NLP-Based Feedback for Prostate Cancer

(NLP RCT Trial)

Overseen By Antwon Chaplin, BA

Age: 18+

Sex: Male

Trial Phase: Academic

Sponsor: Cedars-Sinai Medical Center

Disqualifiers: Under 18, Dementia, Non-English, others

No Placebo Group All trial participants will receive the active study treatment (no placebo)

Compensation: Varies

Number of Visits: 2

220 Participants Needed

Trial Summary

Will I have to stop taking my current medications?

The trial information does not specify whether you need to stop taking your current medications. It seems focused on communication and decision-making rather than medication changes.

What data supports the effectiveness of the treatment NLP-Based Feedback for prostate cancer?

The research suggests that natural language processing (NLP) can accurately assess important patient-centered outcomes like urinary incontinence and bowel dysfunction after prostate cancer treatment, which may help in understanding and improving patient care. ^{1 2 3 4 5}

Is NLP-Based Feedback generally safe for humans?

There is no specific safety data available for NLP-Based Feedback in humans, but NLP is used in healthcare to help identify medication safety issues and adverse events, suggesting it is generally considered safe for use in analyzing health data. ^{6 7 8 9 10}

How does the NLP-based feedback treatment for prostate cancer differ from other treatments?

The NLP-based feedback treatment for prostate cancer is unique because it uses natural language processing (NLP) to analyze clinical notes and assess patient-centered outcomes like urinary incontinence and bowel dysfunction, which are not typically addressed by standard treatments. This approach focuses on improving the understanding of patient experiences and outcomes rather than directly treating the cancer itself. ^{5 11 12 13 14}

What is the purpose of this trial?

The purpose of the research is to assess the impact of a natural language processing + artificial intelligence (NLP+AI)-based risk communication feedback system to improve quality of risk communication of key tradeoffs during prostate cancer consultations among physicians and to improve patient decision making. In this cluster randomized trial, an evaluable 220 patients with newly diagnosed clinically localized prostate cancer will be cluster randomized within an evaluable 22 physicians to: 1. a control arm, in which patients will receive standard of care treatment consultations along with AUA-endorsed educational materials on treatment risks and benefits (for patients) and on SDM (for physicians) or 2. an experimental arm, in which patients and participating physicians will receive NLP+AI-based feedback on what was said about key tradeoffs within approximately 72 hours of the consultation to assist with decision making. Physicians will additionally be provided with grading of their risk communication for each visit based on an a priori defined framework for quality of risk communication and recommendations for improvement. In both study arms, there will be an audio-recorded follow-up phone or video call between the physician and patient to allow for further discussion of risk and clarifying any areas of ambiguity, which will be qualitatively analyzed to see if areas of poor communication were rectified. After the follow-up phone call, patients and participating physicians will be asked to complete a very brief survey about their experience. The study plans to test whether receiving NLP+AI-based feedback improves decisional conflict, shared decision making, and appropriateness of treatment choice over the standard of care in patients undergoing treatment consultations for prostate cancer. Study staff will also test whether providing feedback and grading of risk communication to physicians affects quality of physician risk communication, since providing feedback will promote more accountability for the quality of information provided to patients. The study will also analyze data from the control arm of the randomized controlled trial to understand variation in risk communication of key tradeoffs in relevant subgroups of tumor risk (low-, intermediate-, and high-risk), provider specialty (Urology, Radiation Oncology, Medical Oncology), and patient sociodemographics.

Eligibility Criteria

This trial is for men newly diagnosed with localized prostate cancer. Participants will be under the care of one of the 22 physicians involved in the study. The main goal is to improve how doctors and patients talk about treatment risks and benefits, helping patients make better decisions.

Inclusion Criteria

* Physician Inclusion Criteria (1) Physicians who typically counsel prostate cancer patients (Urology, Radiation Oncology, Medical Oncology)

Patient Inclusion Criteria

1. Patients undergoing initial treatment consultation for clinically localized prostate cancer;

Timeline

Screening

Participants are screened for eligibility to participate in the trial

2-4 weeks

Treatment Consultation

Patients receive standard of care treatment consultations or NLP+AI-based feedback on key tradeoffs to assist with decision making

2-4 weeks

1 visit (in-person)

Follow-up

Audio-recorded follow-up phone or video call between physician and patient to discuss risk and clarify ambiguities

2-4 weeks

1 visit (virtual)

Post-Study Analysis

Analysis of quality of risk communication and shared decision making outcomes

6-9 months

Treatment Details

Interventions

• NLP-Based Feedback

Trial Overview The trial tests a new system that uses NLP+AI to give feedback on doctor-patient conversations about prostate cancer treatments. It compares standard consultations with ones enhanced by AI feedback, aiming to see if this improves decision-making and communication.

Participant Groups

Group I: NLP-Based Feedback Arm Experimental Treatment 1 Intervention

Group II: Standard of Care Arm Active Control 1 Intervention

patients will receive standard of care treatment consultations along with AUA-endorsed educational materials on treatment risks and benefits (for patients) and on SDM (for physicians)

Who Is Running the Clinical Trial?

