Advancing our quantitative understanding of radiotherapy normal tissue morbidity

This and the preceding issue of Acta Oncologica contain several papers adding to our understanding, and ability to predict, normal tissue morbidity in individual patients treated with radiotherapy (RT) [1–7]. The overall goal of normal tissue complication probability (NTCP) modeling is to establish predictive models – or sets of models – of treatment-limiting toxicities, that have general validity and good performance for a wide range of patient, disease, and treatment characteristics. That is, the models should have ‘generalizability’ and ‘validity’. Two of the papers presented in this issue use large institutional databases to test/validate/invalidate dose-volume predictive factors reported elsewhere [1,2]. Short of actual data-pooling, this is an appropriate, and important, methodology. However, due to the (understandable) limited variability of local treatment characteristics – and to some extent patient and disease characteristics as well – variations in NTCP modeling results are not only possible but indeed to be expected. Tucker et al. test for effects of heart dose on radiation pneumonitis and find little evidence for significance when taking the mean dose to the normal lungs as the ‘baseline’ model [1]. This is in contrast to results from the Washington University in St. Louis published previously in this journal [8] – as well as a confirmatory test in an independent cohort from the same institution [9] – showing that heart doses were not only important, but they were as important in multivariate modeling as lung doses. Also other pre-clinical and clinical data, especially from the Groningen group, indicate that radiation damage to the heart can lead to lung damage, and vice versa [10–13]. Obviously, we still do not fully understand the source of the difference in these results.

The picture is clearer when it comes to another important normal tissue in RT, the parotid glands, and the dose limits of this organ aiming at reducing salivary dysfunction. In this issue, Beetz and co-workers present a very nice analysis showing that the QUANTEC guidelines [14] appear to be useful in their local patient cohort [2]. Going beyond that, they demonstrate a striking impact of age that should be tested by other groups. Their study used patient-reported outcomes, whereas the QUANTEC guidelines were developed based on salivary flow measurements; the consistency of results is therefore all the more impressive. It is important to re-emphasize, however, that salivary function is likely even better preserved when the spared parotid gland mean dose can be kept well below the 25 or 20 Gy threshold (say, at 10 or 15 Gy) [14].

Incorporation of clinical factors is the key element of the innovative approach taken by Appelt and collaborators [3]. The paper establishes a new and powerful tool: using the results of meta-analyses as a guide to adding clinical factors, and their odds ratios, to dose-volume predictive models. Starting from a blank canvas for every predictive model now makes little sense. The previous encouragement to adopt a ‘data pooling culture’ [15] is therefore echoed here, as this appears to be essential for further progress in this field. While creation of web-based databases enabling data-mining should be our overall aim [16,17], data such as dose-volume histograms can already now easily and effectively be shared using the features of online appendices/supplementary materials that most journals in our field offer, including Acta Oncologica [18, see appendix E2 therein].

The QUANTEC effort represented a step forward for the NTCP modeling field [19–21], but many of the analyses were somewhere between qualitative and quantitative due to the lack of access to original, high-quality data, and the lack of reporting standards. New follow-up studies underline that we still have many things to learn about the specific tissues responsible for complications from RT [22–30]. The paper by Cella et al. in this issue is another example: the high-dose (30 Gy) volume in the left lung appears to be a better predictor of toxicity than the high-dose volume in both lungs together, for patients receiving radio-chemotherapy for Hodgkins lymphoma [4]. The authors raise the question of whether the left lung itself is the key factor, or does this reflect a connection to heart irradiation? The recent paper by Johansen and colleagues [5] examines the under-studied issue of the cause of arm and shoulder pain in breast cancer patients following surgery and RT. The authors contoured the shoulder joint for each patient, and showed that the corresponding V15 had the strongest association with post-therapy pain and arm stiffness. Again, this raises the question of whether there is a more specific anatomic structure responsible for the symptoms. New methodologies that map a cohort of patient dose distributions to a given ‘reference patient’, whether to an idealized geometry [31,32] or a particular (anonymized) patient [33], to look for dose versus outcome correlation hot spots, may prove very valuable to sort out these questions of anatomic specificity. Experience has taught us that NTCP modeling is Sisyphean, in the sense that every new answer creates new questions. Unlike Sisyphus, however, we are making considerable progress.