Modelling Radiotherapy Side Effects

Practical Applications for Planning Optimisation

Tiziana Rancati, Claudio Fiorino

• Verlag: CRC Press
• Jahr: 2019
• ISBN:
  9781351983105
• Auflage: 1
• Thema: Physical Sciences
• Thema: Physics

The treatment of a patient with radiation therapy is planned to find the optimal way to treat a tumour while minimizing the dose received by the surrounding normal tissues. In order to better exploit the possibilities of this process, the availability of accurate and quantitative knowledge of the peculiar responses of the different tissues is of paramount importance.

This book provides an invaluable tutorial for radiation oncologists, medical physicists, and dosimetrists involved in the planning optimization phase of treatment. It presents a practical, accessible, and comprehensive summary of the field's current research and knowledge regarding the response of normal tissues to radiation. This is the first comprehensive attempt to do so since the publication of the QUANTEC guidelines in 2010.

Features:

• Addresses the lack of systemization in the field, providing educational materials on predictive models, including methods, tools, and the evaluation of uncertainties
• Collects the combined effects of features, other than dose, in predicting the risk of toxicity in radiation therapy
• Edited by two leading experts in the field

chapter 1

The Importance of the Quality of Data

Wilma Heemsbergen and Marnix Witte

Contents

Introduction

Dimensions of Data Quality

Study Design

Cohort Study

Causality

Prospective and Retrospective Study

Studying Long-Term Effects

Case-Control Study

Defining a Complication of Interest

Toxicity Scales

RTOG/EORTC and LENT/SOMA Scales

CTCAE Scale

Subjective Components in Toxicity Scoring

Patient-Reported Outcome Measures (PROMs)

Quality of Life

Electronic Questionnaires

Measuring a Complication

Quantitative Versus Qualitative Measurements

Unstable Uncertainties in Measurements

Timing of Measurements

Disease (Progression) and Complication Scoring

Cross-Sectional versus Longitudinal Measurements

Factors Associated with Complication Rates

Risk Factors versus Prognostic Factors

Recording and Collection of Data

Quality and Validity

The Quality of Dose and Volume Data

Sources of Uncertainties

Collection of Dose and Volume Data

Organs at Risk (OAR)

Organ Delineation

Hollow OAR

Dose Reconstruction in the OAR

Patient Setup and Anatomical Variations

Changes in Volume and/or Location During Treatment

Fractionation Effects

Practical Implications

General Aspects

Complication Registration Procedures

Compilation of Dose and Volume Data

Data Warehouse

Potential Pitfalls

Big Data: Real-Time Monitoring?

Conclusions

References

Introduction

Models quantifying the relationship between dose-volume parameters and normal tissue complications were developed on a large scale once three-dimensional radiotherapy became available in the 1990s. The QUANTEC project (quantitative analysis of normal tissue effects in the clinic) summarized the available clinical data and models on acute and late radiation-induced complications with the goal of improving patient care by providing useful tools. However, this project also revealed the shortcomings of the available models and underlying concepts and data (Bentzen et al., 2010; Deasy et al., 2010; Marks et al., 2010; Jackson et al., 2010). The quality of the data is paramount in Normal Tissue Complication Probability (NTCP) modeling. It is related to several methodological aspects like the nature of the applied study design, the definition of the clinical endpoints, consistency in toxicity scoring, data collection procedures, and inclusion of all relevant variables. Bad data quality inevitably leads to inaccurate analyses and may lead to biased results, bad models, and false conclusions.

The systematic, data-driven collection and evaluation of complications is essential for knowledge-based treatment optimization in radiotherapy. For the development of adequate NTCP models, we need preferably large datasets of high-quality clinical and dose-volume data that represent all relevant information on the dose distribution in the organ(s) at risk of interest, and the radiation-induced complications of interest (also referred to as side effect, adverse event, or toxicity endpoint). The data collection and the underlying concept therefore have to fulfill a number of essential criteria:

1. a. The data collection is part of an appropriately designed study.
2. b. The data recording and collection procedures are well defined and executed.
3. c. The data are collected in relevant patient populations with sufficient numbers.
4. d. The complication of interest is parametrized in an optimal way.
5. e. The complication is measured with a suitable instrument.
6. f. The complication of interest is measured at relevant time points.
7. g. Other factors that might affect the complication probability are registered.
8. h. The origin of the complication (organ/tissue at risk) is known.
9. i. Relevant dose and volume information can be extracted from the treatment plans.
10. j. The available information allows a sufficiently accurate estimate of the actual absorbed dose in tissues.

