Saw Swee Hock School of Public Health, National University of Singapore, 28 Medical Drive, Singapore, 11745, Singapore

Department of Surgery, National University Hospital,Yong Loo Lin School of Medicine, 16 Medical Drive, Singapore, 117597, Singapore

Imaging Division, University Medical Center Utrecht, Utrecht, The Netherlands

Health Promotion Board, Ministry of Health, 3 Second Hospital Avenue, Singapore, 168937, Singapore

Geneva Cancer Registry, Institute for Social and Preventive Medicine, University of Geneva, Geneva, Switzerland

Abstract

Background

Prostate cancer is the most commonly diagnosed malignancy in men in Sweden and Geneva, and the third most common in men in Singapore. This population-based study describes trends in the incidence and mortality rates of prostate cancer in Singapore, Sweden and Geneva (Switzerland) from 1973 to 2006 and explores possible explanations for these different trends.

Methods

Data from patients diagnosed with prostate cancer were extracted from national cancer registries in Singapore (n = 5,172), Sweden (n = 188,783) and Geneva (n = 5,755) from 1973 to 2006. Trends of incidence and mortality were reported using the Poisson and negative binomial regression models. The age, period and birth-cohort were tested as predictors of incidence and mortality rates of prostate cancer.

Results

Incidence rates of prostate cancer increased over all time periods for all three populations. Based on the age-period-cohort analysis, older age and later period of diagnosis were associated with a higher incidence of prostate cancer, whereas older age and earlier period were associated with higher mortality rates for prostate cancer in all three countries.

Conclusions

This study demonstrated an overall increase in incidence rates and decrease in mortality rates in Singapore, Sweden and Geneva. Both incidence and mortality rates were much lower in Singapore. The period effect is a stronger predictor of incidence and mortality of prostate cancer than the birth-cohort effect.

Background

Prostate cancer is the most commonly diagnosed malignancy in men in Sweden

**Country**

**Period**

**Incidence rate**

**Mortality rate**

**(per 100,000 person years)**

**(per 100,000 person years)**

**Total**

**Rate**

**Total**

**Rate**

* Period for Sweden prostate mortality 2003–2005.

Singapore

1973-1977

144

5.1

35

2.0

1978-1982

240

6.8

69

3.4

1983-1987

356

8.0

106

4.1

1988-1992

529

9.6

187

5.7

1993-1997

903

13.9

303

7.7

1998-2002

1357

17.6

434

9.3

2003-2006

1643

23.2

443

6.1

Sweden

1973-1977

17359

44.9

8516

21.6

1978-1982

19202

46.0

8400

19.5

1983-1987

22372

50.6

8795

18.9

1988-1992

25329

55.4

10341

20.9

1993-1997

28985

63.0

11411

21.6

1998-2002

37313

84.5

12287

21.4

2003-2006

38223

109.2

7625*

21.2*

Geneva

1973-1977

407

36.5

249

22.2

1978-1982

524

42.1

270

21.1

1983-1987

585

43.1

289

19.9

1988-1992

724

50.0

317

19.7

1993-1997

955

63.9

316

18.2

1998-2002

1345

86.4

279

14.5

2003-2006

1215

89.3

253

14.4

**Goodness of fit**

**Singapore**

**Sweden**

**Geneva**

**(Negative binomial)**

**(Negative binomial)**

**(Negative binomial)**

^{*} Residual deviance.

^{#} Degree of freedom.

^{§} Akaike information criterion.

