Refining Evaluation Methodology on TNM Stage System: Assessment on HPV-Related Oropharyngeal Cancer

Research Article

Austin Biom and Biostat. 2015;2(1): 1014.

Refining Evaluation Methodology on TNM Stage System: Assessment on HPV-Related Oropharyngeal Cancer

Xu W1,2*, Shen XW1, Su J¹, O’Sullivan B3,4 and Huang SH³

¹Department of Biostatistics, Princess Margaret Cancer Center, Canada

²Dalla Lana School of Public Health, University of Toronto, Canada

³Department of Radiation Oncology, Princess Margaret Cancer Centre, Canada

4Ontario Cancer Institute, University Health Network, Canada

*Corresponding author: Xu W, Department of Biostatistics, Princess Margaret Cancer Center, Dalla Lana School of Public Health, University of Toronto, 10-512, 610 University Ave, Toronto, ON M5G 2M9, Canada.

Received: March 16, 2015; Accepted: March 27, 2015; Published: April 08, 2015


TNM stage is important in treatment decision-making and outcome predicting. The existing Oropharyngeal Cancer (OPC) TNM stages have not made distinction of the two sub sites of HPV+ and HPV- diseases. Besides that, the current TNM stage grouping is generally derived based on literature review and clinical understanding of the disease processes. It is important to have quantitative assessment on the prognostic ability and stability of the TNM stage on HPV-related OPC patients. The current stage evaluation criteria use non-parametric methods and assess the stage performance on limited time points. Even though the current stage systems were mostly developed based on retrospective data, these evaluation criteria don’t consider the important clinical confounders. We developed novel criteria to assess performance of the TNM stage grouping schemes based on parametric modeling adjusting on important clinical factors. These criteria evaluate the TNM stage grouping scheme in five different measures: hazard consistency, hazard discrimination, explained variation, likelihood difference, and balance. The novel criteria were applied to evaluate newly developed TNM stage grouping schemes on HPV+ OPC patients and a few existing TNM stage grouping schemes. The new HPV+ OPC TNM stage outperforms all existing schemes on prognosis and stability.

Keywords: TNM stages; Evaluation criteria; Parametric model; Head and Neck Cancer; Prognosis

Introduction and Background

T (extent of the primary tumor), N (absence or presence and extent of regional lymph node metastasis) and M (absence or presence of distant metastasis) are three components to describe the anatomical tumor extent. In clinic, the TNM categories are combined to classify patients into stage groups with similar survival performance. TNM staging system plays an important role in treatment decision-making and outcome predicting. The most widely used TNM stage grouping scheme is proposed by the Union of International Cancer Control (UICC) and American Joint Committee on Cancer (AJCC) [1].

The TNM staging system is revised periodically to reflect changes in outcomes (generally based on overall survival) as a result of better defining anatomic disease extent from new imaging modalities, improved therapeutic efficacy, and increasing knowledge about tumor biology. The performance of any new staging on outcome prediction should be evaluated objectively.

The commonly used evaluation criteria for TNM stage grouping is proposed by Groome et al. [2] including hazard consistency, hazard discrimination, outcome prediction, and balance. However, most of the current TNM stage systems were developed based on retrospective data; these evaluation criteria don’t consider the potential clinical confounders, especially treatment, that may strongly affect the survival outcomes. Ignoring the important clinical factors in the development and evaluation of stage grouping systems may lower the accuracy of distinguishing risk groups [3]. Furthermore, the existing criteria use non-parametric methods to evaluation hazard consistency and hazard discrimination based on limited time points (monthly rate over 5 years). The use of complete time to event information is needed for the performance evaluation to improve efficiency and accuracy.

