DÄ internationalArchive27-28/2014Patients With Cancer of Unknown Primary

Original article

Patients With Cancer of Unknown Primary

A retrospective analysis of 223 patients with adenocarcinoma or undifferentiated carcinoma

Dtsch Arztebl Int 2014; 111(27-28): 481-7. DOI: 10.3238/arztebl.2014.0481

Löffler, H; Puthenparambil, J; Hielscher, T; Neben, K; Krämer, A

Background: Cancer of unknown primary (CUP) now accounts for 2–3% of all fatal cases of cancer in Germany. Histologically, most of these tumors are either adenocarcinoma or undifferentiated carcinoma. Scant data on their clinical features and prognosis are now available, and the published survival times are highly variable. In this article, we document and analyze our own experience with CUP to date.

Methods: We took all 223 patients with CUP (adenocarcinoma or undifferentiated carcinoma) whom we saw in our CUP clinic from 2006 to 2010 as an unaltered sample for retrospective analysis of clinical data and overall survival. We performed the analysis with Kaplan-Meier plotting, log-rank testing, and Cox regression.

Results: With a median follow-up time of 32.9 months, the median survival from the time of diagnosis was 16.5 months. Metastases were most commonly found in the lymph nodes, followed by the liver, bones, and lungs. The main pre-treatment prognostic variables that remained significant after adjustment for multiple testing were the Eastern Cooperative Oncology Group (ECOG) score for overall state of health and the number of organ systems involved. These variables were used to construct a practice-oriented risk stratification.

Conclusion: In patients with adeno- or undifferentiated CUP syndrome, the ECOG score and the number of organ systems involved are important risk factors.

LNSLNS

By definition, cancers of unknown primary origin (CUP) are histologically confirmed cancers where, when all diagnostic investigations are complete, only metastases have been found, with no evidence of a primary tumor. In terms of all cancer-related deaths in Germany in 2011, CUP was responsible for 2.1% among men and 2.5% among women (1). Older retrospective studies reported median survival at 3 to 6 months, but more recent studies of selected patients give median survival times in the order of 1 year (2–5).

Among CUPs, two special histological categories are neuroendocrine carcinomas and squamous cell carcinomas, which make up respectively 2 to 4% and 5 to 8% of all CUPs (2, 6). The former are treated according to specific protocols for neuroendocrine tumors, while squamous cell CUPs usually iaffect cervical lymph nodes and are treated in a similar way to head and neck tumors of known primary origin (2, 7). Both these subgroups have a significantly better prognosis than adenocarcinomas or undifferentiated carcinomas (which make up the great majority of all CUPs) (2, 3, 6). Generally, the standard treatment for adenocarcinoma or undifferentiated CUP is a combination of two drugs, one of them platinum-based (8), although here again, defined special cases occur that should received other, more specific protocols – but these cases make up only a small minority of this category (2, 8–10).

Current knowledge about CUP is partly based on a limited number of phase II studies, most of them small [they are summarized in (2)], so the evidence level about standard therapies must be classified as low (11). It is also based on a few case series of unselected patients (eTable 1), which have value, particularly for estimating prognosis, but some of which are out of date. Although no standards exist for prognosis-adjusted treatment, from the patient’s point of view statements about prognosis are extremely important. For this reason, we thought it worth presenting a systematic compilation of our own experience of treating patients with CUP.

The aims of this study were to describe our patient population without selecting them for treatability (as required in treatment studies), but in terms of clinically relevant characteristics of their disease; to document their overall survival; and on this basis to identify prognostically relevant variables. Neuroendocrine and squamous cell carcinomas were excluded in order to focus on the most relevant patient group, those with adenocarcinomas and undifferentiated tumors, and to avoid distortion of the results by the biological, clinical, and prognostic features of the former two subtypes.