**Model**

**Res Dev**
^{
*
}

**Df**
^{
#
}

**AIC**
^{§}

**P-value**

**Res Dev**

**DF**

**AIC**

**P-value**

**Res Dev**

**DF**

**AIC**

**P-value**

**Age (A)**

31.5

42

473

0.882

24.4

42

817

0.632

38.3

42

500

0.986

**Age-Drift (AD)**

7.2

41

451

0.999

6.6

41

802

0.999

14.6

41

478

0.999

**Age-Period (AP)**

5.6

36

460

0.999

3.7

36

809

0.999

12.1

36

485

0.999

**Age-Cohort (AC)**

6.4

30

472

0.999

2.9

30

820

0.999

6.1

30

491

0.999

**Full APC**

5.0

25

481

0.999

2.0

25

829

0.999

5.1

25

500

0.999

**Likelihood ratio test**

**Model**

**Dev**

**DF**

**P-value**

**Dev**

**DF**

**P-value**

**Dev**

**DF**

**P-value**

**(H**
_{
0
}**vs H**
_{
1
}**)**

**AP vs APC**

0.6

11

0.999

1.7

11

0.999

7.0

11

0.800

**AC vs APC**

1.4

5

0.925

0.9

5

0.972

1.0

5

0.959

**Goodness of fit**

**Singapore**

**Sweden**

**Geneva**

**(Negative binomial)**

**(Negative binomial)**

**(Poisson)**

^{*} Residual deviance.

^{#} Degree of freedom.

^{§} Akaike information criterion.

**Model**

**Res Dev**
^{
*
}

**Df**
^{
#
}

**AIC**
^{§}

**P-value**

**Res Dev**

**DF**

**AIC**

**P-value**

**Res Dev**

**DF**

**AIC**

**P-value**

**Age (A)**

13.5

42

402

0.999

1.1079

42

641

0.999

72.0

42

282

0.003

**Age-Drift (AD)**

13.3

41

404

0.999

1.1077

41

643

0.999

40.4

41

252

0.495

**Age-Period (AP)**

8.7

36

409

0.999

0.71

36

653

0.999

32.4

36

254

0.642

**Age-Cohort (AC)**

6.9

30

419

0.999

0.50

30

665

0.999

34.1

35

263

0.512

**Full APC**

5.0

25

428

0.999

0.28

25

674

0.999

27.6

30

267

0.594

**Likelihood ratio test**

**Model (H**
_{
0
}**vs H**
_{
1
}**)**

**Dev**

**DF**

**P-value**

**Dev**

**DF**

**P-value**

**Dev**

**DF**

**P-value**

**AP vs APC**

3.7

11

0.979

0.43

11

0.999

4.8

6

0.570

**AC vs APC**

1.9

5

0.869

0.22

5

0.999

6.5

5

0.258

PSA testing received US Food and Drug Administration approval as a monitor for treatment response in 1986 and was subsequently approved as a screening aid for diagnosis in 1994

Despite increasing incidence rates globally, mortality rates of prostate cancer have declined in several countries

Methods

Data source

Incidence and mortality data (1973–2006)

Singapore

Comprehensive population-based cancer registration in Singapore began in January 1968, with the aim of providing current information on cancer patterns and trends. Data was obtained from different sources, namely notifications by physicians, pathology records, hospital records, and death certificates. Population based data for incidence and mortality rates for Singapore was taken from the Singapore Cancer Registry, National Registry of Diseases Office (NRDO) (n = 5,172). The denominators for incidence and mortality were the total number of person-years from the Singapore resident population based on the Singapore Population Census 2000 updated reports

Sweden

Sweden was selected as a country of comparison as it has one of the highest incidence rates of prostate cancer globally

Geneva

Geneva was selected for comparison due the availability of a complete dataset for analysis. Population based data on incidence and mortality for Geneva was used over the period 1973 to 2006. Data on incident cases and deaths were obtained from the Geneva Cancer Registry (Institute for Social and Preventive Medicine, Geneva University, Geneva, Switzerland) and the ‘Office Cantonal de la Population’ respectively (n = 5,755). The Geneva Cancer Registry recorded all incident cancers occurring in the population of the canton (approximately 465,000 inhabitants in 2007) since 1970. Information was collected from various sources

Statistical analysis

Incidence rates and mortality rates were defined as the number of new cases and the number of deaths, respectively, in a given period for a specific population. To adjust for the differences in the age structure and to account for the strong influence of age on the risk of cancer, the incidence rates and mortality rates were age-standardised to the world standard million population and expressed as 100,000 person years for persons aged 45 years and above. Data were analysed using five year age groups (45–49, 50–54, …70-74, 75+ years) for all three countries. As those aged 45 years and above constituted only 26% in the world standard million population, we recalibrated the weighting to 100% based on the respective weights of each age group.

