- Email: [email protected]
Trends in incidence of and mortality from invasive cancer of the uterine cervix in Scotland (1975–1994)
Public Health (1998) 112, 373±378 ß R.I.P.H.H. 1998 http://www.stockton-press.co.uk/ph
Trends in incidence of and mortality from invasive cancer of the uterine cervix in Scotland (1975 ± 1994) JJ Walker1, D Brewster2, A Gould2 and GM Raab1 Department of Mathematics, Napier University, Edinburgh; 2Information and Statistics Division, National Health Service in Scotland, Edinburgh 1
Objective: I. To identify major trends in the incidence of and mortality from invasive cancer of the cervix uteri in Scotland during the twenty year period 1975 ± 1994; II. to consider the extent to which these trends may have been shaped by the introduction of systematic cervical screening. Design: Analysis of annual age standardised and age speci®c rates for incidence and mortality, based on data collected by the Scottish Cancer Registry and the General Register Of®ce for Scotland. Setting: Scotland. Subjects: Women registered with the Scottish Cancer Registry as having developed invasive cancer of the cervix during the period of interest. Results: Annual all ages incidence rates of invasive cervical cancer show little overall change over the period 1975 ± 1989, but exhibit a pronounced decline from 1990 onwards. All-ages mortality rates show clear evidence of decline during the period 1975 ± 1994, the rate for 1994 being some 30% lower than that for 1975. Annual age-speci®c incidence rates show different patterns by age group, with clear evidence of decreasing trends in the age range 50 ± 64 years but different patterns in younger and older age groups. Most age groups show steep declines in incidence from 1990 onwards. Age speci®c mortality rates for 1975 ± 1994 exhibit the most pronounced decreasing trends in the age range 50 ± 64 years. The trends identi®ed are broadly similar to those experienced in England and Wales over an approximately comparable period. Conclusions: The overall (all ages) incidence of invasive cervical cancer in Scotland changed little during the period 1975 ± 1989, but declined sharply from 1990 onwards. The most pronounced decline in incidence across the period 1975 ± 1994 appears to have taken place in the age range 50 ± 64 years. This decline has been accompanied by a commensurate fall in mortality in the same age range. These reductions in incidence and mortality may be attributable in part to increased coverage of cervical screening programmes during the period of interest. Evidence from other studies suggest that, without the increased coverage of cervical screening achieved during this period incidence rates in Scotland might have been seen to increase. Keywords: cervical cancer; incidence; mortality; trends; Scotland
Introduction Invasive cancer of the uterine cervix accounts for some 15% of all cancers among women.1 The scale of incidence of and mortality from the disease quali®es it to be considered as a major international public health issue Ð the total number of new cases worldwide in 1980 was estimated to be in excess of 460 000.1 From a more immediate perspective, an average of almost 450 new cases of the disease were registered in Scotland annually during the period 1981 ± 1990, and more than 200 women died, on average, each year from cervical cancer in Scotland during the same period.2 Primary prevention of cervical cancer is not feasible, because the agents which cause the disease are not as yet fully understood. This is despite considerable research effort, over a period of many years, having been devoted to the problem. Factors which have been associated with development of the disease include sexual activity,3 smoking,4 viral infections of the class known as human papillomaviruses,5 use of oral contraceptives,6 and socioeconomic status.7,8 Of these risk factors, the one whose Correspondence: Mr JJ Walker, University of Edinburgh, Department of General Practice, Levinson House, 20 West Richmond Street, Edinburgh, EH8 9DX Accepted 15 June 1998
in¯uence on the development of cervical cancer is widely accepted is sexual activity Ð it is now generally acknowledged that the age at which a women ®rst engages in sexual intercourse, and the number of sexual partners she has, are signi®cant determinants of development of the disease. A substantial body of research has con®rmed the association between these sexual factors and cervical cancer.9 ± 11 In keeping with a potentially sexually transmissible agent, human papillomavirus is now generally though to be the major causal factor for most cases of cervical cancer.12 The effectiveness of cervical screening programmes based on cervical cytology in reducing incidence of and mortality from invasive cervical cancer is widely accepted.1 Although cervical screening has never been the subject of a published randomised trial, a convincing body of nonexperimental evidence points to its effectiveness, most notably the experience of a group of countries in Northern Europe which introduced organised screening programmes during the 1960s and early 1970s.13 In Scotland, the introduction of cervical screening has been a gradual process. `Opportunistic' screening Ð that is, taking cervical smears in the course of other routine GP or clinic visits Ð has been operating in some form since at least 1960, but organised mass screening based on a computerised call=recall system dates only from the late 1980s. A convenient concise history of the implementation of cervical screening in Scotland is available elsewhere.14
374
Invasive cancer uterine cervix Ð trends JJ Walker et al
In order to assess the effect of screening on the incidence of and mortality from cervical cancer, annual age standardised and age speci®c rates for the disease in Scotland during the period 1975 ± 1994 were examined. The objective was to identify signi®cant time related trends Ð if any Ð and then attempt to relate these to the expected effects of screening. Methods Cancers are recorded Ð `registered' Ð in Scotland by ®ve regional registries (based respectively in Inverness, Aberdeen, Dundee, Glasgow and Edinburgh), which are responsible for the collection of relevant data from a variety of sources: clinics, pathology laboratories, hospital records and death certi®cate records. The data collected by these regional registries are processed, collated and validated by the Scottish Cancer Registry within the Scottish Cancer Intelligence Unit (SCIU), the latter being a component of the Information and Statistics Division (ISD) of the NHS Common Services Agency based in Edinburgh. Currently about 32 000 registrations (of all types of cancer) are made annually in Scotland, and the quality of the registration data is generally high.15 Fuller details of the functions of the SCIU and ISD are available.16 The cancer registration data maintained by ISD are dynamic, in the sense that new cases are continually being added. In certain circumstances, a case may not be registered until some time after the onset of the disease Ð for example, this happens with cases who, due to limited attendance at hospital, fail to come to the attention of the registry until a record of their deaths is received. As a result, any analysis of the registration data held by the SCIU is technically valid only at the time when the data extract on which the analysis is based was prepared. In practice however, registration data for all but the most recent years can be assumed to be `complete'. For this study, registration data covering the period 1975 ± 1994 were examined, based on a data extract as at 20 June 1996. These data comprised 8447 records relating to registrations of invasive cancer of the cervix (corresponding to Rubric 180 in the World Health Organisation's International Classi®cation of Diseases, Ninth Revision, 1977). Mortality data were obtained from annual reports of the Registrar General for Scotland.
