A natural language processing pipeline for pairing measurements uniquely across free-text CT reports.

OBJECTIVE: To standardize and objectivize treatment response assessment in oncology, guidelines have been proposed that are driven by radiological measurements, which are typically communicated in free-text reports defying automated processing. We study through inter-annotator agreement and natural language processing (NLP) algorithm development the task of pairing measurements that quantify the same finding across consecutive radiology reports, such that each measurement is paired with at most one other ("partial uniqueness"). METHODS AND MATERIALS: Ground truth is created based on 283 abdomen and 311 chest CT reports of 50 patients each. A pre-processing engine segments reports and extracts measurements. Thirteen features are developed based on volumetric similarity between measurements, semantic similarity between their respective narrative contexts and structural properties of their report positions. A Random Forest classifier (RF) integrates all features. A "mutual best match" (MBM) post-processor ensures partial uniqueness. RESULTS: In an end-to-end evaluation, RF has precision 0.841, recall 0.807, F-measure 0.824 and AUC 0.971; with MBM, which performs above chance level (P<0.001), it has precision 0.899, recall 0.776, F-measure 0.833 and AUC 0.935. RF (RF+MBM) has error-free performance on 52.7% (57.4%) of report pairs. DISCUSSION: Inter-annotator agreement of three domain specialists with the ground truth (κ>0.960) indicates that the task is well defined. Domain properties and inter-section differences are discussed to explain superior performance in abdomen. Enforcing partial uniqueness has mixed but minor effects on performance. CONCLUSION: A combined machine learning-filtering approach is proposed for pairing measurements, which can support prospective (supporting treatment response assessment) and retrospective purposes (data mining).

Automatically pairing measured findings across narrative abdomen CT reports.

Radiological measurements are one of the key variables in widely adopted guidelines (WHO, RECIST) that standardize and objectivize response assessment in oncology care. Measurements are typically described in free-text, narrative radiology reports. We present a natural language processing pipeline that extracts measurements from radiology reports and pairs them with extracted measurements from prior reports of the same clinical finding, e.g., lymph node or mass. A ground truth was created by manually pairing measurements in the abdomen CT reports of 50 patients. A Random Forest classifier trained on 15 features achieved superior results in an end-to-end evaluation of the pipeline on the extraction and pairing task: precision 0.910, recall 0.878, F-measure 0.894, AUC 0.988. Representing the narrative content in terms of UMLS concepts did not improve results. Applications of the proposed technology include data mining, advanced search and workflow support for healthcare professionals managing radiological measurements.

A phase II study of lapatinib in recurrent/metastatic squamous cell carcinoma of the head and neck.

PURPOSE: This study sought to determine the efficacy and safety profile of lapatinib in patients with recurrent/metastatic squamous cell carcinoma of the head and neck (SCCHN). EXPERIMENTAL DESIGN: This phase II multiinstitutional study enrolled patients with recurrent/metastatic SCCHN into two cohorts: those without (arm A) and those with (arm B) before exposure to an epidermal growth factor receptor (EGFR) inhibitor. All subjects were treated with lapatinib 1,500 mg daily. Primary endpoints were response rate (arm A) and progression-free survival (PFS; arm B). The biologic effects of lapatinib on tumor growth and survival pathways were assessed in paired tumor biopsies obtained before and after therapy. RESULTS: Forty-five patients were enrolled, 27 in arm A and 18 in arm B. Diarrhea was the most frequent toxicity occurring in 49% of patients. Seven patients experienced related grade 3 toxicity (3 fatigue, 2 hyponatremia, 1 vomiting, and 1 diarrhea). In an intent-to-treat analysis, no complete or partial responses were observed, and stable disease was the best response observed in 41% of arm A (median duration, 50 days, range, 34-159) and 17% of arm B subjects (median, 163 days, range, 135-195). Median PFS was 52 days in both arms. Median OS was 288 (95% CI, 62-374) and 155 (95% CI, 75-242) days for arms A and B, respectively. Correlative analyses revealed an absence of EGFR inhibition in tumor tissue. CONCLUSION: Lapatinib as a single agent in recurrent/metastatic SCCHN, although well tolerated, appears to be inactive in either EGFR inhibitor naive or refractory subjects.

Amongst eligible patients, age and comorbidity do not predict for dose-limiting toxicity from phase I chemotherapy.

BACKGROUND: There are no clear predictors clinicians can use to determine who is more likely to experience dose-limiting toxicity (DLT) in phase I chemotherapy clinical trials. Many providers are reluctant to refer older adults to phase I trials because of concerns about the development of toxicity. The goal of this study was to identify clinical and nonclinical factors which were associated with the development of DLT in phase I studies. METHODS: Patients (pts) were included if they were treated at maximally tolerated dose (MTD) and above. Studies were included only if MTD was reached. Data collected included age, comorbidity (Cumulative Illness Rating Score-Geriatrics), labs at enrollment, height, weight, performance status, cancer type, duration of diagnosis, prior treatment, drug level, smoking status, marital status, mean income, percent of population high school educated as determined by ZIP code, and distance to the phase I trial hospital. Those who did and did not have DLT were compared by bivariate and then multivariate analysis. RESULTS: A total of 242 charts were reviewed from 24 cytotoxic chemotherapy studies, and 27 different types of cancer were represented. On bivariate analysis, mean age, household income (higher), weight, body surface area, dose of drug, alkaline phosphatase, hemoglobin, and LDH were significantly associated with DLT (P < 0.05). CIRS-G score was not associated with DLT. In multivariate analysis, dose level (P = 0.004) and distance from the phase I trial hospital (P = 0.04) were still significant predictors of DLT. Age did not predict for severity of DLT. CONCLUSIONS: Age and comorbidity did not predict for development of DLT in phase I chemotherapy trials. Many of these pts were very fit, with relatively low CIRS-G scores, so the impact of comorbidity may not have been fully evaluated. Several social and clinical factors may predict for development of DLT. A prospective study is being planned to confirm these results.