An End-to-End Natural Language Processing System for Automatically Extracting Radiation Therapy Events From Clinical Texts

PURPOSE

Real-world evidence for radiation therapy (RT) is limited because it is often documented only in the clinical narrative. We developed a natural language processing system for automated extraction of detailed RT events from text to support clinical phenotyping.

METHODS AND MATERIALS

A multi-institutional data set of 96 clinician notes, 129 North American Association of Central Cancer Registries cancer abstracts, and 270 RT prescriptions from HemOnc.org was used and divided into train, development, and test sets. Documents were annotated for RT events and associated properties: dose, fraction frequency, fraction number, date, treatment site, and boost. Named entity recognition models for properties were developed by fine-tuning BioClinicalBERT and RoBERTa transformer models. A multiclass RoBERTa-based relation extraction model was developed to link each dose mention with each property in the same event. Models were combined with symbolic rules to create a hybrid end-to-end pipeline for comprehensive RT event extraction.

RESULTS

Named entity recognition models were evaluated on the held-out test set with F1 results of 0.96, 0.88, 0.94, 0.88, 0.67, and 0.94 for dose, fraction frequency, fraction number, date, treatment site, and boost, respectively. The relation model achieved an average F1 of 0.86 when the input was gold-labeled entities. The end-to-end system F1 result was 0.81. The end-to-end system performed best on North American Association of Central Cancer Registries abstracts (average F1 0.90), which are mostly copy-paste content from clinician notes.

CONCLUSIONS

We developed methods and a hybrid end-to-end system for RT event extraction, which is the first natural language processing system for this task. This system provides proof-of-concept for real-world RT data collection for research and is promising for the potential of natural language processing methods to support clinical care.

Prevention of relapse in generalized social phobia: results of a 24-week study in responders to 20 weeks of sertraline treatment

The aim of this study was to evaluate the efficacy, tolerability, and effects on quality of life of sertraline, a selective serotonin reuptake inhibitor, in the prevention of relapse of generalized social phobia. Fifty adult outpatients with generalized social phobia who were rated much or very much improved on the Clinical Global Impression Scale of Improvement (CGI-I) after 20 weeks of sertraline treatment (50-200 mg/day) were randomly assigned in a one-to-one ratio to either continue double-blind treatment with sertraline or immediately switch to placebo for another 24 weeks. The initial 20-week study was placebo-controlled, and 15 responders to placebo also continued to receive double-blind placebo treatment in the continuation study. Eighty-eight percent of patients in the sertraline-continuation group and only 40% of patients in the placebo-switch and placebo-responder groups completed the study. In intent-to-treat endpoint analyses, 1 (4%) of 25 patients in the sertraline-continuation group and 9 (36%) of 25 patients in the placebo-switch group had relapsed at study endpoint (chi2 = 8.0, Fisher exact test, p = 0.01). The relative risk (hazards ratio) for relapse associated with placebo-switch relative to sertraline-continuation treatment was 10.2 (95% confidence interval, 1.3-80.7). Mean CGI-Severity, Marks Fear Questionnaire (MFQ) Social Phobia subscale, and Duke Brief Social Phobia Scale (BSPS) total scores were reduced by 0.07, 0.34, and 1.86 in the Sertraline-Continuation group and increased by 0.88, 4.09, and 5.99 in the Placebo-Switch group (all F > 5.3, p < 0.03), respectively. CGI-Severity, MFQ Social Phobia subscale, and BSPS scores also increased in the Placebo-Responder group. Discontinuations because of lack of efficacy were 4% in the sertraline-continuation group, 28% in the placebo-switch group (chi2 = 5.36, Fisher exact test, p = 0.049), relative to sertraline, and 27% in the placebo-responder group. Sertraline was effective in preventing relapse of generalized social phobia. Future research should assess whether improvements may be maintained or further increased by longer periods of treatment or through the addition of cognitive-behavioral techniques.

Accuracy of subcutaneous fat measurement: comparison of skinfold calipers, ultrasound, and computed tomography

OBJECTIVE

The purpose of this study was to compare skinfold caliper and ultrasound measurement of subcutaneous body fat at three abdominal sites with computed tomography, which is considered to be the gold standard.