Methods

The patients studied were a convenience sample collected with

Selected patient characteristics (whole study group)
Table 1
Selected patient characteristics (whole study group)
Case series of unselected patients with CUP (in chronological order)
eTable 1
Case series of unselected patients with CUP (in chronological order)
no thoughts about representativeness and no planning of case numbers. The group included all patients with adenocarcinoma or undifferentiated CUP who presented to us (H.L., K.N., and A.K.) at the CUP outpatient clinic at the National Cancer Center (NCT, Nationales Centrum für Tumorerkrankung) in Heidelberg, Germany, during the years 2006 to 2010. Inclusion criteria were a histologically confirmed cancer diagnosis and lack of evidence of a primary tumor after completion of diagnostic investigations that included at least cross-sectional imaging of thorax and abdomen, gastroscopy, and, for women, gynecological examination and mammography. The diagnostic classifications and all the findings included in this analysis were validated by one of us (H.L., K.N., or A.K.) as the relevant medical specialist. Neuroendocrine and squamous cell carcinomas were excluded. The group also included the cases of patients attending for a second opinion, and of patients whose relatives consulted us in the absence of the patient concerned, as long as these were consultations in person with one of us during the outpatient clinic and the cases were adequately documented by written findings and details of medical history.

The analysis focused on clinical variables that are regularly or frequently recorded during routine care and of which it was thought possible that they might affect the prognosis. The variables recorded appear in eTables 1 and 2. Tumor markers were classified as raised if they exceeded the upper reference value published by the laboratory in question. Cases were followed up on the basis of the medical records and of systematic data collection from doctors at other institutions and from local residents’ registration offices (Einwohnermeldeämter).

Overall survival was calculated from the time of first diagnosis. The distribution of survival times was estimated using the Kaplan–Meier method. In the primary analysis, hazard ratios (HR) including the 95% confidence interval (CI) of clinical factors were calculated using the univariate Cox regression model. The proportional hazards assumption was tested using the method of Grambsch and Therneau (12). To control the false discovery rate, the univariate Cox regression p-values were adjusted for multiple testing using the Benjamini–Hochberg correction (13). In individual cases, groups were additionally investigated for overall differences using the log rank test and multivariate Cox regression was carried out to correct for further characteristics. Since this was a retrospective observation study that was purely exploratory in nature, the p-values should not be interpreted as confirmatory, but as hypothesis-generating. All p-values are two-tailed. p-Values below 0.05 were regarded as statistically significant. All analyses were carried out using R 3.0.1 software (14).

Results

General patient characteristics

From 2006 to 2010, 218 patients with adenocarcinoma or undifferentiated CUP were treated or advised in the CUP outpatient clinic at the NCT in Heidelberg. In another five cases, patients’ relatives attended for consultations based on written documents and medical history details, resulting in a final total of 223 patients. In 68 cases (30%) patients were treated at the NCT, while 155 patients (70%) were primarily treated elsewhere. Median age at first diagnosis was 59.7 years (range 20 to 86 years); the numbers of men and women were almost equal (Table 1).

Median observation time was 32.9 months, and 145 of the 223 patients died during the observation period. Median survival was 16.5 months (95% CI 12.9 to 19.9 months), with a 1-year survival rate of 57% (50% to 65%) and a 2-year survival rate of 36% (29% to 43%). Men had significantly shorter survival than women (eFigure a, eTable 2), but no age dependence was shown (eFigure b, eTable 2). Highly significant factors were the ECOG (Eastern Cooperative Oncology Group) performance score (15) (eFigure c, eTable 2) and the number of organ systems involved: the relatively large patient group (50% of the whole) (Table 1) with only one organ system involved showed a marked survival advantage (eFigure d, eTable 2).

Factors associated with an unfavorable prognosis were involvement of lungs, liver, or adrenals (Table 2). Among patients with adrenal metastases, median survival was 7.4 months, corresponding to a HR of 2.69 (1.17 to 6.22). However, in the Cox model, adrenal metastases as a risk factor were not independent of the number of organ systems involved, since nine out of ten patients with adrenal metastases had more than two involved organ systems, compared with only 17% of all other patients.

A detailed analysis in terms of tumor markers and immunohistochemical markers as risk indicators yielded little in this group of patients. Initially raised serum CA 19–9 levels (shown in 40 of 89 cases with a recorded serum CA 19–9 concentration) was associated with significantly poorer survival (eTable 2). For further details, see eTable 2. Since a raised level of most tumo

Overall survival and hazard ratios in relation to involvement of different organ systems
Table 2
Overall survival and hazard ratios in relation to involvement of different organ systems
Parameters investigated and their prognostic effect on overall survival
eTable 2
Parameters investigated and their prognostic effect on overall survival
r markers was prognostically unfavorable (although for most markers a significant difference was not reached because of the small number of cases), this is likely to be due to a general association between tumor mass and prognosis. Overall, especially since in the case of most tumor markers information was only available for a small proportion of the patients, the results of the present study relating to tumor markers should be interpreted with caution.

