1) BACKGROUND - sharedocs.ca · Web viewRetinoblastoma is an aggressive childhood cancer of the eye, affecting ~8,000 children annually (approximately 1/17,000 live births)1. Retinoblastoma

Retrospective Retinoblastoma Activity Index Validation (RRAIV): A Digital Solution to Standardize Outcome Measures in Clinical Trials for Intraocular Retinoblastoma.REB# 1000064840Study ProtocolVersion 1November 15, 2019Principal Investigator (PI):Dr. Brenda Gallie, MD, [email protected], Extension: 203287

Co-Investigator:Furqan Shaikh, MD, MSc, [email protected], Extension: 20886

Research Team:Mei Chen, Clinical Project Coordinator MASc; [email protected] Soliman, Research Fellow, MD, [email protected] Gelkopf, Research Student, BSc; [email protected] Janušonis, Research Student; [email protected] Coordinator:Kaitlyn Flegg, [email protected], Extension: 203287Study Site:The Hospital for Sick Children

Funding Sources:Pitblado GrantGarron Family Cancer Centre

mailto:[email protected]







RRAIVREB # 1000064840 PI: Dr. Brenda Gallie

1) BACKGROUND

Retrospective RAI ValidationVersion date: 15/11/2019 Page 2 of 9


Retinoblastoma is an aggressive childhood cancer of the eye, affecting ~8,000 children annually (approximately 1/17,000 live births)1. Retinoblastoma is typically caused by a biallelic mutation in the RB1 gene, with approximately half of all retinoblastoma patients carrying this mutation in their constitutional cells1.

Despite the global nature of this condition, retinoblastoma centres around the world largely operate independently from one another, with few collaborative research trials and data-sharing. These are partially due to the variations in staging systems, treatments, and outcome measurements. In other pediatric solid cancers, standardized measures of tumour response, such as RECIST2 indicate the primary outcome in trials of novel agents. However, RECIST cannot be applied to retinoblastoma due to the uniqueness and difference in scale of retinoblastoma to other types of cancers. The retinoblastoma tumour can be viewed directly through the dilated pupil of the patient, examined under anesthesia (EUA). During this time, clinical staff take sectional pictures of the tumour and hand-draw retinal diagrams to encapsulate the key clinical features of the disease, see figure 1.0.

Figure 1. The hand drawn diagram in the middle cumulatively incapsulates the disease status, clinically interpreted from the tumour snapshots taken during the exam under anesthesia; yellow represents active tumours while green represents vitreous seeds.



The only widely accepted measure of retinoblastoma tumor response is the salvage of an eye without use of external beam radiation3. While eye salvage is a meaningful outcome, it is affected by variables not directly related to the treatment (parental preference, institutional culture, visual potential of either eye, etc.) and does not allow an objective comparison of different treatments. Therefore, there is an urgent need for the development of a standardized response assessment measure, analogous to RECIST but tailored specifically to the unique complexities and features of retinoblastoma.

We propose a novel method to quantify tumour severity by taking the total area of the yellow and green pixels in the hand-drawn images – coined the Retinoblastoma Activity Index (RAI). This will be a clean and simple computer algorithm that will automatically read the labeled drawing and output a number from 0 to 100, indicating the percent of area covered by the yellow tumour relative to the total area of the retina.

Since 2003, these retinal drawings have been stored on an electronic system called the electronic Cancer Care for Retinoblastoma (eCCRB). We propose to use these retrospective drawings in the validation of the RAI severity scoring method.

We will assess if the RAI (based on objective measures, retinal drawings, available images, and clinical expertise of the circle of care in the formulation of the drawings) correlates to clinic impression of treatment outcomes, and final outcome of the eye and life. A secondary purpose is to determine whether if the RAI is a reliable tool for assisting physicians with the decision of whether or not to enucleate.

Having a standardized retinoblastoma response criterion will allow comparisons of different treatment outcomes. This will accelerate the rate of development of new treatments and technologies for treating retinoblastoma. With a RAI score calculated at every patient encounter, clinicians and researchers will have a graphical representation of the tumour activity across time. They would be able to trend the tumour activity as with other biological feature data such as heart rate or blood pressure. This will be key data to share with families in order to help them visually understand their child’s disease status. Lastly, this concept of quantifying tumour status, if validated, has the potential to be applied to solid tumours beyond retinoblastoma.

2) STUDY PURPOSE AND OBJECTIVES

The primary objective of this study is to develop an algorithm that will produce the most disease representative RAI score that predicts eye outcomes.

3) HYPOTHESES

The retinoblastoma activity index score encapsulates the disease status, and it is correlated with treatment outcomes.



4) STUDY DESIGN AND METHODOLOGY

4.1) General Study Design

This study is a retrospective review of eCCRB digital retinal drawings in order to create and validate an effective algorithm which would make up the retinoblastoma activity index.

Retinal drawings and treatment outcomes will be retrieved from files in eCCRB. Missing information will be attempted to be filled in with data from hospital records. All identifying patient information will be removed from the main database and stored in a separate key excel document, on a password protected computer, in a locked office.

Computationally, an image processing script will be developed using python as the programming language. It will be used to quantify tumour area via the total area of the tumour (marked yellow) as well as modifying factors (see data collection sheet A).

