study, using obese mouse models, we investigated the relationship among urokinase plasminogen activator (uPA), metabolic disorders, glomerular filtration rate, and adipose tissues. Two groups, each comprised of C57BL/6J and BALB/c male mice, were fed a chow diet (CD) and a high fat diet (HFD), respectively. Within the two HFD groups, half of each group were euthanized at 8 weeks (W8) or 16 weeks (W16). Blood, urine and adipose tissues were collected and harvested for evaluation of the effects of obesity. In both mouse models, triglyceride with insulin resistance and body weight increased with duration when fed a HFD in comparison to those in the groups on a CD. In both C57BL/6J and BALB/c HFD mice, levels of serum uPA initially increased significantly in the W8 group, and then the increment decreased in the W16 group. The glomerular filtration rate declined in both HFD groups. The expression of uPA significantly decreased in brown adipose tissue (BAT), but not in white adipose tissue, when compared with that in the CD group. The results suggest a decline in the expression of uPA in BAT in obese m models as the serum uPA increases. There is possibly an association with BAT fibrosis and dysfunction, which may need further study.

Deficiency of Urokinase Plasminogen Activator May Impair ß Cells Regeneration and Insulin Secretion in Type 2 Diabetes Mellitus.

: Background: The relationship between urokinase-type plasminogen activator (uPA) and the development of type 2 diabetes mellitus (T2DM) was investigated in the study by using mice and cell models, as well as patients with T2DM. METHODS: In mice models, wild-type and uPA knockout (uPA-/-) BALB/c mice were used for induction of T2DM. In cell models, insulin secretion rate and ß cell proliferation were assessed in normal and high glucose after treating uPA siRNA, uPA, or anti-uPA antibody. In our clinical study, patients with T2DM received an oral glucose-tolerance test, and the relationship between uPA and insulin secretion was assessed. RESULTS: Insulin particles and insulin secretion were mildly restored one month after induction in wild-type mice, but not in uPA-/- mice. In cell models, insulin secretion rate and cell proliferation declined in high glucose after uPA silencing either by siRNA or by anti-uPA antibody. After treatment with uPA, ß cell proliferation increased in normal glucose. In clinical study, patients with T2DM and higher uPA levels had better ability of insulin secretion than those with lower uPA levels. CONCLUSION: uPA may play a substantial role in insulin secretion, ß cell regeneration, and progressive development of T2DM. Supplementation of uPA might be a novel approach for prevention and treatment of T2DM in the future.

Impact of the use of anti-diabetic drugs on survival of diabetic dialysis patients: a 5-year retrospective cohort study in Taiwan.

BACKGROUND: Type 2 diabetes mellitus (DM) and associated complications are common in patients with chronic kidney disease (CKD) and can increase morbidity and mortality. A longitudinal 5-year observational study was conducted to investigate whether the use of anti-diabetic medications or not affected survival rates of diabetic dialysis patients. METHODS: Using a data sample of a million patients from Taiwan's National Health Insurance Database, a retrospective cohort study surveyed patients with type 2 DM who began dialysis between 2002 and 2007. The study population was classified into groups using or not using anti-diabetic drugs. The group using anti-diabetic drugs was then categorized into 3 subgroups, including use of only oral hypoglycemic agents (OHAs), only insulin, and OHAs-combined insulin groups. Subjects of these four groups were followed 5 years or to date of death. Three major areas were analyzed: (1) demographic data and medical history; (2) survival prognosis and causes of death; and (3) effects on survival prognosis of different classes of OHAs. RESULTS: A total of 912 patients fitting inclusion criteria were enrolled and followed-up for 5 years or to date of death. A total 465 patients died, and those not using anti-diabetic drugs (67.34 %) had a higher mortality rate than those using anti-diabetic drugs (46.42 %). After the multivariate analysis, group of OHAs-combined insulin had the lowest risk of death (HR 0.36, 95 % CI 0.27-0.47), followed by OHAs alone (HR 0.49, 95 % CI 0.38-0.63) and then insulin alone (HR 0.67, 95 % CI 0.51-0.88). To clarify four classes of OHAs (sulfonylurea, α-glucosidase inhibitors, meglitinide, and thiazolidinedione) are used in Taiwan for uremia patient with type 2 DM, and in our study, there were no significant differences in survival prognosis for the four drugs. Finally, the most common cause of death was infectious disease and there were no significant differences among the four groups. CONCLUSION: This 5-year observational study results suggested that diabetic dialysis patients with anti-diabetic drugs had a lower risk of death compared with those without anti-diabetic drugs. Despite insulin therapy, appropriate OHAs should play an important role in treating these patients.

