Natural Language Processing to Identify Cancer Treatments With Electronic Medical Records

Abstract

PURPOSE Knowing the treatments administered to patients with cancer is important for treatment planning and correlating treatment patterns with outcomes for personalized medicine study. However, existing methods to identify treatments are often lacking. We develop a natural language processing approach with structured electronic medical records and unstructured clinical notes to identify the initial treatment administered to patients with cancer. METHODS We used a total number of 4,412 patients with 483,782 clinical notes from the Stanford Cancer Institute Research Database containing patients with nonmetastatic prostate, oropharynx, and esophagus cancer. We trained treatment identification models for each cancer type separately and compared performance of using only structured, only unstructured ( bag-of-words, doc2vec, fasttext), and combinations of both ( structured + bow, structured + doc2vec, structured + fasttext). We optimized the identification model among five machine learning methods (logistic regression, multilayer perceptrons, random forest, support vector machines, and stochastic gradient boosting). The treatment information recorded in the cancer registry is the gold standard and compares our methods to an identification baseline with billing codes. RESULTS For prostate cancer, we achieved an f1-score of 0.99 (95% CI, 0.97 to 1.00) for radiation and 1.00 (95% CI, 0.99 to 1.00) for surgery using structured + doc2vec. For oropharynx cancer, we achieved an f1-score of 0.78 (95% CI, 0.58 to 0.93) for chemoradiation and 0.83 (95% CI, 0.69 to 0.95) for surgery using doc2vec. For esophagus cancer, we achieved an f1-score of 1.0 (95% CI, 1.0 to 1.0) for both chemoradiation and surgery using all combinations of structured and unstructured data. We found that employing the free-text clinical notes outperforms using the billing codes or only structured data for all three cancer types. CONCLUSION Our results show that treatment identification using free-text clinical notes greatly improves upon the performance using billing codes and simple structured data. The approach can be used for treatment cohort identification and adapted for longitudinal cancer treatment identification.

Prediction of acute appendicitis among patients with undifferentiated abdominal pain at emergency department

“…We first pre-processed the unstructured data, including word segmentation and removal of stop words. Then we used TaggedDocument in the gensim package to wrap the input sentence and change it to the input sample format required by Doc2Vec [ 25 , 26 ]. After that, we loaded the Doc2vec model with window size of 3 and started training, and finally we mapped the unstructured data into 128-dimensional paragraph vectors and made further predictions.…”

Uncovering interpretable potential confounders in electronic medical records

“…In the end, we have 9 structured covariates for prostate cancer and 7 structured covariates for NSCLC. [7, 8].…”

“…In the end, we have 9 structured covariates for prostate cancer and 7 structured covariates for NSCLC. While billing codes can be used to generate additional structured features for diagnosis and past treatments, existing studies have found these can be unreliable [27, 28]. Hence, we chose to focus mainly on clinical notes to capture additional information that can influence survival time, such as patient symptoms and performance status.…”

“…Our main contribution is presenting an approach to uncover interpretable potential confounders from clinical text. Existing work in observational causal inference rarely employs unstructured data [14, 15, 16], and most NLP studies on clinical text focus on prediction or classification settings [17, 18, 19, 20, 8]. Our paper is the first to uncover interpretable potential confounders from clinical notes for causal analysis on cancer therapies, and one of the few works that combines NLP and causal inference in a time-to-event setting.…”

Uncovering interpretable potential confounders in electronic medical records

“…Observational studies are more reliable when we can better control for these confounders. While structured EMR data, such as billing codes, can be used to encode expert-curated patient characteristics, studies suggest that administrative claims data may contain errors 27 , 28 and expert-curated covariates may not capture all potential confounding 7 , 29 . EMR clinical text is a potential source of additional information about factors that might relate to both treatment assignment and prognosis.…”