chapter 5 · 201 chapter 5 effectiveness of screening in populations this chapter deals with...

201

Chapter 5

Effectiveness of screening in populations

This chapter deals with populationmeasures of effectiveness of cervicalcancer screening. At the populationlevel, effectiveness is ultimatelyassessed by the reduction in mortalitydue to cancer of the cervix. There aretwo reasons to expect mortality todecrease as a result of screening:removal of incident cases throughdetection and treatment of premalig-nant lesions, and diagnosis of inva-sive lesions at earlier, more curablestages. Because screening for cervi-cal cancer can detect precursorlesions that can be treated to preventprogression to invasive disease,reduction of cervical cancer incidencecan be used as a measure of effec-tiveness as well.

Four methods have been used toassess the effectiveness of screening:individual-based studies using case–control or cohort designs (see Chapter4); ecological analyses (correlatingscreening activity with changes in mor-tality or incidence rates across time,place or age group); modelling ofscreening policy and practice to estimate effectiveness; and evaluationof operational parameters of screen-ing. The latter includes screening performance indicators such as partic-ipation, quality and adequacy of follow-up of positive test results. This chapteris concerned with the last three ofthese.

Time trends are of considerable inter-est, in part for the light that may beshed on changes in exposure to etio-logical factors (especially betweenwomen of different generations) and inpart as a means of evaluating the suc-cess, or otherwise, of screening pro-grammes. Because of their compre-hensive coverage and availability, mor-tality data are often used in studies oftime trends; however, care is needed indoing so, on account of the changingproportions of deaths assigned to‘Uterus, unspecified’ (NOS) (seeChapter 1), and possible changes intreatment-induced survival, which maybe quite large if long time series arestudied (Pontén et al., 1995).

A reduction over time in the inci-dence of invasive cervical cancer,especially in those age groups wherescreening is mostly targeted, isanother long-term indicator of effectiveness. However, high-qualitypopulation-based incidence data, asprovided by population-based cancerregistries, are available in relatively fewregions of the world (Parkin et al.,2002) and fewer still have incidencedata covering extended periods oftime.

Time trends by regionUntil quite recently, most studiesfocused on the overall cervical cancertrends rather than looking separately atadenocarcinoma and squamous-cellcancer. However, cytological screeningidentifies mainly the latter. Since mostcervical cancers are squamous-cellcarcinomas, studies of overall cancerincidence and mortality largely reflecttrends in this histological type.

Developed countriesEuropeTrends in cervical cancer incidenceand mortality have been intensivelystudied in the Nordic countries, whereit has also been possible to comparethe trends in the different countries inrelation to the intensity of screeningundertaken (Hakama, 1982; Hakulinenet al., 1986, Läärä et al., 1987;Engeland et al., 1993; Sigurdsson,1993, 1995; Hristova & Hakama, 1997;Anttila & Läärä, 2000; Moller et al.,2002). In these countries, national incidence and mortality data are available from before and after thetimes that screening programmes wereimplemented. Towards the end of the1960s, Finland, Sweden and Icelandhad nationwide, organized screeningprogrammes, and the same was truefor several Danish counties. Norway, incontrast, had organized screening onlyin a single county covering about 5% ofthe population. Throughout the Nordiccountries, opportunistic testing alsoincreased at the same time.

Incidence and mortalitytrends in relation to screening

201-226 (VOIR IMAGES) 24/01/05 11:10 Page 201

From the late 1960s, a decreasewas seen in both the incidence of andmortality from cervical cancer inFinland, Sweden, Iceland and Denmark(Figure 54). The decrease, relative tothe levels before screening, was largestin Finland, where the age-standardizedmortality rate decreased more than80% from 6.6 deaths per 100 000 inearly 1960s to 1.2 deaths per 100 000in the early 1990s (decrease 82%)(rates adjusted for age to the worldstandard population). In the earlierperiod, women 30–55 years of agewere invited with a five-year screeninginterval and it was only in the early1990s that the maximum age for invita-tion was raised to 60 years in Finland.The decreases in the mortality rateswere 65% and 55%, respectively, inSweden and Denmark, with partialcoverage by organized programmes. Areduction in cervical cancer incidencewas also observed in the Danish coun-ties with organized screening com-pared to those without (Lynge et al.,1989). In Norway, the incidence

increased until the mid-1970s, and thedecrease in mortality was considerablyless (41% from the early 1960s to theearly 1990s) than in the other Nordiccountries. At that time, opportunisticscreening had become frequent also inNorway. A national organized pro-gramme of cervical cancer screeningstarted in Norway in 1995 (Nygard etal., 2002). The trend in incidence wasquite similar to the trend in mortalitywithin each country up to the mid-1990s in terms of percentage reduc-tion in the age-standardized rate. Alsothe incidence to mortality ratios werequite stable.

In general, incidence and mortalityhave also declined in the last 20–40years in many other European coun-tries (Coleman et al., 1993; Beral et al.,1994), but in some populationsincreases have been observed amongyounger women (aged under 35years), particularly during the 1970sand 1980s (Figure 55). This was firstnoted in England and Wales, wheregenerations of women born since

about 1935 were observed to be atincreasingly high risk (Hill & Adelstein,1967; Cook & Draper, 1984; Parkin etal., 1985). Similar phenomena havebeen seen in Belgium (Vyslouzilova etal., 1997), Slovenia (Kirn et al., 1992),Slovakia (Vlasak et al., 1991), Spain(Llorca et al., 1999) and in severalother countries of eastern Europe(Beral et al., 1994).

Since the early 1990s, the inci-dence rate has started to increase inFinland among women below 55 yearsof age (Figure 56) (Anttila et al., 1999).This trend is probably due to a combination of changes in sexuallifestyles and increased transmissionof papillomaviruses in younger generations of women, as well as inadequacies in the screening programme such as changes in labo-ratory procedures during this time(Nieminen et al., 2002). Because the effect of increasing incidence hasbeen partly obscured by the protectiveeffect of screening, in some countries,there has been little or no increase in

202

IARC Handbooks of Cancer Prevention Volume 10: Cervix Cancer Screening

Figure 54 Incidence and mortality rates of cervical cancer in the Nordic countries, 1958–97 (mortality available up to1996)Whole female population, adjusted for age to the world standard population (Läärä et al., 1987; Engeland et al., 1993; Hristova &Hakama 1997; Parkin et al., 1997; Moller et al., 2002; EUROCIM (European Network of Cancer Registries) database).

Incidence

Inci

den

ce r

ate

per

105

wo

man

-yea

rs

Mo

rtal

ity

rate

per

105

wo

man

-yea

rs

Mortality

201-226 (VOIR IMAGES) 24/01/05 15:17 02

risk in young women (for example,Sweden (Figure 57); Bergström et al.,1999).

In the United Kingdom, cytologicalscreening was introduced in the1960s, but an organized programme,including a call/recall system and quality assurance, was implementedonly from the 1988 onwards, leading toincreased coverage within the targetedpopulation. A sharp decrease in cervical cancer incidence and mortalityrates since 1990 has been attributed tothis organized programme (Sasieni etal., 1995, Gibson et al., 1997; Quinn etal., 1999; Sasieni & Adams, 1999)(Figure 58). The average drop in theage-adjusted mortality rate was esti-mated as 1–2% per year during

203


Figure 56 Cervical cancer incidence(----) and mortality ( ) trends inFinland, all ages

Figure 55 Cervical cancer incidence (---) and mortality ( ) trends in theUnited Kingdom, all ages

Figure 57 Cervical cancer incidence(----) and mortality ( ) trends inSweden, all ages

Figure 58 Age-standardized incidence of invasive cervical cancer, Englandl,1971–95From Quinn et al. (1999) (reproduced with permission from the BMJ Publishing Group).

Sweden

201-226 (VOIR IMAGES) 27/01/05 16:25 Page 203

1960–88 and 7% since then (Sasieniet al., 1995).

Coding of deaths as due to cancerof the uterus NOS has been commonin many countries and this affects com-parability over time. In Belgium, anattempt has been made to estimatethe proportions of deaths ascribed tocancer of the uterus NOS that shouldbe redistributed to cervix and otheruterine cancer (Arbyn & Geys, 2002).The corrected age-standardized mor-tality rates decreased from 14 per 100 000 in the 1950s to 4.5 in the1990s (68% decrease), while the certi-fied rates decreased from 6.3 to 3(52% decrease).

In a number of eastern Europeancountries such as Bulgaria, Romaniaand the Russian Federation, where little or no screening has taken place,cervical cancer mortality rates arerapidly rising (Figure 59), notablyamong recently born generations, asseen for Bulgarian women. In moreaffluent eastern European countriessuch as the Czech Republic, Hungaryand Poland, there is some evidence ofvery recent declines (Figure 59), andin terms of birth cohort, mortality mayhave peaked among women bornbetween 1945 and 1960 and thendecreased, as observed in Hungary.

North AmericaOverall, cervical incidence and mortal-ity in the USA have declined for manydecades in both black and white popu-lations (Figure 60); this has beenattributed to the effect of cytologicalscreening programmes countering anyincrease due to changes in risk factors(Devesa et al., 1989). Increases atyounger ages have not been observedin white or black women (Devesa et al.,1989; Wang et al., 2004).

In British Columbia, Canada, theage-adjusted incidence rate of squa-mous-cell cervical cancer was 28.4cases per 100 000 woman-years in1955, before the large-scale popula-

tion-based centrally organized screen-ing programme, and decreased to 6.4in 1985 (a 78% decrease) (Boyes et al.,1981; Anderson et al., 1988). The cor-responding mortality rate decreasedfrom 11.4 deaths per 100 000 woman-years in 1958 to 3.1 deaths in 1985 (a72% decrease). The lifetime coverageof cytological testing was estimated at85% from 1970 onwards.