Significance of the urokinase-type plasminogen activator and its receptor in the progression of focal segmental glomerulosclerosis in clinical and mouse models.

BACKGROUND: suPAR biomarker generally considered a pathogenic factor in FSGS. However, studies have been published that dispute this conclusion. The current study was designed to investigate the roles of uPA and suPAR in FSGS in clinical and mouse models. METHODS: Clinical subjects including those with biopsy-proven FSGS and MCD were enrolled. To verify the role of uPA in FSGS, Adriamycin was used to induce FSGS in uPA knockout (uPA(-/-)) and BALB/c (WT) mice. Proteinuria and suPAR, the cleaved/intact forms of the circulating suPAR, and possible proteases involving cleavage of the suPAR were also studied. RESULTS: FSGS clinical cases presented significantly higher serum levels of suPAR and Cr and lower serum levels of uPA. In the mice model, the uPA(-/-) group exhibited faster disease progression and worsening proteinuria than the WT group. In addition, the uPA(-/-) group had higher plasma suPAR levels, glomerular cell apoptosis, and dysregulation of the Th1/Th2 balance. In an analysis of suPAR variants in FSGS, both the intact and cleaved forms of the suPAR were higher in clinical subjects and the mouse model. However, the process of suPAR cleavage was not mediated by enzymatic activities of the uPA, elastase, or cathepsin G. CONCLUSIONS: A deficiency of uPA accelerated the progression of Adriamycin-induced mouse FSGS model. Decrease of serum uPA levels may be an indicator of the progression of FSGS in clinical subjects and animal models.

Association of complete blood cell counts with metabolic syndrome in an elderly population.

BACKGROUND: Metabolic syndrome's (MetS) role in predicting cardiovascular diseases and diabetes has been confirmed in many large cohort studies. Nontraditionally, hematogram components are significantly related to MetS in many different age groups. However, little is known about its role among the elderly. METHODS: We enrolled 18,907 subjects over the age of 65 years who underwent regular health examinations. They were divided into three groups according to age: young old (YO: ≥ 65 and < 74 years old), old old (OO: ≥ 75 and < 84 years old), and oldest old (ODO: ≥ 85 years old). The MetS components were determined, and correlations between MetS and hematogram components were evaluated using Pearson and multivariate linear regression analyses. The hematogram components were the independent variables evaluated separately against the dependent variable (MetS components). RESULTS: While SBP and HDL-C increased, most other MetS and hematogram parameters decreased in men with age. Fewer significant differences were noted among the women. In the YO and OO groups for both genders, the subjects with MetS had higher WBC and Hb. None of the hematogram components were different for subjects with or without MetS in the ODO group. Multiple regression results show that most of the relationships between hematogram and MetS components disappeared in the ODO groups. The WBC levels were mainly correlated with WC and TG. At the same time, Hb was associated with BP, FPG, and LDL-C. Compared to WBC and Hb, PLT was least related to MetS, except in the cases of LDL-C and TG. Among the MetS components, BMI, LDL-C, and TG were consistently related to all the hematogram components in YO and OO men. However, only TG had the same consistency among YO and OO women. CONCLUSIONS: This study's three major findings are as follows: WBC and Hb are associated with MetS, even among the YO and OO groups, regardless of gender; among the three hematogram components, Hb had the strongest and PLT had the weakest correlation with MetS; and TG is not the only component with relatively higher r values, and it is related to all hematogram components.

Using hematogram model to predict future metabolic syndrome in elderly: a 4-year longitudinal study.

OBJECTIVES: The metabolic syndrome (MetS) is proposed to predict future occurrence of cardiovascular diseases and diabetes. There are some other "non-traditional" risk factors such as hematogram components that are also related to the same endpoints as MetS. In this four-year longitudinal study, we used hematogram components to build models for predicting future occurrence of MetS in older men and women separately. METHODS: Subjects above 65 years without MetS and related diseases were enrolled. All subjects were followed up until they developed MetS or until up to four years from the day of entry, whichever was earlier. RESULTS: Among the 4539 study participants, 1327 developed MetS. Models were built for men and women separately and the areas under the receiver operation curves were significant. The Kaplan-Meier plot showed that the models could predict future MetS. Finally, Cox regression analysis showed that the hematogram model was correlated to future MetS with hazard ratios of 1.567 and 1.738 in men and women, respectively. CONCLUSION: Our hematogram models could significantly predict future MetS in elderly and might be more practical and convenient for daily clinical practice.

Use of white blood cell counts to predict metabolic syndrome in the elderly: a 4 year longitudinal study.

