Dataset Information

The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression.

ABSTRACT:

Objective

Methods to correct class imbalance (imbalance between the frequency of outcome events and nonevents) are receiving increasing interest for developing prediction models. We examined the effect of imbalance correction on the performance of logistic regression models.

Material and methods

Prediction models were developed using standard and penalized (ridge) logistic regression under 4 methods to address class imbalance: no correction, random undersampling, random oversampling, and SMOTE. Model performance was evaluated in terms of discrimination, calibration, and classification. Using Monte Carlo simulations, we studied the impact of training set size, number of predictors, and the outcome event fraction. A case study on prediction modeling for ovarian cancer diagnosis is presented.

Results

The use of random undersampling, random oversampling, or SMOTE yielded poorly calibrated models: the probability to belong to the minority class was strongly overestimated. These methods did not result in higher areas under the ROC curve when compared with models developed without correction for class imbalance. Although imbalance correction improved the balance between sensitivity and specificity, similar results were obtained by shifting the probability threshold instead.

Discussion

Imbalance correction led to models with strong miscalibration without better ability to distinguish between patients with and without the outcome event. The inaccurate probability estimates reduce the clinical utility of the model, because decisions about treatment are ill-informed.

Conclusion

Outcome imbalance is not a problem in itself, imbalance correction may even worsen model performance.

SUBMITTER: van den Goorbergh R

PROVIDER: S-EPMC9382395 | biostudies-literature | 2022 Aug

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression.

van den Goorbergh Ruben R van Smeden Maarten M Timmerman Dirk D Van Calster Ben B

Journal of the American Medical Informatics Association : JAMIA 20220801 9

<h4>Objective</h4>Methods to correct class imbalance (imbalance between the frequency of outcome events and nonevents) are receiving increasing interest for developing prediction models. We examined the effect of imbalance correction on the performance of logistic regression models.<h4>Material and methods</h4>Prediction models were developed using standard and penalized (ridge) logistic regression under 4 methods to address class imbalance: no correction, random undersampling, random oversampli ...[more]

PMID: 35686364

Similar Datasets

Project description:BackgroundIn binary classification for clinical studies, an imbalanced distribution of cases to classes and an extreme association level between the binary dependent variable and a subset of independent variables can create significant classification problems. These crucial issues, namely class imbalance and complete separation, lead to classification inaccuracy and biased results in clinical studies.MethodTo deal with class imbalance and complete separation problems, we propose using a fuzzy logistic regression framework for binary classification. Fuzzy logistic regression incorporates combinations of triangular fuzzy numbers for the coefficients, inputs, and outputs and produces crisp classification results. The fuzzy logistic regression framework shows strong classification performance due to fuzzy logic's better handling of imbalance and separation issues. Hence, classification accuracy is improved, mitigating the risk of misclassified conditions and biased insights for clinical study patients.ResultsThe performance of the fuzzy logistic regression model is assessed on twelve binary classification problems with clinical datasets. The model has consistently high sensitivity, specificity, F1, precision, and Mathew's correlation coefficient scores across all clinical datasets. There is no evidence of impact from the imbalance or separation that exists in the datasets. Furthermore, we compare the fuzzy logistic regression classification performance against two versions of classical logistic regression and six different benchmark sources in the literature. These six sources provide a total of ten different proposed methodologies, and the comparison occurs by calculating the same set of classification performance scores for each method. Either imbalance or separation impacts seven out of ten methodologies. The remaining three produce better classification performance in their respective clinical studies. However, these are all outperformed by the fuzzy logistic regression framework.ConclusionFuzzy logistic regression showcases strong performance against imbalance and separation, providing accurate predictions and, hence, informative insights for classifying patients in clinical studies.

Project description:BackgroundGeneralized regression neural network (GRNN) and logistic regression (LR) are extensively used in the medical field; however, the better model for predicting stroke outcome has not been established. The primary goal of this study was to compare the accuracies of GRNN and LR models to identify the most optimal model for the prediction of acute stroke outcome, as well as explore useful biomarkers for predicting the prognosis of acute stroke patients.MethodIn a single-center study, 216 (80% for the training set and 20% for the test set) acute stroke patients admitted to the Shenzhen Second People's Hospital between December 2019 to June 2021 were retrospectively recruited. The functional outcomes of the patients were measured using Barthel Index (BI) on discharge. A training set was used to optimize the GRNN and LR models. The test set was utilized to validate and compare the performances of GRNN and LR in predicting acute stroke outcome based on the area under the receiver operating characteristic curve (AUROC), accuracy, sensitivity, and the Kappa value.ResultThe LR analysis showed that age, the National Institute Health Stroke Scale score, BI index, hemoglobin, and albumin were independently associated with stroke outcome. After validating in test set using these variables, we found that the GRNN model showed a better performance based on AUROC (0.931 vs 0.702), sensitivity (0.933 vs 0.700), specificity (0.889 vs 0.722), accuracy (0.896 vs 0.729), and the Kappa value (0.775 vs 0.416) than the LR model.ConclusionOverall, the GRNN model demonstrated superior performance to the LR model in predicting the prognosis of acute stroke patients. In addition to its advantage in not affected by implicit interactions and complex relationship in the data. Thus, we suggested that GRNN could be served as the optimal statistical model for acute stroke outcome prediction. Simultaneously, prospective validation based on more variables of the GRNN model for the prediction is required in future studies.

Project description:ObjectiveAfter deploying a clinical prediction model, subsequently collected data can be used to fine-tune its predictions and adapt to temporal shifts. Because model updating carries risks of over-updating/fitting, we study online methods with performance guarantees.Materials and methodsWe introduce 2 procedures for continual recalibration or revision of an underlying prediction model: Bayesian logistic regression (BLR) and a Markov variant that explicitly models distribution shifts (MarBLR). We perform empirical evaluation via simulations and a real-world study predicting Chronic Obstructive Pulmonary Disease (COPD) risk. We derive "Type I and II" regret bounds, which guarantee the procedures are noninferior to a static model and competitive with an oracle logistic reviser in terms of the average loss.ResultsBoth procedures consistently outperformed the static model and other online logistic revision methods. In simulations, the average estimated calibration index (aECI) of the original model was 0.828 (95%CI, 0.818-0.938). Online recalibration using BLR and MarBLR improved the aECI towards the ideal value of zero, attaining 0.265 (95%CI, 0.230-0.300) and 0.241 (95%CI, 0.216-0.266), respectively. When performing more extensive logistic model revisions, BLR and MarBLR increased the average area under the receiver-operating characteristic curve (aAUC) from 0.767 (95%CI, 0.765-0.769) to 0.800 (95%CI, 0.798-0.802) and 0.799 (95%CI, 0.797-0.801), respectively, in stationary settings and protected against substantial model decay. In the COPD study, BLR and MarBLR dynamically combined the original model with a continually refitted gradient boosted tree to achieve aAUCs of 0.924 (95%CI, 0.913-0.935) and 0.925 (95%CI, 0.914-0.935), compared to the static model's aAUC of 0.904 (95%CI, 0.892-0.916).DiscussionDespite its simplicity, BLR is highly competitive with MarBLR. MarBLR outperforms BLR when its prior better reflects the data.ConclusionsBLR and MarBLR can improve the transportability of clinical prediction models and maintain their performance over time.

Dataset Information

The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression.

Objective

Material and methods

Results

Discussion

Conclusion

Publications

The harm of class imbalance corrections for risk prediction models: illustration and simulation using logistic regression.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets