Primary aim of this study was to assess changes in basic clinical characteristics, treatment approaches and their impact on survival in a very large cohort of patients followed over a long period of time. Secondly, we analysed patterns of change in the use of radio-(chemo)therapy with or without surgery. For comparison with epidemiological data we included M1 patients but removed them for not compromising the informative value.
In general, there was a clear shift over the observation period regarding fundamental patient characteristics towards higher ages, a higher proportion of adenocarcinoma and towards more advanced tumour stages at diagnosis. Similar tendencies were also visible when analysing the complete epidemiologic data of the Munich Cancer Registry. Stage shift and stage migration is - at least in part - related to an increasing utilisation of improved staging possibilities including Endo-US, MRI and PET-CT [16–20].
The increased rate of adenocarcinoma was already reported by multiple groups world-wide and potentially reflects a completely different pathogenesis [21–27].
However, the shift towards adenocarcinoma is slightly less pronounced when compared to the SEER database (59.2%, n = 16,162; 2004-2008). As opposed to literature [4, 28] we detected a worse OS for AC patients than in patients with SCC, possibly because of the negative patient selection in our cohort. The proportion of synchronous malignancies of 11.9% in our cohort is well in line with published data [29, 30].
In general, the prognosis of oesophageal cancer is dominated by two competing risks: loco-regional relapse and distant metastases. Surgery as well as radiation as single modality approaches yield 5 y survival rates below 10% . With both approaches local control rates are far from being optimal. For this reason surgery, radiotherapy and chemotherapy were combined in a multimodality approach. As expected, in our cohort therapy options shifted towards multimodality. However, only the combination of radiotherapy with chemotherapy seems to impact on overall survival. This has already been shown in the framework of prospective randomized trials [8, 32, 33] and also for patients treated along the given trial protocols but formally outside of the randomized trial . However, no such data have been shown for a non-selected patient cohort. Of special notice in this regard is the fact that most patients in the Munich setting have received mitomycin-C  instead of cisplatin which was used in most of the randomized trials [7, 8, 14, 33]. Interestingly, the addition of chemotherapy only increases local control and was never found to reduce the high rate of loco-regional and distant seeding.
However, for the whole study cohort the improvements in the field of radio-chemotherapy were not prominent enough to significantly increase OS rates. Thus, despite considerable changes in the treatment regimens over the past decades results are still moderate and only a slight improvement in OS was seen in this unselected patient-cohort.
When trying to determine the role of surgery using our patient cohort some limitations have to be considered: The most striking bias is the fact that all patients have been selected to receive radiotherapy based on clinical reasons. Thus, the value of surgery can only be estimated for those patients. In our cohort, the value of surgery for overall survival seems to be limited. This is in accordance with the results of randomized trial showing that the addition of surgery to combined radio-chemotherapy does not increase OS [3, 4, 9]. Despite the fact that the best results regarding survival are achieved in patients with triple modality approaches the impact on a larger cohort is limited. This is related to the fact that the increase in mortality by the addition of surgery counteracts the effects of an increased local control. Thus, no clear contribution of surgery to the outcome is visible in our cohort.
Importantly, radiation dose was related to OS-rates in this large cohort. In a previous randomized trial  no such correlation was documented. However the value of this trial is strongly limited since most of the excess mortality in the higher dose arm occurred early in the treatment course and several protocol violations were documented. Our findings are in accordance with data published by Geh  who has provided evidence for clear dose response relationships for oesophageal cancer based on dose response data compiled from multiple trials.
As already pointed out, the use of a non-selected patient cohort for study purposes is associated with certain shortcomings: It is impossible to control for imbalances due to individual and location specific clinical decisions. In addition, time trends are influenced by general scientific progress but also by centre specific variables. Thus, all interpretations need to be very careful, considering selection bias in this retrospective setting for such a long period of time. The patient characteristics in our unselected cohort are comparable with whole epidemiologic data set of the Munich Cancer Registry (MCR) and published trials. Thus the resulting conclusions are substantiated to some degree. Up to now only very few other trials have tried to approach the value of treatment approaches in oesophageal cancer using population-based data. In this regard a US group supported by the National Institutes of Health used the instrument of a survey to collect demographic data on patient and information on surgical approach for oesophageal carcinoma across the whole country . The evaluation showed that there is a substantial heterogeneity in surgery strategies and emphasized the need of controlled trials to determine best practices. Another study from the US queried the SEER database to prove the benefit of neoadjuvant RT on survival for patients undergoing definitive surgery . Also by SEER-query Chang et al.  found no difference in survival and response between AC and SCC across any of the major treatment modalities.
An investigation on trends in treatment and factors influencing treatment receipt and survival were sourced from the Irish National Cancer Registry  and showed decreased use of surgery, especially in older patients and a considerable difference between the survival observed at population level and in randomised controlled trials.
The MCR and the SEER registry are population-based databases that represent an unselected group of patients without consistently recorded medical or course information. Therefore only statements concerning epidemiology and outcomes in actual clinical practice outside of the controlled setting of research protocols can be obtained. Although course and recurrence information is poor in our data and also in the databases, overall survival is a good surrogate for disease recurrence, because the outcomes with oesophageal cancer are poor and recurrences in this disease are rarely salvaged. Cause of death and mortality data are available and generally corresponds to disease recurrence.
Despite the fact that the results of several large randomized trials are available the general progress in oesophageal cancer is limited. In the future, several crucial aspects are of importance: A key problem is the fact that a considerable number of patients are not suitable for aggressive approaches . Hence, major efforts should be placed on the development of tools for accurate patient selection according to the individual risk situation and estimated prognosis and reducing therapy associated morbidity and mortality [22, 25, 31, 40, 41]. In this regard, important contributions may come from similar disorders including head and neck cancer or lung cancer in which the development of modern imaging approaches as well as biological stratification approaches already dominate the research horizon [42, 43].
In parallel, the fate of patients with oesophageal cancer is largely influenced by early dissemination, thus it is of key importance to test the value of an integration of targeted drugs with proven activity in either SCC or AC into putative new treatment protocols [44–46]. Similarly, new radiation techniques suitable to reduce lung toxicity or increase target volume conformity [47–50] will have a clear role in optimizing the outcomes in oesophageal cancer.