Although screening for cervicalcancer commenced in North Americatowards the middle of the 20th century(in British Columbia, Canada, in 1949),there was a concomitant decline inmortality from the disease that initiallydid not seem to be associated withscreening (Kinlen & Doll, 1973).Therefore, studies were initiated in theUSA and Canada which attempted toevaluate the association between theextent of the decline in mortality fromcancer of the cervix and the intensity ofscreening (Cramer, 1974; Miller et al.,1976). In both countries, a strong asso-ciation was found when regionaldeclines were analysed in relation toscreening data from various sources. InCanada, the cytology data were derivedfrom a national survey and the mortalitydata were rates among women aged30–64 years, the ages at which mortal-ity was expected to be most stronglyassociated with screening (Miller et al.,1976). Mortality from cancer of all partsof the uterus was used, as the extent towhich deaths were attributed to cancerof the uterus NOS varied across thecountry and with time. The associationbetween reduction in mortality andscreening was strong at the nationallevel for declines from 1960–62 to1970–72, and at the census districtlevel, and it was demonstrated that thedecline was not explained by census-derived risk factors. In a further analy-sis, Miller et al. (1981) showed that thedecline was not explained by changesin hysterectomy rates.

Subsequently, Miller (1986) re-examined the correlation between mor-

tality rate and screening intensity in var-ious parts of Canada for later time peri-ods. Although he found consistent neg-ative correlations between screeningintensity and mortality rate at differentpoints in time, he did not find consistentcorrelations with mortality reductionduring time periods after the 1960s.Problems the author noted with thisapproach were possible changes in theunderlying incidence of cervical cancerand the marginal effect expected withmarginal increases in screening activityover time once a certain level of activityhad been established.

Australia and New ZealandAlthough cervical cancer incidencerates in Australia (Figure 61, NewSouth Wales) and New Zealand havenot greatly decreased (Coleman et al.,1993), the mortality rates have beenclearly declining in Australia for manydecades; in women aged under 35years, decreases have been seen sincethe mid-1980s in incidence and fromaround 1990 in mortality.Some increasesin mortality rates during the 1970s werenoted, notably in younger women(Armstrong & Holman, 1981), and amore recent study observed continuingperiod-specific declines in incidence andmortality from 1972 to 1996, alongsideincreasing rates in successive genera-tions (Taylor et al., 2001). Increasingcohort-specific risks in women born in thelate 1930s in New Zealand were reported(Cox & Skegg, 1986), but not confirmedlater (Cox & Borman, 1994).

JapanAlthough incidence and mortality fromcervical cancer in Japan have beenreported to be falling for many decades(Figure 62) (Coleman et al., 1993),there is evidence of some increase(particularly in mortality) during the1980s and 1990s in women agedunder 35 years. In a study of cervicalcancer incidence in Miyagi Prefectureduring 1959–87, an age–period–

204


201-226 (VOIR IMAGES) 24/01/05 11:10 Page 204

205


Figure 59 Age-standardized mortality rates of cervical cancer in Bulgaria, the Czech Republic, Hungary, Poland,Romania and the Russian Federation, ages 0–85+

201-226 (VOIR IMAGES) 24/01/05 11:10 Page 205

cohort model showed that risk haddecreased in recent periods and inyounger generations of women(Minami et al., 1996).

Time trends in developing countriesThere is limited information on timetrends in cervical cancer in developingcountries. In general terms, rates of inci-dence and mortality have been relativelystable or shown rather modest declines(Sankaranarayanan et al., 2001). Theabsence of the declines in incidenceand mortality that have been observedin high-resource populations probablyreflects the lack of screening pro-grammes, or, where they exist, the lowpopulation coverage and poor quality ofcytology (Lazcano-Ponce et al., 1998).

206


Figure 61 Cervical cancer incidence(----) and mortality ( ) trends inAustralia, New South Wales, all ages

Figure 60 Cervical cancer incidence(-----, � black, white) and mortality( ) trends in the USA, all ages

Figure 62 Cervical cancer incidence(----) and mortality ( ) trends inJapan, all ages

Figure 63 Annual cervical cancer mortality rates (per 100 000) in selectedLatin American countries, age-adjusted to the world population, 1960–95. Five-year moving averages

Dea

ths

per

100

000

wom

en

18

16

14

12

10

8

6

4

2

0

201-226 (VOIR IMAGES) 27/01/05 16:26 Page 206

Latin AmericaIn contrast to most developed coun-tries, mortality due to cervical cancer inLatin America increased between 1975and 1985 (Restrepo et al., 1993). A lateranalysis (Robles et al., 1996) showedalmost no significant downward changein mortality in Latin American countriesbetween 1960 and 1993.

Figure 63 shows trends in age-adjusted cervical cancer mortality ineight Latin American countriesbetween 1960 and 1994. In PuertoRico, with rates similar to those ofMexico, Venezuela and Uruguay at thebeginning of the period, a persistentdecline has been observed, that gaveit, by the end of the 1990s, the lowestrisk in the region. This decline parallelsthe introduction of a screening

programme (Robles et al., 1996), theeffect of which can be seen in the pro-gressive decline in age-specific rates,especially in the middle of the agerange (30–69), where screeningshould have the highest effect (Figure64a). In Cali, Colombia, a decline inthe incidence of invasive carcinomawas accompanied by an increase inregistrations of carcinoma in situ fol-lowing the introduction of a screeningprogramme in 1967 (Figure 64b)(Aristizabal et al., 1984). In Chile,Costa Rica, Cuba and Mexico, verylimited changes in mortality from cervi-cal cancer appear to have followed theintroduction of screening. Mortalityincreased from 1965 onward inMexico, where a national cervical can-cer screening programme was initiated

in 1974; although a slight decreasingtrend has been observed since the1990s, the risk remains among thehighest in the region. In Costa Rica,cytology testing has been availablenationwide to women aged over 15years since 1970, but mortality andincidence have remained almostunchanged (Herrero et al., 1992). InCuba likewise, the national screeningprogramme was judged to have had noimpact on either incidence or mortalityin the period 1980–94 (FernandezGarrote et al., 1996). In Chile, mortalityrates increased steadily between 1960and 1975, and then began todecrease, although rather slowly. Thisdecline has been modest despite theoperation of an organized screeningprogramme since the early 1970s

207


Figure 64 Age-specific incidence rates of cervical cancer in successive time periodsa. Puerto Rico; b. Cali, Colombia

201-226 (VOIR IMAGES) 24/01/05 15:18 Page 207

(Taucher et al., 1996; Sankarana-rayanan et al., 2001). However, theproportion of cancer of the uterus NOShas steadily decreased from almost50% at the beginning of the 1960s toaround 10% in the 1990s, and thiswould have masked some of thedecline in mortality from cancer of thecervix, as described above.

AsiaFigure 65 shows trends in cervix can-cer incidence reported by the cancerregistries of Mumbai (India) andSingapore. Declines in incidence arerelatively modest (except for the Indianpopulation of Singapore, among whichthe age-standardized rate declinedfrom 29.8 per 100 000 in 1968–72 to8.2 in 1993–97). In contrast, dramaticdeclines in cervix cancer in China havebeen reported. The age-adjusted inci-dence in Shanghai fell from 26.7 to 2.5

per 100 000 between 1972–74 and1993–94 (Jin et al., 1999) and mortal-ity rates have fallen dramatically, espe-cially in urban populations, althoughthe trend has reversed recently inyounger women (Yang et al., 2003).The declines have been attributed tocytological screening, treatment pro-grammes and improved female genitalhygiene, while the increased ratesamong younger women may reflectchanging economic circumstancesand sexual habits leading to a greaterprevalence of infection with HPV andother agents (Li et al., 2000).

AfricaThere are very few data on time trendsfrom Africa. In Bulawayo, Zimbabwe,the frequency of cervical cancerincreased significantly during theperiod 1963–77 (Parkin et al., 1994).Mortality data from South Africa sug-

gested some increase in rates for the‘coloured’ population between 1949and 1979, but little change in the blackpopulation from 1964 to 1977(Bradshaw & Harington, 1985). Afterabout 1980, the mortality in the‘coloured’ population remained moreor less constant, while in the whitepopulation, mortality declined from themid-1960s (Bailie et al., 1996). The dif-ference was ascribed to the availabilityof screening services, particularly forolder women.

In some registry series, recent inci-dence rates appear to be higher thanearlier ones. In Kampala, Uganda, forexample, there has been a significantincrease since the 1960s (Wabinga etal., 2000). On the other hand, thereseems to have been little change in therecorded rate in Nigeria; it was 20.9 in1960–69 and 19.9 in 1998–99 (Parkinet al., 2003).

Caveats in the evaluation oftime trends in relation to inten-sity of screeningTrends in cancer incidence and mortal-ity are a complex phenomenon to study,having substantial limitations andpotential errors associated with them(Saxen, 1982; Muir et al., 1994). In addi-tion, there are specific issues that con-cern the interpretation of time trends ofcervical cancer, including changes overtime in the proportions of deaths certified as uterus NOS and in theprevalence of hysterectomy (seeChapter 1).

The effect of screening is difficult toseparate from the effects of other factors influencing rates of cancerdiagnosis or death. For example, cervi-cal cancer mortality rates were declin-ing in North America before wide-spread screening was introduced, andthe rate of decline changed little overthe period 1946–74 despite consider-ably increased screening activity(Gardner & Lyon, 1977). This has alsobeen noted in other parts of the world

208


Figure 65 Cervical cancer incidence trends in Mumbai, India and Singapore(Source of data: Cancer Incidence in Five Continents)

Year of incidence

201-226 (VOIR IMAGES) 24/01/05 15:20 Page 208

(e.g., Miller, 1999; Arbyn & Geys,2002). Some studies have foundunderlying risk to be greater for recent-born cohorts (e.g., in Belgium, Arbyn &Geys, 2002; England and Wales,Parkin et al., 1985). The increased inci-dence in young Finnish women in the1990s may also indicate changing risk,although some inadequacies in screen-ing might also be responsible (Anttila etal., 1999). Because screening hasbeen in use for decades in many partsof the world, especially in developedcountries, it is difficult, if not impossible,to accurately estimate the risk of cervixcancer in its absence, in order to esti-mate the true impact of screening (vanBallegooijen et al., 2000). Further diffi-culties in interpreting trends result froma lack of information on the extent andquality of screening, particularly whena large proportion of tests are outsideorganized programmes.