We introduced new criteria to assess performance of the TNM stage grouping schemes based on parametric modeling adjusting for important clinical factors, using HPV-driven (HPV+) Oropharyngeal Cancer (OPC), as an example. HPV+ OPC, a fast emerging disease entity, is different from traditional smoking/alcohol related OPC in many ways. Many studies have shown that HPV+ OPC patients tend to have better survival and lower recurrence rates compared to HPV- OPC patients [4-6]. Currently for OPC, UICC/AJCC stage grouping scheme has not made distinction of these two diseases and is derived from the historical data that are comprised of mainly HPVOPC patients. Other existing stage grouping schemes were proposed to improve the prognostic ability of UICC/AJCC, but still paying little attention on the distinction between HPV+ and HPV- OPC patients [7-12]. Takes et al. [13] discussed about the shortcomings of the current TNM classification system (UICC/AJCC) within head and neck cancer and suggested to improve and expand current TNM system based on tumor, patient and environment-related factors.

In the present study, we firstly applied Recursive Partitioning Analysis (RPA) model in non-metastatic HPV+ OPC patients treated in our institution to derive a new stage grouping scheme using the same TNM classification. Then, we applied both the current and the new criteria on the newly developed TNM stage grouping scheme and a few existing TNM stage grouping schemes including UICC/ AJCC 7th edition [14], TANIS, Synderman, Hart, Berg, Kiricuta, and Hall [7-12]. We also evaluated the performance of the new criteria using bootstrap algorithm and multiple imputations as validation.


Refined TNM stage grouping for HPV+ OPC

After institutional Ethic Board approval, a retrospective review was conducted for 573 newly diagnosed p16 confirmed HPV+ OPC treated in our institution during January 2000 to December 2010. Patients whose p16 status was unknown or who had metastatic disease at presentation were excluded. T- and N-classification was based on UICC/AJCC TNM 7th edition for OPC [1,15]. Median follow up was 3.82 years.

To derive new TNM stage groupings for HPV+ OPC, RPA method was explored using Overall Survival (OS) as outcome endpoint. A new stage scheme was derived using ordinal T- (T1/T2/ T3/T4) and N- (N0/N1/N2a/N2b/N2c/N3) categories objectively. The RPA algorithm is based on the optimized binary partition of T or N categories. It results in subgroups with relatively homogeneous survival performance. The derived stages were termed RPA-stages [16]. The performance of the newly derived TNM stage grouping schemes is evaluated against the following existing TNM stage grouping schemes which were developed for head and neck cancers:

UICC/AJCC 7th edition: It was derived based on clinical understanding of the disease processes, with an interest in maintaining the same scheme across as many head and neck cancer sites as possible. It has been widely used in most head and neck cancer sites now.

The TANIS scheme: It was proposed by Jones et al. [7] as an easyto- remember approach. It was developed based on the observations that, in head and neck cancers, the corresponding T and N categories yield similar prognostic effects and that those effects were linear, thereby allowing the T and N integers to be summed and generating the TANIS-7 scheme. The investigators showed that three TANIS groupings (TANIS-3): 1-3, 4, and 5-7 discriminated better than UICC/AJCC in their study population.

The snyderman scheme: Snyderman and Wagner [8] derived another staging scheme from the TANIS scheme by identifying optimal groupings based on visual examination of the TANIS survival curves. They derived the scheme based on a population consisted of 186 cases of oral cavity carcinoma treated with primary surgery and they used disease-free survival as primary outcome of interest.

Hart scheme: Hart et al. [9] developed a scheme from an analysis of disease-specific survival in 640 OPC cases. A stepwise backward elimination model was applied to determine stage groupings.

Kiricuta scheme: Kiricuta [10] later introduced a scheme with some modification to the one proposed by Hart.

Berg scheme: The scheme proposed by Berg [11] was based on observed survival rates and compared with UICC/AJCC by assessing the degree of discrimination among survival curves.

Hall scheme: Groome and Hall [12] proposed a scheme, which further separated N2 category into N2a and N2bc in the design of the stage grouping.

We compared the TNM stage group schemes of UICC/AJCC 7th edition, TANIS, Synderman, Hart, Kiricuta, Berg, Hall, and the newly derived stage grouping schemes (RPA-staging scheme). An overview of the T and N category combinations and distributions within each of these schemes is provided on Figure 1. We refer to a subgroup as the patients with a specific combination of T and N categories and group as the combination of subgroups within a stage grouping scheme. The terms classification, scheme and system are used interchangeably.