BACKGROUND: The white blood cell (WBC) count was one of the first inflammatory markers associated with metabolic syndrome (MetS). Recently, two longitudinal studies have demonstrated a cause and effect relationship between MetS and WBC counts among middle-aged adults. However, no study has used WBC cutoff values to predict MetS in the elderly. METHODS: Subjects who underwent routine health checkups, and were above 60 years of age, were enrolled. All subjects were followed-up until they developed MetS or until 4 years from the date of entry, whichever came earlier. Of the 4539 subjects eligible for enrollment, 3428 subjects comprised the study group and 1111 subjects comprised the validation group. RESULTS: WBC counts were significantly different between subjects with and without MetS in both genders. Using the ROC curve, WBC cutoff values of 5.7 × 10(3)/µl in males and 5.0 × 10(3)/µl in females were associated with the increased risk of developing MetS (all p values <0.001). Using these WBC cutoff values, the hazard ratio (HR) for females was significant in both the study group and validation group. However, the HR for males failed significance in the validation group. Kaplan-Meier plots and κ coefficients confirmed that the WBC cutoff value could predict development of MetS in women but not in men. CONCLUSIONS: The association between WBC count and MetS was gender specific. A WBC cutoff value greater than 5.0 10(3)/µl may predict the development of MetS in elderly women.

Predicting metabolic syndrome by using hematogram models in elderly women.

BACKGROUND: Low-grade inflammatory status was thought to be a major underlying mechanism in MetS. White blood cell (WBC) count was one of the inflammatory markers identified to be associated with MetS. Moreover, not only WBC but also hemoglobin (Hb) and platelet (PLT) were all associated with MetS. OBJECTIVE: In this study, we tried to build models by the hematogram components. In this way, we can not only predict the occurrence of MetS with a relatively low-cost and routine lab test, but also can understand more about the relationships between low grade inflammation and MetS. METHODS: We randomly collected subjects over 65 years old from MJ Health Screening Center's database between 1999 and 2008. After excluding subjects with medications for hypertension, hyperlipidemia and/or diabetes, 13,132 female were eligible for analysis. RESULTS: All the MetS components, hematogram parameters and age were higher in group with MetS. In the correlation matrix, all these three hematogram parameters (WBC, Hb and PLT) were correlated with MetS components except for the correlation between Hb and HDL-C. The ROC curves showed that the model 3 (PLT + Hb + WBC) had greatest area under the curve of 0.631 with the sensitivity of 58.1% and specificity of 61.4%. CONCLUSIONS: Our findings have shown that all the three hematogram parameters are related to MetS. The results not only shed light on the complex relationships, but also demonstrate a common and easy model to aid clinicians to be more aware of the occurrence of MetS.

Estimation of the disposition index based on components of metabolic syndrome.

Decreased insulin sensitivity (IS) and impaired insulin secretion are major pathological features of type 2 diabetes (T2DM). The product of these factors is the disposition index (DI). We aimed to develop an equation for predicting DI. We enrolled 167 participants in our study. We randomly assigned 126 (75%) of the participants to the study group, whose data would be used to build the equation for estimating the DI. The remaining 41 participants comprised the external validation group. A frequently sampled intravenous glucose-tolerance test was performed for all participants, and the IS, the glucose sensitivity, the acute insulin response to the glucose load, and the DI were determined. Three factors were selected from multiple linear regression analysis, and we constructed the equation log (DI) = 2.449 - 0.113 × fasting plasma glucose + 0.046 × body mass index - 0.612 × high-density lipoprotein cholesterol. Using this equation, the calculated log (DI) significantly correlated with the measured log (DI) in the external validation group (r = 0.428, p = 0.007). By using the equation based on the demographic data and measurements of metabolic syndrome components, the DI could be predicted with acceptable accuracy (r = 0.428). Because of the relationships between the MetS and demographic parameters, this method of predicting DI may help further clinicians' understanding of the underlying pathological mechanisms in T2DM.

Evaluation of second phase insulin secretion with simple surrogates derived from the mixed meal tolerance test in patients with type 2 diabetes.

