145P - Predicting the efficacy of immunotherapy in non-small cell lung cancer using machine learning based on simple clinical characteristics and biochemical indexes

Background

The emergence of immune checkpoint inhibitors (ICIs) in recent years has revolutionised the therapeutic landscape in non-small cell lung cancer (NSCLC), but the overall response rate (ORR) is low, although TMB and MSI show some predictive value for immune efficacy. In real-world clinical practice, however, immunotherapy decisions for many patients must be made on the basis of limited clinical information. Here, we developed a machine learning-based clinical decision support algorithm to synthesise multidimensional clinical information to predict therapeutic response to ICI in NSCLC patients and reduce the rate of first-line therapeutic misdosing.

Methods

A total of 66 clinical characteristics and biochemical indexes were collected from electronic medical records, and correlated with optimal clinical efficacy. Neural network algorithm, was used to predict immune response and then validated in an independent validation set of ICI-treated patients.

Results

A total of 402 patients were enrolled in the study, including 321 patients in the training set (ORR, 57.3%) and 81 patients in the test set (ORR, 58%). Neural network algorithm showed a superior pronounced predictive effect (area under the curve [AUC] 0.71) using only eight common clinical parameters, which included sex, age, tumor subtype, smoking history, lymphocyte, eosinophil, neutrophil, and erythrocyte distribution width. The overall accuracy in the validation set was 70.4%, the positive predictive value (PPV) was 70.2% and the negative predictive value (NPV) was 70.8%. The model was able to identify 50% of these poorly treated patients (17/34), mainly non-responders and hyper-progressors, in a timely manner prior to dosing, resulting in a 50% reduction in the rate of clinical misses.

Conclusions

Overall, our neural network algorithm provides a rapid clinical decision support model for predicting ICI treatment response in NSCLC patients. Furthermore, this model relies only on basic clinical information and biochemical indicators, which can effectively reduce clinical overdosing in the first-line setting.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

OrigiMed Shanghai Co., Ltd.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.