Age-standardised annual rates for incidence and mortality were calculated, standardised with respect to the European Standard Population17 using the direct method.18 Population numbers used in deriving the rates were estimated for mid-year female population totals for Scotland, obtained from the Population Statistics Branch of the General Register Of®ce for Scotland.19 Age speci®c rates for incidence and mortality were calculated as the numbers of cases or deaths divided by the estimated mid-year population for each age group. The age standardised and age speci®c rates were initially examined graphically for evidence of major trends. To provide more formal support for this purely visual approach, linear regression models were ®tted to the overall rates, with time (that is the calendar year in which the case or death was reported) speci®ed as the independent variable.18 For the age-speci®c data, Poisson regression models were ®tted in which the response variable Ð number of cases, or number of deaths Ð was modelled as dependent on (i) subject's age group; and (ii) the crossed effect, or interaction, of age group with (calendar year of reporting or death). Due to very low numbers of cases in younger women, age groups below 25 y were excluded from these models. Finally, to interpret the interactions that were found, smoothed plots of the rates by calendar time were examined.20 Results Overall age-standardised annual incidence and mortality rates for invasive cervical cancer in Scotland during the period 1975 ± 1994, together with ®tted values derived from linear regression models, are presented (Figure 1). The data for incidence show a complex mix of effects, notably: (i) rates increasing at a fairly constant rate during the period 1982 ± 88; (ii) a marked drop in 1989, relative to the preceding and succeeding years; and (iii) a continuing steep decline from 1990 ± 1994. In visual terms, no overall single trend in incidence rates can be discerned; this is re¯ected in the slope of the ®tted regression line, which is all but horizontal. More formally, the slope is not signi®cantly different from zero (slope estimate 7 0.03%; 95% con®dence interval from 7 0.16 to 0.1). The mortality data show less evidence of complex multiple effects. Visually, the impression is one of fairly regular decline during the period examined, with no evidence of
Figure 1 Age standardized incidence and mortality rates for invasive cervical cancer, Scotland, 1975 ± 1994.
Invasive cancer uterine cervix Ð trends JJ Walker et al
Table 1 Invasive cancer of the uterine cervix in Scotland 1975 ± 1994: Poisson regression parameter estimates for crossed effect of year with age group Age group 25 ± 29 30 ± 34 35 ± 39 40 ± 44 45 ± 49 50 ± 54 55 ± 59 60 ± 64 65 ± 69 70 ± 74 75 ± 79 80 ± 84 85
Estimate (incidence)
P value
Estimate (incidence)
P value
0.018 0.019 0.032 0.016 0.005 7 0.026 7 0.026 7 0.041 7 0.009 0.012 7 0.011 7 0.010 7 0.003
0.08 0.01 < 0.001 0.03 NS < 0.001 < 0.001 < 0.001 NS NS NS NS NS
0.035 0.014 0.018 7 0.002 7 0.004 7 0.050 7 0.061 7 0.037 7 0.007 7 0.005 7 0.017 7 0.022 7 0.024
0.09 NS NS NS NS < 0.001 < 0.001 < 0.001 NS NS 0.06 0.04 0.08
NS not signi®cant (P > 0.1).
the kind of `sub trends' apparent in the incidence data. The visual suggestion of mortality rates declining with time is supported by the ®tted regression line (slope estimate 70.13; 95% CI from 70.17 to 70.9). The Poisson regression models of age-speci®c rates indicate that both age group and the age group=time interaction are highly signi®cant for both incidence and mortality (P < 0.001). In other words, these models suggest (i) that, as expected, some age groups experienced higher numbers of cases=deaths than others during the study period; and (ii) that the effect of time (year) differed among age groups. To assess how the effect of time differs among age groups, reference is made to
individual parameter estimates for the crossed effect of year with age group from the Poisson regressions. Table 1 presents these estimates for the incidence and mortality data. Interpretation of the table is straightforward. The columns headed `estimate (incidence)' and `estimate (mortality)' give the value of the Poisson regression linear predictor corresponding to the crossed effect, or interaction, of time (year) with age group for, respectively, the age speci®c incidence and mortality rates. These parameters, explained more fully in the Appendix to this paper, estimate the rate of change in the number of cases or deaths per year, for each age group shown. The parameters can be thought of as analogous to the slope parameter estimate derived from a conventional (least squares) linear regression. In the Poisson models, a positive parameter estimate indicates an increasing trend with time, a negative estimate a decreasing trend. The P values in the columns to the right of the estimates indicate the signi®cance of each parameter to the model. From Table 1, it is apparent that there is strong evidence (P < 0.001) of increased incidence in age group 35 ± 39; decreased incidence in the age range 50 ± 64; and decreased mortality in the age range 50 ± 64. In order to understand the apparently complex patterns of the age group by time (calendar year) interaction, smoothed plots of the age speci®c incidence and death rates were examined, and are illustrated in Figures 2 and 3. In the age groups below 50 y the incidence rate appears to have increased up to the mid 1980s and then started to fall. The death rates show little evidence of change for these age groups. For the age groups in the range 50 ± 69, both incidence rates and death rates appear to be falling steadily throughout the period. For the older age groups (70 )
Figure 2 Incidence rates (smoothed) for invasive cervical cancer per 100 000 of population, Scotland, 1975 ± 1994.
