Optimal design for survival experiments– A simulation study
Başlık çevirisi mevcut değil.
- Tez No: 402847
- Danışmanlar: DR. STEFANIE BIEDERMANN
- Tez Türü: Yüksek Lisans
- Konular: Matematik, Mathematics
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 2015
- Dil: İngilizce
- Üniversite: University of Southampton
- Enstitü: Yurtdışı Enstitü
- Ana Bilim Dalı: Belirtilmemiş.
- Bilim Dalı: Belirtilmemiş.
- Sayfa Sayısı: 56
Özet
Özet yok.
Özet (Çeviri)
Survival experiments consist of analysing longitudinal data on the occurrence of events. Events may include death, injury and onset of illness, recovery from illness or any other meaningful continuous variable. In this study, time-to-event between two treatments or a single treatment with a placebo was used. Time-to-event is the time from entry into a study until a subject has a particular outcome. Designing of survival experiments is an integral part of the scientific process because resources are always scarce and judicious use of the limited resources is essential. Identifying optimal designs for data collection and assessing their performance in realistic scenarios is therefore paramount. Design of those experiments includes different parameters but one of the most important aspects is how to divide patients into two groups. The balanced design which means assigning the same amount of patients in each group is a commonly used method and is used as a baseline value in this study. A lot of studies are made to find the optimal design in survival experiments but when the data is censored the optimal design changes. However, there is not enough information in the literature on how to deal with censoring. Censoring occurs when patients are lost to follow up or drop out of the study before the study finishes. Also when the study finishes before the patients reach the outcome of interest their data is censored. It is therefore important to find plausible scenarios for different censoring methods and compare optimal designs. This paper uses type-I censoring and random censoring to compare different survival experiment scenarios in order to give a guidance for practitioners planning a survival trial as to what designs would be a good choice. By looking at the literature, it is seen that c-optimal designs are the best fit for the kind of survival experiments that is used in this study. Other than the c-optimal designs, standardised maximin and locally c-optimal designs are used to compare the c-efficiencies with the balanced designs. It has been found that whether there was censoring or no censoring, the usage of a c-optimal design increases the efficiency of the parameter estimates. The parameter estimates are the natural logarithm of the baseline hazard and the coefficient of the covariate which can be either a treatment or a placebo. It has been seen that when the data being analysed had a negative baseline hazard and when a c-optimal design is used rather than a balanced design, the increase in the accuracy of the parameter estimates can go up to 9% for the coefficient of the covariate and it can go up to 21% for the estimate of the natural logarithm of the baseline hazard. This increase in the accuracy reduces when coefficient of the covariate in the data decreases. As this coefficient goes towards 0, the mean squared errors of the parameter estimates decrease but they are still 2% better for the coefficient and 10% better for natural logarithm of the baseline hazard compared to the balanced design. The reason for the change in the accuracy of the mean squared error of the estimates can be explained with the fact that c-optimal design uses different weights to assign patients in each treatment. C-optimal designs tend to assign less patients to support point 0, meaning they are not taking any treatment and in the placebo group when the coefficient is negative, whereas when the coefficient is positive more patients are assigned to support point 0. When this coefficient gets smaller, c-optimal design tends to have a similar patient distribution as the balanced design. If the data is heavily censored, the number of patients assigned to support point 0 increases compared with the low censoring cases. With the increase of proportion of censoring the design assigns more patients where it is more difficult to estimate. The calculations are made using a baseline hazard very close to zero meaning that the baseline hazard does not have as much effect as the coefficient of the covariate. The reason behind this is that it is a realistic scenario in practice as in (Freireich, et al., 1963). Same calculations can also be made with a positive baseline hazard to see the changes. Also another software instead of R can be used, to get better results for different distributions because when there is heavy censoring R cannot seem to converge in hazard coefficients and this results in some wrong values.
Benzer Tezler
- Healthcare network design and inventory decisions for disaster management under uncertainty
Afet yönetiminde belirsizlik altında afet yönetimi için sağlık ağ tasarımı ve envanter kararları
MÜGE ACAR
Doktora
İngilizce
2020
Endüstri ve Endüstri MühendisliğiEskişehir Teknik ÜniversitesiEndüstri Mühendisliği Ana Bilim Dalı
PROF. DR. ONUR KAYA
- Functional chocolate development: Enrichment of dark chocolate with nano-liposome encapsulated antioxidants, and/or pro- and pre-biotics, bioavailability studies
Fonksiyonel çikolata geliştirilmesi: Bitter çikolatanın nano-lipozomla enkapsüle edilen antioksidanlarla, ve/veya pro- ve pre-biyotiklerle zenginleştirilmesi, biyoyararlılık çalışmaları
MİNE ÖZGÜVEN
Doktora
İngilizce
2014
Gıda Mühendisliğiİstanbul Teknik ÜniversitesiGıda Mühendisliği Ana Bilim Dalı
PROF. DR. BERAAT ÖZÇELİK
- Yolcu güvenliğini güvence altına alan E.C.E regülasyonuna göre otobüs gövde analizi
Bus superstructure analysis according to the E.C.E regulation securing passenger safety
FATİH HAN AVCI
Yüksek Lisans
Türkçe
2016
Makine Mühendisliğiİstanbul Teknik ÜniversitesiMakine Mühendisliği Ana Bilim Dalı
PROF. DR. CEVAT ERDEM İMRAK
- Optimal design and expansion of water distribution systems using genetic algorithm
Su dağıtım şebekelerinin genetik algoritma ile optimum dizaynı ve genişletilmesi
ABDULLAH MURAT KAHRAMAN
Yüksek Lisans
İngilizce
2003
Çevre MühendisliğiDokuz Eylül ÜniversitesiÇevre Mühendisliği Ana Bilim Dalı
PROF. DR. DAVUT ÖZDAĞLAR
- Sürdürülebilir toplu konut yerleşmesi tasarımı için Pareto genetik algoritmaya dayalı bir model önerisi: SSPM
A model for sustainable site layout design with pareto genetic algorithm: SSPM
YAZGI AKSOY