Treatments received

For CUP, the therapeutic algorithm involves first excluding defined special cases whose treatment differs from the general standard of a broad-acting palliative cytostatic combination therapy (2, 9, 10). In particular, patients with local involvement should be referred for local radical treatment, typically with curative intent – i.e., radical resection and/or irradiation. In our study, this applied to 30 patients (13%), of whom

  • nine (4%) underwent resection only;
  • fourteen (6%) underwent irradiation (in three cases combined with chemotherapy);
  • seven (3%) underwent a combination of resection and irradiation.

The treatment algorithm also requires excluding other specific special cases, but these are rare in the group of adenocarcinomas and undifferentiated CUPs (2). For the remaining great majority of CUP patients, the standard treatment is the combination of a platinum derivative together with another cytostatic with palliative intent; the German guideline, in agreement with international recommendations, favors the combination of carboplatin and paclitaxel (9, 10, 16), which is also the first choice in our center. Another frequently used standard combination is cisplatin with gemcitabine. If a milder cytostatic therapy appears indicated, often monotherapy with gemcitabine is used, as is also preferred by ourselves in this situation. In our patient group, 138 patients (62%) received cytostatic therapy alone; for 55 patients (25%) no information about treatment received was available or no treatment was received (Table 1). Of the patients treated with cytostatics, the majority (110 of 138, corresponding to 80%) received a platinum-based treatment (Table 1).

The breakdown of overall survival according to treatment received shows that local treatment was associated with a significantly better prognosis (Figure 1a, eTable 2). Another breakdown of patients treated with cytostatics (Figure 1b, eTable 2) shows best results for the combination of carboplatin and paclitaxel, which in pairwise comparisons did significantly better than cisplatin with gemcitabine (HR for cisplatin + gemcitabine: 2.28, p = 0.02). However, after adjusting for risk factors (ECOG score, number of organ systems involved, etc.), the difference was no longer significant (HR: 1.17; p = 0.78).

A practical approach to risk stratification

Overall survival was highly significantly associated with ECOG score and number of involved organ systems, allowing easy classification of patients on the basis of these parameters into those with a favorable prognosis (ECOG ≤1 and 1 organ system involved, 67 patients, 36%), those with an intermediate prognosis (either ECOG >1 or >1 organ system involved, 86 patients, 46%), and those with an unfavorable prognosis (both ECOG >1 and >1 organ system involved, 33 patients, 18%). Within the overall study group, highly significant prognostic differences were shown between these groups (Figure 2a). It might be objected that the group with a favorable prognosis also contained the patients in the categories given treatment with curative intent, thus falsifying the result. However, the differences between the prognostic groups remained highly significant even after the exclusion of patients who underwent surgery and/or irradiation – even then, the favorable prognosis group with only one involved organ system and an ECOG score of 0 or 1 (50 patients) showed a median survival of 28.2 months (Figure 2b).

Discussion

Overall survival in relation to treatment
Figure 1
Overall survival in relation to treatment

The results of this study in regard to the prognostic significance of the ECOG score, the number of organ systems affected, and adrenal metastases confirm similar results of earlier studies (eTable 1) (6, 17). The median age in our study group agrees well with that in previous publications (3–6, 17, 18), which supports the assumption that our study group was representative. One difference from older data, which usually show a predominance of male patients, is the equal numbers of men and women in our study. An obvious possible explanation of the rise in numbers of women with CUP, which has been confirmed in independent epidemiological data, is the increased tobacco consumption among women, which in the past few decades has led to a general rise in the occurrence of tobacco-related cancers in women (1, 19).

Naturally, our patient study group is not representative of the general population, because no planning was carried out to ensure that a representative sample was included. The relatively high proportion of patients (41%) with only one involved organ system and good physical performance (ECOG ≤1) suggests that some patien

A practical approach to risk stratification
Figure 2
A practical approach to risk stratification
t selection must have taken place, perhaps due to the fact that the patients most likely to attend a university center are those for whom every possible attempt at therapy seems indicated, whereas those with multiple involved organ systems and poorer physical performance are more often treated palliatively nearer home. These factors could have contributed to the fact that, at 16.5 months, the median survival in our study was better than that in most previous studies. Despite these reservations, however, we do believe that, today, for mobile and treatable patients who can be managed on an outpatient basis, a longer median survival may be expected than the 3 to 6 months shown in older studies.