Age, period and cohort effectsfor squamous-cell carcinomaTrend studies generally fail to distin-guish adenocarcinomas from squa-mous-cell carcinomas, although theiretiology may be rather different, andtheir susceptibility to detection by cyto-logical screening certainly is (Mitchellet al., 1995b, 2003). In an attempt tofurther evaluate the effectiveness ofscreening, the trends in incidence ofthe squamous-cell carcinoma by ageat diagnosis, period of diagnosis, andbirth cohort have been examined (Brayet al., 2004) using an age–period–cohort model (Case, 1956; Holford,1983; Clayton & Schifflers, 1987a,b).

An examination of cancer ratesaccording to birth cohort may provideinsight into the nature and intensity ofdisease-correlated exposures that mayvary across successive generations.Cohort effects may relate to birth itself,or may appear to be related to birthonly as a result of influences that areshared in the same group as they agetogether. Temporal changes in environ-

mental risk factors tend to affect partic-ular generations of individuals in thesame way as they age together, andare more likely to exert particular influ-ence on earlier stages of carcinogenesis.

Cancer rates by time, on the otherhand, may act as surrogate measuresof events that quickly change inci-dence or mortality with the same orderof magnitude in all age groups understudy. These effects may be the resultof planned interventions that act atlater stages of carcinogenesis, such asnew therapies that improve survival in allage groups. More frequently, they aredue to influences that artificially raise orlower the number of observed events(e.g., changes in classification orimprovements in diagnostic procedures).

It is likely that any major effect of ageneral screening policy will be more visible in the period than in the cohortparameters. Such an interpretation iscrude and obviously subject to uncer-tainties; thus, a screening policy mayfocus on a narrow window of ages andbe of short duration, corresponding to acohort.

Age is a powerful determinant ofcancer risk, since it parallels the cumu-lative exposure to carcinogens overtime and the accumulation of theseries of mutations necessary for theunregulated cell proliferation that leadsto cancer (Peto et al., 1985).

In presenting the age, period andcohort effects for squamous-cell cervi-cal cancer incidence, the effect of agewas fixed a priori as a biological con-stant. Two characteristic age curvesthat related the time before screeningdistorted the age–incidence pattern.Here the Gustafsson et al. (1997a)proposal was applied and the finalchoice for each population took intoaccount the credibility of the curvesfrom a biological point of view andempirical evidence that the subse-quent period and cohort effects were inreasonable agreement with theobserved trends.

Figure 66 provides estimates ofsquamous-cell carcinoma trends fromage–period–cohort models for womenaged 30–64 years. In Finland, thedeclines observed since screening wasintroduced in the 1960s have recentlyreversed, rates having steadilyincreased in cohorts of women bornafter 1945 (Figure 66a) and diagnosedin the 1990s These model-based esti-mates are consistent with the observedoverall time trends (Figure 56) (Anttila etal., 1999). Similar declines in periodtrend and fluctuation in cohort parame-ters are seen in Sweden (Figure 66b),although notable changes in rates inyounger generations are not clear in theobserved trend (Figure 57) (Bergstromet al., 1999). In England, increasingrates are seen in generations born after1935 (Figure 66c). In the observedtrends, a deceleration in the rise hastaken place among very recent genera-tions (Vizcaino et al., 2000). The periodparameters for England have reversedsince the late 1980s, a finding which isconsistent with the overhaul of thescreening programme from 1988(Walker et al., 1998; Quinn et al., 1999).In Estonia, where little screening hastaken place (Aareleid et al., 1993), theperiod parameters have no trend, andthere have been clear cohort-drivenrises in women born since the mid-1930s (Figure 66d). In the USA, thereare clear and uniform cross-sectionaldeclines in period parameter, observedin the rates from the 1970s in both blackand white women (Figure 60). Rateswere relatively stable among succes-sive generations of white women(Figure 66e) and steadily decreasingcohort trends among black women(Figure 66f).

In conclusion, the age–period–cohort modelling seems to confirmwhat is known on screening activities,effectively summarizing the data, aswell as shedding additional light on theeffects of etiological exposures.

209


201-226 (VOIR IMAGES) 24/01/05 11:10 Page 209

210


1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000Birth cohort/period

1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 19902000



Effe

ct–1

.5

–1

–0.5

0

0.5

1

1.

5

Effe

ct–1

.5

–1

–0.5

0

0.5

1

1.

5

Effe

ct–1

.5

–1

–0.5

0

0.5

1

1.

5

Effe

ct–1

.5

–1

–0.5

0

0.5

1

1.

5 E

ffect

–1.5

–1

–0

.5

0

0.

5

1

1

.5

Effe

ct–1

.5

–1

–0.5

0

0.5

1

1.5



Figure 66 Estimates of squamous-cell carcinoma trends from age–period–cohort models for women aged 30–64 years.Left-hand curve: cohort parameters. Right-hand curve: period parameters

(a) Finland

(c) England

(e) USA, SEER white (f) USA, SEER black

(b) Sweden

(d) Estonia

201-226 (VOIR IMAGES) 24/01/05 11:11 Page 210

creo

Statistical models have been devel-oped to explore the effect of screeningtest, policy and programme character-istics on the expected reductions inincidence and mortality (and derivativequantities such as years of life saved).These have led to improved under-standing of the relative importance ofvarious screening parameters, whichin turn has made it possible to inferwhat changes in screening pro-grammes might be most effective. Thequality of the models has improvedover time as the underlying parame-ters (natural history, test sensitivity,etc.) have become better understood.The models have also become morewidely used, as the contribution of thesophisticated methodology hasbecome better appreciated and thestatistical techniques more widely dis-seminated. As with any model, theydepend on the availability of data andon the accuracy of the assumptions.

The pooling of several case–con-trol and cohort studies (IARC WorkingGroup on Cervical Cancer Screening,1986) was used to estimate the reduc-tion in incidence in a cohort of womenas they age from 20 to 64 years underdifferent assumptions of the ages oftesting and its frequency (WHO, 1986).The results have been widely quoted andused as input data in various models.

In the absence of direct observa-tions, models can examine the influ-ence of variation in the underlying riskof disease over time or between popu-lation subgroups, and in the quality ofscreening procedures, on the effective-ness of screening as measured by inci-dence. Simulation models have beendeveloped. These use observed dataon the natural history of the disease,screening test performance and effec-tiveness of different options for treat-

ment of precancerous lesions, and canallow for variation in risk, accessibility,compliance and feasibility. The earlymodels of this type were used to exam-ine the relative effectiveness of differ-ent programmes and were relativelysimple computer simulations (Knox,1976; Eddy 1980). More complex mod-els use Monte Carlo simulation meth-ods to provide greater flexibility andmore realistic simulation of diseasenatural history (van Oortmarssen etal., 1981; Goldie, 2002). Gustafssonand Adami (1992) developed a differ-ential equation describing natural his-tory based on the use of computerizedidentification techniques.

The main findings derived fromthese models were that, with increas-ing numbers of tests, the marginalgains become smaller with each addi-tional test (or unit cost). With few tests,the optimal age to start screening isaround the age of 35 years, and asmore tests are added to the schedule,the optimum age at start diminishesbut less than the addition of years forexamination at older ages. Attendance,test sensitivity and completeness of fol-low-up, at moderate levels of screening

intensity, improve effectiveness morethan increasing numbers of tests.

Van Ballegooijen et al. (2000) usedthe more general MISCAN simulationmodel to compare the impact of poli-cies and characteristics of screeningprogrammes across Europe on themodelled reduction in life-years lostdue to cervical cancer. They did nottake into account possible regionalvariations in, for example, natural his-tory, underlying risk, prognosis or testsensitivity. They estimated reductionsfrom 21% to almost 100% for differentscreening policies and coverage ratesoperative in European countries, underthe most conservative assumptionabout round-to-round participation inscreening (Table 72). They also esti-mated the reductions in incidence andmortality for each country in the light oftheir screening policy and assumingcomplete coverage. This work is con-tinuing.

Using modelling techniques with-out simulation, Goel et al. (1998) esti-mated the impact of various potentialimprovements to screening in Canadaon the incidence of cervical cancer. Theresults suggested that the number of

211


Netherlands, Belgium, France, GermanyFinland Greece, Italy, Spain

Starting age 30 y 25 y 20 yInterval 5 y 3 y 1 yEnding age 60 y 64 ya 72 yLifetime number of tests 7 14 53

Interval coverage (%) % Reduction in life-years lost

25 21 24 2550 42 47 5075 63 71 75100 84 94 99.9

a For France, the stopping age is 65 yearsAdapted from Van Ballegooijen et al. (2000)

Table 72. Percentage reduction in life-years lost according to policy andcoverage

Use of modelling in thedesign and evaluation ofscreening

201-226 (VOIR IMAGES) 24/01/05 11:11 Page 211

creo

cervical cancer cases could be reducedby 15% if Canada achieved full cover-age with three-yearly screening. If,instead, smear quality were improved sothat all smears were satisfactory for eval-uation (versus 5% inadequacy assumedby the model), cases would decline byabout half this amount.

Hakama and Hristova (1997) usedage–period–cohort modelling to pro-ject mortality to 2017 across the Nordiccountries under three scenarios: withno screening, with present levels ofscreening (and projections based onrecent trends) and with improvedscreening as for Finland, that wastermed optimal screening. They esti-mated that 91% of cervical cancerdeaths were prevented in Finland byscreening, i.e., could be prevented withoptimal screening. Further, they notedthat greater declines than observedcould have occurred in the Nordiccountries other than Finland had theFinnish (optimal) screening practicesbeen adopted. It was predicted thatscreening would prevent the loss of 10 000 woman years in the Nordiccountries in 2010 and that the costs ofthe health services were less withscreening than without, assuming theorganization practised in Finland.

