Prediction Of Toxicity In Irradiated Head And Neck Cancer Patients Based On The Geometry Of High/Middle/Low Ptvs To Surrounding Oars

Purpose/Objective(s)The quality of life of the irradiated head and neck cancer (HNC) patient can be significantly limited by xerostomia, dysphagia, and painful mucositis leading to weight loss. The ability to predict and iteratively reduce these toxicities by applying a learning health system (LHS) model is thus an important long-term goal in our Oncospace project. We hypothesize that the geometry of planning target volumes (PTVs) and organs at risks (OARs) are correlated with these outcomes. This study aims to assess the relationship between toxicities and the distance from high/middle/low dose PTVs to OARs in simultaneous-integrated boost intensity-modulated radiation therapy (SIB-IMRT).Materials/MethodsOncospace is an integrated analytic relational database systematically captures outcome results and all aspects of a radiation therapy treatment plan. 479 HNC patients treated from 2007 to 2014 were available for analysis of the geometric relationship between PTVs and OARs with toxicity outcomes. Minimum distances between the contoured surfaces of the high (HD) / middle (MD) / low dose (LD)-PTVs and each a prior hypothesized toxicity OAR were analyzed with an in-house developed overlap volume histogram (OVH) algorithm. The relationship between toxicities and distances from PTV to OARs was modeled by known Nadaraya-Watson density estimation method with a Gaussian kernel using weighted squared sum of the distance differences among patients.ResultsA total of 77 individual OARs were assessed. Combined distance from LD-PTV to parotids/submandibular (AUC=.699, p-value=0.0002 for the alternative AUC>.5) and all the OARs (AUC=.731, p-value<0.0001) was a strong predictor of grade ≥2 vs. <2 xerostomia, demonstrating the potential clinical impact of both the parotid and submandibular glands. It was also discovered that dysphagia (grade ≥3 vs. <3) was most strongly predicted by the combined distance of the MD-PTV to the larynx, constrictor muscles and esophagus (AUC=.856, p-value<0.0001). Weight loss (≥5kg vs. <5kg) at 3 months post-RT was predicted by the combined distance from the HD-PTV to all the OARs (AUC=.664, p-value= 0.0015), indicating the existence of high risk PTV locations relative to OARs.ConclusionThe informatics framework combined with data mining tools can facilitate large-scale analysis of toxicity outcomes and is encouraging for the development of a LHS model to reduce the risk of radiation therapy toxicities. Our analysis suggests that the OVH and attention to the distance between PTVs and OARs may be a promising approach to model the risk of developing treatment complications warranting further investigations. Purpose/Objective(s)The quality of life of the irradiated head and neck cancer (HNC) patient can be significantly limited by xerostomia, dysphagia, and painful mucositis leading to weight loss. The ability to predict and iteratively reduce these toxicities by applying a learning health system (LHS) model is thus an important long-term goal in our Oncospace project. We hypothesize that the geometry of planning target volumes (PTVs) and organs at risks (OARs) are correlated with these outcomes. This study aims to assess the relationship between toxicities and the distance from high/middle/low dose PTVs to OARs in simultaneous-integrated boost intensity-modulated radiation therapy (SIB-IMRT). The quality of life of the irradiated head and neck cancer (HNC) patient can be significantly limited by xerostomia, dysphagia, and painful mucositis leading to weight loss. The ability to predict and iteratively reduce these toxicities by applying a learning health system (LHS) model is thus an important long-term goal in our Oncospace project. We hypothesize that the geometry of planning target volumes (PTVs) and organs at risks (OARs) are correlated with these outcomes. This study aims to assess the relationship between toxicities and the distance from high/middle/low dose PTVs to OARs in simultaneous-integrated boost intensity-modulated radiation therapy (SIB-IMRT). Materials/MethodsOncospace is an integrated analytic relational database systematically captures outcome results and all aspects of a radiation therapy treatment plan. 479 HNC patients treated from 2007 to 2014 were available for analysis of the geometric relationship between PTVs and OARs with toxicity outcomes. Minimum distances between the contoured surfaces of the high (HD) / middle (MD) / low dose (LD)-PTVs and each a prior hypothesized toxicity OAR were analyzed with an in-house developed overlap volume histogram (OVH) algorithm. The relationship between toxicities and distances from PTV to OARs was modeled by known Nadaraya-Watson density estimation method with a Gaussian kernel using weighted squared sum of the distance differences among patients. Oncospace is an integrated analytic relational database systematically captures outcome results and all aspects of a radiation therapy treatment plan. 479 HNC patients treated from 2007 to 2014 were available for analysis of the geometric relationship between PTVs and OARs with toxicity outcomes. Minimum distances between the contoured surfaces of the high (HD) / middle (MD) / low dose (LD)-PTVs and each a prior hypothesized toxicity OAR were analyzed with an in-house developed overlap volume histogram (OVH) algorithm. The relationship between toxicities and distances from PTV to OARs was modeled by known Nadaraya-Watson density estimation method with a Gaussian kernel using weighted squared sum of the distance differences among patients. ResultsA total of 77 individual OARs were assessed. Combined distance from LD-PTV to parotids/submandibular (AUC=.699, p-value=0.0002 for the alternative AUC>.5) and all the OARs (AUC=.731, p-value<0.0001) was a strong predictor of grade ≥2 vs. <2 xerostomia, demonstrating the potential clinical impact of both the parotid and submandibular glands. It was also discovered that dysphagia (grade ≥3 vs. <3) was most strongly predicted by the combined distance of the MD-PTV to the larynx, constrictor muscles and esophagus (AUC=.856, p-value<0.0001). Weight loss (≥5kg vs. <5kg) at 3 months post-RT was predicted by the combined distance from the HD-PTV to all the OARs (AUC=.664, p-value= 0.0015), indicating the existence of high risk PTV locations relative to OARs. A total of 77 individual OARs were assessed. Combined distance from LD-PTV to parotids/submandibular (AUC=.699, p-value=0.0002 for the alternative AUC>.5) and all the OARs (AUC=.731, p-value<0.0001) was a strong predictor of grade ≥2 vs. <2 xerostomia, demonstrating the potential clinical impact of both the parotid and submandibular glands. It was also discovered that dysphagia (grade ≥3 vs. <3) was most strongly predicted by the combined distance of the MD-PTV to the larynx, constrictor muscles and esophagus (AUC=.856, p-value<0.0001). Weight loss (≥5kg vs. <5kg) at 3 months post-RT was predicted by the combined distance from the HD-PTV to all the OARs (AUC=.664, p-value= 0.0015), indicating the existence of high risk PTV locations relative to OARs. ConclusionThe informatics framework combined with data mining tools can facilitate large-scale analysis of toxicity outcomes and is encouraging for the development of a LHS model to reduce the risk of radiation therapy toxicities. Our analysis suggests that the OVH and attention to the distance between PTVs and OARs may be a promising approach to model the risk of developing treatment complications warranting further investigations. The informatics framework combined with data mining tools can facilitate large-scale analysis of toxicity outcomes and is encouraging for the development of a LHS model to reduce the risk of radiation therapy toxicities. Our analysis suggests that the OVH and attention to the distance between PTVs and OARs may be a promising approach to model the risk of developing treatment complications warranting further investigations.