Artificial intelligence can predict cancer survival using data from a single visit

According to a study published in JAMA network open.

The researchers found that the natural language processing models could predict survival at 6 months, 36 months, and 60 months at least as well as previous models, without focusing on a specific type of cancer.

“These results suggest that it is possible to predict cancer patient survival using a common cancer document without additional data and without forming separate models for specific cancer types,” the authors wrote. researchers.

Continue Reading

Researchers trained and evaluated conventional and neural models to predict whether patients would survive 6 months, 36 months, or 60 months based on an initial consultation with an oncologist. The main outcome was the performance of the predictive models. The secondary result was the words used by the models.

Data from 47,625 cancer patients who had a first oncologist consultation within 180 days of diagnosis were included. The consultation documents for the dataset were generated by oncologists practicing at 6 centers located in various geographies.

Survival was calculated as the number of months between the selected document and either the patient’s death or the death cut-off date of April 6, 2022.

The researchers compared 4 models: the non-neural bag-of-words (BoW) algorithm and 3 neural models: convolutional neural networks (CNN), long-short-term memory (LSTM), and bidirectional encoder representations at from transformers (BERT).

The results showed numerically similar performance for BoW, CNN and LSTM (Balanced Accuracy [BAC] > 0.800; area under curve [AUC] > 0.900). BERT underperformed in all areas.

BoW obtained the best results for predicting survival at 6 months (BAC, 0.856; AUC, 0.928). CNN had the best performance in predicting survival at 36 months (BAC 0.842; AUC 0.918) and survival at 60 months (BAC 0.837; AUC 0.918).

There was a lack of clear performance gain using neural models compared to BoW, suggesting that survival prediction depends on the presence of certain words in initial oncology consultation documents.

Similar tokens (words with endings removed) have the top 10 importance for BoW models used to predict 6-month and 60-month survival. For example, “palliat“ (short for “palliative” or “palliative”) is the more important feature of the two.

Additionally, words for different types of cancer are important for survival at 6 and 60 months. The words ‘breast’ and ‘prostate’ are positive predictors of 6-month survival, while ‘lung’, ‘liver’ and ‘glioblastoma’ are negative predictors of 60-month survival. “N0” (negative lymph node involvement) was one of the top 10 positive predictors for 60-month survival.

“These results suggest that the models have comparable or better performance than previous models predicting cancer survival and that they may be able to predict survival using readily available data without focusing on one type of cancer,” concluded the researchers.

“Our results suggest that it is possible to predict cancer patient survival without having to construct structured datasets or limiting predictions to specific types or locations of cancer. Given the widespread availability of the papers initial consultation of oncologists, this opens the possibility of training and more easily using such models for all types of cancer in different cancer centers.

Disclosures: This research was partially funded by the Pfizer Innovation Fund. Some study authors have declared affiliations with biotechnology, pharmaceutical and/or device companies. Please see the original reference for a full list of disclosures.