The Papanicolaou test was never rig-orously evaluated in a randomizedcontrolled trial such as those to whichnew screening techniques are subjecttoday. Observational data have beenused to demonstrate the efficacy andeffectiveness of screening in control-ling cervical cancer (see Chapter 4and above) and it would now be uneth-ical to conduct a randomized study inthe presence of existing cytology-based screening.Where screening hasfailed to work, the blame can be laid onthe design or delivery of the screeningservice (Zapka et al., 2003).

Reaching the women at riskThe many organized screening pro-grammes around the world aredescribed in Chapter 3, but much cervi-cal screening is also undertaken spon-taneously. In the USA, where screeningis opportunistic or spontaneous, about80% of women over the age of 25 yearsreport having had a test in the last threeyears (Breen et al., 2001; Swan et al.,2003). In England, where screening isorganized, 81.5% of women aged25–64 are reported as having had a testin the last five years (Statistical Bulletin,2003). These are clearly very compara-ble rates. Gustafsson et al. (1995) sug-gested that a screening test can beequally effective whether performed inan organized or opportunistic setting.[The Working Group noted thatGustafsson et al. inappropriately con-sidered the prevalence of carcinoma insitu and microinvasive cancer in esti-mating the effectiveness of screening,rather than the incidence of clinical inva-sive cancer.] However, while oppor-tunistic screening may avoid the costsof the central call/recall bureaucracy, itcan be considerably more expensive(Schaffer et al., 1995).

A further issue is whether screen-ing reaches the population at high risk.In the United Kingdom, before orga-nized call/recall was introduced in1988, only around a quarter of womenhad had a recent test and these werelargely lower-risk women (Farmery &Gray, 1994). In the context of a gener-ally organized and centrally fundedhealth service, opportunistic screeningwas failing a large proportion of thepopulation and, in addition, cervicalcancer rates were beginning to rise,particularly among younger women(Beral & Booth, 1986). Organization ofthe screening programme in Englandand Wales raised the coverage ratesand led directly to a 42% drop in cervi-cal cancers between 1988 and 1999,after an initial increase in the numberof cases diagnosed (Quinn et al.,

2001). Following a case–control study,Nieminen et al. (1999) concluded thatthe substantial decrease in the inci-dence of and mortality from cervicalcancer in Finland was due mainly tothe organized mass screening that hadtaken place rather than to any oppor-tunistic testing and that opportunistictesting was far less efficient.

The part of a population that ishardest to reach generally includesmany of the high-risk women (Davey-Smith et al., 1994), for reasons thatare surprisingly similar despite the different health systems observed.These include socioeconomic depriva-tion, cultural and language barriers,often being from a minority ethnicgroup, being highly mobile in resi-dence and not having a ‘usual careprovider’ (Lawson et al., 2000). Manystrategies have been employed toattract such women for screening, withvarying degrees of success. It hasbeen found that use of nurses to takesmears improves acceptance, particu-larly among deprived women (Baker & Middleton, 2003) and non-medicalsmear-takers may also be used(National Cervical Screening Pro-gramme (NZ), 1998). Even when theinitial test and any follow-up requiredis provided without charge, difficultiesin reaching deprived women can persist (Chiu, 2003). Organizedcall/recall systems are more likely toreach these women, although theaccuracy of the register is a key factorin the degree of success (Baker &Middleton, 2003).

Screening is best organized on apopulation basis as a public health pro-gramme. Evidence from the Nether-lands illustrates the difficulties in orga-nizing cervical screening within a gen-eral practice setting (Hermens et al.,1998). Even when professional thinkingon policies is clear, external assistanceis required to bridge the gap betweenpolicy and effective delivery of the pro-gramme (Hanselaar, 2002).

212


Issues in the implementa-tion of screening

201-226 (VOIR IMAGES) 24/01/05 11:11 Page 212

Most cervical cancer screeningnow takes place in the developedworld, although 80% of the cases arefound in developing countries. Thescarcity of the skills and resourcesrequired to report cervical cytology indeveloping countries together with thedifficulty of finding and treating womenhave led to interest in investigating alter-native techniques for cervical screeningin these areas, such as visual inspec-tion of the cervix (see Chapter 4).

Age and frequency of screeningThe ages at which screening takesplace vary considerably. In some coun-tries, such as the USA, screening is rec-ommended from the age of 21 or threeyears from the onset of sexual activity.However, in others, such as theNetherlands, screening does not com-mence until the age of 30 (Coleman etal., 1993). A study in the UnitedKingdom found that cytology screeningwas less effective in young women, butgrew in effectiveness as women aged(Sasieni et al., 2003). This led to thedecision in England to move from a rec-ommended age of 20 years for the initi-ation of screening to the age of 25 years.

The frequency of screening alsovaries widely. In the USA, screeninghas generally been recommended onan annual basis. In several Europeancountries, five-yearly screening is rec-ommended. In England, based on thefindings of Sasieni et al. (2003) on thevariable effectiveness of screening withage, there has recently been a move tothree-yearly screening for women aged25–49 and five-yearly screening forwomen aged 50–64 years.

In low-resource settings whereorganized screening programmes arebeing developed, optimizing thescreening intervals may be less impor-tant than ensuring that each woman inthe target demographic groups isscreened once before any is screeneda second time (Suba et al., 2004).

Identifying abnormalitiesThe process of performing and report-ing the original test has a number ofdistinct phases. The first of these isobtaining the sample. When cervicalscreening was first implemented in astructured way in British Columbia,Canada, in 1949, the objective was todemonstrate the effectiveness of thetechnique first reported by Papanico-laou in the 1940s, and the majority ofsmears were taken by general practi-tioners during examinations. Most cer-vical screening still takes place in theprimary-care setting, but the nature ofthe individual who actually performs thetest varies from one country to another(Boyes & Worth, 1976)

The sampling device used in theoriginal Canadian system was theAyre’s spatula. This has remained inuse to the present day, often in combi-nation with an endocervical brush toensure sampling of the endocervicalcanal. Cotton swabs have also some-times been used [the Working Groupnoted that this is not an efficient sam-pling technique]. Extended-tip spatulae,such as the British ‘Aylesbury’ spatula,have come into use over the last 15years and more recently plasticbrooms. These are almost always usedwhere a liquid-based specimen is to betaken. Buntinx and Brouwers (1996)conducted a meta-analysis looking forany relationship between samplingdevice and detection of abnormalityand concluded that either theextended-tip spatula, a combination ofany spatula plus the Cytobrush or cot-ton swab, or the plastic broom shouldbe used for cervical screening.

The most common screening testremained the conventional Papanicolaousmear until relatively recently, when liquid-based cytology (LBC) was intro-duced. LBC testing is now used for themajority of cervical screening in theUSA (Noller et al., 2003) and this isspreading elsewhere. The UnitedKingdom is now converting its entire

programme (NICE, 2003), as a result ofimprovements in efficiency due to thedramatic drop in the number of testsreported as inadequate for diagnosisand improved laboratory productivity.LBC may increase the detection rate ofcervical screening (see Chapter 4),although studies are generally basedon findings at one test, rather than in apopulation over time, so there is a lack oflong-term data (Payne et al., 2000). Inthe United Kingdom, the introduction ofLBC was modelled to be cost-effective,as discussed below (Moss et al., 2003).Changing to LBC facilitates a move tousing automated devices to assist inreporting, although few studies have yetprovided data obtained with currentlyavailable equipment.

The place of testing for high-riskHPV DNA in a cervical cancer controlprogramme is not yet defined. Testingwomen for high-risk HPV as triage forborderline or ASCUS cytologicalresults is becoming common followingthe publication of data from studiesconducted primarily in the USA (Manoset al., 1999; Solomon et al., 2001;ANAES, 2002; Arbyn et al., 2004).Studies are also being conducted intothe possibility of using HPV DNA testing as a primary screening test (seeChapters 2 and 4). Adding HPV testingto cervical cytology allows the intervalto be increased for HPV-negativewomen with normal cytological results(van den Akker-van Marle et al., 2003).

Quality assurance is an integralpart of most screening programmes,although in practice it varies from thecomprehensive quality assurance pro-gramme seen in the United Kingdomto systems which cover the laboratoryonly, such as the Clinical LaboratoryImprovement Amendments (CLIA) inthe USA. European guidelines, origi-nally produced in 1993 (Coleman etal., 1993), are currently being revised.Evaluation of screening programmesin the longer term requires monitoringof cervical cancer incidence and mor-

213


201-226 (VOIR IMAGES) 24/01/05 11:11 Page 213

tality rates and comparison of data inthe screened population with whatmight have been seen in unscreenedpopulations (Day, 1986).

Follow-up and treatment ofabnormalitiesSasieni et al. (1996) calculated that inEngland, 21% of the cervical cancerswith inadequate screening history in1992 were due to failure to follow upabnormalities according to the thencurrent guidelines. Failure to investi-gate and treat women with cytologicalabnormalities and loss to follow-upafter treatment are well documentedpitfalls in the operation of cervicalscreening programmes (see Chapter3). The majority of preinvasive cervicallesions can today be treated undercolposcopic guidance and with no oronly a local anaesthetic. This has considerably lessened the harmcaused by cervical screening com-pared with the early days when radicalsurgical techniques were the treat-ment of choice (Boyes & Worth, 1976).Treatment is now extremely success-ful, with a complication rate of lessthan 2% (Luesley & Leeson, 2004).However, due to the risk of recurrentdisease, women who have beentreated for cervical intraepithelial neoplasia (CIN) are generally recom-mended to have annual cytologicalscreening for around 5–10 yearsbefore returning to a longer cycle.

A systematic review on HPV DNAtesting in the follow-up after treatmentof CIN indicates that a positive HPVtest can pick up treatment failure morequickly than cytology (Paraskevaidis etal., 2004).

Demonstration projectsEach population to which screeningwill be applied has different character-istics, priorities and health systems. Inorder to determine the optimal servicedesign and delivery for a given popula-tion, a demonstration project should be

undertaken (Miller et al., 2000). Thisshould consider the feasibility of theproposed arrangements and testing ofthose arrangements in practice.

