碩士 / 國立高雄應用科技大學 / 電子工程系 / 101 / Purpose: To improve the predict performance of logistic regression based on model parameters selection for the incidence of xerostomia after radiotherapy in nasopharyngeal carcinoma (NPC) and head and neck squamous cell carcinoma (HNSCC).
Materials and methods: A prospective survey of quality of life (QoL) using the EORTC QLQ-C30 and QLQ-H&N35 questionnaires were used. The patients were asked to complete the questionnaire prior to treatment and 3 months, and 12 months after radiotherapy. Prior analysing the questionnaires, we used EM (Expectation maximization algorithm) to impute missing data, and excluded the samples with moderate and severe xerostomia before treatment. Two kinds of input-data modules: Ipd (parotid dose data) and Irobust (selected) were used in this study. Among them, Irobust method was using Carmer’s V and point-biserial correlation (rpb) to select highly relevant variables. The binary logistic regression was used to describe the normal tissue complication probability (NTCP) model for the incidence of xerostomia. Model performance was assessed by using statistical diagnosis parameters, the receiver operating characteristic curve (ROC) and the area under the curve (AUC).
Results: The parameters selected by the Omnibus test are statistical significant for system performance improvement with a p < 0.05. Hosmer-Lemeshow test shows that there are no statistical significant differences between observed and predicted values with a p > 0.05. The prediction accuracy in HNSCC are Ipd65.00% (3 months), Irobust 72.50% (3 months); Ipd82.61% (12 months), Irobust 84.00% (12 months) , respectively. For NPC cases, the prediction accuracy are Ipd 63.33% (3 months); Irobust 70.00% (3 months); Ipd 72.73%(12 months), Irobust 77.27% (12 months), respectively. AUC (area under the receiver operating characteristic curve) presents that the Irobust has better performance in HNSCC. AUC values are 0.75 (3 months) and 0.91 (12 months), respectively; while the best results for NPC is Irobust: AUC values are 0.73 (3 months) and 0.74 (12 months), respectively.
Conclusion: We show that the predict performance of logistic regression for the incidence of xerostomia can be improved by using model parameters selection for patients with NPC and HNSCC after radiotherapy. Whether such superiority in prediction accuracy could transfer into clinical advantages needs further investigation.
| Identifer | oai:union.ndltd.org:TW/101KUAS8393004 |
| Date | January 1900 |
| Creators | Yen-Yu Liu, 劉彥佑 |
| Contributors | Tsair-Fwu Lee, 李財福 |
| Source Sets | National Digital Library of Theses and Dissertations in Taiwan |
| Language | zh-TW |
| Detected Language | English |
| Type | 學位論文 ; thesis |
| Format | 71 |
Page generated in 0.022 seconds