NLP-Based Feedback to Improve Risk Comms and Informed Shared Decision Making

Natural Language Processing-Based Feedback to Improve Physician Risk Communication and Informed Shared Decision Making in Men With Clinically Localized Prostate Cancer

In this pilot study, the investigators will show feasibility of the NLP-based feedback system in 20 consultations of men with newly diagnosed prostate cancer. The investigators will recruit from the practices of up to 10 physicians who typically see these patients. The investigators will report the top five sentences from each consultation across key content areas (cancer prognosis, life expectancy, erectile dysfunction, urinary incontinence, and irritative urinary symptoms) to both patients and physicians within 2 weeks of the consultation.

The primary research procedures are: Audio recording and transcribing treatment counseling discussions for 20 men with newly diagnosed clinically localized prostate cancers and utilize NLP to extract key content using the system described above. Reports including the top five sentences by NLP probability for key content areas will be generated and will be provided to patients and providers within 2 weeks after each case. For patients, decisional conflict and risk perception will be assessed before and after receiving the NLP-based feedback. For physicians, the investigators will assess baseline quality of risk communication, any changes in individual physician communication over time, and accuracy of risk estimates for key content areas. Within 2 weeks of receiving the NLP-based feedback, the investigators will conduct a 30-minute semi-structured interview with patients to obtain their opinions on the utility and ideal implementation strategy for the NLP-based feedback. At the conclusion of the pilot trial, the investigators will conduct 30-minute semi-structured interview with counseling physicians to obtain their opinions on the utility and ideal implementation strategy for the NLP-based feedback.

20 men with newly diagnosed clinically localized prostate cancers and utilize NLP to extract key content using the top five sentences by NLP probability for key content areas will be generated and will be provided to patients and providers within 2 weeks after each case.

Audio recordings will be made using either digital recorders or telehealth platform-generated transcripts. Patient reports will include only the extracted sentences related to content areas. Physician reports will note the extracted statements across each content area, the quality scores for individual statements based on the pre-specified hierarchy, the statements that achieved the highest score across each content area, and feedback on what could be improved. For patients, decisional conflict and risk perception will be assessed before and after receiving the NLP-based feedback. For physicians, the investigators will assess baseline quality of risk communication by pre-specified hierarchy, any changes in individual physician communication over time, and accuracy of risk estimates for key content areas.

The investigators will employ the validated Decisional Conflict Scale (DCS), to estimate uncertainty associated with treatment choice. Effect sizes of 0.3 to 0.4 are considered meaningful. Variability (standard deviation) in DCS scores before and after receiving NLP-based feedback will be assessed and used in planning a larger trial.

Measured directly after treatment consultation and after NLP-based feedback given to patients within 2 weeks of consultation

The investigators will evaluate concordance of cancer risk perception with actual cancer risk at the patient level before and after the intervention. Cancer risk perception will be assessed by multiple-choice questions. Concordance of patient answers with actual cancer risk as estimated by outcomes of the SPCG-4 randomized trial comparing surgery versus watchful waiting at the patient's PCCI-predicted life expectancy will be assessed as a binary outcome. Risk perception will be assessed before and after their consultation. Variability (standard deviation) in risk perception scores before and after receiving NLP-based feedback will be assessed and used in planning a larger trial.

Measured directly after treatment consultation and after NLP -based feedback given to patients within 2 weeks of consultation

30-minute semi-structured interviews with counseling physicians will be conducted within 2 weeks of the intervention to obtain their opinions on the utility and ideal implementation strategy for the NLP-based feedback.

Difference between reported risk of side effects and prognosis with gold standard (physician-level outcome)

The difference in reported risk estimates given by physicians during the consultation as compared with the gold standards for these risks (i.e. for side effects, estimates from the CAESAR study; for cancer risk with and without treatment, risks of cancer mortality in the WW group of SPCG-4 trial at the patient's life expectancy as determined by the prostate cancer comorbidity index). Variability (standard deviation) in accuracy of estimates will be assessed and used in planning a larger trial. Accuracy of estimates for the interventional period will be compared with physician-specific historical references from a previously conducted trial using the standard of care (i.e. no NLP-based intervention).

Quality of composite physician risk communication score in treatment consultation (physician-level outcome)

Quality of risk communication scores will be calculated by qualitatively analyzing treatment consultation transcripts to assess the highest quality of communication used to transmit information regarding all key tradeoffs (cancer prognosis, life expectancy, erectile dysfunction, urinary incontinence, and irritative urinary symptoms). The quality of risk communication scale ranges from 0 to 5 for each outcome, with 0 representing the lowest score and 5 representing the highest score (Daskivich et al, J Urol 2022; Naser-Tavakolian et al, J Urol 2022). Scores for all key tradeoffs will be averaged to yield a composite quality of risk communication score. Variability (standard deviation) in quality scores will be assessed and used in planning a larger trial.

30-minute semi-structured interviews with patients will be conducted at the conclusion of the study period to obtain their opinions on the utility and ideal implementation strategy for the NLP-based feedback

Inclusion Criteria: Men undergoing initial treatment consultation for clinically localized prostate cancer; Men with upgraded prostate cancer on active surveillance considering conversion to definitive local therapy. Cedars-Sinai patient. Ability to read and write in English. Exclusion Criteria: Under 18 years of age; Subjects with difficulty communicating or dementia; Non-English speakers, given that our NLP-based tools cannot be used with languages other than English; Men with locally advanced or metastatic prostate cancer; Men who have already been treated for clinically localized prostate cancer