Screening is an unusual medical inter-vention in that it is an "investigation,which does not arise from a patient’srequest for advice for a specific com-plaint" (McKeown, 1968). While thereare excellent data supporting theimplementation of mass cervical can-cer screening programmes, there arealso negative consequences ofscreening large numbers of healthywomen in order to prevent significantdisease in a few. These include:

• Psychological consequences of apositive screening result, withincreased anxiety and fear amongwomen;

• Misunderstanding by women andhealth-care providers of the mean-ing of a positive screening test,such that a positive result is inter-preted as a ‘cancer diagnosis’;

• Misunderstanding by women andhealth-care providers of themeaning of a negative test asimplying no risk rather than lowrisk for cervical cancer, whichmay lead to underinvestigation ofsymptoms;

• False positive screening resultsleading to unnecessary interven-tions, with both human and finan-cial cost implications;

• False negative screening resultsgiving false reassurance;

• Overtreatment of preinvasivelesions that left alone would neitherprogress nor cause any clinicallysignificant disease, particularly asthere are still no reliable markers todetermine which high-grade cervi-cal cancer precursors will progress

to cancer or will remain clinicallyinsignificant;

• Complications of treatment such ascervical stenosis, cervical incom-petence and infertility, as well asthe results of more radical thera-pies, such as hysterectomy, with arange of potentially negativesequelae related to the surgicalintervention;

• Opportunity costs to the health-care system of introducing ascreening programme;

• Impact of incidental findings duringscreening.

Psychological consequences ofscreeningThere have been few studies directedspecifically at evaluation of the psycho-logical impact of participation in a cer-vical cancer screening programme. Insuch programmes, women who arewell and asymptomatic are required toundergo a gynaecological examina-tion, which for many women is uncom-fortable and experienced as a rela-tively invasive procedure in a privateand intimate part of their bodies. Aftera delay of varying intervals, dependingon the quality of the screening service,the woman receives her result.Approximately 1–10% of all smearsare considered abnormal. In mostscreening services, low-grade cervicalabnormalities are managed byrepeated testing at defined intervals,while for high-grade abnormalities,women are referred for colposcopicevaluation, another uncomfortable andinvasive procedure. Both approachesmay cause significant anxiety inwomen (Marteau et al., 1990).

The notification of an abnormal resultmay cause anxiety and fear amongwomen. For many, the concept of a ‘pre-cancerous’ lesion is difficult to grasp andthe assumption may be that they eitherhave or are at great risk of having anestablished cancer (Posner & Vessey,1988). Despite the markedly improved

214


Hazards of screening pro-grammes

201-226 (VOIR IMAGES) 24/01/05 11:11 Page 214

outcome after treatment for cancer in thepast 20 years, many women still equatea ‘cancer diagnosis’ with a ‘death sen-tence’ (Greer, 1985).The word precanceritself causes anxiety, because whatwomen hear is the word ‘cancer’ withoutthe qualification of the medical meaningthat it is a precursor lesion that maynever develop into an invasive lesion(Kavanagh & Broom, 1998).

In addition, in many colposcopyservices there is considerable delay foran appointment and the waiting periodmay be associated with acute anxiety,particularly if the woman believes thatshe has cancer, even though, due tothe long latent period in the natural his-tory of cervical cancer, there may beno clinically significant consequence ofthis delay.

Posner and Vessey (1988) used asemi-standardized interview to study153 women from the time they werereferred to the clinic for colposcopy toafter their final check-up. Of thesewomen, 65% described feeling ‘wor-ried or alarmed’ after receiving notifica-tion of their abnormal test and 27%used words such as ‘shocked’,‘stunned’ and ‘devastated’. Women’sanxiety was related principally to thebelief that the positive result impliedcancer and death. Further, even afterappropriate treatment, 35% of thewomen still felt afraid of the possibilityof cancer and 43% worried aboutrecurrence of disease. Women alsoreported having a different view of theirbodies and a different attitude to sexafter a positive test.

There is very little good informationon whether women who attend col-poscopy clinics after an abnormal testresult differ in perception of cancer riskfrom those who fail to attend, as sug-gested by Posner and Vessey. Funkeand Nicholson (1993) found no differ-ence in such risk perception betweenwomen who attended for colposcopyand non-attenders. Lerman et al.(1990) found that women who had

been screened in the past three yearshad less fear of cancer than those whohad not been screened in the sameperiod. McKee et al. (1999) found nodifference between women compliantwith colposcopy attendance withregard to fear of cancer compared withthose who were non-compliant. Theyalso found no difference in the percep-tion of the gynaecological examinationas embarrassing between attenders andnon-attenders at colposcopy clinics.

In evaluating women’s responsesto an abnormal test result, factorsother than fears associated with can-cer and death also need to be takeninto account. For instance, McKie(1993) found that women made negative links between cervical cancerand sexual promiscuity. The epidemio-logical findings that having multiplesexual partners or a partner with multi-ple partners increases the risk of cervi-cal cancer may be interpreted bywomen with abnormal tests as imply-ing that they or their partners havebeen promiscuous; the comment hasbeen made that ‘their character, as wellas their cervix, is smeared’(McCormick, 1989).

In most screening programmes,information given about screening isaimed at achieving high uptake ofscreening, in keeping with the welldocumented benefit of wide coverageon reduction in cervical cancer.However, giving information thatemphasizes only the positive aspectsof screening may have negative con-sequences for some women who feellet down by the screening process,particularly women who receive falsenegative or false positive results. Abelief that screening gives full protec-tion may in itself have negative conse-quences and it is important thatwomen understand that screening willnot prevent all deaths related to cervi-cal cancer. In addition, many screen-positive women will be treated for anasymptomatic condition that, if left

undetected, would never have pro-gressed to a clinically significantlesion.

Concern has been raised that giving realistic information to the pub-lic, including explaining that certainindividuals can suffer adverse out-comes, can have a negative impact onuptake of screening, and brings up theissue of the ‘public good’ versus‘individual autonomy’. Wardle andPope (1992), commented that atten-tion to the psychological costs ofscreening had lagged far behind thetechnical and organizational aspects ofscreening services. Research into thisqualitative aspect of screening hassuggested a substantial toll of emotional turmoil, but most studieshave been uncontrolled and involvedsubjective evaluation.

Unnecessary treatment,overtreatment and adverse con-sequencesThe appropriateness of screening forthe prevention of cervical cancershould be seen as a balance betweenthe beneficial health effects and theadverse effects and costs of interven-tion. Unnecessary referrals and diag-nostic and therapeutic procedures areoften cited as the major adverse effectof cervical cancer screening, althoughfew studies have attempted to quantifythem. The Dutch Evaluation Com-mission (Evaluation CommissionCervical Cancer Screening, 1988,reported in Van Ballegooijen et al.,1990) evaluated the amount of diag-nostic and treatment proceduresinduced by cervical cancer screeningprospectively and in relation to mortal-ity reduction using data from the Dutchscreening programme. A model-basedanalysis led to the following estimates:(1) a mean duration of preinvasive dis-ease of 17 years, with the shortest atolder ages; (2) a regression rate ofpreinvasive disease of 60% on aver-age, with the highest at young ages;

215


201-226 (VOIR IMAGES) 28/01/05 12:16 Page 215

and (3) a sensitivity of cytology ofaround 70% for CIN 3. The false posi-tive rate of cytology was assumed tobe 0.4%. The group calculated that forfive invitations for screening amongwomen aged 37–70 years at eight-yearly intervals, 13 deaths wereavoided per million women per screen-ing year. Each death avoided is balanced by 2800 preventive tests,nine women referred for a gynaecolog-ical assessment and four for minortreatment procedures (e.g., conizationof the cervix). Increasing the invitationsto 25 from five would avoid 27 deathsper million women per screening year,but would require 7300 preventivetests, 22 referrals to gynaecology andeight minor treatment procedures.These data clearly showed that moreintensive screening greatly increasesthe need for intervention with diminish-ing returns for the extra effortsrequired.

CIN lesions have been treatedusing a variety of ablative and exci-sional techniques over the past 40years (Martin-Hirsch et al., 2004), witheach method having its own range ofcomplications and consequences(although the same efficacy). Ablativetechniques rely on a histological diag-nosis provided by colposcopicallydirected punch biopsy, which may bothundercall (leading to missed diagnosisof microinvasive cancers and under-treatment of these conditions) or over-call (leading to over or unnecessarytreatment of the cervix). The mostwidely used ablative techniques includecryotherapy and laser therapy (seeChapter 1).

In addition to the potential forunnecessary interventions, the wide-spread use of excisional procedures totreat preinvasive lesions of the cervixmay have led to considerableovertreatment. Data on negative histological findings in tissue obtainedduring loop electrosurgical excisionprocedure (LEEP) have been reported

in a number of studies using a ‘see-and-treat’ approach, i.e., treat-ment of CIN after a colposcopic diag-nosis without prior histological confir-mation. Murdoch et al. (1991) reportedan overall 41% rate of negative histol-ogy after LEEP in a highly selectedgroup of women attending a colposcopy clinic because of abnormalcytology. In women who had had priorhistological sampling, negative LEEPhistology was found in 43%, comparedwith 38% of women treated with LEEPon a see-and-treat basis. Of thewomen who had negative LEEP histol-ogy and who were treated on a see-and-treat basis, the majority (53%)had index cytology of CIN 1. As a con-sequence, the authors cautionedagainst the see-and-treat approach inwomen with low-grade referral cytology.

In a retrospective analysis of LEEPperformed at a colposcopy clinic inSouth Africa (Denny et al., 1995), 18%of LEEPs performed after prior histolog-ical sampling (21 out of 116) yieldedhistologically negative findings com-pared with 14% of those treated on asee-and-treat basis (16 out of 114). Thewomen were referred to this colposcopyclinic as a result of persistent LSIL (2–3LSIL cytological results over 12–18months) or one result of HSIL or suspi-cious of malignancy. An additional find-ing in this study was that 25% of punchbiopsies were falsely negative; theauthors emphasized that a punchbiopsy is only as reliable as the colpo-scopist’s ability to identify the mostabnormal area for biopsy.