We will treat each eye of the patient as its own entity, due to the fact that clinical decision of each eye is considered separately according to its independent disease presentation.

4.2) Total Number of Sites and Participants This study will only be conducted on retrospective data from patients treated at SickKids, entered into eCCRB. There are 595 patient records in the eCCRB database; we will be collecting all data that fit within our inclusion criteria. Approximately 60% of patient present with bilateral disease.

4.3) Inclusion Criteria

- SickKids patients that have been treated for retinoblastoma and are included in the eCCRB database, diagnosed since 2003-01-01 - The eye is included if it had a minimum of 4 completed high quality retinal drawings available in eCCRB (one of which must be from staging examination under anesthesia); high quality is defined as a completed retinal drawing

4.4) Exclusion Criteria

- Patients who received cancer staging outside of SickKids – excluded due to treatment prior to arrival at SickKids.- Patients added to eCCRB retrospectively, excluded due to variations in illustration of retinal drawings- Patients who have been bilaterally enucleated before their 4th retinal drawing- Patients who have extraocular disease

4.5) Study Procedures Retrospective RAI ValidationVersion date: 15/11/2019 Page 5 of 9


Clinical data will be retrieved from the eCCRB platform, recorded in a password protected excel document, stored on a password locked computer in a secure SickKids office. The data will be de-identified on collection; a key will be created to link the patient ID to the patients’ hospital number – this key will be kept separately from the data collection sheet (see appendix A).

The image processing script will be built with python as the programming language. The script will read the retinal images and quantify the amount of yellow area (indicating tumour area) on the drawing.

Modifying variables to be collected from eCCRB: - Whether tumour is touching or covering the optic nerve (this is a common route of metastasis) - Whether tumour is touching or covering the macula (indicative of low vision potential)- Bilateral or unilateral of disease

Outcome variables:- Presence of metastatic disease- Salvage of other eye (# of eyes left in patient)- Whether eye is saved (success)

The RAI will be calculated by optimizing the variables that will predictably correlate it to the outcome variables (the saving of the eye).

4.6) Date Range

This study will include charts ranging from 01-01-2003 to 01-11-2019.

4.7) Participant Selection

Records will be identified by a member of the research team. Records will be drawn from eCCRB by a member of the study team who is listed on the REB application.

4.8) Data Sources



This study will use retinal drawings from eCCRB, both SickKids exclusive databases for retinoblastoma patients. No tissue samples or lab values will be included in the data.

Please see data collection sheet for more information (Appendix A).

4.9) Primary and Secondary Endpoints The primary objective of the study is to determine the optimal weighting to assign each modifying variable (conclude with a representative RAI equation), determine the association between the RAI score and patient outcome, and determine a predictive threshold to make a clinical suggestion on whether or not to enucleate the eye.

The primary endpoint will be the algorithm used to calculate the RAI score. The development of this tool will be done using python, leveraging python libraries: TensorFlow, OpenCV, Numpy, and Scikit-Learn.

The secondary endpoints will to determine the RAI correlations with patient outcomes. This will be completed via statistic tools on python and SPSS.

4.10) Statistical Methods/Data Analysis

Tumour area calculation, modifying variables and eye outcomes will be summarized using descriptive statistics (percentages and categorical variables). We will tune the weightings assigned to each modifier and optimize the RAI score to output a predictive cluster weighting for the outcome variables. Clustering techniques (i.e. K-means) will be used to assess and optimize the weightings assigned to the modifiers.

5) DATA PRIVACY

Proper safeguarding techniques to protect confidentiality of the data will be employed; data collected from eCCRB will be protected and safeguarded as per SickKids policy:- Issue a unique study ID code for each participant. - Protect identifiable information (i.e., participant name) in a code-breaking log, which will be stored separately from study data.- Limit access to identifying information.



- Store all data using a two-lock system.- Store all data for 7 years from last publication.- Securely destroy all data: - Electronic records will be destroyed by contacting the SickKids help desk- Paper records will be disposed in SickKids confidential bins.- Old CDs, DVDs, videos, USB keys, external hard drives and other technology will be sent to the repair center for destruction

REFERENCES

1) Dimaras H, Corson TW, Cobrinik D, et al: Retinoblastoma. Nat Rev Dis Primers 1:15021, 20152) Eisenhauer EA, Therasse P, Bogaerts J, et al: New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer 45:228-47, 20093) Chan HS, Deboer G, Thiessen JJ, et al: Combining cyclosporin with chemotherapy controls intraocular retinoblastoma without requiring radiation. Clin Cancer Res 2:1499-508, 1996



Appendix A

Data collection sheet

Participant ID

Bilateral/ Unilateral

Study Eye [OD/OS]

Encounter Date [yy/mm/dd]

Tumour touching/covering optic nerve

Tumour touching/covering macula

Salvage [Y/N]

Metastatic disease [Y/N]

Notes

Master Linking Log (Confidential - kept separately from data collection sheet)

Study ID Patient MRN #


Documents

1) BACKGROUND - sharedocs.ca · Web viewRetinoblastoma is an aggressive childhood cancer of the eye, affecting ~8,000 children annually (approximately 1/17,000 live births)1. Retinoblastoma