DESIGN

This was a cross-sectional study in which computed-tomography, ultrasound, and skinfold caliper measurements were made at three distinct abdominal sites. All body composition and anthropometric measurements were performed on each subject on one occasion.

SUBJECTS

Twenty-two subjects were recruited (13 men and 9 women). Mean ages (+/- standard deviation) were 43 +/- 4 years for the women and 51 +/- 18 years for the men. All subjects had been previously scheduled for an abdominal or pelvic computed-tomography scan at the Department of Radiology, St Paul's Hospital, Vancouver, British Columbia, Canada, and participated in the study on a volunteer basis.

MAIN OUTCOME MEASURES

A better agreement was found between the skinfold calipers and computed-tomography methods than between the ultrasound and computed-tomography method for the measurement of subcutaneous body fat. This was observed when the data were analyzed for both correlational agreement and for graphical interpretation.

STATISTICAL ANALYSES PERFORMED

The relationships among skinfold, ultrasound, and computed-tomography measurements were analyzed by determining Pearson correlation coefficients. A graphical method described by Bland and Altman was also used to assess agreement among the three methods.

RESULTS

Significant correlation coefficients were observed between skinfold calipers and computed tomography at all three abdominal sites (site 1, r = .60, P = .003; site 2, r = .70, P = .0001; site 3, r = .73, P = .0001). Ultrasound and computed-tomography methods only showed a significant correlation at site 3 (r = .54; P = .009). The graphical method revealed that the variation in the ultrasound measurements was much greater than that of the skinfold measurements when compared to computed-tomography values.

APPLICATIONS/CONCLUSIONS

The results of this study indicated that relative agreement in the measurement of subcutaneous body fat between skinfold and computed-tomography measurements was superior to that exhibited between ultrasound and computed-tomography measurements. This finding enhances the potential use of skinfold calipers in the clinical setting, particularly in view of the fact that measurement of subcutaneous body fat at different body sites is becoming increasingly important for the characterization of risk of certain disease states.

Effect of morphine on cortisol and prolactin secretion in anorexia nervosa and depression

Endogenous opioid peptides are involved in feeding regulation, and alterations in opioidergic regulation have been implicated in the pathophysiology of eating disorders. To investigate further this hypothesis, we conducted a placebo-controlled study of the effect of the opiate alkaloid morphine on cortisol and prolactin secretion in six patients with anorexia nervosa and six age-matched healthy volunteers, and compared the results with those obtained in nine depressed patients. Basal cortisol but not basal prolactin levels were elevated in patients with anorexia nervosa and patients with depression. Following the administration of morphine plasma concentrations of cortisol levels declined progressively and at a similar rate in all three groups. The prolactin response to morphine was attenuated significantly in patients with depression. Neither the cortisol and prolactin response to morphine in the anorectic patients nor the cortisol response in the depressed patients we observed in this study suggests altered opiate receptor sensitivity. However, the decreased prolactin response to morphine in depressed patients remains compatible with this hypothesis.

Aperture averaging effects on the two-color correlation of scintillations

The scintillation two-color correlation coefficient was measured using a variable aperture. The correlation, at constant aperture, vs turbulence strength and the aperture averaging effects at constant turbulence strength were obtained. The results show a strong decrease or correlation with increasing turbulence strength and a moderate increase with increasing aperture. These results cannot be accounted for by a constant shape turbulence spectrum. It is shown that a turbulence strength-dependent inner scale can explain most of the results using the Rytov approximation.

Sample size influence on optical scintillation analysis. Analytical treatment of the higher-order irradiance moments

The statistical distribution of the experimentally obtained higher-order moments of optical scintillation probability density is studied. It is shown that this distribution is strongly dependent on the size of the data sample. At reasonable sample sizes the correct estimation of the theoretical value is improbable. At practically available sample sizes the region of the most probable values of the estimated higher-order moment is almost independent of the scintillation probability density function (PDF). The distinction between the candidate PDFs is almost impossible at reasonable sample sizes.

Sample size influence on optical scintillation analysis.2: Simulation approach

This paper is the second part of the sample size influence analysis. We estimate this influence on different methods of irradiance probability density function (PDF) diagnostics. The estimation is achieved by a Monte Carlo-based simulation. The problem of the fastest effective sampling rate is also considered. The results indicate the low resolving power of conventional irradiance PDF integrating methods.