These aspects of data quality concerning clinical, dose, and volume parameters will be further discussed in the next sections.

Dimensions of Data Quality

Study Design

Cohort Study

Collecting dose and complication data for a defined group of patients treated with radiotherapy can be considered as an observational cohort study. Which patients have to be included in the cohort of interest is specified with in- and exclusion criteria. Inclusion criteria could, for instance, be based on the diagnosis of the patient and the applied treatment protocol. Exclusion criteria could refer to specific conditions and circumstances, like re-irradiation, multiple tumors, or not speaking the native language. Defining the patient group, one should keep in mind to which patient group the results will be applied, in order to obtain a representative study population.

In a cohort study, typically the association between exposure and the development of health-related events is studied, including the assessment of (potential) risk factors and effect modifiers. Translated into the radiotherapy setting, the exposure is the radiotherapy course and the event is the complication. The key factor for demonstrating causal relationships in an epidemiological observational cohort is a valid comparison group. A cohort lacking a control group is referred to as a descriptive cohort or a case-series, which has a lower level of evidence.

With data from descriptive cohorts, patterns over time and among cases and subgroups can be described without providing evidence for a causal relationship between observed events and factors. Establishing the outcomes of interest in descriptive patient cohorts defined by its disease and treatment is a broadly accepted and applied study design in medicine, since the treating physician is mainly interested in knowledge about the prognosis of treated patients rather than a comparison with an untreated control group.

Causality

Consider the following situation: We observe 35 events of urinary obstruction during ­follow-up in a prostate cancer patient group treated with radiotherapy, and we observe 12 events in a similar patient group without radiotherapy (a valid control group). Now (a) we can estimate the true treatment effect, and (b) we have level II evidence that a causal relationship exists between radiotherapy for prostate cancer and urinary obstruction. As mentioned earlier, in radiotherapy we usually evaluate descriptive cohorts (all exposed) for NTCP modeling. However, if we are able to establish a dose–effect relationship in a descriptive cohort, this is also regarded as a piece of evidence for causal relationships, as described in Austin Bradford’s famous paper (Bradford, 1965), which describes eight criteria to assess causality for an observed association between a cause and an effect: temporality, strength, dose-response, reversibility, consistency, biological plausibility, specificity, and analogy. In studies concerning NTCP modeling, the identified correlation is usually similar to as reported in other studies (analogy); the underlying mechanism of radiation-induced damage may be known (biologic plausibility); for some endpoints, it has been demonstrated that the effect can be reduced by reducing exposure, i.e., lower dose levels (reversibility); and obviously the exposure preceded the observed effect (temporality). We should however keep in mind that without a control group we have no estimate of the number of non-radiation-induced events among the observed events.

Prospective and Retrospective Study

There are two conceptual types of observational cohort studies: prospective and retrospective. A cohort study may also have retrospective and prospective phases. The main essence of a prospective cohort is that a subject is recruited for the cohort prior to developing the complication of interest, according to a predefined set of in-/exclusion criteria. Retrospective studies are considered as having a relatively low level of evidence, but this depends on the specific design of the retrospective study. In a hospital setting, retrospective data collection from medical patient records is often of poor-quality. In case of frequent follow-up visits, limited dropout of patients, and the availability of extensive information in the patient files, a dataset of reasonable quality may be achieved as long as it concerns clinically relevant information that is reported in the patient records in a reasonably consistent and systematic way by the treating physicians. However, obtaining consistent and systematic data on relevant baseline information (especially potential prognostic factors and effect modifiers) remains an issue in such studies.