375
376
Invasive cancer uterine cervix Ð trends JJ Walker et al
incidence appears to be fairly stable, but there is some evidence of a decline in mortality. Discussion The observed changes in the overall (all ages) incidence and mortality rates for invasive cervical cancer in Scotland during the period studied may be compared with those experienced in England and Wales over an approximately comparable period. One recent study21 of the latter identi®es little change in overall incidence rates for the years 1971 ± 1989 despite a 20% fall in mortality over the same period. Another source22 identi®es a substantial decrease in mortality rates in England and Wales from 88 per million in 1972 to 63 per million in 1992 but little change in incidence rates. Corresponding mortality rate values for Scotland are 83 per million in 1975 and 58 per million in 1994. There is thus some evidence that the trends in all-ages incidence and mortality rates observed for Scotland in the present study are broadly in line within those in England and Wales over an approximately comparable time period. Turning to the age speci®c rates, it has been observed that incidence rates in England and Wales have decreased in older age groups, particularly for women aged 45 ± 54 and 55 ± 64 y, during the period 1974 ± 1988; but that rates for younger women have shown signi®cant increases.22 This again agrees fairly closely with the Scottish experience Ð information given in Table 1 suggests increasing trends in all age groups up to the 40 ± 44 y range, though the evidence is strongest for the 35 ± 39 y group. The same
source presents graphical evidence of steep declines in mortality among the 45 ± 54 and 55 ± 64 age groups over the same period it thus appears that the main age-speci®c trends identi®ed by the present study are generally similar to those experienced in England and Wales, although perhaps rather less clearly de®ned due to the smaller population numbers involved. One major area of interest is the extent to which the trends can be seen to re¯ect an effect of screening for cervical cancer. However, efforts to identify causal links between the introduction of screening in Scotland and the observed trends are beset by dif®culties. In contrast to the implementation of screening for cancer of the female breast Ð which was introduced over a relatively short space of time Ð cervical screening in Scotland has developed gradually, over a period of many years. It is thus not possible to identify a single cut-off point which can be de®ned as a boundary between `unscreened' and `screened' time periods. This uncertainty surrounding the rate at which screening has developed inevitably hinders the identi®cation of relationships between screening rates and incidence=mortality trends. Cervical screening in Scotland began in the 1960s when general practitioners were paid an item of service fee for performing cervical smears in women aged 35 ± 64 who had not had a smear taken in the preceding ®ve years. However, many cervical smears were performed in other settings (e.g. antenatal clinics, gynaecology clinics) with little or no central co-ordination. During the 1980s it was increasingly felt that the programme was often failing to screen those women at highest risk. The Strong Report23 published in 1987 recommended a systematic screening programme with a
Figure 3 Death rates (smoothed) for invasive cervical cancer, per 100 000 of population, Scotland, 1975 ± 1994.
Invasive cancer uterine cervix Ð trends JJ Walker et al
computerised call±recall system and attention to quality assurance. In 1990, a new general practitioner contract was introduced under which GPs are now encouraged to screen 80% of their eligible population (women aged 21 ± 60) every ®ve years. National ®gures for uptake of cervical screening are not available prior to 1994 Ð the ®nal year covered by the present paper. However, laboratory data for the number of cervical smears performed annually (see Table 2) certainly suggest an increase in screening coverage during the period under investigation. These data should be interpreted with caution, since they do not relate to individual women (they include repeat smears) and they do not convey anything about the appropriateness of the activity (for example, the ages of the women being screened and the screening intervals). Nevertheless the ®gures do suggest a gradual increase in coverage since the late 1960s with a more dramatic step up in activity in 1990. Interpreting the incidence rates shown in Figure 1 in the light of these ®gures, the rise in incidence evident in 1990 may be partly explained by the detection of prevalent cases resulting from increasing cervical screening Ð perhaps analogous to the detection of prevalent cases during the ®rst round of breast screening. While the value of organised cervical screening programmes is now very widely acknowledged, a number of commentators have suggested that screening in the UK (including Scotland) has been less effective than in other parts of the world. If this explanation is accepted, effort requires to be focused on why this has occurred; some comments on the topic can be found elsewhere.22,24,25 The notion that there has been a general failure of the UK screening programme is initially appealing in the context of the present study, in that it provides a convenient explanation for the relatively stable all-ages incidence rates, and for the apparent increases observed in younger women. However, this study has produced strong evidence of substantial declines in incidence and mortality in the `middle' age range (50 ± 64 y), together with a general decline in all-ages mortality. In an observational study such as the present one a causal connection between screening and the changes observed cannot be inferred with certainty. However, we are not aware of any other events during the Table 2 Cervical smears processed at NHS laboratories in Scotland 1967 ± 1994; annual numbers of smears (including repeat smears) Year
No. of smears
Year
No. of smears
1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981
503521 507275 485368 509026 567136 442150 448752 420361 361626 356747 322484 315044 285105 293870
1980 1979 1978 1977 1976 1975 1974 1973 1972 1971 1970 1969 1968 1967
282510 252896 253270 239572 235582 253461 248610 238752 224837 210364 188233 174231 175155 151620
Source: Information and Statistics Division of the NHS in Scotland.
period under investigation that might have contributed to these changes. For example, the National Survey of Sexual Attitudes and Lifestyles26 does not suggest that there has been any major decrease in sexual activity for the women in these cohorts Ð rather, the reverse seems to be true. Since there have been no major advances in the treatment of cervical cancer since at least 1971,21 the decreased mortality cannot be explained by improved treatment. This then raises the question of why screening appears to have been effective in certain age groups but not in others. One answer to this question, suggested by a number of commentators, is that screening has been effective across a wider range of ages, but that its effect in younger women has, by implication, been countered by other in¯uences. One report27 suggests that without screening an increase in mortality from cervical cancer would have occurred in the UK, and estimates that screening in the UK has resulted in a reduction of some 20 ± 30% in the risk of and mortality from the disease. This view, that screening has prevented rises in incidence and=mortality which might otherwise have occurred, is supported by one recent study.21 Using case-control methods, the authors while recognising certain limitations of their work, such as the fact that the District Health Authorities and Health Boards involved were selfselected Ð present evidence for a very substantial effect of screening, their results suggesting that screening prevented between 1100 and 3900 cases of invasive cervical cancer in the UK in 1992. If the authors of this study are correct, screening in the UK (including Scotland) has effectively suppressed potential increases in incidence (and hence mortality) among younger women arising from other factors. Possible explanations are that phenomena such as changes in the prevalence of smoking among younger women, or altered patterns of sexual behaviour may have contributed to underlying increases which have been partly contained by screening. Informal support for a somewhat different effect of screening can reasonably be argued on the basis of effects evident in Figure 1. Mindful of the fact that screening has been implemented gradually, the general increase in incidence evident during the 1980s might be due in part to higher detection rates arising from increasing screening activity over a number of years. Similarly, a portion at least of the consistent decline in mortality rates may be attributable to the bene®cial effect of early treatment made possible by this increasing activity. Interpretation of the steep decline in incidence from 1990 onwards is more problematic. This effect is unlikely to be due to the introduction of systemic (call and recall) screening in 1990: the expected effect of this development would actually be higher detection and increased incidence until at least the ®rst two screening `rounds' had been completed. However, it may be that the fairly long period of steadily increasing screening activity implied by Table 2 has led to a situation where, for an appreciable portion of the target population, the initial effect of increasing screening activity Ð i.e. more detection and thus higher incidence Ð is over, the effect of the screening programme now being to produce a decline in incidence that may yet change as future cohorts experience screening through their lifetimes. Though conjectural, this interpretation of the results obtained in the study is intuitively appealing. While there is no room for complacency, the results of this analysis are reassuring insofar as they suggest that screening is ®nally making an impact on cervical cancer in Scotland. More recent ®gures from the
377
378
Invasive cancer uterine cervix Ð trends JJ Walker et al
General Register Of®ce (Scotland) show that deaths from cervical cancer fell from 154 in 1994 to 147 in 1995.19 This should not detract from efforts to develop and maintain an effective cervical screening programme. Research should also continue into the role of human papillomavirus and possible cofactors; not only might this lead to a more effective screening test, it also holds out the exciting prospect of primary prevention through immunisation. Acknowledgements Thanks are due to members of staff from the Information and Statistics Division of the National Health Service in Scotland, in particular S Black, V Harris, C Thomson, C Tyack and J Warner. In addition, K Davidson and I Abdallah from the Department of Mathematics at Napier University provided useful suggestions and assistance. Appendix The parameter estimates shown in Table 1 are used to compute predicted numbers of cases=deaths by substitution into a Poisson regression model of the form log m log N a0 age 1a1 age 2a2 age 3a3 . . . year age 1b1 year age 2b2 year 3b3 . . .