One potentially useful feature of the risk stratification presented here is that it counteracts the way in which selection of the patient population at our center, as referred to above, distorts prognosis estimation, by separating out the patient groups who may be under- or over-represented. It is worth mentioning that several systems for risk stratification of patients with CUP have already been published, but have not yet come in wide use, partly because they require the use of complicated decision trees (eTable 1) (6, 17). Compared to these, the risk stratification method we present here is easy to use on all patients who have a complete diagnosis, without any extra work. Because our study does not contain a separate validated patient sample, however, prospective validation of the method in an independent cohort would be desirable. As a further limitation of the validity of our results, it could be objected that diagnoses were not generally confirmed by histological reference assessment, which was only done if initial classification was not sufficient; however, it must be emphasized that histological confirmation with assessment by a board-certified pathologist was obtained in every case.

A valid comparison of the efficacy of individual treatments is not possible in our study, as the lack of randomization could confound the results, e.g., different ranges of indications associated with” the choice of treatment. Our results relating to this should therefore be interpreted with caution. They do appear to confirm that the guideline standard combination treatment with carboplatin and paclitaxel should normally be offered as the first choice, and that gemcitabine monotherapy is a good alternative if carboplatin+paclitaxel appears too toxic (10). It also appears relevant that patients treated primarily with local radical procedures, especially radical resection with or without postsurgical radiotherapy, have a comparatively good prognosis. It must be emphasized that this likely does not result from the chosen mode of therapy, but rather from rational selection of patients to receive it: local radical procedures are only a rational option for locally limited disease. To this extent, the data in our study support the guideline recommendation that, for patients with locally limited disease, local radical therapy should be considered first (9, 10).

Conclusions for clinical practice

In the group of patients with adenocarcinoma or undifferentiated CUP seen at a university cancer center and presented here, median survival was between 1 and 2 years. A practical approach to risk stratification is to take together the patient’s ECOG score for physical performance and the number of organ systems involved in the cancer. The guideline recommendation that patients with local involvement should be referred for a local radical procedure (surgery with or without radiation) is supported by the data, which show that these patients have a relatively good prognosis.

Conflict of interest statement

Professor Krämer received funding for the PACET-CUP study from Merck.

The other authors declare that no conflict of interest exists.

Manuscript received on 25 October 2013, revised version accepted on
5 May 2014.

Translated from the original German by Kersti Wagstaff, MA.

Corresponding author:
Prof. Dr. med. Alwin Krämer
Medizinische Klinik V
Universität Heidelberg
Im Neuenheimer Feld 410
69120 Heidelberg, Germany
Alwin.Kraemer@med.uni-heidelberg.de