The major contributors to the pathogenesis of type 2 diabetes are impaired insulin action and insulin secretion, including second phase insulin secretion (2nd ISEC). This study aimed to compare surrogates derived from the mixed meal tolerance test (MTT) with 2nd ISEC derived from modified low-dose graded glucose infusion (M-LDGGI) in patients with type 2 diabetes. We were subsequently able to decide which surrogate would be performed easily and accurately. Twenty type 2 diabetes patients were enrolled. They received both MTT and M-LDGGI. The standardized MTT meals were provided at 8:00 A.M. and 12:00 P.M. The M-LDGGI was a simplified version of the Polonsky method; only two 80-min stages of glucose infusion (2 and 6 mg/kg/min) were given. The slopes of the insulin to glucose curve during the test were regarded as the 2nd ISEC. First, we used the area under the insulin curve (AUC(IN)) during MTT to quantify the 2nd ISEC. The best correlated AUC(IN) was from 60-240 min. Second, the slopes between any two time points of the plasma insulin to glucose level (SLOPE(I/G)) were also assessed. The time period best correlated with 2nd ISEC was from 0-120 min (SLOPE0₋120). Finally, the insulin-to-glucose ratio (IGr) of each time point was used to estimate the 2nd ISEC, and the best correlation was observed at 180 min. In conclusion, estimating 2nd ISEC surrogates derived from MTT proved to be possible. The most accurate surrogate is the SLOPE0₋120, while IG(r180) is another less precise but more convenient method.

Redrawing Urokinase Receptor (uPAR) Signaling with Cancer Driver Genes for Exploring Possible Anti-Cancer Targets and Drugs.

During tumorigenesis, urokinase (uPA) and uPA receptor (uPAR) play essential roles in mediating pathological progression in many cancers. To understand the crosstalk between the uPA/uPAR signaling and cancer, as well as to decipher their cellular pathways, we proposed to use cancer driver genes to map out the uPAR signaling. In the study, an integrated pharmaceutical bioinformatics approach that combined modulator identification, driver gene ontology networking, protein targets prediction and networking, pathway analysis and uPAR modulator screening platform construction was employed to uncover druggable targets in uPAR signaling for developing a novel anti-cancer modality. Through these works, we found that uPAR signaling interacted with 10 of 21 KEGG cancer pathways, indicating the important role of uPAR in mediating intracellular cancerous signaling. Furthermore, we verified that receptor tyrosine kinases (RTKs) and ribosomal S6 kinases (RSKs) could serve as signal hubs to relay uPAR-mediated cellular functions on cancer hallmarks such as angiogenesis, proliferation, migration and metastasis. Moreover, we established an in silico virtual screening platform and a uPAR-driver gene pair rule for identifying potential uPAR modulators to combat cancer. Altogether, our results not only elucidated the complex networking between uPAR modulation and cancer but also provided a paved way for developing new chemical entities and/or re-positioning clinically used drugs against cancer.

Using Machine Learning to Predict Abnormal Carotid Intima-Media Thickness in Type 2 Diabetes.

Carotid intima-media thickness (c-IMT) is a reliable risk factor for cardiovascular disease risk in type 2 diabetes (T2D) patients. The present study aimed to compare the effectiveness of different machine learning methods and traditional multiple logistic regression in predicting c-IMT using baseline features and to establish the most significant risk factors in a T2D cohort. We followed up with 924 patients with T2D for four years, with 75% of the participants used for model development. Machine learning methods, including classification and regression tree, random forest, eXtreme gradient boosting, and Naïve Bayes classifier, were used to predict c-IMT. The results showed that all machine learning methods, except for classification and regression tree, were not inferior to multiple logistic regression in predicting c-IMT in terms of higher area under receiver operation curve. The most significant risk factors for c-IMT were age, sex, creatinine, body mass index, diastolic blood pressure, and duration of diabetes, sequentially. Conclusively, machine learning methods could improve the prediction of c-IMT in T2D patients compared to conventional logistic regression models. This could have crucial implications for the early identification and management of cardiovascular disease in T2D patients.

Changes in insulin resistance, glucose effectiveness, and first and second phases of insulin secretion in women aged 45-60 years old in Taiwan.

BACKGROUND: In women after menopause, the incidence of diabetes mellitus increases. Increased insulin resistance (IR), decreased glucose effectiveness (GE), and the first and second phases of insulin secretion (FPIS and SPIS), are the four most important factors that trigger glucose intolerance and diabetes (diabetogenic factor [DF]). In the cross-sectional study, we enrolled nondiabetic women between the ages of 45 and 60 years to observe the changes in DFs during the perimenopausal period and to elucidate the underlying mechanisms of diabetes in menopausal women. METHODS: We randomly enrolled 4194 women who underwent health checkups. Using demographic and biochemical data, IR, FPIS, SPIS, and GE were calculated using previously published equations. The relationship between the DFs and age was evaluated using a simple correlation. RESULTS: Body mass index, blood pressure, fasting plasma glucose, low-density lipoprotein cholesterol, triglyceride, and SPIS were higher, and GE was lower in older women (≥52 years old). A significant decrease in GE and increased SPIS were observed with age. However, no changes were observed in IR or FPIS. CONCLUSION: The IR and FPIS did not change during perimenopause. Increased SPIS may compensate for the decrease in GE, which is probably one of the reasons for the higher incidence of diabetes in menopausal women.

