Annals of Surgical Oncology - This study aimed to assess the performance of the pre- and postoperative early recurrence after surgery for liver tumor (ERASL) models at external validation.... 相似文献
Background. Neural networks are nonparametric, robust, pattern recognition techniques that can be used to model complex relationships.
Methods. The applicability of multilayer perceptron neural networks (MLP) to coronary artery bypass grafting risk prediction was assessed using The Society of Thoracic Surgeons database of 80,606 patients who underwent coronary artery bypass grafting in 1993. The results of traditional logistic regression and Bayesian analysis were compared with single-layer (no hidden layer), two-layer (one hidden layer), and three-layer (two hidden layer) MLP neural networks. These networks were trained using stochastic gradient descent with early stopping. All prediction models used the same variables and were evaluated by training on 40,480 patients and cross-validation testing on a separate group of 40,126 patients. Techniques were also developed to calculate effective odds ratios for MLP networks and to generate confidence intervals for MLP risk predictions using an auxiliary “confidence MLP.”
Results. Receiver operating characteristic curve areas for predicting mortality were approximately 76% for all classifiers, including neural networks. Calibration (accuracy of posterior probability prediction) was slightly better with a two-member committee classifier that averaged the outputs of a MLP network and a logistic regression model. Unlike the individual methods, the committee classifier did not overestimate or underestimate risk for high-risk patients.
Conclusions. A committee classifier combining the best neural network and logistic regression provided the best model calibration, but the receiver operating characteristic curve area was only 76% irrespective of which predictive model was used. 相似文献
Decreased kidney function, determined using a serum creatinine–based estimation of GFR, is associated with a higher risk for mortality from cardiovascular disease. Equations incorporating cystatin C improve the estimation of GFR, but whether this improves the prediction of risk for mortality is unknown. We measured cystatin C on 6942 adult participants in the Third National Health and Nutrition Examination Survey Linked Mortality File, including all participants who had high serum creatinine (>1.2 mg/dl for men; >1.0 mg/dl for women) or were older than 60 yr and 25% random sample of participants who were younger than 60 yr. We estimated GFR using equations that included standardized serum creatinine, cystatin C, or both. Participant data were linked to the National Death Index. A total of 1573 (22.7%) deaths (713 deaths from cardiovascular disease) occurred during a median of 8 yr. Lower estimated GFR based on cystatin C was strongly associated with higher risk for overall and cardiovascular mortality across the range of normal to moderately decreased estimated GFR. Creatinine-based estimates of GFR resulted in weaker associations, with the association between estimated GFR and all-cause mortality reversed at higher levels of estimated GFR. An equation using both creatinine and cystatin C (in addition to age, race, and gender) resulted in weaker associations than equations using only cystatin C (with or without age, race, and gender). In conclusion, despite better performance in terms of estimating GFR, equations based on both cystatin C and creatinine do not predict mortality as well as equations based on cystatin C alone.Moderately decreased kidney function, as estimated from equations based on serum creatinine, is associated with an elevated risk for mortality, both in individuals with existing cardiovascular disease (CVD) and in the general population.1–3 Serum levels of creatinine, however, are affected by other factors in addition to GFR, most importantly, variations in creatinine generation as a result of differences in muscle mass and turnover.4 Muscle wasting as a result of chronic illness is associated with lower creatinine generation, leading to an overestimation of GFR in such individuals. Because these same individuals are at an elevated risk for mortality, this systematic bias would result in an underestimation of the association between decreased GFR and mortality risk. This may be a particular problem in the elderly because of their higher prevalence of chronic illness and higher risk for mortality. In addition, currently available GFR estimates based on serum creatinine are less accurate at higher levels of kidney function, probably reflecting a greater proportional contribution of creatinine generation to variation in the serum creatinine level than at lower levels of kidney function.5Cystatin C is a marker of kidney function that is less sensitive to differences in muscle mass than is serum creatinine.4 Cystatin C predicted total and cardiovascular mortality risk more strongly than creatinine-based estimates of GFR in prospective studies of older adults.6,7 Data on younger individuals are lacking. Recent data allow GFR estimation from serum cystatin C and showed that equations using both serum creatinine and cystatin C, in addition to age, race, and gender, are more closely correlated with directly measured GFR among individuals with chronic kidney disease (CKD) than equations based on either marker alone.8 The associations with mortality risk of eGFR based on serum creatinine, cystatin C, or the combination of the two, however, has not been studied. As the use of cystatin C as a marker of cardiovascular risk increases, it is critical to understand how one should combine information on serum creatinine with cystatin C for risk prediction. We analyzed up to 13 yr of mortality follow-up on 6942 participants in the Third National Health and Nutrition Examination Survey (NHANES III) to assess the association of eGFR based on equations using creatinine, cystatin C, or both markers with the risk for all-cause and cardiovascular mortality in a representative sample of US adults. 相似文献