Cedars-Sinai Medical Center

Lead Sponsor

Trials

523

Recruited

165,000+

National Cancer Institute (NCI)

Collaborator

Trials

14,080

Recruited

41,180,000+

Findings from Research

Natural language processing (NLP) and machine learning (ML) techniques are effectively used to analyze unstructured patient-reported outcome (PRO) data from electronic health records (EHRs), with 79 studies reviewed highlighting their potential in clinical care.

The majority of studies focused on extracting PROs from clinical narratives and using them for purposes like disease progression prediction and phenotyping, indicating a growing trend towards integrating advanced data analysis methods in healthcare.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Natural language processing with machine learning methods to analyze unstructured patient-reported outcomes derived from electronic health records: A systematic review. Sim, JA., Huang, X., Horan, MR., et al.[2023]

Nomograms are effective tools that can help patients with prostate cancer and their doctors make informed treatment decisions by predicting individual outcomes based on multiple variables.

Currently, nomograms can predict progression-free survival after various treatments for localized prostate cancer, but there is a need for more models that also consider survival and quality of life.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Predicting clinical end points: treatment nomograms in prostate cancer. Di Blasio, CJ., Rhee, AC., Cho, D., et al.[2022]

Natural language processing (NLP) can significantly enhance cancer research by extracting valuable clinical data from unstructured text in electronic medical records, which can help in personalizing treatment options.

Oncologists can actively participate in developing NLP systems, which not only aids in cancer case identification and outcomes measurement but also promotes evidence-based research in oncology, potentially leading to improved cancer care.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Natural Language Processing in Oncology: A Review. Yim, WW., Yetisgen, M., Harris, WP., et al.[2021]

Artificial neural networks and clinical prediction tools can effectively forecast various cancer-related outcomes, such as disease stage and survival rates, providing valuable information for physicians and patients.

There is a need for more models that also consider quality of life outcomes, and combining these predictions into a comprehensive table could help patients make better-informed treatment choices.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Prediction tools in surgical oncology. Isariyawongse, BK., Kattan, MW.[2019]

A new natural language processing (NLP) pipeline effectively identifies patient-centered outcomes (PCOs) like urinary incontinence and bowel dysfunction in clinical notes, achieving an impressive average F1 score of 0.86.

This machine learning approach outperformed existing rule-based models, demonstrating that it can accurately classify clinical notes without the need for manual review, making it a more efficient tool for assessing patient outcomes after prostate cancer treatment.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Weakly supervised natural language processing for assessing patient-centered outcome following prostate cancer treatment. Banerjee, I., Li, K., Seneviratne, M., et al.[2020]

A new standardized Patient-Reported Outcomes version of the Common Terminology Criteria for Adverse Events (PRO-CTCAE) subset was developed specifically for monitoring adverse events in prostate cancer patients, based on input from 30 patients and 16 healthcare providers.

The final subset includes 12 key adverse events such as erectile dysfunction, urinary issues, and emotional distress, with 90% consensus among participants, highlighting the importance of capturing patient experiences for better care.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Selecting a PRO-CTCAE-based subset for patient-reported symptom monitoring in prostate cancer patients: a modified Delphi procedure. Feldman, E., Pos, FJ., Smeenk, RJ., et al.[2023]

Natural language processing (NLP) can significantly enhance medication safety by automating the identification of adverse events from various data sources, such as electronic health records and social media, which can lead to quicker detection and response to potential risks.

Despite its potential, the effectiveness of NLP in monitoring adverse events is currently limited by challenges in data sharing and interoperability between healthcare organizations, which hinders comprehensive surveillance across populations.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Natural Language Processing and Its Implications for the Future of Medication Safety: A Narrative Review of Recent Advances and Challenges. Wong, A., Plasek, JM., Montecalvo, SP., et al.[2019]

A retrospective study involving 400 cancer patients identified 790 clinical triggers in medical records, leading to the detection of 304 unique adverse events (AEs), highlighting the tool's potential to improve patient safety in oncology.

The original screening tool had a positive predictive value (PPV) of 0.40 for total AEs and 0.15 for preventable AEs, with modifications increasing the PPV to 0.48 for total AEs, suggesting that refining this tool could enhance its effectiveness in real-time AE detection.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Performance of a Trigger Tool for Identifying Adverse Events in Oncology. Lipitz-Snyderman, A., Classen, D., Pfister, D., et al.[2021]

In a study of 322,887 cancer patients undergoing treatment, 19% experienced oncology-specific adverse events, with higher rates in those with metastatic cancers (41% for colorectal and 50% for lung cancer).

The most common triggers for these adverse events included abnormal serum bicarbonate levels and blood transfusions, highlighting the need for targeted monitoring in cancer care to improve patient safety.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Developing a cancer-specific trigger tool to identify treatment-related adverse events using administrative data. Weingart, SN., Nelson, J., Koethe, B., et al.[2021]

The MOSES method, which incorporates the timing and severity of adverse events, was found to be more accurate than the traditional CTCAE system in predicting the impact of symptoms on quality of life in cervical cancer patients undergoing radiotherapy, with a study involving 300 patients.