Comparison between linear regression and four different machine learning methods in selecting risk factors for osteoporosis in a Chinese female aged cohort.

BACKGROUND: Population aging is emerging as an increasingly acute challenge for countries around the world. One particular manifestation of this phenomenon is the impact of osteoporosis on individuals and national health systems. Previous studies of risk factors for osteoporosis were conducted using traditional statistical methods, but more recent efforts have turned to machine learning approaches. Most such efforts, however, treat the target variable (bone mineral density [BMD] or fracture rate) as a categorical one, which provides no quantitative information. The present study uses five different machine learning methods to analyze the risk factors for T-score of BMD, seeking to (1) compare the prediction accuracy between different machine learning methods and traditional multiple linear regression (MLR) and (2) rank the importance of 25 different risk factors. METHODS: The study sample includes 24 412 women older than 55 years with 25 related variables, applying traditional MLR and five different machine learning methods: classification and regression tree, Naïve Bayes, random forest, stochastic gradient boosting, and eXtreme gradient boosting. The metrics used for model performance comparisons are the symmetric mean absolute percentage error, relative absolute error, root relative squared error, and root mean squared error. RESULTS: Machine learning approaches outperformed MLR for all four prediction errors. The average importance ranking of each factor generated by the machine learning methods indicates that age is the most important factor determining T-score, followed by estimated glomerular filtration rate (eGFR), body mass index (BMI), uric acid (UA), and education level. CONCLUSION: In a group of women older than 55 years, we demonstrated that machine learning methods provide superior performance in estimating T-Score, with age being the most important impact factor, followed by eGFR, BMI, UA, and education level.

Comparison between multiple logistic regression and machine learning methods in prediction of abnormal thallium scans in type 2 diabetes.

BACKGROUND: The prevalence of type 2 diabetes (T2D) has been increasing dramatically in recent decades, and 47.5% of T2D patients will die of cardiovascular disease. Thallium-201 myocardial perfusion scan (MPS) is a precise and non-invasive method to detect coronary artery disease (CAD). Most previous studies used traditional logistic regression (LGR) to evaluate the risks for abnormal CAD. Rapidly developing machine learning (Mach-L) techniques could potentially outperform LGR in capturing non-linear relationships. AIM: To aims were: (1) Compare the accuracy of Mach-L methods and LGR; and (2) Found the most important factors for abnormal TMPS. METHODS: 556 T2D were enrolled in the study (287 men and 269 women). Demographic and biochemistry data were used as independent variables and the sum of stressed score derived from MPS scan was the dependent variable. Subjects with a MPS score ≥ 9 were defined as abnormal. In addition to traditional LGR, classification and regression tree (CART), random forest, Naïve Bayes, and eXtreme gradient boosting were also applied. Sensitivity, specificity, accuracy and area under the receiver operation curve were used to evaluate the respective accuracy of LGR and Mach-L methods. RESULTS: Except for CART, the other Mach-L methods outperformed LGR, with gender, body mass index, age, low-density lipoprotein cholesterol, glycated hemoglobin and smoking emerging as the most important factors to predict abnormal MPS. CONCLUSION: Four Mach-L methods are found to outperform LGR in predicting abnormal TMPS in Chinese T2D, with the most important risk factors being gender, body mass index, age, low-density lipoprotein cholesterol, glycated hemoglobin and smoking.

Gamma-Glutamyltransferase Is a Predictor for Future Changes of Diabetogenic Factors in Aged Chinese-A Four-Year Follow-Up Study.

Glucose homeostasis in the body is determined by four diabetes factors (DFs): insulin resistance (IR), glucose effectiveness (GE), and the two phases of insulin secretion-first phase (FPIS) and second phase (SPIS). Previous research points to a correlation between elevated levels of gamma-glutamyl transferase (γGT) and an increased risk of type 2 diabetes. This study investigates the relationship between γGT and the four DFs in older Chinese individuals. This study involved 2644 men and 2598 women, all of whom were relatively healthy Chinese individuals aged 60 years or more. The DFs were calculated using formulas developed by our research, based on demographic data and factors related to metabolic syndrome. Pearson's correlation was utilized to assess the relationship between γGT and the four DFs. The findings suggested a positive correlation between γGT and IR, FPIS, and SPIS, but a negative correlation with GE in men. Among women, only SPIS and GE were significantly correlated with γGT. The factors showed varying degrees of correlation, listed in descending order as follows: GE, SPIS, FPIS, and IR. This study confirms a significant correlation between γGT and DFs in this population, highlighting the noteworthy role of GE.