Rates of negative LEEP histologyranging from 5 to 41% have beenreported. All of these studies were performed in colposcopy clinicswhere women had been referred withabnormal cytology (Prendiville et al.,1989; Luesley et al., 1990; Whiteley &Olah, 1990; Bigrigg et al., 1991;Hallam et al., 1993; Denny et al.,1995). False positive cytology or col-

poscopy, false negative histology inthe LEEP specimen and possiblecomplete excision or spontaneousresolution of the lesion after priorbiopsy are possible explanations fornegative LEEP histology. In addition,some series have reported negativeLEEP histology where there has beenextensive thermocoagulation prevent-ing a histological diagnosis.

While excisional procedures of thecervix are performed under localanaesthetic in an outpatient settingand complications are considered rela-tively benign, this is not always thecase. In a randomized trial of LEEP,cryotherapy and laser vaporization ofthe cervix, Mitchell et al. (1998)reported complications from LEEP in7.6% (10/130) of women treated. Themajority of the complications related tobleeding (70%); there was one case ofinfection, one woman complained ofsevere pain and required treatmentand one woman subsequently devel-oped cervical stenosis.

Ferenczy et al. (1996) reported on1070 women who underwent LEEPand who returned for post-LEEP fol-low-up. Complications were recordedin 71 women (7%); 37 had significantintra- or post-operative bleeding, andone required admission to hospital tocontrol bleeding. Seven women devel-oped a purulent vaginal discharge andpelvic pain within one week of treat-ment. A further 13 women (1.2%)developed cervical stenosis, of whom11 were over the age of 45 and not onhormone replacement therapy.

Cervical stenosis, while ratherrarely reported, is a significant compli-cation of treatment of the cervix andmay result in infertility or menstrualcomplications such as haematometria,making follow-up with cytology and/orcolposcopy difficult if not impossible.These complications may necessitatehysterectomy, with all the potentialsequelae associated with this moreradical surgical intervention. Hysterec-

216


201-226 (VOIR IMAGES) 24/01/05 11:11 Page 216

tomy may also be the result of clinicaluncertainty as to the meaning of per-sistently low-grade cytology or persis-tently inadequate tests, neither ofwhich can be resolved satisfactorily.

In some women, if the excisionalprocedure removes large amounts ofcervical stroma, a complication oftreatment may be cervical incompe-tence. This has been associated withpre-term delivery due to premature rupture of membranes (Sadler et al.,2004)

In addition to the complicationsassociated with treatment, the risk ofpersistence or recurrence of lesionsafter treatment makes long-term follow-up of treated women essential. Reportsfrom both non-randomized and ran-domized trials of treatment for cervicalcancer precursors using a variety oftreatment modalities indicate that mosttreatments have about a 90% successrate (Martin-Hirsch et al., 2004).

Opportunity costs to the healthsystemSetting up a screening programmedesigned to detect disease in healthyindividuals necessitates diversion ofhuman and financial resources fromother health interventions, in particular,treatment of already existing or appar-ent disease. Thus a screening pro-gramme and its hazards and benefitsneed to be evaluated in the context ofthe competing health needs of thespecific country to ensure thatresources are used to the maximumbenefit of the entire population.

One feature of cervical cancerscreening programmes in low-resourcecountries with endemic HIV infection isthe diagnosis of CIN lesions in as manyas 20–30% of HIV-infected women.Thismay lead to the consumption of scarcehealth resources for treatment of CIN.However, in countries such asZimbabwe where HIV infection hasbecome pandemic, HIV-infectedwomen succumb to opportunistic infec-

tions long before invasive cervical can-cer arises (Chokunonga et al., 1999). Insituations where it is not possible to pro-vide any form of treatment for HIV-related conditions (e.g., treatment ofopportunistic infections and/or anti-retroviral therapy), cervical cancerscreening may not be a priority.

Incidental findingsWhile screening is an activity designedfor healthy, asymptomatic women,there can be unexpected incidentalfindings at the time of screening thatmay cause harm as well as benefit towomen and the health system. Forinstance, the discovery of underlyingdiabetes due to the diagnosis of dia-betic vulvitis at the time of performingthe screening test may be of benefit tothe woman, but the discovery of alethal co-existent cancer may onlyincrease suffering without offering anyimprovement in quality of life, if there isno effective treatment for that cancer.

In low-resource countries wherewomen generally have little access tohealth care, screening may identify asignificant number of women with co-morbid health conditions which it isimpossible to manage with the avail-able resources.

One possible incidental finding ofparticular significance is identificationof a woman as HIV-positive.The preva-lence of CIN among women infectedwith HIV is nearly five times that inHIV-negative controls (Chirenje et al.,2002). In low-resource countries wherecervical cancer screening is mainlyopportunistic, there is a large burdenof women harbouring CIN lesions withconcomitant HIV infection that has notbeen identified.

Cervical cancer screening inHIV-positive womenIt is estimated that there are now up to42 million people worldwide living withHIV infection or AIDS. About 70% ofthese individuals live in sub-Saharan

Africa (UNAIDS, 2003) and more thanhalf of the infected people are women;cervical cancer screening has beenwidely unavailable in this area up to now.

In many HIV-endemic countries,CIN lesions may be detected at thesame time that HIV infection is firstdiagnosed. The discovery of CINlesions in an HIV-infected woman maycreate major psychological and moralechallenges, not only to the woman, butalso to health workers.

The diagnosis of HIV infection stillcarries a high level of stigmatizationand fear of disclosure of an incurablesexually transmissible disease, withsome women being exposed to violentresponse from their male partners andpermanent psychological isolationfrom the community. In the absence ofantiretroviral treatment, some womenbecome suicidal and in this group anadded diagnosis of a CIN lesionthrough cervical cancer screeningcauses special concern. Linkage of acervical cancer screening programme toHIV testing may present a new barrier toparticipation in cervical screening.

Another concern is how to interpreta positive cervical screening test resultin the presence of HIV-positivity, bear-ing in mind that natural progression ofboth conditions is dependent on avail-ability of effective treatment. Treatmentof CIN lesions in HIV-positive womenby ablative or excision proceduresresults in epithelial disruption whichcan theoretically enhance viral acquisi-tion or transmission.

Health workers treating HIV-posi-tive women for cervical disease shouldtake universal precautions as for allhealth interventions.

The minimal essential elements of acervical screening programme are: adefined population to screen; invitationto this population to participate;

217


Performance evaluation

201-226 (VOIR IMAGES) 24/01/05 11:11 Page 217

assessment of coverage; a quality con-trol system; and treatment for test-pos-itive women. The means of achievingthese, e.g., population registers orgeographical location to define thepopulation; personal invitation letters(call/recall) or mass education to invitewomen to participate; population-linked cervical cancer registry or sam-ple surveys to assess coverage) willdepend on the local circumstances.

The recognition that the effective-ness of screening "is determined bythe proportion of progressive lesionsthat are successfully detected andtreated" (Pontén et al., 1995), whichdepends in turn on screening policyand its implementation, has led to thedevelopment of indicators of screeningprogramme performance for routinemonitoring. The determinants of thisproportion represent the essential elements of a good screening pro-gramme (Hakama et al., 1985). Theyinclude coverage of the target popula-tion (participation); attendance forrescreening; adequacy of smear-tak-ing; quality of interpretation of smears;and follow-up of abnormal results.Specific indicators for these pro-gramme performance areas have beendeveloped in the context of cytologyand more specifically Pap smears.

The Council of Europe recentlyrecommended all Member States tooffer organized screening for threecancers including cervical cancer, andstressed that this should be managedin such a way that the performancecan be evaluated fully (Council of theEuropean Union, 2003). Organizedscreening requires adequate data col-lection systems to be set up concern-ing invitation and participation of thetarget population, registration ofscreen test results and follow-up ofscreen positives (Arbyn et al., 1999;Advisory Committee on CancerPrevention, 2000). Screening data-bases, including personal records,should be linkable with cancer and

mortality registers, in order to allow fullevaluation of the programme. Thisneeds to be done with full respect fornational legislation. Opportunisticscreening systems are in general lesscost-effective and do not allow evalua-tion (Advisory Committee on CancerPrevention, 2000).

Screening policyUnlike breast cancer screening, whereoptimal screening policy (in terms ofage range, frequency and modality)has been determined by randomizedtrials and most programmes followsimilar policies, results of observa-tional studies of protection offered bycytological screening have producedestimates of effectiveness for a varietyof alternative policies. Thus, the maxi-mal effectiveness of a programme willdepend on what policy is adopted. Theimpact of policy on effectiveness hasbeen modelled recently for Europeancountries by van Ballegooijen et al.(2000) (see Table 72).

Most screening programmes rec-ommend the same screening intervalfor the entire target age range.However, a recent audit of screeningin the United Kingdom suggested thatthe policy of one screening intervalacross the entire target age rangemay not yield optimal effectivenessand recommended screening womenaged 25–49 more frequently thanolder women (Sasieni et al., 2003).This further illustrates how effective-ness can be determined in part by policy.

Many factors influence what policyis adopted. For example, a nationalworkshop in Canada recommendedthat screening intervals not be raisedto three years without the security pro-vided by an information system (Milleret al., 1991). Since most Canadianjurisdictions have not in the past hadinformation systems, this almost certainly resulted in over-screening.Medical legal issues may also influ-

ence screening policy (and thereforeeffectiveness) in some countries. Forexample, a shorter screening intervalis safer than a longer one and abroader age range safer than a nar-rower one. However, the marginal util-ity of each extra test diminishes rapidly.This is of particular concern in developing countries.

