Sağlık verilerinde veri madenciliği teknikleri ile sağkalımı etkileyen faktörlerin seçimi, performanslarının değerlendirilmesi
Küçük Resim Yok
Tarih
2022
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Ege Üniversitesi
Erişim Hakkı
info:eu-repo/semantics/openAccess
Özet
Giriş ve Amaç: Bu çalışmada, ağaç tabanlı veri madenciliği yöntemi Rastgele Sağkalım Orman'ın araştırılması ve tıp alanındaa toplanan veri üzerinde analizler yaparak sonuçların incelenmesi amaçlanmıştır. Yöntem: Çalışmada Mendeley Data açık kaynak platformunda yayınlanan Awodutire, Kolawole, Ilori tarafından 2017 yılındaki çalışmalarında uygulanmış, Osogbo'daki Ladoke Akintola Teknoloji Eğitim Hastanesi'ndeki meme kanseri hastasından alınan klinik verilerle Cox Regresyon ve Rastgele Sağkalım Orman yöntemleri ile analizler yapılmıştır. Analizlerde RStudio Version 1.4.1717 programı kullanılmıştır. Bulgular: Çalışmada 89 hasta vardır. Hastaların yaş ortalaması 50.29±10.848, ortalama menarş yaşı 15.72±2.326 ve ortalama emzirme yılı 1.39±0.547'dir. Hastalara ait sağkalım sürelerinin %42.7'si (n=38) sağdan sansürlüdür. Hastaların %53.9'u (n=48) doğum kontrol hapı kullanmakta, teşhis anında %67.4'ünün (n=60) erken evre(Evre I. Ve II), %32.6'sının (n=29) geç evredir (III.Evre ve IV.Evre). Tedavi sırasında %43.8'ine (n=39) neoadjuvant uygulanmıştır. Hastaların medyan sağkalım süresi 251± 52.82 [%95 G.A. 147.5-354.5] gün olarak hesaplanmıştır. Cox regresyon yönteminde tedavi süresinde neoadjuvan kullanımının tehlike oranı üzerindeki etkisi istatistiksel olarak anlamlıdır (p<0.05). Cox regresyon modeli uyum indeksi olan C-Index 0.648 olarak hesaplandı. Rastgele Sağkalım orman yönteminde veri seti %70 test, %30 analiz dışı tutulmuş, out-of-box veri olarak ayrılmıştır. Buna göre önemli değişkenler ortalama emzirme yılı ve doğum kontrol hapı kullanımı olarak belirlenmiştir. Rastgele sağkalım yöntemine ait C-index 0.60 olarak hesaplanmıştır. Sonuç: Her iki yöntemde benzer performans göstermiştir, Örneklem ve değişken sayısı arttırılarak analiz tekrarlanmalıdır
Introduction and Aim: In this study, it is aimed to examine the Random Survival Forests (RSF) method, which is one of the tree-based data mining methods, and to discuss the results by applying it to a data set obtained from the health field. Methods: In the study, analyzes were made by Cox Regression and Random Survival Forest methods with clinical data obtained from a breast cancer patient at Ladoke Akintola Technology Teaching Hospital in Osogbo, which was applied by Awodutire, Kolawale, Ilori, published on the Mendeley Data open source platform, in their study in 2017. RStudio Version 1.4.1717 program was used in the analysis. Results: There were 89 patients in the study. The mean age of the patients was 50.29±10.848, the mean age at menarche was 15.72±2.326, and the mean breastfeeding year was 1.39±0.547. 42.7% (n=38) of the patients' survival times were right-censored. 53.9% (n=48) of the patients were using birth control pills, at the time of diagnosis, 67.4% (n=60) were in the early stage (Stage I and II), and 32.6% (n=29) was in the late stage (III. Stage and Stage IV). Neoadjuvant was applied to 43.8% (n=39) of them during treatment. Median survival of patients 251±52.82 [95% G.A. 147.5-354.5] days. In the Cox regression method, the effect of neoadjuvant use on the hazard ratio was statistically significant (p<0.05). The C-Index, which is the Cox regression model fit index, was calculated as 0.648. In the Random Survival forest method, the data set was divided as 70% tested, 30% excluded, out-of-box data. Accordingly, the important variables were determined as the mean year of breastfeeding and the use of birth control pills. The C-index of the random survival method was calculated as 0.60. Conclusion: Both methods showed similar performance. The analysis should be repeated by increasing the number of samples and variables.
Introduction and Aim: In this study, it is aimed to examine the Random Survival Forests (RSF) method, which is one of the tree-based data mining methods, and to discuss the results by applying it to a data set obtained from the health field. Methods: In the study, analyzes were made by Cox Regression and Random Survival Forest methods with clinical data obtained from a breast cancer patient at Ladoke Akintola Technology Teaching Hospital in Osogbo, which was applied by Awodutire, Kolawale, Ilori, published on the Mendeley Data open source platform, in their study in 2017. RStudio Version 1.4.1717 program was used in the analysis. Results: There were 89 patients in the study. The mean age of the patients was 50.29±10.848, the mean age at menarche was 15.72±2.326, and the mean breastfeeding year was 1.39±0.547. 42.7% (n=38) of the patients' survival times were right-censored. 53.9% (n=48) of the patients were using birth control pills, at the time of diagnosis, 67.4% (n=60) were in the early stage (Stage I and II), and 32.6% (n=29) was in the late stage (III. Stage and Stage IV). Neoadjuvant was applied to 43.8% (n=39) of them during treatment. Median survival of patients 251±52.82 [95% G.A. 147.5-354.5] days. In the Cox regression method, the effect of neoadjuvant use on the hazard ratio was statistically significant (p<0.05). The C-Index, which is the Cox regression model fit index, was calculated as 0.648. In the Random Survival forest method, the data set was divided as 70% tested, 30% excluded, out-of-box data. Accordingly, the important variables were determined as the mean year of breastfeeding and the use of birth control pills. The C-index of the random survival method was calculated as 0.60. Conclusion: Both methods showed similar performance. The analysis should be repeated by increasing the number of samples and variables.
Açıklama
Anahtar Kelimeler
Biyoistatistik, Biostatistics
