• Open Access

Exploring the epidemiological characteristics of cancers of unknown primary site in an Australian population: implications for research and clinical care

Authors


Correspondence to:
Dr Colin Luke, Epidemiology Branch, Department of Health, PO Box 6, Rundle Mall, Adelaide, South Australia 5001. Fax: (08) 8226 6291; e-mail: Colin.Luke@health.sa.gov.au

Abstract

Objectives: To investigate incidence, mortality and case survival trends for cancer of unknown primary site (CUP) and consider clinical implications.

Method: South Australian Cancer Registry data were used to calculate age-standardised incidence and mortality rates from 1977 to 2004. Disease-specific survivals, socio-demographic, histological and secular predictors of CUP, compared with cancers of known primary site, and of CUP histological types, using multivariable logistic regression were investigated.

Results: Incidence and mortality rates increased approximately 60% between 1977-80 and 1981-84. Rates peaked in 1993-96. Male to female incidence and mortality rate ratios approximated 1.3:1. Incidence and mortality rates increased with age. The odds of unspecified histological type, compared with the more common adenocarcinomas, were higher for males than females, non-metropolitan residents, low socio-economic areas, and for 1977-88 than subsequent diagnostic periods. CUP represented a higher proportion of cancers in Indigenous patients. Case survival was 7% at 10 years from diagnosis. Factors predictive of lower case survival included older age, male sex, Indigenous status, lower socio-economic status, and unspecified histology type.

Conclusion: Results point to poor CUP outcomes, but with a modest improvement in survival. The study identifies socio-demographic groups at elevated risk of CUP and of worse treatment outcomes where increased research and clinical attention are required.

Cancers of unknown primary site (CUP) account for about: 4% of invasive cancers recorded by Australian cancer registries and 8% of cancer deaths.1 The reported proportion of cancers comprising CUP generally has approximated 2% to 5% in North America and Europe,2–5 although higher figures of up to 10% have been cited.5 CUP has received little attention in epidemiological studies other than as a marker of sub-optimum site specification in cancer registration.5,6

Although a heterogeneous entity, CUP is increasingly thought to include a distinct clinical sub-group of tumours characterised by rapid progression, atypical metastatic spread, and multiple metastases that commonly involve the lung, bone, lymph nodes and liver.7–10 Although prognosis is poor, with a five-year survival of approximately 11%,11,12 some sub-sets of CUP are responsive to multi-agent chemotherapy and other systemic treatments.13–17

CUP cases include a disproportionately high number of adenocarcinomas and related cell types.3,11 Undisclosed primary sites are thought to frequently include the lung and pancreas, and less so the breast, prostate, stomach, liver and large bowel.7,8,10,11 While identifying the primary site was regarded originally to be of key importance for optimum treatment,18 the benefits are now thought to be minor and exhaustive searching is frequently discouraged to avoid burdening the patient unnecessarily.8,11

In this study, we describe the epidemiological characteristics of CUP in South Australia. This is one of Australia's eight states and territories, with a population of 1.5 million residents. The study complements CUP data routinely available from Australian cancer registries by further exploring differences in CUP incidence and survival by socio-demographic sub-group and histological type,1,11,12 and by considering clinical implications. Indications of results for research and service delivery are discussed.

Methods

Data collection

The South Australian Cancer Registry is population-based and has received statutory notifications of invasive cancers diagnosed in South Australia since 1977.19–21 Death data are collected through routine notifications, electronic searches of official State death records and the National Death Index at the Australian Institute of Health and Welfare, and from interstate registries.19–21 Under-ascertainment has been checked through active follow-up, and with deaths reported independently, and found to be minimal, with little effect on calculated survival.21,22

This study included 5,705 CUP cases (ICDO3: C80) diagnosed between 1977 and 2004, plus 4,940 deaths. These were metastatic cancers where the primary organ site of origin could not be identified. They were classified as defined in ICDO3, with reliance on clinical and histopathology reporting, additional consultation with treating clinicians where indicated and where feasible and indicated, review of clinical notes.19,21

SNOMED II histological codes were grouped into five broad categories for this study, i.e., as not otherwise specified (NOS) (80003-80043), epithelial neoplasms NOS (80103-80413), papillary and squamous cell neoplasms (80503—80823), adenocarcinomas, cystic/mucinous/serous and related glandular lesions (81403-85503), and “other” cancers (remaining histological types).21,23

Socio-demographic descriptors included age at diagnosis; sex; region of residence, classified as 20 statistical sub-divisions and as metropolitan or non-metropolitan;21 country of birth (expressed using World Health Organization criteria);24 Indigenous status; and relative socio-economic disadvantage, as inferred from residential postcode characteristics using the SEIFA index and expressed as four ordinal categories.25

Statistical analyses

Mean annual incidence and mortality rates were determined for four-year periods from 1977 to 2004, directly standardising by five-year age group (open-ended category from 85 years) to the 2001 Australian reference population.26,27 Ninety-five per cent confidence limits were calculated assuming a Poisson distribution, as described previously.27,28 Rates were calculated for all ages combined and for age categories under 30 years and 30-39, 40-49, 50-59, 60-69, 70-79, and 80 years or more respectively, to assist visualisation of trends.

Differences in epidemiological characteristics of CUP and cancers of known primary site were investigated, as were differences by histological category of CUP. Initially each characteristic was analysed as a univariate predictor, using the Pearson chi-square test for nominal variables and the Mann-Whitney U test or Kruskal-Wallis ANOVA for ordinal variables, depending on whether the outcome variable was dichotomous (e.g., CUP yes/no) or multinomial (e.g., histological categories).26,27

Epidemiological differences between CUP and cancers of known primary site were further explored using multiple logistic regression analysis.26,27 All socio-demographic variables were entered as predictors, with backwards elimination of those where the fit of the model did not reduce as a consequence (p>0.05). Assumptions underlying the analysis, including an absence of co-linearity, were found to be satisfied.26 Differences in epidemiological characteristics of CUP by histological category also were investigated using multinomial logistic regression.26,27

Case survivals were calculated, with a date of censoring of live cases of 31 December, 2004. Kaplan-Meier product-limit estimates of disease-specific survival were calculated, treating deaths from other causes as censored observations.26,27 Multivariable Cox proportional hazards regression also was undertaken to assess socio-demographic and histological predictors of survival outcomes from CUP,26,27 employing the same censoring criteria as for the Kaplan-Meier analyses. Again, backwards elimination was used, as described for the logistic regression. Assumptions underlying the analysis, including proportionality and an absence of co-linearity, were found to be satisfied.

Disease-specific survival was preferred to relative survival or excess mortality methods because life tables were not readily available for all sub-groups. Also, survival estimates were more precise (i.e., smaller confidence intervals), assisting comparisons where numbers of cancers were low. Analyses have shown very similar survival estimates in South Australia, irrespective of whether relative survivals or disease-specific survivals were used.29

Results

Time trends in incidence and mortality

Annual incidence and mortality rates for CUP increased from1977-80 to 1981-84, both in males and females (Table 1). Rates tended to peak in 1993-96 before declining. The M/F incidence and mortality rate ratios were 1.3: 1.

Table 1.  Mean annual age-standardised (Australia, 2001) incidence and mortality rates (95% confidence limits) for invasive CUP per 100,000 South Australians by sex and calendar year period.a
Year1977-801981-841985-881989-921993-961997-20002001-041977-2004
  1. Note:

