Time-dependent personalized prognostic analysis by machine learning in biochemical recurrence after radical prostatectomy: a retrospective cohort study
Abstract Background For biochemical recurrence following radical prostatectomy for prostate cancer, treatments such as radiation therapy and androgen deprivation therapy are administered. To diagnose postoperative recurrence as early as possible and to intervene with treatment at the appropriate tim...
Saved in:
| Main Authors: | , , , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
BMC
2024-11-01
|
| Series: | BMC Cancer |
| Subjects: | |
| Online Access: | https://doi.org/10.1186/s12885-024-13203-8 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850216514922217472 |
|---|---|
| author | Kodai Sato Shinichi Sakamoto Shinpei Saito Hiroki Shibata Yasutaka Yamada Nobuyoshi Takeuchi Yusuke Goto Sazuka Tomokazu Yusuke Imamura Tomohiko Ichikawa Eiryo Kawakami |
| author_facet | Kodai Sato Shinichi Sakamoto Shinpei Saito Hiroki Shibata Yasutaka Yamada Nobuyoshi Takeuchi Yusuke Goto Sazuka Tomokazu Yusuke Imamura Tomohiko Ichikawa Eiryo Kawakami |
| author_sort | Kodai Sato |
| collection | DOAJ |
| description | Abstract Background For biochemical recurrence following radical prostatectomy for prostate cancer, treatments such as radiation therapy and androgen deprivation therapy are administered. To diagnose postoperative recurrence as early as possible and to intervene with treatment at the appropriate time, it is essential to accurately predict recurrence after radical prostatectomy. However, postoperative recurrence involves numerous patient-related factors, making its prediction challenging. The purpose of this study is to accurately predict the timing of biochemical recurrence after radical prostatectomy and to analyze the risk factors for follow-up of high-risk patients and early detection of recurrence. Methods We utilized the machine learning survival analysis model called the Random Survival Forest utilizing the 58 clinical factors from 548 patients who underwent radical prostatectomy at Chiba University Hospital. To visualize prognostic factors and assess accuracy of the time course probability, we employed SurvSHAP(t) and time-dependent Area Under Cureve(AUC). Results The time-dependent AUC of RSF was 0.785, which outperformed the Cox proportional hazards model (0.704), the Cancer of the Prostate Risk Assessment (CAPRA) score (0.710), and the D’Amico score (0.658). The key prognostic factors for early recurrence were Gleason score(GS), Seminal vesicle invasion(SV), and PSA. The contribution of PSA to recurrence decreases after the first year, while SV and GS increase over time. Conclusion Our prognostic model analyzed the time-dependent relationship between the timing of recurrence and prognostic factors. Our study achieved personalized prognosis analysis and its rationale after radical prostatectomy by employing machine learning prognostic model. This prognostic model contributes to the early detection of recurrence by enabling clinicians to conduct appropriate follow-ups for high-risk patients. |
| format | Article |
| id | doaj-art-0acbb301de4c43c2896d13286fe6b29c |
| institution | OA Journals |
| issn | 1471-2407 |
| language | English |
| publishDate | 2024-11-01 |
| publisher | BMC |
| record_format | Article |
| series | BMC Cancer |
| spelling | doaj-art-0acbb301de4c43c2896d13286fe6b29c2025-08-20T02:08:16ZengBMCBMC Cancer1471-24072024-11-0124111010.1186/s12885-024-13203-8Time-dependent personalized prognostic analysis by machine learning in biochemical recurrence after radical prostatectomy: a retrospective cohort studyKodai Sato0Shinichi Sakamoto1Shinpei Saito2Hiroki Shibata3Yasutaka Yamada4Nobuyoshi Takeuchi5Yusuke Goto6Sazuka Tomokazu7Yusuke Imamura8Tomohiko Ichikawa9Eiryo Kawakami10Department of Urology, Graduate School of Medicine, Chiba UniversityDepartment of Urology, Graduate School of Medicine, Chiba UniversityDepartment of Urology, Graduate School of Medicine, Chiba UniversityDepartment of Urology, Graduate School of Medicine, Chiba UniversityDepartment of Urology, Graduate School of Medicine, Chiba UniversityDepartment of Urology, Graduate School of Medicine, Chiba UniversityDepartment of Urology, Graduate School of Medicine, Chiba UniversityDepartment of Urology, Graduate School of Medicine, Chiba UniversityDepartment of Urology, Graduate School of Medicine, Chiba UniversityDepartment of Urology, Graduate School of Medicine, Chiba UniversityDepartment of Artificial Intelligence Medicine, Graduate School of Medicine, Chiba UniversityAbstract Background For biochemical recurrence following radical prostatectomy for prostate cancer, treatments such as radiation therapy and androgen deprivation therapy are administered. To diagnose postoperative recurrence as early as possible and to intervene with treatment at the appropriate time, it is essential to accurately predict recurrence after radical prostatectomy. However, postoperative recurrence involves numerous patient-related factors, making its prediction challenging. The purpose of this study is to accurately predict the timing of biochemical recurrence after radical prostatectomy and to analyze the risk factors for follow-up of high-risk patients and early detection of recurrence. Methods We utilized the machine learning survival analysis model called the Random Survival Forest utilizing the 58 clinical factors from 548 patients who underwent radical prostatectomy at Chiba University Hospital. To visualize prognostic factors and assess accuracy of the time course probability, we employed SurvSHAP(t) and time-dependent Area Under Cureve(AUC). Results The time-dependent AUC of RSF was 0.785, which outperformed the Cox proportional hazards model (0.704), the Cancer of the Prostate Risk Assessment (CAPRA) score (0.710), and the D’Amico score (0.658). The key prognostic factors for early recurrence were Gleason score(GS), Seminal vesicle invasion(SV), and PSA. The contribution of PSA to recurrence decreases after the first year, while SV and GS increase over time. Conclusion Our prognostic model analyzed the time-dependent relationship between the timing of recurrence and prognostic factors. Our study achieved personalized prognosis analysis and its rationale after radical prostatectomy by employing machine learning prognostic model. This prognostic model contributes to the early detection of recurrence by enabling clinicians to conduct appropriate follow-ups for high-risk patients.https://doi.org/10.1186/s12885-024-13203-8Prostate cancerProstatectomyRecurrenceMachine learningPrognostic model |
| spellingShingle | Kodai Sato Shinichi Sakamoto Shinpei Saito Hiroki Shibata Yasutaka Yamada Nobuyoshi Takeuchi Yusuke Goto Sazuka Tomokazu Yusuke Imamura Tomohiko Ichikawa Eiryo Kawakami Time-dependent personalized prognostic analysis by machine learning in biochemical recurrence after radical prostatectomy: a retrospective cohort study BMC Cancer Prostate cancer Prostatectomy Recurrence Machine learning Prognostic model |
| title | Time-dependent personalized prognostic analysis by machine learning in biochemical recurrence after radical prostatectomy: a retrospective cohort study |
| title_full | Time-dependent personalized prognostic analysis by machine learning in biochemical recurrence after radical prostatectomy: a retrospective cohort study |
| title_fullStr | Time-dependent personalized prognostic analysis by machine learning in biochemical recurrence after radical prostatectomy: a retrospective cohort study |
| title_full_unstemmed | Time-dependent personalized prognostic analysis by machine learning in biochemical recurrence after radical prostatectomy: a retrospective cohort study |
| title_short | Time-dependent personalized prognostic analysis by machine learning in biochemical recurrence after radical prostatectomy: a retrospective cohort study |
| title_sort | time dependent personalized prognostic analysis by machine learning in biochemical recurrence after radical prostatectomy a retrospective cohort study |
| topic | Prostate cancer Prostatectomy Recurrence Machine learning Prognostic model |
| url | https://doi.org/10.1186/s12885-024-13203-8 |
| work_keys_str_mv | AT kodaisato timedependentpersonalizedprognosticanalysisbymachinelearninginbiochemicalrecurrenceafterradicalprostatectomyaretrospectivecohortstudy AT shinichisakamoto timedependentpersonalizedprognosticanalysisbymachinelearninginbiochemicalrecurrenceafterradicalprostatectomyaretrospectivecohortstudy AT shinpeisaito timedependentpersonalizedprognosticanalysisbymachinelearninginbiochemicalrecurrenceafterradicalprostatectomyaretrospectivecohortstudy AT hirokishibata timedependentpersonalizedprognosticanalysisbymachinelearninginbiochemicalrecurrenceafterradicalprostatectomyaretrospectivecohortstudy AT yasutakayamada timedependentpersonalizedprognosticanalysisbymachinelearninginbiochemicalrecurrenceafterradicalprostatectomyaretrospectivecohortstudy AT nobuyoshitakeuchi timedependentpersonalizedprognosticanalysisbymachinelearninginbiochemicalrecurrenceafterradicalprostatectomyaretrospectivecohortstudy AT yusukegoto timedependentpersonalizedprognosticanalysisbymachinelearninginbiochemicalrecurrenceafterradicalprostatectomyaretrospectivecohortstudy AT sazukatomokazu timedependentpersonalizedprognosticanalysisbymachinelearninginbiochemicalrecurrenceafterradicalprostatectomyaretrospectivecohortstudy AT yusukeimamura timedependentpersonalizedprognosticanalysisbymachinelearninginbiochemicalrecurrenceafterradicalprostatectomyaretrospectivecohortstudy AT tomohikoichikawa timedependentpersonalizedprognosticanalysisbymachinelearninginbiochemicalrecurrenceafterradicalprostatectomyaretrospectivecohortstudy AT eiryokawakami timedependentpersonalizedprognosticanalysisbymachinelearninginbiochemicalrecurrenceafterradicalprostatectomyaretrospectivecohortstudy |