where m is the expected mean number of events (cases or deaths); N is the population under observation (an offset variable); a0 represents the ®tted value at year 0 for a `baseline' age group (here arbitrarily selected by the computer procedure used28 as the age group 85 years). Year values are adjusted such that the ®rst year (1975) is represented as having value 0, the second year as having value 1, and so on; age( j) is an indicator variable corresponding to the jth level of age group, for example age( j ) 1 if age group j, otherwise age ( j ) 0. Note that as 13 age groups are featured, the js range from 1 ± 13, that is age (1) ages 25 ± 29; age (13) ages 85 ; the remaining a terms give the ®tted values at year 0 for each age group, relative to the ®tted value for the `baseline' age group. Since age group 13 has been selected as the `baseline' group, a13 0; the b terms give the increase in the dependent variable (cases=deaths) with calendar time for each age group; these are the values shown in the `estimate' columns of Table 1. References 1 Parkin DM, Laara E, Muir CS. Estimates of the world-wide frequency of sixteen major cancers in 1980. Int J Cancer 1988; 41: 184 ± 197. 2 Sharp L et al. Cancer Registration Statistics Scotland 1981 ± 1990: Information and Statistics Division, Directorate of Information Services, The National Health Service in Scotland: Edinburgh, 1993. 3 Brinton LA, Hammon RF. Sexual and reproductive risk factors for invasive squamous cell cervical cancer. J Natl Cancer Inst 1987; 79: 23 ± 30. 4 Brinton LA et al. Cigarette smoking and invasive cervical cancer. JAMA 1986; 255: 3265 ± 3269. 5 Schiffman MH. Recent progress in de®ning the epidemiology of human papillomavirus infection and cervical neoplasia. J Natl Cancer Inst 1992; 84: 394 ± 398.
6 Beral V, Hannaford P, Kay C. Oral contraceptives use and malignancies of the genital tract. Lancet 1988; (ii): 1331 ± 1334. 7 Stocks P. Cancer of uterine cervix and social conditions. Br J Cancer 1955; 9: 487 ± 494. 8 Wynder EL, Corn®eld J, Schroff PD, Doraiswami KR. Study of environmental factors in carcinoma of cervix. Am J of Obstetrics and Gynaecology 1954; 68: 1016 ± 1052. 9 Fraumeni JF, Lloyd JW, Smith EM, Wagoner JK. Cancer mortality among nuns: role of marital status in etiology of neoplastic disease in women. J Nat Cancer Inst 1969; 42: 455 ± 468. 10 Kennaway EL. The racial and social incidence of cancer of the uterus. Br J Cancer 1948; 2: 177 ± 212. 11 Cross HE, Kennel EE, Lilien®eld AM. Cancer of the cervix in an Amish population. Cancer 1968; 21: 102 ± 108. 12 IARC Working Group. IARC Monographs on the Evaluation of Carcinogenic Risks in Humans, Vol 64: Human Papillomaviruses. World Health Organization, International Agency for Research on Cancer: Lyon, 1995. 13 Laara E, Day NE, Hakama M. Trends in mortality from cervical cancer in the Nordic countries: association with organised screening programmes. Lancet 1987; (i): 1247 ± 1249. 14 The Scottish Of®ce. Report of the Inquiry into Cervical Cytopathology at Inverclyde Royal Hospital, Greenock. HMSO: Edinburgh, 1993. 15 Brewster D, Crichton J, Muir CS. How accurate are Scottish cancer registration data? Br J Cancer 1994; 70: 954 ± 960. 16 Information and Statistics Division, Directorate of Information Services. ISD Guide: An A ± Z of the Work of the Information and Statistics Division. The National Health Service in Scotland: Edinburgh, 1995. 17 Waterhouse J, Muir CS, Correa P, Powell J (Eds). Cancer Incidence in Five Continents Volume III. (IARC Scienti®c Publication No. 15): World Health Organisation, International Agency for Research on Cancer: Lyon, 1976. 18 Boyle P, Parkin DM. Statistical methods for registries. In Jensen OM et al (Eds). Cancer Registration Principles and Methods: (IARC Scienti®c Publication No. 95) World Health Organisation, International Agency for Research on Cancer: Lyon, 1991. 19 General Register Of®ce for Scotland. Annual Reports. Her Majesty's Stationery Of®ce: Edinburgh, 1975 ± 1994. 20 Cleveland WS. Robust locally weighted regression and smoothing scatterplots. J Am Stat Ass 1979; 74: 829 ± 836. 21 Sasieni PD, Cuziek J, Lynch-Farmery E, and the National Coordinating Network for Cervical Cancer Working Group. 1996. Estimating the ef®cacy of screening by auditing smear histories of women with and without cervical cancer. Br J Cancer 1996 73: 1001 ± 1005. 22 Cancer Research Campaign. Factsheets Ð Series 12. Cancer of the Cervix Uteri and Series 13. Cervical Cancer Screening, 1994. 23 Report by Ad Hoc Group of the Histopathology SubCommittee of the Scienti®c Services Advisory Group on the Cervical Cytology Service in Scotland (`The Strong Report'). The Scottish Home and Health Department: Edinburgh, 1987. 24 Coleman MP et al. Trends in cancer incidence and mortality. World Health Organisation, International Agency for Research on Cancer (IARC Scienti®c Publication No. 121): Lyon, 1993. 25 Day NE. Screening for cancer of the cervix. J Epidemiol Community Health 1989; 43: 103 ± 106. 26 Johnson AM, Wadsworth J, Wellings K, Field J. Sexual Attitudes and Lifestyles. Blackwell: Oxford, 1994. 27 Hakama M et al. Evaluation of screening programmes for gynaecological cancer. Br J Cancer 1985; 52: 669 ± 673. 28 SAS Technical Report P-243 Ð SAS=STAT Software: The GENMOD Procedure. SAS Institute Inc., Cary North Carolina, 1993.