Incidence
Malesn=213n=340n=355n=380n=504n=536n=510n=2,838
 12.117.616.915.719.318.916.516.7
 [10.6, 13.8][15.8, 19.5][15.2, 18.7][14.2, 17.3][17.7, 21.0][17.4, 20.6][15.1, 18.0][16.1, 17.3]
Femalesn=171n=332n=351n=403n=520n=555n=535n=2,867
 7.513.112.412.915.415.013.212.8
 [6.4, 8.7][11.7, 14.6][11.2, 13.7][11.7, 14.2][14.1, 16.8][13.8, 16.3][12.1, 14.4][12.3, 13.3]
Totaln=384n=672n=706n=783n=1,024n=1,091n=1,045n=5,705
 9.415.114.314.217.216.814.814.5
 [8.5, 10.4][14.0, 16.3][13.3, 15.4][13.2, 15.2][16.2, 18.3][15.8, 17.8][13.9, 15.7][14.1, 14.9]
Mortality
Malesn=188n=285n=332n=320n=443n=441n=426n=2,435
 11.014.716.013.417.015.713.814.5
 [9.5, 12.7][13.1, 16.5][14.3, 17.8][12.0, 14.9][15.5, 18.6][14.3, 17.2][12.5, 15.2][13.9, 15.1]
Femalesn=142n=290n=320n=353n=454n=471n=475n=2,505
 6.211.411.311.313.412.911.711.2
 [5.2, 7.3][10.1, 12.8][10.1, 12.6][10.2, 12.5][12.2, 14.7][11.8, 14.1][10.7, 12.8][10.8, 11.6]
Totaln=330n=575n=652n=673n=897n=912n=901n=4,940
 8.112.813.312.215.014.012.712.6
 [7.3, 9.0][11.8, 13.9][12.3, 14.4][11.3, 13.1][14.0, 16.0][13.1, 14.9][11.9, 13.5][12.3, 13.0]

Incidence by histological and socio-demographic characteristic

The highest incidence of CUP applied to adenocarcinomas and the NOS histological category, irrespective of sex (Table 2). Mean annual incidence rates for all histological types combined increased more than 100 fold between ages less than 50 years and those of 80 years or more, both in males and females. The strongest upward gradient by age was for the NOS histology category. A higher incidence applied to males than females in each histological group (Table 2).

Table 2.  Mean annual age-standardised (Australia, 2001) incidence (95% confidence limits) for invasive CUP per 100,000 South Australians in 1977-2004 by sex and histology type.a
Histology type
SexAge at diagosis (year)Adenocarc.Epithelial NOSNOSOtherPapillary Squam. Cell CarcinomaTotal
  1. Note:

  2. (a) Data source: South Australian Cancer Registry.

Males<50n=96n=32n=19n=5n=19n=171
  0.70.20.20.00.11.2
  (0.6, 0.9)(0.1, 0.3)(0.1, 0.3)(0.0, 0.1)(0.1, 0.2)(1.0, 1.4)
 50-59n=184n=78n=41n=4n=46n=353
  8.53.61.80.22.116.2
  (7.3, 9.8)(2.9, 4.5)(1.4, 2.6)(0.1, 0.5)(1.6, 2.8)(14.6, 18.0)
 60-69n=364n=159n=173n=5n=74n=775
  21.79.410.30.34.446.1
  (19.6, 24.0)(8.0, 10.9)(8.8, 11.9)(0.1, 0.7)(3.5, 5.5)(42.9, 49.5)
 70-79n=434n=193n=297n=1n=70n=995
  40.718.529.30.36.595.3
  (37.0, 44.7)(16.0, 21.3)(26.9, 33.9)(0.0, 1.5)(5.1, 8.2)(89.5, 101.4)
 80+n=161n=73n=267n=2n=41n=544
  40.420.973.30.711.6146.9
  (34.5, 47.0)(16.5, 26.1)(64.8, 82.6)(0.1, 2.3)(8.4, 15.6)(134.9, 159.6)
 Totaln=1,239n=535n=797n=17n=250n=2,838
  6.93.15.20.11.416.7
  (6.5, 7.3)(2.8, 3.4)(4.8, 5.6)(0.1, 0.2)(1.2, 1.6)(16.1, 17.3)
Females<50n=91n=35n=16n=1n=12n=155
  0.70.30.10.00.11.2
  (0.6, 0.9)(0.2, 0.4)(0.1, 0.2)(0.0, 0.0)(0.1, 0.2)(1.0, 1.4)
 50-59n=170n=54n=26n=3n=22n=275
  7.92.51.10.11.012.7
  (6.8, 9.2)(1.9, 3.2)(0.7, 1.6)(0.0, 0.3)(0.6, 1.5)(11.2, 14.3)
 60-69n=360n=82n=87n=5n=24n=558
  19.84.64.90.11.430.9
  (17.8, 21.9)(3.7, 5.7)(3.9, 6.0)(0.0, 0.3)(0.9, 2.0)(28.4, 33.5)
 70-79n=447n=144n=262n=3n=46n=902
  32.510.820.20.13.166.8
  (29.6, 35.6)(9.1, 12.7)(17.8, 22.8)(0.0, 0.2)(2.3, 4.1)(62.5, 71.3)
 80+n=318n=86n=543n=1n=29n=977
  40.912.071.60.13.8128.4
  (36.6, 45.6)(9.7, 14.7)(65.8, 77.8)(0.0, 0.5)(2.6, 5.4)(120.5, 136.6)
 Totaln=1,386n=401n=934n=13n=133n=2,867
  6.21.94.00.00.612.8
  (5.9, 6.5)(1.7, 2.1)(3.7, 4.3)(0.0, 0.0)(0.5, 0.7)(12.3, 13.3)

Percentage of cancers classified as CUP

Overall, 3.5% of cancers recorded for 1977-2004 were CUP. Univariate analyses indicated that this percentage was higher for people born outside Australia (p=0.001); Indigenous residents (p<0.001); and potentially those in low socio-economic areas (p=0.087). Little difference was evident by statistical subdivision with similar CUP percentages applying to metropolitan and non-metropolitan regions (p=0.644).

Differences presented in the distribution of CUP by histological category for the following univariate predictors: country of birth (p<0.001), with papillary and squamous lesions reported less frequently in cases born outside Australia; socio-economic status (p=0.015), with papillary and squamous lesions reported more frequently in lower socio-economic areas; and metropolitan/non-metropolitan place of residence (p<0.001), with NOS histology reported more frequently in non-metropolitan areas.

Multiple logistic regression analysis indicated that compared with cancers of known primary site, the relative odds of CUP was higher in older patients, females, Indigenous patients, and those diagnosed after 1977-80 (Table 3). Country of birth was not predictive in this multivariable context (p>0.200). The relative odds of CUP were higher for the NOS and epithelial NOS categories than for adenocarcinomas, whereas they were lower for papillary and squamous cell lesions and other histological types.

Table 3.  Relative odds (95% confidence limits) of invasive CUP compared with a cancer of known primary site by socio-demographic and histological characteristics; South Australia, 1977-2004.
Multiple logistic regression
Predictors Relative odds
  1. Notes:

  2. (a) Data in brackets (column 1) show sub-totals for all invasive cancers

  3. Data source: South Australian Cancer Registry.

Age at diagnosis (yrs.):
 80+ (reference) (n=26,048)1.00
 70-79 (n=46,785)0.93 (0.87, 1.01)
 60-69 (n=41,541)0.80 (0.74, 0.87)
 50-59 (n=25,256)0.69 (0.62, 0.76)
 40-49 (n=13,291)0.58 (0.50, 0.67)
 Under 40 (n=11,250)0.44 (0.35, 0.55)
Sex:
 Male (reference) (n=89,317)1.00
 Female (n=74,854)1.25 (1.18, 1.32)
Race:
 Non-Indigenous (reference) (n=163,731)1.00
 Indigenous (n=440)2.07 (1.41, 3.04)
Histology type:
 Adenocarcinoma (reference) (n=87,928)1.00
 Epithelial NOS (n=8,266)4.37 (4.04, 4.73)
 NOS (n=8,901)5.90 (5.49, 6.33)
 Other (n=41,883)0.02 (0.02, 0.03)
 Papillary/squamous cell carcinomas (n=17,193)0.79 (0.71, 0.88)
Diagnostic period:
 1977-80 (reference) (n=14,259)1.00
 1981-84 (n=17,099)2.31 (1.97, 2.70)
 1985-88 (n=19,930)2.13 (1.83, 2.49)
 1989-92 (n=23,449)2.00 (1.72, 2.33)
 1993-96 (n=27,788)2.21 (1.90, 2.56)
 1997-2000 (n=29,898)2.17 (1.87, 2.52)
 2001-04 (n=31,748)1.94 (1.67, 2.25)

Multinomial logistic regression revealed socio-demographic differences by histological type of CUP (Table 4). Compared with adenocarcinomas, the relative odds of the:

Table 4.  Relative odds (95% confidence limits) of specified histology type compared with adenocarcinoma among invasive CUP by socio-demographic characteristics; South Australia, 1977-2004.a
Multinomial logistic regression
 Histology type
 Epithelial NOS (n=936)NOS (n=1,731)Other (n=30)Papillary/Squamous cell carcin. (n=383)
  1. Notes:

  2. (a) Data in brackets (column 1) show CUP sub-totals

  3. (b) Includes “not specified”

  4. Data source: South Australian Cancer Registry.

Age at diagnosis (yrs.):
   Under 60 (reference) (n=954)1.001.001.001.00
   60-69 (n=1,333)0.89 (0.71, 1.12)1.88 (1.45, 2.43)0.61 (0.33, 1.13)0.73 (0.54, 1.00)
   70-79 (n=1,897)1.06 (0.85, 1.31)3.58 (2.81, 4.55)0.18 (0.07, 0.49)0.74 (0.55, 0.99)
   80+ (n=1,521)0.99 (0.75, 1.31)10.75 (8.39, 13.78)0.24 (0.09, 0.67)0.91 (0.65, 1.28)
Sex:
   Male (reference) (n=2,838)1.001.001.001.00
   Female (n=2,867)0.67 (0.58, 0.79)0.81 (0.71, 0.93)0.74 (0.37, 1.46)0.47 (0.37, 0.59)
Race:
   Non-Indigenous (reference) (n=5,674)1.001.001.00
   Indigenous (n=31)0.67 (0.19, 2.41)2.06 (0.88, 4.84) 1.89 (0.52, 6.90)
Residence:
   Metropolitan (reference) (n=4,273)1.001.001.001.00
   Non-metropolitan (n=1,432)0.82 (0.68, 0.97)1.55 (1.33, 1.81)0.76 (0.33, 1.79)0.94 (0.73, 1.23)
Socio-economic status:
   Not low (reference) (n=3,861)1.001.001.001.00
   Low (n=1,844)1.09 (0.93, 1.29)1.16 (1.00, 1.34)1.21 (0.55, 2.64)1.55 (1.23, 1.94)
Country of birth:
   Australiab (reference) (n=3,902)1.001.001.001.00
   Other (n=1,803)1.03 (0.88, 1.21)0.95 (0.82, 1.10)0.82 (0.45, 1.49)0.69 (0.53, 0.88)
Diagnostic period:
   1977-88 (reference) (1,762)1.001.001.001.00
   1989-92 (n=783)0.90 (0.70, 1.15)0.70 (0.57, 0.87)1.11 (0.42, 2.96)1.09 (0.77, 1.54)
   1993-96 (n=1,024)1.12 (0.90, 1.39)0.70 (0.57, 0.85)1.23 (0.40, 3.75)1.01 (0.77, 1.32)
   1997-2000 (n=1,091)1.04 (0.84, 1.29)0.66 (0.55, 0.80)1.76 (0.54, 5.68)1.04 (0.75, 1.42)
   2001-04 (n=1,045)0.91 (0.73, 1.15)0.70 (0.57, 0.85)3.54 (1.46, 8.62)1.38 (1.02, 1.87)
  • 1) NOS category were higher in the older age groups, males, non-metropolitan suburbs, low socio-economic areas, and in patients diagnosed in 1977-88 compared with 1989-2004. While the relative odds doubled for Indigenous people (i.e., RO=2.06), numbers were small and the result not statistically significant (p=0.100).
  • 2) Epithelial NOS were higher in males and metropolitan areas.
  • 3) Papillary and squamous cell neoplasms were higher in younger patients, males, low socio-economic areas, and in 2001-04 compared with 1977-88.
  • 4) Other histological types were higher in patients under 60 years of age and in 2001-04.

Survivals

Survival (± standard error) from CUP was 8.5% (± 0.4) at five years from diagnosis, 7.1% (± 0.4) at 10 years, 6.5% (±0.4) at 15 years, and 5.8% (± 0.6) at 20 years. Multivariable proportional hazards regression indicated a higher risk of case fatality from CUP in older patients, males, Indigenous patients, lower socio-economic areas, and patients with tumours in the NOS histological category or adenocarcinomas (Table 5). Case fatality was lower in 1989-2004 compared with 1977-88 and tended to decrease across the 1989-2004 period (Table 5). Trends towards a lower case fatality in 1989-2004 did not vary by histological type (p=0.322).

Table 5.  Relative risk (95% confidence limits) of death from CUP among cases with these invasive cancers by socio-demographic and histological characteristics; South Australia, 1977-2004.a
Multiple proportional hazards regression
PredictorsRelative odds
  1. Notes:

  2. (a) Date of censoring of live cases: 31 December 31, 2004.

  3. Data in brackets show CUP sub-totals. Data source: SA Cancer Registry.

Age at diagnosis (years)
    Under 40 (reference) (n=87)1.00
    40-49 (n=239)1.32 (1.00, 1.76)
    50-59 (n=628)1.50 (1.16, 1.95)
    60-69 (n=1,333)1.87 (1.45, 2.42)
    70-79 (n=1,897)2.06 (1.60, 2.65)
    80+ (n=1,521)2.38 (1.85, 3.08)
Sex
    Male (reference) (n=2,838)1.00
    Female (n=2,867)0.92 (0.87, 0.98)
Race
    Non-Indigenous (reference) (n=5,674)1.00
    Indigenous (n=31)1.82 (1.28, 2.60)
Socio-economic status
    Not high (reference) (n=4,241)1.00
    High (n=1,464)0.91 (0.85, 0.97)
Histology type
    Adenocarcinoma (reference) (n=2,625)1.00
    Epithelial NOS (n=936)0.87 (0.80, 0.94)
    NOS (n=1,731)1.25 (1.17, 1.34)
    Other (n=30)0.50 (0.31, 0.81)
    Papillary/squamous cell carcinoma (n=383)0.37 (0.32, 0.42)
Diagnosis period
    1977-88 (reference) (n=1,762)1.00
    1989-92 (n=783)0.88 (0.80, 0.96)
    1993-96 (n=1,024)0.85 (0.78, 0.92)
    1997-2004 (n=2,136)0.75 (0.70, 0.80)

Discussion

Our study shows that age-standardised incidence and mortality rates for CUP tended to peak in 1993-96 before decreasing. National data for unknown and ill-defined primary cancers showed a similar peak, although occurring earlier in 1989-92.1

The recent decline in CUP rates was accompanied by population-based mortality reductions for cancers commonly thought to be responsible for CUP, such as cancers of the lung and pancreas in males.30,31 In addition, other cancers less commonly associated with CUP such as female breast and prostate have experienced a mortality decline.30,31 Another factor potentially contributing to declines in CUP would be gains in histopathology techniques, and increased use of diagnostic imaging and colonoscopy, increasing opportunities to discover primary organ sites.32–36

The abrupt increase in CUP rates between 1977-80 and 1981-84 is thought to reflect an increased awareness of CUP as a diagnostic entity, and potentially a greater availability of treatment options for these and other cancers, with the emergence of novel cytoxic agents as well as greater availability of diagnostic and imaging methods. In addition, it is thought, at least partly, to reflect methodological issues. During the early years of the registry, a lower CUP case ascertainment likely applied, due to a lack of electronic search routines, and some of these cancers may have been assigned on the basis of less conclusive evidence to a primary site. In New South Wales, an increase in CUP incidence rates also was apparent in the late 1970s.11

The upward age gradient in incidence of CUP was pronounced in this study, exceeding that seen for invasive cancers in general.37 In particular, a steep gradient was observed for the CUP NOS histological category. While this may reflect less intensive searching for the primary site in older than younger cases, another possible explanation is that CUP includes a distinct clinical entity that becomes more common with increasing age.7–10

The male/female incidence rate ratio for CUP of 1.3:1 was similar to the 1.4:1 observed for unknown and ill-defined primary cancers nationally in 1982-2003,31 and may reflect the excess of cancers in males more generally.

Adenocarcinoma was the largest histological category of CUP, contributing proportionally more to these lesions than to other invasive cancers. Confirmatory evidence is available from New South Wales and other studies.3,11 Again, this may be a function of histological composition of the organ sites responsible for CUP, including the pancreas, and less commonly, the breast, prostate, stomach and large bowel.7,8,10,11

As a proportion of all invasive cancers, CUP was more common in Indigenous patients, which is probably due to the more advanced stages of their cancers at diagnosis and the poorer access in regional areas to advanced diagnostic facilities.38 A relatively high prevalence of deaths has also been attributed to CUP in the NT Indigenous population,39 where similar factors are likely to have contributed. CUP also was more common among cases where histological categories were ill defined (i.e. NOS and Epithelial NOS categories), possibly due to a less intensive investigation of these cases.

Female cancer cases had a higher proportion of CUP than male cases, although incidence rates for CUP and invasive cancers of known organ site were both lower for females than males.37 Meanwhile, the proportion of cases with a NOS histology among CUP cases was higher among older cases, non-metropolitan cases, low socio-economic groups, and cases diagnosed in the early years of the registry (i.e. 1977-80). Again, this may reflect a lower intensity of investigation of CUP for these cases. For reasons unknown, male CUP cases were more likely than female CUP cases to be assigned to the NOS histological category.

By definition, CUP cases had distant metastases. It would have been desirable to undertake an additional comparison of their socio-demographic and histological characteristics with those of cancers of known primary site with distant metastases, but metastatic stage was not recorded on the Cancer Registry. When we repeated the comparative analysis, limiting the data to cancer deaths (i.e. using death as a proxy for advanced stage), similar findings presented. Again, CUP deaths comprised a higher proportion of cancer deaths in older patients, females, Indigenous patients, and in 1981-2004 than 1977-80. After adjusting for these characteristics, the distribution of CUP deaths by histological category was similar to that reported for incidence data.

Survival from CUP was higher for cases diagnosed in 1989-2004 than in earlier periods and a trend towards higher survival was suggested across the 1989-2004 period. This may reflect an increased clinical attention to CUP and gains in response to platinum-based chemotherapy and other systemic care.13–17 The trend for lower CUP survivals among older cases, Indigenous patients, and lower socio-economic groups may reflect less intensive care due to higher co-morbidity and poorer access to service.21,38 These results and the lower survivals for male than female CUP cases is consistent with outcome differences observed for many cancers of known organ site.21,29,38 One reason for better outcomes in women may be the greater contribution to CUP in women from cancers that are relatively sensitive to systemic therapies, such as breast cancer.32

The clinical detail available to us in this study was limited, preventing analyses of sub-groups of CUP of interest, such as metastatic squamous cell carcinomas affecting lymph nodes of the neck and groin; peritoneal carcinomas in females (generally assumed to be ovarian like); adenocarcinomas of the female axilla (generally assumed to be breast cancer); osteoblastic bone metastases in males (generally regarded as prostate cancer); small-cell and other tumours with neuroendocrine differentiation; and mediastinal or retroperitoneal masses in young males with raised AFP or HCG (usually treated as germ cell tumours).32 As these sub-sets of cancers may benefit from systemic treatment, the extent of their representation in CUP could impact on outcomes.32

This study provides a broad overview, which should be complemented with more in-depth focused research of CUP epidemiology and related health-service provision in Australia. Also, further immunohistochemical characterisation, particularly of CUP adenocarcinomas, has suggested additional therapeutic targets such as epidermal growth factor receptor (EGFR) pathway.40 The South Australian Registry data, allied with clinical databases,17 provide an opportunity to define retrospectively the prognostic significance of EGFR mutations and overexpression in CUP adenocarcinomas, which is no longer possible in the study of Massard et al.40 In addition, further clinical research is warranted to assess the benefits of more intensive exploration of primary sites of specified sub-groups of CUP and other treatment strategies.

Conclusion

Results point to improvements in survival from CUP, despite continuing poor outcomes. They identify socio-demographic groups with higher incidence of CUP and population groups at higher risk of poorer outcomes of care, who warrant further research and increased clinical attention.

Acknowledgements

Staff members of the South Australian Cancer Registry are thanked for the care taken over so many years in the collection of data used in this study.

Ancillary