Screening deliveryIt is generally accepted that, for opti-mal effectiveness, cervical cancerscreening should be offered within anorganized programme (see Chapter3). The programme should furtherinclude the specific elements asdescribed by Hakama et al. (1985)(see above). Hakama and others haveconcluded that organized screeningprogrammes, as practised to varyingdegrees in the Nordic countries andparticularly in Finland, are more effec-tive than opportunistic screening activ-ities. This conclusion was basedlargely on comparison of incidenceand mortality trends (e.g., Hakama,1982; Läärä et al., 1987) and cohortstudies, as discussed previously in thisvolume. A case–control study byNieminen et al. (1999) suggested thattests in an organized programme maybe more effective than those deliveredoutside the programme, within thesame jurisdiction. However, sincecountries differ greatly in their screen-ing practices as well as in the level oforganization, it is difficult to attributehigher effectiveness definitively to amore organized programme.

The premise that a totally orga-nized programme is significantly moreeffective than a programme based onlargely opportunistic screening withsome central coordination and ele-ments of organized screening, such asis practised in many parts of the devel-oped world, has been challenged(Madlensky et al., 2003). Finland maybe one of the very few countries in theworld whose screening programme

218


201-226 (VOIR IMAGES) 24/01/05 11:11 Page 218

meets all the criteria outlined byHakama et al. (1985). Although it maynot be feasible to adopt all these ele-ments of organized screening, manyjurisdictions have incorporated someof them. For example, recruitmentstrategies may be targeted to specificpopulation groups in the absence of apopulation register that allows per-sonal invitations to screening and thiscan lead to achievement of low ratesof cervical cancer. For example, inci-dence rates for cervical cancer (uncor-rected for hysterectomy prevalence)and trends over the past 30 years arealmost identical in different provincesof Canada: British Columbia, whichhas had centralized cytologicalscreening and a cytology informationsystem for several decades (see, forexample, Morrison et al., 1996);Ontario, which has a recent informa-tion system including about 80% of theprovince’s screening tests and a pro-gramme that sets policy and stan-dards but where smears are taken andread in a totally decentralized system;and Quebec which has no organizedprogramme and no information sys-tem but opportunistic screening (seeFigure 67). However, unless historicalpatterns of screening and cancer inci-dence rates are taken into account,inferences from such data regardingeffectiveness are uncertain (seeChapter 4).

On the other hand, poor organiza-tion of a screening programme can produce poor outcome. In theUnited Kingdom during the 1970s and1980s, population coverage was low;low-risk groups were over-screenedand the technical quality of screeningprocess parameters was moderate.Imple-mentation of call/recall systemsand targeted rewards for primary careproviders, achieving high levels of coverage among eligible women(Patnick, 2000), resulted in a substan-tial decline in incidence and mortalityin all age groups of the target popula-

tion (Sasieni et al., 1995; Quinn et al.,1999; Sasieni & Adams, 1999, 2000).

There are, however, several com-mon problems with opportunisticscreening versus organized screening,and opportunistic screening shouldtherefore be discouraged:

1. It is less cost-effective (see below)2. Hard-to-reach women are less

likely to be adequately screened3. In many settings, especially in

developing countries, there is disproportionate representation ofwomen who are in contact with thehealth-care system for other healthinterventions such as reproductivecare, so that those in some agegroups are inadequately screened(Were & Buziba, 2001)

4. It can create sporadic work flow,which can lead to reduced profi-ciency, etc.

5. It can result in greater chances ofharm due to over-screening.

6. It may be difficult to ensure quality.7. Screen-positive women may not

have easy access to diagnosticand treatment services.

Performance indicatorsAn integrated information system or aset of systems that can be linked asrequired is recommended as the idealsupport for performance monitoring;such a system can also support pro-gramme operation (Miller, 1992).These systems should permit identifi-cation of each woman as well as eachtest and link them. A model for a com-prehensive information system isshown in Figure 68.

For performance monitoring, thesystem should ideally contain ascreening database including resultsof cytology and follow-up (colposcopy,histopathology, treatment) with periodic linkage to a population regis-ter, tumour registry, mortality file andhysterectomy data. However, even inareas where population registers

and/or other files do not exist or are notaccessible, information systems can bedeveloped that permit estimation ofmany indicators. Others can be estimated periodically by special stud-ies. A number of indicators are basedon negative test results. This generallyassumes exclusion of negative resultsfor women who are under special surveillance (e.g., following colposcopy,previous positive history, etc.).

Table 73 outlines performance indi-cators coinciding with the determinantsidentified by Pontén et al. (1995) andHakama et al. (1985), along with theprogramme area they are designed tospecifically evaluate and required datawhere relevant. These have beenadapted from Coleman et al. (1993),who proposed a menu of specific indi-cators with targets for the EuropeAgainst Cancer Programme.

All indicators should be evaluated,reviewed and published annually.Some of the indicators require multiplesources of information, sometimeslinked at the individual level.Where thisis not possible, alternative methods,such as periodic special studies (e.g.,the health and fertility surveys noted inChapter 3), should be used.

Participation (or coverage)The participation rate is the proportionof eligible women in the target popula-tion who participate in screening withinthe time interval specified by localscreening policy. It can alternatively bedefined in terms of some chosenperiod of time (e.g., tested in the lastthree years). Its estimation requires, ata minimum, counts of screenedwomen in the target age range and ofthe target population. If estimates ofthe prevalence of hysterectomy areavailable, they should be used toreduce population counts to reflectmore accurately the population at risk.Women with only inadequate smearsin the interval should not be countedas having been screened in that

219


201-226 (VOIR IMAGES) 24/01/05 11:11 Page 219

interval. Participation should be evalu-ated according to age group, geogra-phy and other locally relevant indicators(e.g., health-care provider, ethnicity).

The effect of participation in reduc-ing mortality and incidence has beendemonstrated descriptively. For exam-ple, in the United Kingdom, Quinn etal. (1999) showed that incidencedeclined dramatically starting in thelate 1980s after the introduction of acall/recall programme which resulted ingreatly increased coverage. Miller et al.(2000) stated that "… the programmemust focus on achieving the highestpossible coverage rate. To support this,indicators such as number of women

220


Figure 67 Cervical cancer incidence in three large provinces of Canada, withdifferent intensity of programmatic components, 1970–99( ) Quebec, ( ), British Columbia, ( ) Ontario

Comments

Adjust denominator for prevalence of hysterectomy.Estimate for age, region and other risk indicators

‘Recommendations’ are those set by the pro-gramme and include follow-up action (e.g., repeattest, colposcopy) and time to follow-up.Report according to reason for follow up (e.g., inad-equate, ASCUS, etc.) and person/institution respon-sible for ensuring follow-up.Reasons for non-compliance should be noted.

‘Recommendations’ include type of treatment andtime to treatment.Report and investigate as for follow-up compliance.

Requires population-based cancer registry.

Person years calculated from date of test to date ofdiagnosis, date of next expected screen or dateremoved from population at risk.Reasons for individual cases should be investigatedas a kind of audit, along with other cancers that arenot interval cancers

Table 73. Short-term performance indicators for assessing programme effectiveness and efficiency

Definition

Percentage of women in the target populationwith at least one test within the recommend-ed interval

Percentage of women (or tests) with specificpositive (or unsatisfactory) results with follow-up action according to recommendations

Percentage of women requiring treatment whoreceive it according to recommendations

Percentage distribution of stage at diagnosisfor all invasive cervical cancers in region

Number of invasive cancers diagnosed follow-ing negative test result and before next expect-ed screen per 100 000 person years at riskb

Percentage of women with negative test whoare screened again before end of screeningintervalb

Measure

Participation (orcompliance) ratea

in relation to pro-gramme policy

Compliance withrecommenda-tions for follow-upof unsatisfactorysmears and posi-tive test results

Compliance withtreatment rec-ommendations

Stage of inva-sive cancers

Interval cancers

Over-screening

Programmecomponent

Attendance forscreening

Adequacy ofreferral andtreatment sys-tems

Overall

Efficiency

a Also generally referred to as ‘coverage’. However, this is an ambiguous term as it can also refer to the percentage of population targeted for screening.b May also be expressed per 100 000 negatively screened women

Year 1970 1975 1980 1985 1990 1995 2000

201-226 (VOIR IMAGES) 24/01/05 15:21 Page 220

creo

screened, as opposed to number ofPap smears done, should be pro-moted".

Modelling has also shown that par-ticipation rate is the most importantprogrammatic determinant of effective-ness. Van Ballegooijen et al. (2000)modelled the expected reduction in life-years lost for a number of Europeancountries as a function of screening

policy and coverage. They found thatdifferences in coverage resulted inmore or less proportional differences ineffectiveness, essentially independentof screening policy (see Table 72).

High rates of participation are,however, not sufficient to ensure higheffectiveness, if other essential elements of a screening programmeare suboptimal.This is evident from the

situation in some Latin American countries (see Chapter 3 and thischapter).

In populations where the preva-lence of prior hysterectomy is substan-tial, participation rates will be underes-timated, especially in women agedover 50 years, if adjustment is notmade. Because rates of hysterectomyvary across time and place, lack of

221


Figure 68 Model for a comprehensive cervical screening information and reporting system1 Periodic linkages would uptake addresses, identify unscreened women, ascertain ‘failures’ (cancers) and remove women who no longerneed screeningFrom Marrett et al. (2002)

Data inputs

Cytology laboratories• personal identifiers• other personal data• basic smear data• added smear data

Histopathology laboratories• personal identifiers• basic biopsy/treatment data• added specimen data

Colposcopy/treatment centres• personal identifiers• colposcopic impression• treatment data

Population register• list of target population

Tumour registry• cervical cancers

Mortality file• deaths

Hospital separations• hysterectomies

Processing Database Reporting/retrieval

SelectEdit

StandardizeLink

Periodic linkages withexternal databases1

Health centres, physicians

Cytology laboratories

Histopathology laboratories

Screening programme

Government, researchers

Women

CervicalScreeningDatabase

201-226 (VOIR IMAGES) 24/01/05 11:11 Page 221

such adjustment may invalidate com-parisons of participation rates overtime and between populations.