Trends in incidence of and mortality from invasive cancer of the uterine cervix in Scotland (1975 ± 1994) JJ Walker1, D Brewster2, A Gould2 and GM Raab1 Department of Mathematics, Napier University, Edinburgh; 2Information and Statistics Division, National Health Service in Scotland, Edinburgh 1
Objective: I. To identify major trends in the incidence of and mortality from invasive cancer of the cervix uteri in Scotland during the twenty year period 1975 ± 1994; II. to consider the extent to which these trends may have been shaped by the introduction of systematic cervical screening. Design: Analysis of annual age standardised and age speci®c rates for incidence and mortality, based on data collected by the Scottish Cancer Registry and the General Register Of®ce for Scotland. Setting: Scotland. Subjects: Women registered with the Scottish Cancer Registry as having developed invasive cancer of the cervix during the period of interest. Results: Annual all ages incidence rates of invasive cervical cancer show little overall change over the period 1975 ± 1989, but exhibit a pronounced decline from 1990 onwards. All-ages mortality rates show clear evidence of decline during the period 1975 ± 1994, the rate for 1994 being some 30% lower than that for 1975. Annual age-speci®c incidence rates show different patterns by age group, with clear evidence of decreasing trends in the age range 50 ± 64 years but different patterns in younger and older age groups. Most age groups show steep declines in incidence from 1990 onwards. Age speci®c mortality rates for 1975 ± 1994 exhibit the most pronounced decreasing trends in the age range 50 ± 64 years. The trends identi®ed are broadly similar to those experienced in England and Wales over an approximately comparable period. Conclusions: The overall (all ages) incidence of invasive cervical cancer in Scotland changed little during the period 1975 ± 1989, but declined sharply from 1990 onwards. The most pronounced decline in incidence across the period 1975 ± 1994 appears to have taken place in the age range 50 ± 64 years. This decline has been accompanied by a commensurate fall in mortality in the same age range. These reductions in incidence and mortality may be attributable in part to increased coverage of cervical screening programmes during the period of interest. Evidence from other studies suggest that, without the increased coverage of cervical screening achieved during this period incidence rates in Scotland might have been seen to increase. Keywords: cervical cancer; incidence; mortality; trends; Scotland
Introduction Invasive cancer of the uterine cervix accounts for some 15% of all cancers among women.1 The scale of incidence of and mortality from the disease quali®es it to be considered as a major international public health issue Ð the total number of new cases worldwide in 1980 was estimated to be in excess of 460 000.1 From a more immediate perspective, an average of almost 450 new cases of the disease were registered in Scotland annually during the period 1981 ± 1990, and more than 200 women died, on average, each year from cervical cancer in Scotland during the same period.2 Primary prevention of cervical cancer is not feasible, because the agents which cause the disease are not as yet fully understood. This is despite considerable research effort, over a period of many years, having been devoted to the problem. Factors which have been associated with development of the disease include sexual activity,3 smoking,4 viral infections of the class known as human papillomaviruses,5 use of oral contraceptives,6 and socioeconomic status.7,8 Of these risk factors, the one whose Correspondence: Mr JJ Walker, University of Edinburgh, Department of General Practice, Levinson House, 20 West Richmond Street, Edinburgh, EH8 9DX Accepted 15 June 1998
in¯uence on the development of cervical cancer is widely accepted is sexual activity Ð it is now generally acknowledged that the age at which a women ®rst engages in sexual intercourse, and the number of sexual partners she has, are signi®cant determinants of development of the disease. A substantial body of research has con®rmed the association between these sexual factors and cervical cancer.9 ± 11 In keeping with a potentially sexually transmissible agent, human papillomavirus is now generally though to be the major causal factor for most cases of cervical cancer.12 The effectiveness of cervical screening programmes based on cervical cytology in reducing incidence of and mortality from invasive cervical cancer is widely accepted.1 Although cervical screening has never been the subject of a published randomised trial, a convincing body of nonexperimental evidence points to its effectiveness, most notably the experience of a group of countries in Northern Europe which introduced organised screening programmes during the 1960s and early 1970s.13 In Scotland, the introduction of cervical screening has been a gradual process. `Opportunistic' screening Ð that is, taking cervical smears in the course of other routine GP or clinic visits Ð has been operating in some form since at least 1960, but organised mass screening based on a computerised call=recall system dates only from the late 1980s. A convenient concise history of the implementation of cervical screening in Scotland is available elsewhere.14
374
Invasive cancer uterine cervix Ð trends JJ Walker et al
In order to assess the effect of screening on the incidence of and mortality from cervical cancer, annual age standardised and age speci®c rates for the disease in Scotland during the period 1975 ± 1994 were examined. The objective was to identify signi®cant time related trends Ð if any Ð and then attempt to relate these to the expected effects of screening. Methods Cancers are recorded Ð `registered' Ð in Scotland by ®ve regional registries (based respectively in Inverness, Aberdeen, Dundee, Glasgow and Edinburgh), which are responsible for the collection of relevant data from a variety of sources: clinics, pathology laboratories, hospital records and death certi®cate records. The data collected by these regional registries are processed, collated and validated by the Scottish Cancer Registry within the Scottish Cancer Intelligence Unit (SCIU), the latter being a component of the Information and Statistics Division (ISD) of the NHS Common Services Agency based in Edinburgh. Currently about 32 000 registrations (of all types of cancer) are made annually in Scotland, and the quality of the registration data is generally high.15 Fuller details of the functions of the SCIU and ISD are available.16 The cancer registration data maintained by ISD are dynamic, in the sense that new cases are continually being added. In certain circumstances, a case may not be registered until some time after the onset of the disease Ð for example, this happens with cases who, due to limited attendance at hospital, fail to come to the attention of the registry until a record of their deaths is received. As a result, any analysis of the registration data held by the SCIU is technically valid only at the time when the data extract on which the analysis is based was prepared. In practice however, registration data for all but the most recent years can be assumed to be `complete'. For this study, registration data covering the period 1975 ± 1994 were examined, based on a data extract as at 20 June 1996. These data comprised 8447 records relating to registrations of invasive cancer of the cervix (corresponding to Rubric 180 in the World Health Organisation's International Classi®cation of Diseases, Ninth Revision, 1977). Mortality data were obtained from annual reports of the Registrar General for Scotland.