In comparing two radiotherapy techniques, 3DCRT showed a higher cumulative MOSES score than IMRT, indicating potentially greater late toxicity, despite no significant difference in overall gastrointestinal toxicity grades between the two methods.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Months and Severity Score (MOSES) in a Phase III trial (PARCER): A new comprehensive method for reporting adverse events in oncology clinical trials. Ranjan, N., Chopra, S., Mangaj, A., et al.[2022]

A deep learning model using BERT achieved a tumor response category (TRC) classification F1 score of 0.70, outperforming conventional algorithms and radiology technologist students, indicating its potential for effective analysis of free-text oncology reports.

While the BERT model performed comparably to medical students, it was still less effective than experienced radiologists, highlighting the challenges posed by lexical complexity and semantic ambiguities in oncology reports.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Deep Learning-based Assessment of Oncologic Outcomes from Natural Language Processing of Structured Radiology Reports. Fink, MA., Kades, K., Bischoff, A., et al.[2022]

A natural language processing (NLP) algorithm was developed and validated to identify metastatic prostate cancer (mPCa) patients from clinical notes, successfully identifying 76,082 out of 1,144,610 diagnosed veterans, demonstrating high specificity (0.979) and sensitivity (0.919).

The use of NLP significantly enhances the identification of mPCa patients compared to traditional methods, providing a reliable tool to improve cancer registry data and potentially facilitate access to innovative treatments for mPCa.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Ascertainment of Veterans With Metastatic Prostate Cancer in Electronic Health Records: Demonstrating the Case for Natural Language Processing. Alba, PR., Gao, A., Lee, KM., et al.[2021]

Natural language processing (NLP) can effectively convert unstructured clinical information from patient reports into structured data, making it accessible for automated applications that can enhance patient care and reduce costs.

Recent advancements in NLP systems have shown promising results in clinical settings, with some systems already deployed, indicating a significant step forward in utilizing electronic clinical information for better healthcare outcomes.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Natural language processing and its future in medicine. Friedman, C., Hripcsak, G.[2019]

ChatGPT was able to answer all 59 queries related to prostate cancer, but its accuracy and quality were suboptimal, with an average F1 score of 0.426 and a General Quality Score of 3.62 out of 5, indicating room for improvement.

Given the limitations in accuracy and potential for misunderstandings, caution is advised when using ChatGPT as a source of patient information on prostate cancer, highlighting the need for further research on AI-generated content.

NCBI (Pubmed)

pubmed.ncbi.nlm.nih.gov

Can ChatGPT, an Artificial Intelligence Language Model, Provide Accurate and High-quality Patient Information on Prostate Cancer? Coskun, B., Ocakoglu, G., Yetemen, M., et al.[2023]

References

1. Netherlands pubmed.ncbi.nlm.nih.gov

Natural language processing with machine learning methods to analyze unstructured patient-reported outcomes derived from electronic health records: A systematic review. [2023]

2. United States pubmed.ncbi.nlm.nih.gov

Predicting clinical end points: treatment nomograms in prostate cancer. [2022]

3. United States pubmed.ncbi.nlm.nih.gov

Natural Language Processing in Oncology: A Review. [2021]

4. United States pubmed.ncbi.nlm.nih.gov

Prediction tools in surgical oncology. [2019]

5. United States pubmed.ncbi.nlm.nih.gov

Weakly supervised natural language processing for assessing patient-centered outcome following prostate cancer treatment. [2020]

6. England pubmed.ncbi.nlm.nih.gov

Selecting a PRO-CTCAE-based subset for patient-reported symptom monitoring in prostate cancer patients: a modified Delphi procedure. [2023]

7. United States pubmed.ncbi.nlm.nih.gov

Natural Language Processing and Its Implications for the Future of Medication Safety: A Narrative Review of Recent Advances and Challenges. [2019]

8. United States pubmed.ncbi.nlm.nih.gov

Performance of a Trigger Tool for Identifying Adverse Events in Oncology. [2021]

9. United States pubmed.ncbi.nlm.nih.gov

Developing a cancer-specific trigger tool to identify treatment-related adverse events using administrative data. [2021]

10. England pubmed.ncbi.nlm.nih.gov

Months and Severity Score (MOSES) in a Phase III trial (PARCER): A new comprehensive method for reporting adverse events in oncology clinical trials. [2022]

11. United States pubmed.ncbi.nlm.nih.gov

Deep Learning-based Assessment of Oncologic Outcomes from Natural Language Processing of Structured Radiology Reports. [2022]

12. United States pubmed.ncbi.nlm.nih.gov

Ascertainment of Veterans With Metastatic Prostate Cancer in Electronic Health Records: Demonstrating the Case for Natural Language Processing. [2021]

13. United States pubmed.ncbi.nlm.nih.gov

Natural language processing and its future in medicine. [2019]

14. United States pubmed.ncbi.nlm.nih.gov

Can ChatGPT, an Artificial Intelligence Language Model, Provide Accurate and High-quality Patient Information on Prostate Cancer? [2023]