Evaluation of conventional and quantum computing for predicting mortality based on small early-onset colorectal cancer data

Abstract

Background: Quantum computing integrated with machine learning (ML) offers novel solutions in various fields, including healthcare. The synergy between quantum computing and ML in classification exploits unique data patterns. Despite theoretical advantages, the empirical application and effectiveness of quantum computing on small medical datasets remains underexplored. Method: This retrospective study from a tertiary hospital used data on early-onset colorectal cancer with 93 features and 1501 patients from 2008 to 2020 to predict mortality. We compared quantum support vector machine (QSVM) models with classical SVM models in terms of number of features, number of training sets, and outcome ratio. We evaluated the model based on the area under the curve in the receiver operating characteristic curve (AUROC). Results: We observed a mortality rate of 7.6 % (96 of 1253 subjects). We generated the mortality prediction model using 11 clinical variables, including cancer stage and chemotherapy history. We found that the AUROC difference between the conventional and quantum methods was the maximum for the top 11 variables. We also showed the AUROC in QSVM (mean [standard deviation], 0.863 [0.102]) outperformed all the number of trials in the conventional SVM (0.723 [0.231]). Compared to the conventional SVM, the QSVM showed robust performance, consistent with the AUROC, even in the unbalanced case. Conclusion: Our study highlights the potential of quantum computing to improve predictive modeling in healthcare, especially for rare diseases with limited available data. The advantages of quantum computing, such as the exploration of Hilbert space, contributed to the superior predictive performance compared to conventional methods.