Age-standardised annual rates for incidence and mortality were calculated, standardised with respect to the European Standard Population17 using the direct method.18 Population numbers used in deriving the rates were estimated for mid-year female population totals for Scotland, obtained from the Population Statistics Branch of the General Register Of®ce for Scotland.19 Age speci®c rates for incidence and mortality were calculated as the numbers of cases or deaths divided by the estimated mid-year population for each age group. The age standardised and age speci®c rates were initially examined graphically for evidence of major trends. To provide more formal support for this purely visual approach, linear regression models were ®tted to the overall rates, with time (that is the calendar year in which the case or death was reported) speci®ed as the independent variable.18 For the age-speci®c data, Poisson regression models were ®tted in which the response variable Ð number of cases, or number of deaths Ð was modelled as dependent on (i) subject's age group; and (ii) the crossed effect, or interaction, of age group with (calendar year of reporting or death). Due to very low numbers of cases in younger women, age groups below 25 y were excluded from these models. Finally, to interpret the interactions that were found, smoothed plots of the rates by calendar time were examined.20 Results Overall age-standardised annual incidence and mortality rates for invasive cervical cancer in Scotland during the period 1975 ± 1994, together with ®tted values derived from linear regression models, are presented (Figure 1). The data for incidence show a complex mix of effects, notably: (i) rates increasing at a fairly constant rate during the period 1982 ± 88; (ii) a marked drop in 1989, relative to the preceding and succeeding years; and (iii) a continuing steep decline from 1990 ± 1994. In visual terms, no overall single trend in incidence rates can be discerned; this is re¯ected in the slope of the ®tted regression line, which is all but horizontal. More formally, the slope is not signi®cantly different from zero (slope estimate 7 0.03%; 95% con®dence interval from 7 0.16 to 0.1). The mortality data show less evidence of complex multiple effects. Visually, the impression is one of fairly regular decline during the period examined, with no evidence of
Figure 1 Age standardized incidence and mortality rates for invasive cervical cancer, Scotland, 1975 ± 1994.
Invasive cancer uterine cervix Ð trends JJ Walker et al
Table 1 Invasive cancer of the uterine cervix in Scotland 1975 ± 1994: Poisson regression parameter estimates for crossed effect of year with age group Age group 25 ± 29 30 ± 34 35 ± 39 40 ± 44 45 ± 49 50 ± 54 55 ± 59 60 ± 64 65 ± 69 70 ± 74 75 ± 79 80 ± 84 85
Estimate (incidence)
P value
Estimate (incidence)
P value
0.018 0.019 0.032 0.016 0.005 7 0.026 7 0.026 7 0.041 7 0.009 0.012 7 0.011 7 0.010 7 0.003
0.08 0.01 < 0.001 0.03 NS < 0.001 < 0.001 < 0.001 NS NS NS NS NS
0.035 0.014 0.018 7 0.002 7 0.004 7 0.050 7 0.061 7 0.037 7 0.007 7 0.005 7 0.017 7 0.022 7 0.024
0.09 NS NS NS NS < 0.001 < 0.001 < 0.001 NS NS 0.06 0.04 0.08
NS not signi®cant (P > 0.1).
the kind of `sub trends' apparent in the incidence data. The visual suggestion of mortality rates declining with time is supported by the ®tted regression line (slope estimate 70.13; 95% CI from 70.17 to 70.9). The Poisson regression models of age-speci®c rates indicate that both age group and the age group=time interaction are highly signi®cant for both incidence and mortality (P < 0.001). In other words, these models suggest (i) that, as expected, some age groups experienced higher numbers of cases=deaths than others during the study period; and (ii) that the effect of time (year) differed among age groups. To assess how the effect of time differs among age groups, reference is made to
individual parameter estimates for the crossed effect of year with age group from the Poisson regressions. Table 1 presents these estimates for the incidence and mortality data. Interpretation of the table is straightforward. The columns headed `estimate (incidence)' and `estimate (mortality)' give the value of the Poisson regression linear predictor corresponding to the crossed effect, or interaction, of time (year) with age group for, respectively, the age speci®c incidence and mortality rates. These parameters, explained more fully in the Appendix to this paper, estimate the rate of change in the number of cases or deaths per year, for each age group shown. The parameters can be thought of as analogous to the slope parameter estimate derived from a conventional (least squares) linear regression. In the Poisson models, a positive parameter estimate indicates an increasing trend with time, a negative estimate a decreasing trend. The P values in the columns to the right of the estimates indicate the signi®cance of each parameter to the model. From Table 1, it is apparent that there is strong evidence (P < 0.001) of increased incidence in age group 35 ± 39; decreased incidence in the age range 50 ± 64; and decreased mortality in the age range 50 ± 64. In order to understand the apparently complex patterns of the age group by time (calendar year) interaction, smoothed plots of the age speci®c incidence and death rates were examined, and are illustrated in Figures 2 and 3. In the age groups below 50 y the incidence rate appears to have increased up to the mid 1980s and then started to fall. The death rates show little evidence of change for these age groups. For the age groups in the range 50 ± 69, both incidence rates and death rates appear to be falling steadily throughout the period. For the older age groups (70 )
Figure 2 Incidence rates (smoothed) for invasive cervical cancer per 100 000 of population, Scotland, 1975 ± 1994.
375
376
Invasive cancer uterine cervix Ð trends JJ Walker et al
incidence appears to be fairly stable, but there is some evidence of a decline in mortality. Discussion The observed changes in the overall (all ages) incidence and mortality rates for invasive cervical cancer in Scotland during the period studied may be compared with those experienced in England and Wales over an approximately comparable period. One recent study21 of the latter identi®es little change in overall incidence rates for the years 1971 ± 1989 despite a 20% fall in mortality over the same period. Another source22 identi®es a substantial decrease in mortality rates in England and Wales from 88 per million in 1972 to 63 per million in 1992 but little change in incidence rates. Corresponding mortality rate values for Scotland are 83 per million in 1975 and 58 per million in 1994. There is thus some evidence that the trends in all-ages incidence and mortality rates observed for Scotland in the present study are broadly in line within those in England and Wales over an approximately comparable time period. Turning to the age speci®c rates, it has been observed that incidence rates in England and Wales have decreased in older age groups, particularly for women aged 45 ± 54 and 55 ± 64 y, during the period 1974 ± 1988; but that rates for younger women have shown signi®cant increases.22 This again agrees fairly closely with the Scottish experience Ð information given in Table 1 suggests increasing trends in all age groups up to the 40 ± 44 y range, though the evidence is strongest for the 35 ± 39 y group. The same
source presents graphical evidence of steep declines in mortality among the 45 ± 54 and 55 ± 64 age groups over the same period it thus appears that the main age-speci®c trends identi®ed by the present study are generally similar to those experienced in England and Wales, although perhaps rather less clearly de®ned due to the smaller population numbers involved. One major area of interest is the extent to which the trends can be seen to re¯ect an effect of screening for cervical cancer. However, efforts to identify causal links between the introduction of screening in Scotland and the observed trends are beset by dif®culties. In contrast to the implementation of screening for cancer of the female breast Ð which was introduced over a relatively short space of time Ð cervical screening in Scotland has developed gradually, over a period of many years. It is thus not possible to identify a single cut-off point which can be de®ned as a boundary between `unscreened' and `screened' time periods. This uncertainty surrounding the rate at which screening has developed inevitably hinders the identi®cation of relationships between screening rates and incidence=mortality trends. Cervical screening in Scotland began in the 1960s when general practitioners were paid an item of service fee for performing cervical smears in women aged 35 ± 64 who had not had a smear taken in the preceding ®ve years. However, many cervical smears were performed in other settings (e.g. antenatal clinics, gynaecology clinics) with little or no central co-ordination. During the 1980s it was increasingly felt that the programme was often failing to screen those women at highest risk. The Strong Report23 published in 1987 recommended a systematic screening programme with a
Figure 3 Death rates (smoothed) for invasive cervical cancer, per 100 000 of population, Scotland, 1975 ± 1994.
