Dataset Information

Penalization and shrinkage methods produced unreliable clinical prediction models especially when sample size was small.

ABSTRACT:

Objectives

When developing a clinical prediction model, penalization techniques are recommended to address overfitting, as they shrink predictor effect estimates toward the null and reduce mean-square prediction error in new individuals. However, shrinkage and penalty terms ('tuning parameters') are estimated with uncertainty from the development data set. We examined the magnitude of this uncertainty and the subsequent impact on prediction model performance.

Study design and setting

This study comprises applied examples and a simulation study of the following methods: uniform shrinkage (estimated via a closed-form solution or bootstrapping), ridge regression, the lasso, and elastic net.

Results

In a particular model development data set, penalization methods can be unreliable because tuning parameters are estimated with large uncertainty. This is of most concern when development data sets have a small effective sample size and the model's Cox-Snell R² is low. The problem can lead to considerable miscalibration of model predictions in new individuals.

Conclusion

Penalization methods are not a 'carte blanche'; they do not guarantee a reliable prediction model is developed. They are more unreliable when needed most (i.e., when overfitting may be large). We recommend they are best applied with large effective sample sizes, as identified from recent sample size calculations that aim to minimize the potential for model overfitting and precisely estimate key parameters.

SUBMITTER: Riley RD

PROVIDER: S-EPMC8026952 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:This study aims to determine how randomly splitting a dataset into training and test sets affects the estimated performance of a machine learning model and its gap from the test performance under different conditions, using real-world brain tumor radiomics data. We conducted two classification tasks of different difficulty levels with magnetic resonance imaging (MRI) radiomics features: (1) "Simple" task, glioblastomas [n = 109] vs. brain metastasis [n = 58] and (2) "difficult" task, low- [n = 163] vs. high-grade [n = 95] meningiomas. Additionally, two undersampled datasets were created by randomly sampling 50% from these datasets. We performed random training-test set splitting for each dataset repeatedly to create 1,000 different training-test set pairs. For each dataset pair, the least absolute shrinkage and selection operator model was trained and evaluated using various validation methods in the training set, and tested in the test set, using the area under the curve (AUC) as an evaluation metric. The AUCs in training and testing varied among different training-test set pairs, especially with the undersampled datasets and the difficult task. The mean (±standard deviation) AUC difference between training and testing was 0.039 (±0.032) for the simple task without undersampling and 0.092 (±0.071) for the difficult task with undersampling. In a training-test set pair with the difficult task without undersampling, for example, the AUC was high in training but much lower in testing (0.882 and 0.667, respectively); in another dataset pair with the same task, however, the AUC was low in training but much higher in testing (0.709 and 0.911, respectively). When the AUC discrepancy between training and test, or generalization gap, was large, none of the validation methods helped sufficiently reduce the generalization gap. Our results suggest that machine learning after a single random training-test set split may lead to unreliable results in radiomics studies especially with small sample sizes.