Validating the PredicTR treatment response classifier for oropharyngeal cancer (PredicTR 2)

Lead Research Organisation: University of Birmingham
Department Name: Institute of Cancer and Genomic Sciences


Oropharyngeal cancer (OPC) - throat cancer - is one of the most rapidly rising cancers in the West, with ~115,000 new cases per year. This is due mainly to increased human papillomavirus (HPV)-related incidence. Standard treatment for OPC involves chemotherapy and radiotherapy (CRT). Adding surgery may improve outcomes for those patients most at risk of recurrence, but may also result in increased complications and poorer function - including swallowing/eating and speech - as well as increased cost.

Currently, there are no markers to guide treatment selection, therefore the decision whether or not to operate is made according to clinician preference and patient choice.

We have developed a 'biomarker classifier' that can predict which OPC patients will receive most benefit from additional surgery, and so could guide treatment decision-making. This 'biomarker classifier' comprises 4 factors: (i) detection of HPV DNA within tumour cells, (ii) the presence (or absence) of a subset of immune cells called lymphocytes within the tumour microenvironment (tumour infiltrating lymphocytes; TILs), (iii) the expression of a protein called Survivin, which helps to protect cells from programmed cell death, and (iv) p16, a protein whose expression within tumour cells is linked to HPV infection.

The classifier categorises patients into 'low-risk' or 'high-risk' subgroups, using an algorithm (formula) to interpret results from the combined analysis of these 4 factors. Low-risk patients do not benefit from additional surgery and so can be treated with CRT alone. High-risk patients demonstrate ~20% improvement in overall survival if surgery is added to CRT.

In this proposal, we want to complete clinical validation of our biomarker classifier and make the preparations necessary for converting it into a standardised assay which can be adopted into clinical practice. To achieve this we will undertake the following steps:

1. During development, we used a Survivin antibody stain that was not clinical grade. We will test a clinical grade (CE-marked) Survivin antibody to show that it is as effective as the one used previously.

2. We have developed an artificial intelligence-based automated method for scoring TILs in mouth cancer. We want to develop it to score TILs in OPC, and test it to see if it is as good as, or even better than, manual scoring by pathologists.

3. Confirm that the assays are reproducible when tested under clinical conditions, i.e. the same results are obtained when testing of the above 4 markers is carried out on whole tissue sections, at different times in the same laboratory, or in different laboratories, and independently scored by different pathologists.

4. Verify the predictive value of the classifier by testing it on a new group of OPC patients and using whole tissue sections (the original study used very small tissue cores ~1mm in diameter). Samples will be stained and scored for the 4 markers in 6 different NHS labs. The performance of the classifier will be evaluated by comparing the overall survival of patients who received the best treatment option as predicted by the classifier with those who did not, in particular, high-risk patients treated with surgery versus no surgery. The aim is to reproduce the 20% difference in 3-year overall survival for high-risk patients receiving optimal treatment (surgery + CRT).

If the predictive nature of the biomarker classifier is confirmed, our next step would be to use it to determine treatment as part of a clinical trial to provide definitive proof of its efficacy, and to seek a partner to commercialise it so that it can be made available to clinicians and patients.

Technical Summary

We have undertaken preclinical development and internal validation on a retrospective cohort of our PredicTR biomarker classifier, comprising p16, HPV ISH, Survivin and Tumour Infiltrating Lymphocyte score (TILs). This classifier strongly PREDICTS which oropharyngeal cancer (OPC) patients will benefit from adding surgery to current standard of care chemo-radiotherapy (20% improvement in 3-year overall survival (OS) for high risk patients), so could be used to decide treatment.

To progress validation along the NCI-Translational Research Working Group Developmental Pathway, and prepare for a definitive clinical trial, we aim to:

1. Optimise and implement the CE-marked Survivin reagent into our classifier. We will demonstrate equivalence with the Research Use Only reagent used during classifier development, by staining 200 samples from the previously tested cohort, in a bridging study.

2. Assess reproducibility and prognostic performance of an automated algorithm for scoring TILs in OPC. We developed an AI-based algorithm for scoring TILs on digitised images of H&E stained slides in ORAL cancer. We will optimise, then compare its ability to predict OS versus scoring by pathologists on 96 OPC samples.

3. Test assay intra- and inter-laboratory reproducibility for the 4 biomarkers, using CE-marked reagents, on whole tissue sections in routine NHS clinical laboratories. In ring experiments, 25 randomised reference cases will be stained blind, in duplicate, in 6 separate NHS labs, and scored by trained pathologists blinded to outcome. If poorly reproducible, a qualitative study and further experiments will be done to address causes.

4. Validate the biomarker classifier in a prospective, independent external cohort under routine clinical conditions. 504 OPC samples, ALREADY collected PROSPECTIVELY within the Head and Neck 5000 cohort, will be stained and scored blind at 6 NHS laboratories. Classifier performance and prediction accuracy will be evaluated.