Invasive cancer uterine cervix Ð trends JJ Walker et al
computerised call±recall system and attention to quality assurance. In 1990, a new general practitioner contract was introduced under which GPs are now encouraged to screen 80% of their eligible population (women aged 21 ± 60) every ®ve years. National ®gures for uptake of cervical screening are not available prior to 1994 Ð the ®nal year covered by the present paper. However, laboratory data for the number of cervical smears performed annually (see Table 2) certainly suggest an increase in screening coverage during the period under investigation. These data should be interpreted with caution, since they do not relate to individual women (they include repeat smears) and they do not convey anything about the appropriateness of the activity (for example, the ages of the women being screened and the screening intervals). Nevertheless the ®gures do suggest a gradual increase in coverage since the late 1960s with a more dramatic step up in activity in 1990. Interpreting the incidence rates shown in Figure 1 in the light of these ®gures, the rise in incidence evident in 1990 may be partly explained by the detection of prevalent cases resulting from increasing cervical screening Ð perhaps analogous to the detection of prevalent cases during the ®rst round of breast screening. While the value of organised cervical screening programmes is now very widely acknowledged, a number of commentators have suggested that screening in the UK (including Scotland) has been less effective than in other parts of the world. If this explanation is accepted, effort requires to be focused on why this has occurred; some comments on the topic can be found elsewhere.22,24,25 The notion that there has been a general failure of the UK screening programme is initially appealing in the context of the present study, in that it provides a convenient explanation for the relatively stable all-ages incidence rates, and for the apparent increases observed in younger women. However, this study has produced strong evidence of substantial declines in incidence and mortality in the `middle' age range (50 ± 64 y), together with a general decline in all-ages mortality. In an observational study such as the present one a causal connection between screening and the changes observed cannot be inferred with certainty. However, we are not aware of any other events during the Table 2 Cervical smears processed at NHS laboratories in Scotland 1967 ± 1994; annual numbers of smears (including repeat smears) Year
No. of smears
Year
No. of smears
1994 1993 1992 1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981
503521 507275 485368 509026 567136 442150 448752 420361 361626 356747 322484 315044 285105 293870
1980 1979 1978 1977 1976 1975 1974 1973 1972 1971 1970 1969 1968 1967
282510 252896 253270 239572 235582 253461 248610 238752 224837 210364 188233 174231 175155 151620
Source: Information and Statistics Division of the NHS in Scotland.
period under investigation that might have contributed to these changes. For example, the National Survey of Sexual Attitudes and Lifestyles26 does not suggest that there has been any major decrease in sexual activity for the women in these cohorts Ð rather, the reverse seems to be true. Since there have been no major advances in the treatment of cervical cancer since at least 1971,21 the decreased mortality cannot be explained by improved treatment. This then raises the question of why screening appears to have been effective in certain age groups but not in others. One answer to this question, suggested by a number of commentators, is that screening has been effective across a wider range of ages, but that its effect in younger women has, by implication, been countered by other in¯uences. One report27 suggests that without screening an increase in mortality from cervical cancer would have occurred in the UK, and estimates that screening in the UK has resulted in a reduction of some 20 ± 30% in the risk of and mortality from the disease. This view, that screening has prevented rises in incidence and=mortality which might otherwise have occurred, is supported by one recent study.21 Using case-control methods, the authors while recognising certain limitations of their work, such as the fact that the District Health Authorities and Health Boards involved were selfselected Ð present evidence for a very substantial effect of screening, their results suggesting that screening prevented between 1100 and 3900 cases of invasive cervical cancer in the UK in 1992. If the authors of this study are correct, screening in the UK (including Scotland) has effectively suppressed potential increases in incidence (and hence mortality) among younger women arising from other factors. Possible explanations are that phenomena such as changes in the prevalence of smoking among younger women, or altered patterns of sexual behaviour may have contributed to underlying increases which have been partly contained by screening. Informal support for a somewhat different effect of screening can reasonably be argued on the basis of effects evident in Figure 1. Mindful of the fact that screening has been implemented gradually, the general increase in incidence evident during the 1980s might be due in part to higher detection rates arising from increasing screening activity over a number of years. Similarly, a portion at least of the consistent decline in mortality rates may be attributable to the bene®cial effect of early treatment made possible by this increasing activity. Interpretation of the steep decline in incidence from 1990 onwards is more problematic. This effect is unlikely to be due to the introduction of systemic (call and recall) screening in 1990: the expected effect of this development would actually be higher detection and increased incidence until at least the ®rst two screening `rounds' had been completed. However, it may be that the fairly long period of steadily increasing screening activity implied by Table 2 has led to a situation where, for an appreciable portion of the target population, the initial effect of increasing screening activity Ð i.e. more detection and thus higher incidence Ð is over, the effect of the screening programme now being to produce a decline in incidence that may yet change as future cohorts experience screening through their lifetimes. Though conjectural, this interpretation of the results obtained in the study is intuitively appealing. While there is no room for complacency, the results of this analysis are reassuring insofar as they suggest that screening is ®nally making an impact on cervical cancer in Scotland. More recent ®gures from the
377
378
Invasive cancer uterine cervix Ð trends JJ Walker et al
General Register Of®ce (Scotland) show that deaths from cervical cancer fell from 154 in 1994 to 147 in 1995.19 This should not detract from efforts to develop and maintain an effective cervical screening programme. Research should also continue into the role of human papillomavirus and possible cofactors; not only might this lead to a more effective screening test, it also holds out the exciting prospect of primary prevention through immunisation. Acknowledgements Thanks are due to members of staff from the Information and Statistics Division of the National Health Service in Scotland, in particular S Black, V Harris, C Thomson, C Tyack and J Warner. In addition, K Davidson and I Abdallah from the Department of Mathematics at Napier University provided useful suggestions and assistance. Appendix The parameter estimates shown in Table 1 are used to compute predicted numbers of cases=deaths by substitution into a Poisson regression model of the form log m log N a0 age 1a1 age 2a2 age 3a3 . . . year age 1b1 year age 2b2 year 3b3 . . .