@eFigure, eTables:
www.aerzteblatt-international.de/14m0481

1.
Löffler H, Neben K, Krämer A: Cancer of unknown primary : Epidemiology and pathogenesis. Radiologe 2014; 54: 107–11. CrossRef MEDLINE
2.
Neben K, Hübner G, Folprecht G, Jäger D, Krämer A: Metastases in the absence of a primary tumor: Advances in the diagnosis and treatment of CUP syndrome. Dtsch Arztebl Int 2008; 105: 733–40. VOLLTEXT
3.
Altman E, Cadman E: An analysis of 1539 patients with cancer of unknown primary site. Cancer 1986; 57: 120–4. CrossRef MEDLINE
4.
Levi F, Te VC, Erler G, Randimbison L, La Vecchia C: Epidemiology of unknown primary tumours. Eur J Cancer 2002; 38: 1810–2. CrossRef MEDLINE
5.
van de Wouw AJ, Janssen-Heijnen ML, Coebergh JW, Hillen HF: Epidemiology of unknown primary tumours; incidence and population-based survival of 1285 patients in Southeast Netherlands, 1984–1992. Eur J Cancer 2002; 38: 409–13. CrossRef MEDLINE
6.
Hess KR, Abbruzzese MC, Lenzi R, Raber MN, Abbruzzese JL: Classification and regression tree analysis of 1000 consecutive patients with unknown primary carcinoma. Clin Cancer Res 1999; 5: 3403–10. CrossRef
7.
Strojan P, Ferlito A, Langendijk JA, et al.: Contemporary management of lymph node metastases from an unknown primary to the neck: II. a review of therapeutic options. Head Neck 2013; 35: 286–93. CrossRef MEDLINE
8.
Löffler H, Neben K, Krämer A: Structured diagnostics and therapy of the CUP syndrome. Radiologe 2014; 54: 112–6. CrossRef MEDLINE
9.
Fizazi K, Greco FA, Pavlidis N, Pentheroudakis G, Group EGW: Cancers of unknown primary site: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol 2011; 22: 64–8. CrossRef MEDLINE
10.
Hübner G, Bokemeyer C: CUP-Syndrom – Tumorerkrankung mit unbekanntem Primärtumor. www.dgho-onkopedia.de/de/onkopedia/archiv/cup-syndrom/cup-syndrom-stand-dez.-2005 (last accessed on 14 April 2014)
11.
Golfinopoulos V, Pentheroudakis G, Salanti G, Nearchou AD, Ioannidis JP, Pavlidis N: Comparative survival with diverse chemotherapy regimens for cancer of unknown primary site: multiple-treatments meta-analysis. Cancer Treat Rev 2009; 35: 570–3. CrossRef MEDLINE
12.
Grambsch P, Therneau T: Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 1994; 81: 515–26. CrossRef
13.
Benjamini Y, Hochberg Y: Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society Series B 1995; 57: 289–300.
14.
R Core Team R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria 2013. www.R-project.org/.
15.
Oken MM, Creech RH, Tormey DC, et al.: Toxicity and response criteria of the Eastern Cooperative Oncology Group. Am J Clin Oncol 1982; 5: 649–55. CrossRef MEDLINE
16.
National Comprehensive Cancer Network: Occult Primary (Cancer of Unknown Primary [CUP]). www.nccn.org/professionals/physician_gls/pdf/occult.pdf (last accessed on 14 April 2014)
17.
Petrakis D, Pentheroudakis G, Voulgaris E, Pavlidis N: Prognostication in cancer of unknown primary (CUP): Development of a prognostic algorithm in 311 cases and review of the literature. Cancer Treat Rev 2013; 39: 701–8. CrossRef MEDLINE
18.
Pentheroudakis G, Golfinopoulos V, Pavlidis N: Switching benchmarks in cancer of unknown primary: from autopsy to microarray. Eur J Cancer 2007; 43: 2026–36. CrossRef MEDLINE
19.
Kaaks R, Sookthai D, Hemminki K, et al.: Risk factors for cancers of unknown primary site: Results from the prospective EPIC cohort. International journal of cancer Journal international du cancer 2014. Int J Cancer 2014 Apr 2. doi: 10.1002/ijc.28874. [Epub ahead of print]. CrossRef MEDLINE
Department of Internal Medicine V: Hematology, Oncology and Rheumatology, Heidelberg University Hospital: PD Dr. med. Löffler, Joe Puthenparambil, Prof. Dr. med. Krämer
Clinical Cooperation Unit Molecular Hematology/Oncology, Department of Internal Medicine V: Heidelberg University Hospital and German Cancer Research Center (DKFZ), Heidelberg: PD Dr. med. Löffler; Prof. Dr. med. Krämer
German Cancer Research Center, Department of Biostatistics, Heidelberg: Thomas Hielscher
Medical Clinic 2, Baden-Baden City Hospital, Klinikum Mittelbaden, Baden-Baden: Prof. Dr. med. Neben
Overall survival in relation to treatment
Figure 1