Planned Impact

This research proposal seeks to validate a predictive treatment response classifier for oropharyngeal cancer (OPC) which identifies those patients who will receive most benefit from adding surgery to standard-of-care chemoradiotherapy. It has the potential to impact on clinicians, pathologists, the NHS, the commercial sector, the public and most importantly patients.

IMPACT ON PATIENTS: For the first time, patients with OPC will be able to receive treatment recommendations based on the biology of their disease (personalised therapy). The survival benefit for high-risk patients receiving surgery in addition to chemoradiotherapy is an additional 20% absolute survival benefit (63% vs. 42.5% 3 year overall survival), which, if proven, would be considerably more than any recent development in the treatment of head and neck cancer. OPC patients and their families would also gain from the social, psychological and economic benefits of their improved survival. In addition, patients would benefit from avoiding the pain, potential complications and poor outcomes of unnecessary or ineffective treatment.

Currently, there is uncertainty regarding the beneficial role of adding surgery in the treatment of OPC. Our PredICTR classifier will change clinical guidelines making them more evidence based, and enabling them to specify treatment for patient sub-groups, for example surgery plus CRT for high-risk patients.

Pathologists would be able to provide a risk-score based on 4 easily applied relatively inexpensive tests. For the first time, clinicians would then be able to recommend treatment based on the likelihood of patients responding to that treatment. This will be a major advance from the current situation, where treatment decisions are made based in part on clinician preference and patient choice.

IMPACT ON NHS: By ensuring that the treatment option delivered to each OPC patient is the one most likely to be successful, both efficacy and hence cost effectiveness of therapies delivered by the NHS (currently ~£15,000 per course) could increase considerably resulting in significant savings to the NHS. Furthermore, the treatment of patients who fail initial therapy is extremely expensive (~£30K per course) and of limited efficacy (improved survival of 2-3 months on average). Therefore, an improvement in the cure rates of patients with OPC would result in significant savings on palliative therapies. Combined, the effects of these improvements could result in considerable savings to the NHS -between £3.5-4.5 million per year.

IMPACT ON INDUSTRY: Our future commercialisation partners are likely to be industry, who would benefit financially from selling the classifier to clinicians. For broad clinical application, we also envisage future commercial development of an online website and app risk calculator.

IMPACT ON ACADEMIC INSTITUTIONS and CHARITIES: The University of Birmingham, and Cancer Research UK (through its Commercial Partnerships - CRUK-CP) would benefit from income as a result of licensing or selling the technology to industry. This indirectly benefits society through increased revenues for spending on research by these organisations.

IMPACT ON THE GENERAL PUBLIC: The impact of significant cost savings and greater efficiency by the NHS could translate into improved/increased health services delivered to the general public, with better value-for-money for the tax-payer. Wider economic benefits could also result from more patients (who are usually within working age) surviving and returning to work.

Additional benefit will come from public involvement in our engagement programmes, which will increase awareness of OPC and of the importance of research in healthcare.
Description University of Bristol - HN5000 
Organisation University of Bristol
Country United Kingdom 
Sector Academic/University 
PI Contribution InHANSE contributed translational, clinical, and biological expertise, intellectual input, access to samples and digitised pathology and data associated with them as well as protected intellectual property.
Collaborator Contribution Our collaboration added pathology expertise and assay development to the project. Out collaborators provided access to samples and associated data, intellectual input in the data analysis and interpretation.
Impact Not yet resulted in any outputs. This is a multidisciplinary project with expertise in clinical pathology, cellular and molecular biomedical sciences, bioinformatics, biostatistics, and computer sciences.
Start Year 2019
Description University of Warwick - PredictR2 TASIL automated scoring 
Organisation University of Warwick
Department Department of Computer Science
Country United Kingdom 
Sector Academic/University 
PI Contribution InHANSE contributed translational, clinical and biological expertise, intellectual input, access to samples and digitised pathology slides and data associated with them as well protected intelectual property.
Collaborator Contribution Professor Rajpoot's lab has developed a method to score TILS on digitised images of stained slides and a novel AI-based deep learning algorithm Machine learning and digital pathology expertise. Intelectual input and access to staff expertise.
Impact This collaboration has not resulted in any outputs yet. This is a multidisciplinary project involving bioinformatics, Artificial intelligence - machine deep learning expertise, clinical pathology, clinical medicine, translational medical sciences, and biostatistics.
Start Year 2020
Description Presentation at the 3rd International Symposium on Tumor-Host interaction in Head and Neck Cancer 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach International
Primary Audience Professional Practitioners
Results and Impact Presentation title: Biomarkers for treatment selection in oropharyngeal cancer
Year(s) Of Engagement Activity 2022