where m is the expected mean number of events (cases or deaths); N is the population under observation (an offset variable); a0 represents the ®tted value at year 0 for a `baseline' age group (here arbitrarily selected by the computer procedure used28 as the age group 85 years). Year values are adjusted such that the ®rst year (1975) is represented as having value 0, the second year as having value 1, and so on; age( j) is an indicator variable corresponding to the jth level of age group, for example age( j ) 1 if age group j, otherwise age ( j ) 0. Note that as 13 age groups are featured, the js range from 1 ± 13, that is age (1) ages 25 ± 29; age (13) ages 85 ; the remaining a terms give the ®tted values at year 0 for each age group, relative to the ®tted value for the `baseline' age group. Since age group 13 has been selected as the `baseline' group, a13 0; the b terms give the increase in the dependent variable (cases=deaths) with calendar time for each age group; these are the values shown in the `estimate' columns of Table 1. References 1 Parkin DM, Laara E, Muir CS. Estimates of the world-wide frequency of sixteen major cancers in 1980. Int J Cancer 1988; 41: 184 ± 197. 2 Sharp L et al. Cancer Registration Statistics Scotland 1981 ± 1990: Information and Statistics Division, Directorate of Information Services, The National Health Service in Scotland: Edinburgh, 1993. 3 Brinton LA, Hammon RF. Sexual and reproductive risk factors for invasive squamous cell cervical cancer. J Natl Cancer Inst 1987; 79: 23 ± 30. 4 Brinton LA et al. Cigarette smoking and invasive cervical cancer. JAMA 1986; 255: 3265 ± 3269. 5 Schiffman MH. Recent progress in de®ning the epidemiology of human papillomavirus infection and cervical neoplasia. J Natl Cancer Inst 1992; 84: 394 ± 398.
6 Beral V, Hannaford P, Kay C. Oral contraceptives use and malignancies of the genital tract. Lancet 1988; (ii): 1331 ± 1334. 7 Stocks P. Cancer of uterine cervix and social conditions. Br J Cancer 1955; 9: 487 ± 494. 8 Wynder EL, Corn®eld J, Schroff PD, Doraiswami KR. Study of environmental factors in carcinoma of cervix. Am J of Obstetrics and Gynaecology 1954; 68: 1016 ± 1052. 9 Fraumeni JF, Lloyd JW, Smith EM, Wagoner JK. Cancer mortality among nuns: role of marital status in etiology of neoplastic disease in women. J Nat Cancer Inst 1969; 42: 455 ± 468. 10 Kennaway EL. The racial and social incidence of cancer of the uterus. Br J Cancer 1948; 2: 177 ± 212. 11 Cross HE, Kennel EE, Lilien®eld AM. Cancer of the cervix in an Amish population. Cancer 1968; 21: 102 ± 108. 12 IARC Working Group. IARC Monographs on the Evaluation of Carcinogenic Risks in Humans, Vol 64: Human Papillomaviruses. World Health Organization, International Agency for Research on Cancer: Lyon, 1995. 13 Laara E, Day NE, Hakama M. Trends in mortality from cervical cancer in the Nordic countries: association with organised screening programmes. Lancet 1987; (i): 1247 ± 1249. 14 The Scottish Of®ce. Report of the Inquiry into Cervical Cytopathology at Inverclyde Royal Hospital, Greenock. HMSO: Edinburgh, 1993. 15 Brewster D, Crichton J, Muir CS. How accurate are Scottish cancer registration data? Br J Cancer 1994; 70: 954 ± 960. 16 Information and Statistics Division, Directorate of Information Services. ISD Guide: An A ± Z of the Work of the Information and Statistics Division. The National Health Service in Scotland: Edinburgh, 1995. 17 Waterhouse J, Muir CS, Correa P, Powell J (Eds). Cancer Incidence in Five Continents Volume III. (IARC Scienti®c Publication No. 15): World Health Organisation, International Agency for Research on Cancer: Lyon, 1976. 18 Boyle P, Parkin DM. Statistical methods for registries. In Jensen OM et al (Eds). Cancer Registration Principles and Methods: (IARC Scienti®c Publication No. 95) World Health Organisation, International Agency for Research on Cancer: Lyon, 1991. 19 General Register Of®ce for Scotland. Annual Reports. Her Majesty's Stationery Of®ce: Edinburgh, 1975 ± 1994. 20 Cleveland WS. Robust locally weighted regression and smoothing scatterplots. J Am Stat Ass 1979; 74: 829 ± 836. 21 Sasieni PD, Cuziek J, Lynch-Farmery E, and the National Coordinating Network for Cervical Cancer Working Group. 1996. Estimating the ef®cacy of screening by auditing smear histories of women with and without cervical cancer. Br J Cancer 1996 73: 1001 ± 1005. 22 Cancer Research Campaign. Factsheets Ð Series 12. Cancer of the Cervix Uteri and Series 13. Cervical Cancer Screening, 1994. 23 Report by Ad Hoc Group of the Histopathology SubCommittee of the Scienti®c Services Advisory Group on the Cervical Cytology Service in Scotland (`The Strong Report'). The Scottish Home and Health Department: Edinburgh, 1987. 24 Coleman MP et al. Trends in cancer incidence and mortality. World Health Organisation, International Agency for Research on Cancer (IARC Scienti®c Publication No. 121): Lyon, 1993. 25 Day NE. Screening for cancer of the cervix. J Epidemiol Community Health 1989; 43: 103 ± 106. 26 Johnson AM, Wadsworth J, Wellings K, Field J. Sexual Attitudes and Lifestyles. Blackwell: Oxford, 1994. 27 Hakama M et al. Evaluation of screening programmes for gynaecological cancer. Br J Cancer 1985; 52: 669 ± 673. 28 SAS Technical Report P-243 Ð SAS=STAT Software: The GENMOD Procedure. SAS Institute Inc., Cary North Carolina, 1993.