Studying Long-Term Effects

Radiotherapy can cause long-term effects with a long latency time that are impossible to capture during a standard follow-up period at the outpatient clinics. Well-known examples of such long-term effects are heart failure and secondary tumors. Such effects are mainly studied in retrospectively established cohorts. Such cohorts can be established in existing registrations with a prospective nature, such as national cancer registries, to avoid a selection bias. The endpoint of (secondary) cancer or diagnosis of disease is of such a nature that it can be extracted from medical patient records and/or existing prospective registries in a reliable way. However, obtaining consistent and systematic data on relevant baseline information (i.e., potential prognostic factors and effect modifiers) remains an issue in such studies.

Case-Control Study

For rare complications or complications with a long latency, a case-control study can be an attractive, fast, and effective method to study relevant prognostic factors and dose-effect relationships. The level of evidence is considered lower compared to cohort studies because causality cannot be assessed, and the validity of the study can be seriously affected by selection bias, information bias, confounding, and selective choices for cases and controls. However, carefully designed nested case-control studies within established radiotherapy cohorts can be a very useful tool to generate hypotheses and estimate dose-effect relationships for rare complications. Examples of such nested case-control studies are the quantification of the effect of radiation dose to the heart for developing coronary heart disease (van Nimwegen et al., 2016), and assessing the effects of the delivered radiation dose to the lung and the risk of second primary lung cancer (Grantzau et al., 2014).

Defining a Complication of Interest

Irradiation of healthy tissue may eventually lead to (temporary or chronic) clinical symptoms that can be experienced by the patient, diagnosed by the physician, and/or observed through a function test or blood test. The level of the health damage can be observed as an event (qualitative endpoint: present, not present), as a ranked ordinal outcome (none, mild, moderate, severe), or as a continuous, quantitative outcome (e.g., loss of organ function). Furthermore, an event (present, not present) can be scored as “yes” or “no” regardless the time, or it can be studied in relation to the timing (time-to-event), censoring patients with limited follow-up. A quantitative endpoint can be defined at a fixed time point (e.g., function loss after one year) or it can be followed longitudinally, ...

Inhaltsverzeichnis

• Cover
• Half-Title
• Series
• Title
• Copyright
• Contents
• About the Series
• The International Organization for Medical Physics
• Preface
• Contributors
• Introduction
• CHAPTER 1  ▪  The Importance of the Quality of Data
• CHAPTER 2  ▪  Building a Predictive Model of Toxicity: Methods
• CHAPTER 3  ▪  Potentials and Limits of Phenomenological Models
• CHAPTER 4  ▪  Pelvis: Rectal and Bowel Toxicity
• CHAPTER 5  ▪  Pelvis: Urinary Toxicity
• CHAPTER 6  ▪  Stomach, Duodenum, Liver, and Central Hepatobiliary Tract
• CHAPTER 7  ▪  Central Nervous System (Brain, Brainstem, Spinal Cord), Ears, Ocular Toxicity
• CHAPTER 8  ▪  Head and Neck: Parotids
• CHAPTER 9  ▪  Head and Neck: Larynx and Structures Involved in Swallowing/Nutritional Problems and Dysphonia
• CHAPTER 10  ▪  Thorax: Lungs and Esophagus
• CHAPTER 11  ▪  Heart and Vascular Problems
• CHAPTER 12  ▪  Adverse Effects to the Skin and Subcutaneous Tissue
• CHAPTER 13  ▪  Bone Marrow and Hematological Toxicity
• CHAPTER 14  ▪  Predicting Toxicity in External Radiotherapy: A Critical Summary
• CHAPTER 15  ▪  Data Sharing and Toxicity Modelling: A Vision of the Near Future
• CHAPTER 16  ▪  Quantitative Imaging for Assessing and Predicting Toxicity
• CHAPTER 17  ▪  Beyond DVH: 2D/3D-Based Dose Comparison to Assess Predictors of Toxicity
• CHAPTER 18  ▪  Radiobiological Models in (Automated) Treatment Planning
• CHAPTER 19  ▪  Including Genetic Variables in NTCP Models Where Are We? Where Are We Going?
• Index