Extracting number of trial participants from abstracts of randomized controlled trials

Centre for Health Informatics, University of New South Wales, and Department of Health Science and Technology, Aalborg University y Telemedicine - Aid to decision-making in healthcare y The amount of information available to answer questions Æ The use of keyword-based search methods to locate the answers y These methods tend to overload with a lot of irrelevant Æ The precision in the search result should be improved What is the primary disease? What are the characteristics of the patient? What is the main intervention compared to? y 223 labeled abstracts achieved from PubMed y Manually detected number of trial participants for each Example :We determine whether a continuous infusion of ketamine can decrease the severity of a moderately severe acute asthma exacerbation by a clinically significant 2 points using a 15 point Pulmonary Index scoring scale. 0 STUDYOBJECTIVE A A double-blinded, randomized, placebo-controlled trial was performed to evaluate patients aged 2 to 18 years who presented to a pediatric emergency department with an acute asthma exacerbation. 2 METHODS P Exclusion criteria included temperature greater than 39 degrees C (102 degrees F), focal infiltrate on radiograph, or any glucocorticoid Eligible patients received 3 treatments with albuterol, ipratropium bromide, and a dose of oral or parenteral glucocorticoids. 2 If the Pulmonary Index score remained 8 to 14, enrollment proceeded. 2 METHODS MAll enrolled patients received continuous nebulized albuterol at 10 mg/hour and were randomized to receive an intravenous bolus of 0.2 mg/kg of ketamine, followed by a 2 hour ketamine infusion at 0.5 mg/kg per hour or an equal-volume regimen with normal- A Pulmonary Index score was performed on patients at 0, 30, 60, 90, and 120 minutes. 3 METHODS ESixty-eight patients were enrolled, with 33 patients randomized to the ketamine infusion and 35 patients randomized to placebo. 3 Mean ages of patients enrolled, chronic severity of asthma, and duration of symptoms before presentation were similar between At enrollment, the mean Pulmonary Index score in the placebo group was 10.3 +- 1.1 versus 10.5 +- 1.5 for the ketamine group (difference of means 0.2,, 95% confidence interval [CI] -0.5 to 0.8). 4 RESULTS R Sixty-two patients completed the entire 2 hour infusion protocol. 4 RESULTS RNo significant difference between groups was seen in rate of improvement in the Pulmonary Index score at completion. 4 RESULTS RThe mean decrease in the Pulmonary Index scores at the end of the infusion was 3.6 +- 1.3 in the placebo group versus 3.2 +- 2.0 in the ketamine group (difference of means 0.4,, 95% CI -0.4 to 1.3). 5 RESULTS R No short-term adverse effects necessitating discontinuation of the infusion or adverse behavioral impacts at 48 hours after discharge We conclude that ketamine given at 0.2 mg/kg followed by an infusion of 0.5 mg/kg per hour for 2 hours provided no incremental benefit to standard therapy in this cohort of children with a moderately severe asthma exacerbation. 7 CONCLUSION O Natural Language Processing on the abstracts : WORD STEM
randomized randomize VBD to to TO the the DT ketamine ketamine NN infusion infusion NN and and CC 35 randomized randomize VBN to to TO placebo placebo NN .
Features
Centre of Attention
y Three classification algorithms are tested y Reduce the number of false positive predictions y 68 patients were enrolled, with 33 patients randomized to the ketamine infusion and 35 patients randomized to placebo.
y According to definition the highest number is chosen y Rule-based baseline system returns the largest interger that is: { Followed by [Adjective]* [Participant-related Noun] Classifier
Precision
F-measure
Accuracy
Baseline
y ’State of the Art’-algorithm: F-measure = 0.85 Predicted
Trial Participants
Actual
Trial Participants
{ Two numbers of trial participants present in abstract y Correction of NLP-preprocessing errors Classifier
F-measure
Accuracy

Source: http://www2.telemed.no/ttec2008/presentations/tuesday/ps05/04Hansen_OK.pdf