The number of incident cases and deaths were assumed to follow a Poisson distribution and hence we used log-linear Poisson modeling of cases based on age and year of diagnosis (period) and year of birth (birth-cohort). When over-dispersion (variance greater than the mean) was detected while modeling the incidence or mortality, a negative binomial distribution was used

A limitation of using only age-standardised rates to describe the trends in incidence and mortality rates is that this does not account for period and birth-cohort effects. Use of the age-period (AP) and age-birth-cohort (AC) models was applied to disentangle the separate effects of period of diagnosis and birth-cohort on incidence and mortality. Period effects tend to influence all individuals simultaneously during a particular time period regardless of their age, whilst birth-cohort effects are attributable to certain factors related to the birth year. The analysis was based on the generalized linear model approach. We also considered the scenario where the effects of period and birth-cohort in AP and AC models, respectively, were assumed to be linear and hence inseparable. In such cases, the combined linear model is referred to as the age-drift model

The deviance statistic was used to assess the goodness of fit of the models. A non-significant p-value (> 0.05) indicated a good fit. Difference in deviance and the likelihood ratio test were used to compare different models, where a significant p-value indicates that the more complicated model has significant improvement over the simpler model. The deviance and change in deviance due to cohort or period effects were compared to the age-period-cohort (APC) model. The cohort effect was tested in comparison to the AP and APC models and likewise, the period effect was tested in comparison to the AC and APC models.

The Akaike Information Criterion (AIC) was also calculated to compare models with different complexities that are not required to be nested within each other, with smaller AIC values suggestive of a better model fit

Ethical approval for this study was granted by the National University of Singapore IRB (approval number: NUS-747).

Results

Incidence

Overall, incidence rates increased during the period 1973 to 2006 (Figure

Age-standardised (world standard million population) prostate cancer incidence rates per 5-year period (1973–2006) stratified by country for Singapore, Sweden and Geneva.

**Age-standardised (world standard million population) prostate cancer incidence rates per 5-year period (1973–2006) stratified by country for Singapore, Sweden and Geneva.**

Higher age-specific incidence rates were found in older age-groups and increased in the later years of diagnosis for all three countries (Figure

Age-specific prostate cancer incidence rates (per 100,000 person-years) per 5-year period stratified by 5-year age group for Singapore, Sweden and Geneva.

**Age-specific prostate cancer incidence rates (per 100,000 person-years) per 5-year period stratified by 5-year age group for Singapore, Sweden and Geneva.**

Mortality

Overall, Singapore had the lowest age-standardised mortality rates over all periods measured (Figure

Age-standardised (world standard million population) prostate cancer mortality rates per 5-year period (1973–2006) stratified by country for Singapore, Sweden and Geneva.

**Age-standardised (world standard million population) prostate cancer mortality rates per 5-year period (1973–2006) stratified by country for Singapore, Sweden and Geneva.**

Higher age-specific mortality rates were found in the older age groups for all three countries (Figure

Age-specific prostate cancer mortality rates per 5-year (per 100,000 person-years) period stratified by 5-year age group.

**Age-specific prostate cancer mortality rates per 5-year (per 100,000 person-years) period stratified by 5-year age group.**

APC modeling

The negative binomial distribution was preferred based on the goodness of fit test when modeling incidence in all three countries and mortality for Singapore and Sweden. The Poisson distribution was preferred when modeling for mortality for Geneva. The likelihood ratio test for incidence rates in all countries indicated that both the AP model and the AC models were sufficient in explaining the variation in the full APC model for Singapore (P-value = 0.999 (AP vs APC), 0.925 (AC vs APC)), Sweden (P-value = 0.999, 0.975) and Geneva (P-value = 0.800, 0.959) (Table