Overall survival in relation to treatment
A practical approach to risk stratification
Figure 2
A practical approach to risk stratification
Key messages
Selected patient characteristics (whole study group)
Table 1
Selected patient characteristics (whole study group)
Overall survival and hazard ratios in relation to involvement of different organ systems
Table 2
Overall survival and hazard ratios in relation to involvement of different organ systems
Case series of unselected patients with CUP (in chronological order)
eTable 1
Case series of unselected patients with CUP (in chronological order)
Parameters investigated and their prognostic effect on overall survival
eTable 2
Parameters investigated and their prognostic effect on overall survival
1.Löffler H, Neben K, Krämer A: Cancer of unknown primary : Epidemiology and pathogenesis. Radiologe 2014; 54: 107–11. CrossRef MEDLINE
2.Neben K, Hübner G, Folprecht G, Jäger D, Krämer A: Metastases in the absence of a primary tumor: Advances in the diagnosis and treatment of CUP syndrome. Dtsch Arztebl Int 2008; 105: 733–40. VOLLTEXT
3.Altman E, Cadman E: An analysis of 1539 patients with cancer of unknown primary site. Cancer 1986; 57: 120–4. CrossRef MEDLINE
4.Levi F, Te VC, Erler G, Randimbison L, La Vecchia C: Epidemiology of unknown primary tumours. Eur J Cancer 2002; 38: 1810–2. CrossRef MEDLINE
5.van de Wouw AJ, Janssen-Heijnen ML, Coebergh JW, Hillen HF: Epidemiology of unknown primary tumours; incidence and population-based survival of 1285 patients in Southeast Netherlands, 1984–1992. Eur J Cancer 2002; 38: 409–13. CrossRef MEDLINE
6.Hess KR, Abbruzzese MC, Lenzi R, Raber MN, Abbruzzese JL: Classification and regression tree analysis of 1000 consecutive patients with unknown primary carcinoma. Clin Cancer Res 1999; 5: 3403–10. CrossRef
7.Strojan P, Ferlito A, Langendijk JA, et al.: Contemporary management of lymph node metastases from an unknown primary to the neck: II. a review of therapeutic options. Head Neck 2013; 35: 286–93. CrossRef MEDLINE
8.Löffler H, Neben K, Krämer A: Structured diagnostics and therapy of the CUP syndrome. Radiologe 2014; 54: 112–6. CrossRef MEDLINE
9.Fizazi K, Greco FA, Pavlidis N, Pentheroudakis G, Group EGW: Cancers of unknown primary site: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol 2011; 22: 64–8. CrossRef MEDLINE
10.Hübner G, Bokemeyer C: CUP-Syndrom – Tumorerkrankung mit unbekanntem Primärtumor. www.dgho-onkopedia.de/de/onkopedia/archiv/cup-syndrom/cup-syndrom-stand-dez.-2005 (last accessed on 14 April 2014)
11.Golfinopoulos V, Pentheroudakis G, Salanti G, Nearchou AD, Ioannidis JP, Pavlidis N: Comparative survival with diverse chemotherapy regimens for cancer of unknown primary site: multiple-treatments meta-analysis. Cancer Treat Rev 2009; 35: 570–3. CrossRef MEDLINE
12.Grambsch P, Therneau T: Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 1994; 81: 515–26. CrossRef
13.Benjamini Y, Hochberg Y: Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society Series B 1995; 57: 289–300.
14.R Core Team R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria 2013. www.R-project.org/.
15.Oken MM, Creech RH, Tormey DC, et al.: Toxicity and response criteria of the Eastern Cooperative Oncology Group. Am J Clin Oncol 1982; 5: 649–55. CrossRef MEDLINE
16.National Comprehensive Cancer Network: Occult Primary (Cancer of Unknown Primary [CUP]). www.nccn.org/professionals/physician_gls/pdf/occult.pdf (last accessed on 14 April 2014)
17.Petrakis D, Pentheroudakis G, Voulgaris E, Pavlidis N: Prognostication in cancer of unknown primary (CUP): Development of a prognostic algorithm in 311 cases and review of the literature. Cancer Treat Rev 2013; 39: 701–8. CrossRef MEDLINE
18.Pentheroudakis G, Golfinopoulos V, Pavlidis N: Switching benchmarks in cancer of unknown primary: from autopsy to microarray. Eur J Cancer 2007; 43: 2026–36. CrossRef MEDLINE
19.Kaaks R, Sookthai D, Hemminki K, et al.: Risk factors for cancers of unknown primary site: Results from the prospective EPIC cohort. International journal of cancer Journal international du cancer 2014. Int J Cancer 2014 Apr 2. doi: 10.1002/ijc.28874. [Epub ahead of print]. CrossRef MEDLINE