Introduction. Screening for prostate cancer (PCa) through Prostate-Specific Antigen (PSA) test widespread in high-income countries long before definitive results about its efficacy. Actually, even though PSA testing do contribute to reduce PCa mortality, the harm to benefit ratio remains controversial, given that PSA testing can detect cancers that may otherwise go undiagnosed during a man’s life (i.e., overdiagnosis and, consequently, overtreatment with severe side effects). To evaluate the impact of PSA testing diffusion in the Friuli Venezia Giulia (FVG) region, population-based methods for estimating and projecting cancer morbidity and mortality indicators were applied to PCa and results were compared to observed data. Taking advantage of the complete coverage of the FVG population of both the regional cancer registry and the digital health archive, data on PCa cases and on PSA testing use were concurrently analyzed. Estimates of prostate cancer burden in Friuli Venezia Giulia. Methods. The Mortality and Incidence Analysis Model (MIAMOD) (Verdecchia, 1989) is a regression of mortality on observed mortality data (from official statistics) to back-calculate age-period-cohort incidence. The MIAMOD method was applied for estimating and projecting incidence, mortality, and prevalence for selected cancers in FVG region in the period 1970-2015 (Zucchetto et al, 2013). For PCa, a specific procedure was used to better capture recent variations: preliminary mortality estimation up to 2010 was performed using regional mortality data for the period 1970-2007; relative survival estimates were calculated using data from EUROCARE-4 in the period 1985-2002 and modeled by means of mixture cure models of the Weibull type with power function at macro-area level (North-East) for the period 2003-2005, and then assumed to be constant. Results. MIAMOD estimates showed a high goodness of fit with the observed incidence for all cancer sites, except for PCa. Although both increasing with time, PCa incidence rates estimates were much lower than observed incidence rates, especially in the period 1996-2007 and they seemed to converge in 2008-2009. Conversely, in the same period, mortality rates were almost stable. Analysis of PSA testing rates and PCa incidence rates trends in Friuli Venezia Giulia. Methods. Data on PSA tests performed in men aged 40+ years were retrieved from FVG digital health archive for the period 1998-2012. The overall PSA testing rates were calculated as the number of tested men each year (multiple prescriptions to the same man were counted once) over the male population. PCa incidence rates (from the FVG cancer registry) among men aged 40+ years or more were analyzed for the period 1995-2009. Joinpoint regression analysis was performed to identify knots where a statistically significant change over time in the log-slope of the rates occurred. Age-period-cohort analyses were also performed. In order to solve the unidentifiability problem, the following assumptions have been made (based on preliminary plots of rates by periods, birth cohorts, and age classes): a period was selected as reference; cohort effects were constrained to be 0 on average with 0 slope. Age effects represented age-specific rates in the reference period, after adjustment for the cohort effects; period effects were interpretable as rate ratios (RRs) relative to the reference period; cohort effects were interpretable as residual RRs relative to the age-period prediction. Results. PSA testing increased from 12,792 per 100,000 men in 1998, up to 30,407 in 2009, and then slightly decreased. Significant changes emerged in 2002 and 2008, with a high increase of rates in the period 1998-2002 (annual percent change, aPC=16.9; 95% confidence interval, CI: 12.9 to 21.2), a smaller increase in the period 2002-2008 (aPC=3.6; 95% CI: 1.7 to 5.5), and a subsequent stabilization (aPC= -0.7; 95% CI: -3.1 to 1.8). Similar patterns emerged by age strata. Compared to reference period 1998-2002, PSA testing rates were significantly higher in 2003-2007 and stabilized thereafter. Age effects indicated sharp increasing rates up to age 70-74 years and then a reduction. No particular cohort effects emerged, except for a tendency of more recently born men to undergo PSA testing. Age-drift was equal to 4.7% (95% CI: 4.7% to 4.7%). The overall crude incidence rate of PCa increased from 219.8 per 100,000 men in 1995, up to 385.5 in 2007, and then decreased down to 328.3 in 2009. Joinpoint analysis estimated statistically significant changes in PCa incidence rates log-slopes in 1998 and 2007: the aPCs in the periods 1995-1998, 1998-2007, and 2007-2009 were 12.1 (95% CI: 6.6 to 