login
Home / Papers / Natural language processing (NLP) of Internet conversations to evaluate prostate...

Natural language processing (NLP) of Internet conversations to evaluate prostate cancer (PC) patients' perceptions of active surveillance (AS).

2 Citations•2012•
M. Bennett, Armon Vincent, O. T. Lee
Journal of clinical oncology : official journal of the American Society of Clinical Oncology

The potential utility of NLP to analyze ICs in order to provide insight into patient preferences and decision-making is demonstrated and recommended that multidisciplinary clinics consider including an unbiased specialist to present treatment options and that future decision tools for AS include quantitative data regarding outcomes after AS.

Abstract

14 Background: Less than 10% of qualifying PC patients receive primary management with AS. Qualitative research aimed at identifying patient acceptance of AS has been identified as a national health research priority. The primary objective of this study was to determine if NLP of anonymous internet conversations (ICs) could be utilized to identify unmet public needs regarding AS. METHODS After obtaining IRB approval, English-language ICs regarding PC treatment with AS from 2002-2012 were identified using a novel internet search methodology. Web spiders were developed to identify, mine, and gather content from the internet for ICs centered on AS. All ICs identified were screened programmatically to remove any not-on-topic ICs. Collection of ICs was not restricted to any specific geographic region of origin. NLP was used to evaluate content and perform a sentiment analysis. Conversations were scored as positive, negative, or neutral. A sentiment index (SI) was subsequently calculated according to the following formula to compare temporal trends in public sentiment towards AS: [(#Positive IC/#Total IC) - (#Negative IC/#Total IC) x 100]. RESULTS A total of 464 ICs were identified. Sentiment increased from -13 to +2 over the study period. The increase sentiment has been driven by increased patient emphasis on quality-of-life factors and endorsement of AS by national medical organizations. Unmet needs identified in these ICs include: a gap between quantitative data regarding long-term outcomes with AS versus conventional treatments, desire for treatment information from an unbiased specialist, and absence of public role models managed with AS. CONCLUSIONS This study demonstrates the potential utility of NLP to analyze ICs in order to provide insight into patient preferences and decision-making. Based on our findings, we recommend that multidisciplinary clinics consider including an unbiased specialist to present treatment options and that future decision tools for AS include quantitative data regarding outcomes after AS, so that patients can make decisions with an amount of information that is more similar to the resources available regarding radiation or surgery.