The age-specific prostate cancer incidence between 1975 and 2004 was obtained from the MB1 series published by the Office of Population Censuses and Surveys (OPCS) until 1996 and the Office of National Statistics (ONS) thereafter. These volumes cover cancer registrations for both England and Wales until 1994, and England only from 1995 onwards. The validity of these data has been established for comparisons of cancer risk over time .
Age-specific prostate cancer mortality in England and Wales between 1975 and 2004 was obtained from the ONS Series DH2, Mortality statistics: Cause. The process of collecting and coding death registrations changed over the study period. These changes included: updates of International Classification of Diseases (ICD) codes from ICD-8 to ICD-9 in 1979 and from ICD-9 to ICD-10 in 2001 [23,24], introduction of automated cause of death coding in 1993  and changes to the interpretation of WHO Rule 3 for selecting underlying cause of death in 1984, 1993 [26,27] and 2001 . Rule 3 allows a condition reported in either Part I or Part II of the death certificate to take precedence over a condition selected using other coding rules if the latter is obviously a direct consequence of that condition. Between 1984 and 1992, a revised interpretation of WHO Rule 3 was introduced by the OPCS. Consequently, deaths from causes such as pneumonia declined steeply in 1984, whereas deaths from causes often mentioned in part II of the certificate increased . This change resulted in an increase in the death rate from prostate cancer in 1984, which was most marked in the elderly. The change in 1993 was a move back to the internationally accepted interpretation of Rule 3 operating in England and Wales before 1984. Under ICD-10 adopted in January 2001, the interpretation of Rule 3 is similar to that adopted by the OPCS for deaths in 1984–92. A bridge-coding exercise showed that for cancers coded by ICD-9 Rule 3 between 1993 and 2000, the application of ICD-10 Rule 3 would have selected prostate cancer more often as the underlying cause of death [23,24], the ratio of the number of deaths coded to prostate cancer using ICD-10 compared with applying ICD-9 rules being 1.008, 1.031 and 1.358 at ages <75, 75–84 and ≥85 years, respectively . The influence of these procedural changes on the mortality data is investigated by applying the multipliers 1.008 (for deaths in men aged <75 years) and 1.031 (for deaths >75 years) to the data between 1993 and 2000, to give an expected number of deaths that would have been coded to prostate cancer in ICD-10.
The Hospital Episode Statistics (HES) database for England holds information on the care provided to those admitted to NHS hospitals and for NHS hospital patients treated elsewhere. This includes details of surgical procedures carried out, coded using OPCS-4 codes. HES records were extracted from the database held by the Department of Social Medicine, Bristol using the OPCS procedure codes M 61 (RP) and N051, N052, N061, N063 (orchidectomy) when the underlying diagnosis was prostate cancer (ICD9, 185; and ICD10, C61). This information was available from 1991 to 2004.
Data on overall prescribing in England and Wales (1975–2004) of hormonal therapy for prostate cancer (i.e. oestrogens, LHRH analogues and antiandrogens) were obtained from Intercontinental Medical Statistics (IMS) Health Medical Data Index . Age-specific prescribing data were not available. Since 1967, IMS Health has collected quarterly data on drug prescribing in the UK. A prescription is defined as every drug item on a prescription form given as a result of a consultation. Since 1994, anonymized prescribing data have been collected electronically every day from a stratified sample of 500 GPs, giving a total of 26 000 doctor-weeks per year. Sample data are projected to the whole of the UK, weighted by a regional factor, and the figures adjusted to reflect the total number of prescriptions dispensed in the UK as indicated by data published by the UK Prescription Pricing Authority.
Age-specific rates for prostate cancer incidence and mortality, and RP, were estimated for the age groups <55, 55–74 and ≥75 years using the year-by-year information on the mid-year resident population of England and/or Wales, as appropriate, from 1975 to 2004 as provided by the Population Estimates Unit of the ONS. The mid-year population of men resident in England only was used for calculating age-specific incidence rates from 1995 to 2004, and age-specific rates for orchidectomy and RP from 1991 to 2004. The population of men aged ≥55 years in England in 2004 was 6.26 million, compared to 0.40 million in Wales, so if trends differ in the two countries this would only have a minor impact on the results. Unless otherwise noted, all rates are expressed per 100 000 population per year.
Analysis of prostate cancer incidence and mortality trends was conducted by join-point regression, in which trend data are described by a number of contiguous linear segments and ‘join points’ (points at which trends change). Join-point regression was used to estimate the annual change in incidence and mortality rates and the number and location of join points . Models were based on linear regression, with incidence and mortality rates as the dependent variables and year as the independent variable. To identify the best-fitting combination of line segments and join points, a series of permutation tests was used, first testing the null hypothesis (Ho= no join points) vs the alternative hypothesis (Ha= three join points). Hypothesis testing proceeded sequentially, increasing the number of join points under Ho by one if the null hypothesis was rejected and decreasing the number of join points under the alternative hypothesis if Ho was accepted. The maximum number of join points tested was three in each analysis. For each model, the locations (i.e. years and 95% CI) of the best-fitting join points were identified using a grid search algorithm . A Bonferonni correction was applied by conducting each test at the α/3 level, ensuring that the probability of a type I error (i.e. concluding that there are one or more join points when there are in fact none) was at most 0.05. Analyses were conducted using Joinpoint software (version 3 April 2005) made available by the National Cancer Institute (srab.cancer.gov/joinpoint).