-- WilliamDupont - 04 Mar 2022

Tumor Mutation Burden and Survival in Cancer Patients

This application accompanies our published manuscript revealing that tumor mutation burden is a key independent index of survival, of analogous clinical utility to cancer type, anatomic stage, and histologic grade(1). Survival is affected through a non-linear function of the number of pathogenic short base substitution and insertion-deletion mutations (SM count) as well as copy number alteration (CNA) fraction. Results are based upon survival among 10,844 patients from the Pan-Cancer Atlas who were diagnosed with 32 different cancer types. I have published an app that estimates the hazard ratio for a patient of a given age, cancer type, stage, grade, absolute SM count (or tumor mutation burden (TMB)), and CNA fraction. The application draws the 15-year survival plot for the user-described patient based upon a proportional hazards regression model and Pan-Cancer Atlas reference data. These models use three-knot restricted cubic spline covariates derived from the patient's log SM count, and 4-knot restricted cubic spline covariates derived from her/his CNA fraction. Indicator covariates were entered into these models for cancer-type, stage, and grade. Age at diagnosis was entered into the models as a raw covariate.


Pan-Cancer Atlas pathogenic SM counts are derived from the entire sequenced exome of each primary tumor, rather than from a candidate gene set. A closely related measure is TMB, commonly defined as the rate of nonsynonymous single nucleotide and short insertion-deletion mutations per coding megabase. SM count and TMB are highly correlated in the Pan-Cancer Atlas (r = .98); they are not perfectly correlated due to subtle differences of exome sequenced across contributing TCGA sites. Given some TCGA TMB value, the corresponding SM count can be estimated by multiplying the TMB value by 29.8. The app permits users to enter either SM counts or TMBs. When TMB is entered this approximation of SM count is used to derive hazard ratios and survival curves.

Complete mutation burden data is available on 91% of evaluated patients in the Pan-Cancer Atlas. Hence, our results concerning the effects of SM count and CNA fraction on survival are unlikely to be meaningfully affected by related missing reference data. The estimation of survival curves can also be affected by missing values of age, stage and grade. Survival curves cannot be drawn for select combinations of these variables where all patients with the cancer of interest are missing one of the required values. For some cancers and combinations of variables these missing values make sense (e.g. hematologic cancers do not have a grade). In other TCGA cancers, select data categories are missing without ready explanation. Table 1 shows the variables that may be grossly missing for a given TCGA cancer. When this happens, the Shiny app gives the message “Pan-cancer Atlas does not contain the data needed for this survival curve.” For example, the app will generate a survival curve for a breast cancer patient corresponding to a user-entered age and stage, but fails when grade is also entered because none of the Pan-Cancer Atlas breast cancer patients have a recorded grade.

Table 1
  Cancers that have the indicated prognostic variable for some patients
------------------------ Cancer --------------------- SM count CNA fraction Age Stage Grade
Adrenocortical Carcinoma Yes Yes Yes Yes No
Bladder Cancer Yes Yes Yes Yes Yes
Breast Cancer Yes Yes Yes Yes No
Cervical Squamous Cell Carcinoma Yes Yes Yes Yes Yes
Cholangiocarcinoma Yes Yes Yes Yes Yes
Colorectal Adenocarcinoma Yes Yes Yes Yes Yes
Diffuse Large B-Cell Lymphoma Yes Yes Yes No No
Esophageal Adenocarcinoma Yes Yes Yes Yes Yes
Glioblastoma Multiforme Yes Yes Yes No No
Head and Neck Squamous Cell Carcinoma Yes Yes Yes Yes Yes
Kidney Chromophobe Yes Yes Yes Yes No
Kidney Renal Clear Cell Carcinoma Yes Yes Yes Yes Yes
Kidney Renal Papillary Cell Carcinoma Yes Yes Yes Yes No
Acute Myeloid Leukemia Yes Yes No No No
Brain Lower Grade Glioma Yes Yes Yes No Yes
Liver Hepatocellular Carcinoma Yes Yes Yes Yes Yes
Lung Adenocarcinoma Yes Yes Yes Yes No
Lung Squamous Cell Carcinoma Yes Yes Yes Yes No
Mesothelioma Yes Yes Yes Yes No
Ovarian Serous Cystadenocarcinoma Yes Yes Yes No Yes
Pancreatic Adenocarcinoma Yes Yes Yes Yes Yes
Pheochromocytoma and Paraganglioma Yes Yes Yes No No
Prostate Adenocarcinoma Yes Yes Yes No No
Sarcoma Yes Yes Yes No No
Skin Cutaneous Melanoma Yes Yes Yes Yes No
Stomach Adenocarcinoma Yes Yes Yes Yes Yes
Testicular Germ Cell Tumors Yes Yes Yes Yes No
Thyroid Carcinoma Yes Yes Yes Yes No
Thymoma Yes Yes Yes No No
Uterine Corpus Endometrial Carcinoma Yes Yes Yes No Yes
Uterine Carcinosarcoma Yes Yes Yes No No
Uveal Melanoma Yes Yes No No No
Table 2 gives the proportion of patients with one or more missing covariates in the models evaluated by the Shiny app. This app evaluates seven different models in which survival is regressed against seven different combinations of covariates. The sets of covariates in these models are labeled A through G in Table 2. They are

A: SM count and CNA fraction
B: SM count, CNA fraction and cancer
C: SM count CNA fraction, cancer and age at diagnosis,
D: SM count CNA fraction, cancer and stage,
E: SM count CNA fraction, cancer, age at diagnosis and stage,
F: SM count CNA fraction, cancer, stage and grade, and
G: SM count CNA fraction, cancer, age at diagnosis, stage and grade.
The percent of patients with each cancer who are missing one or more of the covariates in these models is given in Table 2. These percentages vary considerably from cancer to cancer and model to model. For many cancers and models the proportion of patients with the needed covariates is high. For a few, the proportion of patients with the required covariates is low but not zero. For example, only 44% of patients with Glioblastoma Multiforme have recorded values of the TMB covariates and age at diagnosis. The survival plots from Model B for this cancer could be biased if the survival for patients with missing age at diagnosis differs from that of patients whose age is available. See Parl et al.(1) for additional details about these models.

Table 2
  ------------------------------ Model -----------------------------
------------------------ Cancer --------------------- A B C D E F G
Adrenocortical Carcinoma 97.8 97.8 97.8 95.6 95.6 0.0 0.0
Bladder Cancer 98.3 98.3 98.3 97.8 97.8 97.3 97.3
Breast Cancer 91.7 91.7 91.7 90.0 90.0 0.0 0.0
Cervical Squamous Cell Carcinoma 93.3 93.3 93.3 72.1 72.1 66.2 66.2
Cholangiocarcinoma 100.0 100.0 100.0 100.0 100.0 100.0 100.0
Colorectal Adenocarcinoma 87.0 87.0 87.0 84.9 84.9 84.7 84.7
Diffuse Large B-Cell Lymphoma 76.6 76.6 76.6 0.0 0.0 0.0 0.0
Esophageal Adenocarcinoma 100.0 100.0 100.0 87.9 87.9 70.3 70.3
Glioblastoma Multiforme 64.3 64.3 44.0 0.0 0.0 0.0 0.0
Head and Neck Squamous Cell Carcinoma 95.0 95.0 95.0 82.8 82.8 79.7 79.7
Kidney Chromophobe 100.0 100.0 100.0 100.0 100.0 0.0 0.0
Kidney Renal Clear Cell Carcinoma 68.6 68.6 68.6 68.6 68.6 67.3 67.3
Kidney Renal Papillary Cell Carcinoma 96.8 96.8 96.1 87.9 87.1 0.0 0.0
Acute Myeloid Leukemia 95.4 95.4 0.0 0.0 0.0 0.0 0.0
Brain Lower Grade Glioma 98.4 98.4 98.4 0.0 0.0 0.0 0.0
Liver Hepatocellular Carcinoma 94.8 94.8 94.8 89.9 89.9 89.3 89.3
Lung Adenocarcinoma 96.2 96.2 94.2 95.8 93.8 0.0 0.0
Lung Squamous Cell Carcinoma 96.2 96.2 94.8 95.6 94.6 0.0 0.0
Mesothelioma 94.1 94.1 94.1 94.1 94.1 0.0 0.0
Ovarian Serous Cystadenocarcinoma 69.5 69.5 59.5 0.0 0.0 0.0 0.0
Pancreatic Adenocarcinoma 94.5 94.5 94.5 93.4 93.4 91.8 91.8
Pheochromocytoma and Paraganglioma 90.4 90.4 90.4 0.0 0.0 0.0 0.0
Prostate Adenocarcinoma 98.4 98.4 98.4 0.0 0.0 0.0 0.0
Sarcoma 90.6 90.6 90.6 0.0 0.0 0.0 0.0
Skin Cutaneous Melanoma 98.1 98.1 98.1 90.4 90.4 0.0 0.0
Stomach Adenocarcinoma 98.6 98.6 97.8 96.4 95.7 94.2 93.5
Testicular Germ Cell Tumors 95.5 95.5 95.5 58.6 58.6 0.0 0.0
Thyroid Carcinoma 96.4 96.4 96.4 96.0 96.0 0.0 0.0
Thymoma 100.0 100.0 100.0 0.0 0.0 0.0 0.0
Uterine Corpus Endometrial Carcinoma 95.4 95.4 95.1 0.0 0.0 0.0 0.0
Uterine Carcinosarcoma 98.2 98.2 48.2 0.0 0.0 0.0 0.0
Uveal Melanoma 100.0 100.0 0.0 0.0 0.0 0.0 0.0
Reference

1. Jeffrey R. Smith, Fritz F. Parl, William D. Dupont. Mutation Burden Independently Predicts Survival in The Pan-Cancer Atlas. JCO Precision Oncology 2023; Jun;7:e2200571. doi: 10.1200/PO.22.00571.
Topic revision: r14 - 06 Jun 2023, WilliamDupont
 

This site is powered by FoswikiCopyright © 2013-2022 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Vanderbilt Biostatistics Wiki? Send feedback