Biologically based mathematical models which have the potential to predict radiation-induced cancer risks caused by modern radiotherapy protocols allow the risks of second cancers to be estimated and minimized during treatment plan optimization. This task is becoming increasingly important because of the increasing number of younger individuals undergoing radiation therapy and improving survival times after treatment. For example, the 5-y relative survival rate for prostate cancer in the USA has increased from ~67 to almost 98 % and the mean age at diagnosis decreased from 72 to 69 over the past few decades(1, 2)
. Ten-year survival rates have also improved substantially over the same time.
Patients exposed to radiotherapy as children are probably inherently more sensitive to radiation-induced carcinogenesis than adults and have a longer life expectancy. Consequently, radiotherapy-induced second cancers are a particularly important issue for childhood cancer survivors. Relative risks of cancer in exposed children can be very high, on the order of 10–100 after typical radiotherapy treatments(3)
Many retrospective epidemiological studies of second-cancer risks after radiation therapy have been conducted(3–11)
. However, radiotherapy treatment techniques are changing quite rapidly, especially in terms of escalating treatment dose, altered dose fractionation and altered normal-tissue dose distributions such as from intensity-modulated radiation therapy(12–14)
. Radiation-induced second cancers typically develop after a long latency period of a decade or more following exposure(15, 16)
. For these reasons, risks estimated based on decades-old radiotherapy methods cannot generally be directly applied to modern or prospective protocols. This problem can potentially be solved by developing mathematical models which can predict the second-cancer risk of any given radiotherapy protocol tusing target organ dose distributions (dose–volume histograms (DVHs)). Such models can also provide insight into the underlying mechanisms of radiation carcinogenesis and, as argued, represent a useful initial step towards the reduction of radiotherapy-induced second-cancer risks.
Many radiation carcinogenesis models have been proposed and used over the past several decades(17–35)
. Some can be called short-term
models, meaning that they focus on processes such as cell killing, mutagenesis and chromosome aberrations which occur during and shortly after irradiation(30, 34, 36–38)
. They typically provide a detailed dose–response relationship for the selected endpoints, but do not directly address the complexity of those processes that take place before exposure and many years to decades after exposure, up to the time when cancer develops. The short-term models help to understand the effects of dose/fluence, radiation quality, dose rate and dose fractionation.
A very different approach is employed in long-term
models(17, 20, 21, 24, 39)
, which encompass the entire lifetime, e.g. tracking the kinetics of pre-malignant cell clones. They concentrate on background carcinogenesis processes, but generally treat radiation exposure as a simple (e.g. linear as function of dose) modulation of the background rates. The long-term models help to understand the effects of age at exposure, time since exposure and modulation of background cancer rates.
Because radiation-specific effects are treated simplistically by most long-term models, predictions from such models are typically limited to exposure conditions where a simple dose–response relationship holds. In situations where this relationship is more complex, such as high fractionated doses to organs located in proximity to the radiotherapy target volume, current long-term models have limited utility. Conversely, the detailed dose–responses produced by short-term models can be used to estimate cancer risks only by considering the effects of factors such as background risks, age at exposure and time since exposure, which are not directly taken into account by short-term formalisms. A unified approach of integrating short- and long-term methods is needed, where a detailed initial dose–response for pre-malignant cell numbers is produced over a wide range of radiation doses, and changes to the shape of this dose–response during the latency period before the development of cancer are also analysed in detail.