17.9), 1.9 (95% CI: 0.9 to 2.8), and -7.0 (95% CI: -14.3 to 0.9), respectively. The results of age-period-cohort analysis with natural splines showed, as compared to the reference period 1995-1999, higher incidence rates in the subsequent 5-year period and then a plateau (the analysis performed using 1-year time spans, highlighted an increase up to 2007 followed by a reduction). Age-specific PCa incidence rates were sharply increasing up to 70-75 years and then reduced. Residual cohort effects highlighted high increasing risks for the most recently born men (i.e., after 1950). The age-drift was equal to 2.3% (95% CI: 1.9%-2.7%). Discussion. The diffusion of PSA testing in FVG has inflated the incidence of PCa without affecting the overall mortality. Given that MIAMOD estimates are based on mortality data, which have not been so heavily modified by the introduction of PSA test as PCa incidence rates, this could explain the difference between MIAMOD estimates and observed incidence rates, especially in the period 1996-2007. Though the period was not totally coincident, trend over time of PCa incidence rates resembled to some extent those of PSA-testing rates and the age, period, and cohort effects were somewhat similar (though PCa incidence was more affected by cohort effects). The PCa incidence estimates produced using MIAMOD could be considered as the rates that would be observed in FVG in the absence of such a great increase of PSA testing use (observed since 1998, but reasonably started before). Therefore, the difference between observed and estimated incidence rates (ranging between 7% and 18% in the period 1996-2007, median 11%), could be attributable to screening with PSA and, reasonably, to overdiagnosis. These results are in line with estimates of overdiagnosis derived from micro-simulation models based on randomized trials results. Conclusion. Estimates of PCa incidence and prevalence based on mortality data should be carefully evaluated taking into account of the age, period, and cohort trends of PSA testing data, which are available in several areas, including those not yet covered by cancer registration.
THE IMPACT OF THE WIDESPREAD SCREENING THROUGH PROSTATE-SPECIFIC ANTIGEN (PSA) TEST ON POPULATION-BASED MODELS FOR ESTIMATING PROSTATE CANCER BURDEN
ZUCCHETTO, ANTONELLA
2015
Abstract
Introduction. Screening for prostate cancer (PCa) through Prostate-Specific Antigen (PSA) test widespread in high-income countries long before definitive results about its efficacy. Actually, even though PSA testing do contribute to reduce PCa mortality, the harm to benefit ratio remains controversial, given that PSA testing can detect cancers that may otherwise go undiagnosed during a man’s life (i.e., overdiagnosis and, consequently, overtreatment with severe side effects). To evaluate the impact of PSA testing diffusion in the Friuli Venezia Giulia (FVG) region, population-based methods for estimating and projecting cancer morbidity and mortality indicators were applied to PCa and results were compared to observed data. Taking advantage of the complete coverage of the FVG population of both the regional cancer registry and the digital health archive, data on PCa cases and on PSA testing use were concurrently analyzed. Estimates of prostate cancer burden in Friuli Venezia Giulia. Methods. The Mortality and Incidence Analysis Model (MIAMOD) (Verdecchia, 1989) is a regression of mortality on observed mortality data (from official statistics) to back-calculate age-period-cohort incidence. The MIAMOD method was applied for estimating and projecting incidence, mortality, and prevalence for selected cancers in FVG region in the period 1970-2015 (Zucchetto et al, 2013). For PCa, a specific procedure was used to better capture recent variations: preliminary mortality estimation up to 2010 was performed using regional mortality data for the period 1970-2007; relative survival estimates were calculated using data from EUROCARE-4 in the period 1985-2002 and modeled by means of mixture cure models of the Weibull type with power function at macro-area level (North-East) for the period 2003-2005, and then assumed to be constant. Results. MIAMOD estimates showed a high goodness of fit with the observed incidence for all cancer sites, except for PCa. Although both increasing with time, PCa incidence rates estimates were much lower than observed incidence rates, especially in the period 1996-2007 and they seemed to converge in 2008-2009. Conversely, in the same period, mortality rates were almost stable. Analysis of PSA testing rates and PCa incidence rates trends in Friuli Venezia Giulia. Methods. Data on PSA tests performed in men aged 40+ years were retrieved from FVG digital health archive for the period 1998-2012. The overall PSA testing rates were calculated as the number of tested men each year (multiple prescriptions to the same man were counted once) over the male population. PCa incidence rates (from the FVG cancer registry) among men aged 40+ years or more were analyzed for the period 1995-2009. Joinpoint regression analysis was performed to identify knots where a statistically significant change over time in the log-slope of the rates occurred. Age-period-cohort analyses were also performed. In order to solve the unidentifiability problem, the following assumptions have been made (based on preliminary plots of rates by periods, birth cohorts, and age classes): a period was selected as reference; cohort effects were constrained to be 0 on average with 0 slope. Age effects represented age-specific rates in the reference period, after adjustment for the cohort effects; period effects were interpretable as rate ratios (RRs) relative to the reference period; cohort effects were interpretable as residual RRs relative to the age-period prediction. Results. PSA testing increased from 12,792 per 100,000 men in 1998, up to 30,407 in 2009, and then slightly decreased. Significant changes emerged in 2002 and 2008, with a high increase of rates in the period 1998-2002 (annual percent change, aPC=16.9; 95% confidence interval, CI: 12.9 to 21.2), a smaller increase in the period 2002-2008 (aPC=3.6; 95% CI: 1.7 to 5.5), and a subsequent stabilization (aPC= -0.7; 95% CI: -3.1 to 1.8). Similar patterns emerged by age strata. Compared to reference period 1998-2002, PSA testing rates were significantly higher in 2003-2007 and stabilized thereafter. Age effects indicated sharp increasing rates up to age 70-74 years and then a reduction. No particular cohort effects emerged, except for a tendency of more recently born men to undergo PSA testing. Age-drift was equal to 4.7% (95% CI: 4.7% to 4.7%). The overall crude incidence rate of PCa increased from 219.8 per 100,000 men in 1995, up to 385.5 in 2007, and then decreased down to 328.3 in 2009. Joinpoint analysis estimated statistically significant changes in PCa incidence rates log-slopes in 1998 and 2007: the aPCs in the periods 1995-1998, 1998-2007, and 2007-2009 were 12.1 (95% CI: 6.6 to 17.9), 1.9 (95% CI: 0.9 to 2.8), and -7.0 (95% CI: -14.3 to 0.9), respectively. The results of age-period-cohort analysis with natural splines showed, as compared to the reference period 1995-1999, higher incidence rates in the subsequent 5-year period and then a plateau (the analysis performed using 1-year time spans, highlighted an increase up to 2007 followed by a reduction). Age-specific PCa incidence rates were sharply increasing up to 70-75 years and then reduced. Residual cohort effects highlighted high increasing risks for the most recently born men (i.e., after 1950). The age-drift was equal to 2.3% (95% CI: 1.9%-2.7%). Discussion. The diffusion of PSA testing in FVG has inflated the incidence of PCa without affecting the overall mortality. Given that MIAMOD estimates are based on mortality data, which have not been so heavily modified by the introduction of PSA test as PCa incidence rates, this could explain the difference between MIAMOD estimates and observed incidence rates, especially in the period 1996-2007. Though the period was not totally coincident, trend over time of PCa incidence rates resembled to some extent those of PSA-testing rates and the age, period, and cohort effects were somewhat similar (though PCa incidence was more affected by cohort effects). The PCa incidence estimates produced using MIAMOD could be considered as the rates that would be observed in FVG in the absence of such a great increase of PSA testing use (observed since 1998, but reasonably started before). Therefore, the difference between observed and estimated incidence rates (ranging between 7% and 18% in the period 1996-2007, median 11%), could be attributable to screening with PSA and, reasonably, to overdiagnosis. These results are in line with estimates of overdiagnosis derived from micro-simulation models based on randomized trials results. Conclusion. Estimates of PCa incidence and prevalence based on mortality data should be carefully evaluated taking into account of the age, period, and cohort trends of PSA testing data, which are available in several areas, including those not yet covered by cancer registration.File | Dimensione | Formato | |
---|---|---|---|
phd_unimi_R09744.pdf
Open Access dal 23/06/2016
Dimensione
1.54 MB
Formato
Adobe PDF
|
1.54 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/81854
URN:NBN:IT:UNIMI-81854