Similarly, the likelihood ratio test for mortality rates indicated that both the AP model and AC model were also sufficient in explaining the variation in the full APC model for Singapore (P-value = 0.979, 0.869), Sweden (P-value = 0.999, 0.999) and Geneva (P-value = 0.570, 0.258) (Table

Discussion

In general, the age-standardised incidence rates of prostate cancer above the age of 50 years in all three countries increased between 1973 and 2006, and occurred at a faster rate in Sweden and Geneva than in Singapore. The sharper rise occurring in Sweden after the early 1990s is consistent with the increasing availability of PSA testing

As the upper limit for screening in Singapore

The age-standardised mortality rates declined in the later periods for all three countries. From our results, mortality rates for Geneva declined steadily from 1973 onwards whereas for Singapore there was a steady rise in mortality rates from 1968 to 1992 with a recent decline from 1993 onwards. This raises the question of whether this could be an effect of PSA testing. To date there is little conclusive evidence that PSA-based screening reduces prostate cancer mortality. However, recent randomised controlled trials have shown contradicting results. The Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial in the US showed a marginal increase in the incidence and concluded that there was no mortality reduction with combined PSA and digital rectal examination screening over an 11 year median follow-up

Treatment of prostate cancer does not differ greatly between the three countries yet the mortality rate is declining much faster in Singapore and Geneva compared to Sweden. It is uncertain whether this observation can be explained by genetic differences in the populations, different environmental factors or a combination of both (gene-environment interactions)

From the APC modeling, the incidence and mortality rates of prostate cancer appeared to be more strongly associated with the age and period effect than with the birth-cohort effect in all three countries, based on the lower calculated AIC criteria and likelihood ratio test results. Older males tended to be at higher risk of developing prostate cancer and in the later periods. In addition, mortality was higher in older males and across the earlier periods, which could be confounded by PSA testing. Males of older age and in the later period of diagnosis had a higher incidence, whereas males of older age and in the earlier period had higher mortality rates for prostate cancer.

A limitation of this study was that it extended over a relatively long time period during which changes in diet, environmental and diagnostic factors are likely to have occurred. It would be a further challenge to identify the independent factors influencing the change in the trends of incidence and mortality as it could be due to individual factors acting independently or in combination. Age-standardized rates for incidence and mortality in our study are not comparable with other published data as different weightings were used. Accurate interpretation of the incidence and mortality trends in the three countries would be incomplete without data on PSA screening as a potential confounder. As data on individual PSA testing was not available, it was not possible to separate the effect of the real increment in incident cases from over-diagnosis due to increased screening, as we would need to establish whether patients had PSA screening prior to diagnosis.

Conclusion

Our analysis showed that overall age-standardised incidence rates of prostate cancer increased over the period 1973 to 2006 and that the mortality rates declined over the later period (1998 to 2006) in all three countries. Both incidence and mortality rates were much lower in Singapore than in Sweden and Geneva. The mortality rates for Singapore followed an inverted U-shape whereas the mortality rates for Sweden remained relatively unchanged and Geneva experienced a steady decline over the period 1973 to 2006. Based on APC modeling, the age and period effects were shown to be more strongly associated with incidence and mortality than the birth cohort effect.

Competing interests

The authors declare that they have no competing interests.

Authors’ contributions

CC carried out all statistical analysis and drafted the manuscript. NN and QY were involved in drafting and critical review of the manuscript. MH was involved in drafting the manuscript and provided the Swedish population data. HMV drafted the manuscript and provided the population data from Geneva. EYL provided the Singapore population data. CB provided the Geneva population data and was involved in manuscript writing. KSC conceived the study and provided critical review of the manuscript. SEC conceived the study, and participated in the study design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript.

Acknowledgements

We thank the Singapore Cancer Registry, Geneva Cancer Registry and NORDCAN for permission to use their data. This study has been approved by the National University of Singapore Institutional Review Board (Reference code 09–097).

Pre-publication history

The pre-publication history for this paper can be accessed here: