Introduction

Predicting clinical outcomes and survival of cancer patients treated by given chemotherapy can assist in choosing the course of treatment. We developed a methodology for predicting clinical outcome and progression-free survival (PFS) of canine lymphoma patients treated by (L-)CHOP chemotherapy.

Methods

We collected live cancer cells from fresh FNA taken from affected lymph nodes, as well as the response and prognosis of 242 canine lymphoma patients treated by (L)-CHOP for at least 4 weeks. We used three types of data from ex vivo chemosensitivity, flow cytometry, and bloodwork to train a machine learning model that predicts the probability of achieving complete remission at the 4th, 8th, or 12th week of the protocol. The same set of data were also used to predict PFS by utilizing the Cox proportional hazards model.

Results

The predictive accuracy of machine learning models was as high as 80.4%, 89.1%, or 82.7% when predicting the clinical outcome after 4th, 8th, or 12th week. The performance of the Cox hazards model for predicting PFS was also high, featuring the C-statistic of 0.850. The stratification of the patients based on both the subtype (B- vs. T-cell) and the Cox hazards model outperformed the one based on only the subtype when analyzing PFS.

Conclusion

The results demonstrate substantial enhancement in the predictive accuracy by incorporating a greater variety of data. They also highlight superior performance in predicting survival compared to the conventional stratification method. We believe that the proposed methodology can contribute to improving and personalizing the care of canine lymphoma patients.

Funding Information

This work was self-supported by ImpriMed, Inc.