It is most appropriate to estimateparticipation based on all screeningtests, whether from an organized pro-gramme or opportunistic.

Quality indicatorsFor programmes to be effective, allparts should be quality assured, withindicators for each part of the pro-gramme. Furthermore, the programmeneeds to have access to such qualityassurance information and to ensurethat it is fed back, along with standardsor comparisons, to those providing theservice (e.g., primary care providers,laboratories, etc.). Quality indicatorsshould cover test-taking (e.g., inade-quacy rates), interpretation (e.g., posi-tive predictive value), treatment and fol-low-up and programme organization.Acceptable ranges for any such indica-tors should be specified and values thatare outside the range should be inves-tigated. Many of these are discussed inmore detail in Chapter 3. Programmesshould regularly audit cases of cervicalcancer to identify possible shortcom-ings of the programme.

Modelling of the impact of qualityimprovements suggests that they haveless impact than improvements in cov-erage. However, in a programme withhigh coverage, improvements in quality can increase effectiveness.Goel et al. (1998) estimated that theimpact of reducing the false negativerate from 0.25 to 0.10 (i.e., increasingsensitivity from 75% to 90%), whileleaving screening otherwise un-changed, might result in 25% fewerincident cases. As noted by Fahey et al. (1995), however, even the lowerof these sensitivity values may behigher than is generally achieved.Modelling of effectiveness of screeningprogrammes must use realistic estimates of test quality if it is to provide valid estimates of effectiveness.

Follow-upPreinvasive (or early invasive) lesionsidentified via screening must be treatedif development of invasive (lethal) dis-ease is to be avoided. Thus, referral forand presentation at follow-up of a posi-tive test result as well as post-treatmentfollow-up of confirmed precancerouslesions, in accordance with treatmentpolicy, are important. It is also essentialthat test results be provided to thehealth-care provider and woman,including specification of the need forfollow-up, in a timely fashion.

Follow-up consistent with recommendationsA screening programme should have apolicy for follow-up of unsatisfactorytests and positive test results. Policyshould specify the action required andthe time frame within which this actionshould take place for specific testresults and patient history. Actual fol-low-up should then be monitored inrelation to programme policy. An unsat-isfactory test means that a woman hasnot been adequately screened and theappropriate follow-up recommendationin this situation is for a repeat test.Compliance with this recommendationshould also be monitored.

The proportions of women who arefollowed up according to programmepolicy (in terms of both action and timeframe) should be calculated accordingto reason for follow-up. Reasons for follow-up failure should be documented.

Computation of indicators of com-pliance with follow-up for positive testresults requires linkage between thesetests and follow-up data. This oftenincludes information on colposcopyvisits.

Treatment consistent with recommendationsScreening programmes should alsohave policies regarding treatment ofconfirmed abnormalities, again specifying both action and time frame.

As with abnormal or inadequatescreening test results, the proportionsof women receiving adequate treat-ment should be calculated accordingto reason for treatment. Reasons fornon-compliance should be docu-mented.

Goel et al. (1998) estimated thatimproving efficacy of follow-up andtreatment following a positive testresult from 0.8 to 0.9 might reduce thenumber of cancer cases by 2%. Theoverall impact is relatively smallbecause this improvement affects onlywomen with positive results, which rep-resent a very small fraction of all tests.Pontén et al. (1995) estimated that in aprogramme with fewer lifetime screen-ings than in the Canadian context,about half of the ultimate reduction inmortality might result from detectionand treatment of early invasive diseaseand half from removal of screen-detected precursors. In such situa-tions, an improvement in the efficacy offollow-up might have greater impact onmortality.

Overall short-term indicatorsBecause cervical screening can detectasymptomatic invasive disease, aneffective programme would beexpected to lead to a shift towardsmore microinvasive and early invasivedisease. Thus the major changes inincidence that occurred in Iceland following introduction of screeningwere among advanced cancers(Johannesson et al., 1982). Evidencefor the effectiveness of screening dueto early detection of disease can alsobe obtained by examining trends in survival (see, for example, Adami etal., 1994). However, observedincreases in survival can be due toimprovements in treatment. In general,a trend in survival is not a good indica-tor of an effect of screening (seeChapter 1).

222


201-226 (VOIR IMAGES) 24/01/05 11:11 2

Stage distribution of incident cervicalcancersThe distribution of all newly diagnosedcervical cancers in the programmearea according to stage at diagnosisshould be calculated annually. Thisrequires a population-based cancerregistry that includes standardizedstaging information on all or at least ahigh proportion of newly diagnosedcases. If such data are not routinelyavailable, periodic special studies canbe conducted. This indicator is particu-larly important in areas where screen-ing is not yet well established andscreening intervals are relatively long.In such situations, a relatively largepart of the effect of screening on mor-tality will be due to earlier stage atdiagnosis.

Interval cancersInterval cancers are those that arisefollowing a negative test and before thenext scheduled screen. These canarise for two reasons: either the previ-ous result was a false negative or the(pre)cancer was not detectable at thetime of the previous screen (for exam-ple, because it was fast-growing or didnot go through detectable preclinicalstages). A test repeated every threeyears on women aged 35–64 yearshas been estimated to ‘prevent’84–91% of invasive cancers (Day,1989; Sasieni et al., 2003) if all abnor-malities are effectively treated. In theUnited Kingdom, testing at three-yearintervals was estimated to be capableof preventing about 70% of all cancersoccurring in women aged 40–69 years,after allowing for cancers that developin women with positive test results(Sasieni et al., 2003).

The interval cancer rate is calcu-lated as the number of interval cancersper 100 000 person years at risk.Person-years at risk are estimated bysumming time from the date of the lastnegative test to the end of the recom-mended screening interval or to diag-

nosis of cancer (or until a womanbecomes ineligible due to emigration,death, etc.) for all women having anegative test. Use of this rate to assessprogramme effectiveness is difficult inregions where screening is well established, because the expected rateof cervical cancer in the absence ofscreening is unknown. However, theinterval cancer rate can be followedover time and compared across programmes with similar screeningpolicies. It may be closely paralleled bythe ratio of the number of cancers diag-nosed in screen-negative women dur-ing the screening interval divided by thenumber of screen-negative women,expressed per 100 000 women.

Calculation of the interval cancerrate requires knowledge of cancersoccurring in negatively screenedwomen. In general, this requires link-age between a population-based can-cer registry and the screening testdata. In many regions, no cancer reg-istry exists or linkage between screen-ing data and the cancer registry, atleast on a routine basis, is not permit-ted. In such cases, audit studies ofinterval cancers should be performed.

Screening histories (preceding testand other follow-up results) of all inva-sive cancers, whether they are trueinterval cancers or not, should beexamined routinely to identify areaswhere programme improvements maybe required.

Indicators of efficiencyThere are a variety of parameters thatindicate the efficiency of a screeningprogramme. The most important ofthese relates to over-screening. Thiscan be assessed by the proportion ofwomen with a negative result having asubsequent test before the end of thescreening interval. Retention ofscreened women for rescreening canbe assessed by the proportion ofscreen-negative women returning forrescreen at about the right time. Both

of these can be examined in relation tolength of interval between tests,region, smear taker, etc. The averagenumber of tests per woman during therecommended screening interval isanother indicator of the extent of over-screening.

Process quality indicatorsA number of process indicators shouldbe monitored to ensure that screeningis operating as it should. Those thatare chosen will depend on the issuesthat are important in a particular set-ting. These might include elapsedtimes (e.g., between test-taking andreporting; between reporting of a posi-tive result and follow-up colposcopy),results of laboratory proficiency test-ing, etc. Acceptable ranges for anysuch indicators should be specifiedand values that are outside the rangeshould be investigated.

The basic principle of a decision analytic approach is that all conse-quences of decisions (e.g., individualclinical outcomes, population-basedoutcomes and costs) should be identi-fied, measured and valued. When adecision analysis formally comparesthe relationship between the healthand economic consequences associ-ated with different public health careinterventions, it is considered a cost-effectiveness analysis. The applicationof economics to public policy does notnecessarily mean that less moneyshould be spent, but rather that the useof resources might be more efficient.

Different types of economic evalua-tion are commonly confused. Forexample, there are distinct differencesbetween cost-minimization analysis(how much money can be saved?) and

223


Economic evaluation andcost-effectiveness of cervical cancer screening

201-226 (VOIR IMAGES) 24/01/05 11:11 Page 223

cost-effectiveness analysis (how muchhealth improvement can be gained,per unit expenditure?). The results of acost-effectiveness analysis are sum-marized using an incremental cost-effectiveness ratio. In this ratio, allhealth outcomes associated with aparticular strategy (compared with analternative) are included in the denom-inator, and all costs or changes inresource use with a particular strategy(compared with an alternative) areincluded in the numerator. This type ofanalysis defines the ‘opportunity cost’of choosing one clinical or publichealth approach over another.

Advances in the various technolo-gies and preventive modalities for cer-vical cancer screening mean that pol-icy-makers in national and interna-tional agencies are confronted withvarious strategies from which tochoose. However, there are importantdifferences between developed anddeveloping countries in the policyquestions that are most relevant to cer-vical cancer control. Scarce resources,limited infrastructure and competinghealth priorities have prevented mostlow-resource countries from imple-menting successful cervical cancerscreening programmes.

In countries classified as low-income economies (gross nationalincome per capita equal to or less thanUS $755 in 2000), the key problem ishow to implement a sustainablescreening programme in the setting ofcompeting health priorities and limitedresources. If a cytology-based screen-ing programme is to be introduced,cost-effectiveness modelling usinglocally derived information about costsand the age–incidence curve for can-cer in the population in question,together with internationally accepteddata on efficacy, will assist in decidingon the number of screening roundsand the age group to be targeted.

For a developed country where aconventional cytology screening pro-

gramme already exists, information oneffectiveness may be available, butwhether it is cost-effective may nothave been fully established. However,it is possible to assess the likely cost-effectiveness of a new technology indetecting precursor lesions of cervicalcancer relative to conventional cytol-ogy. For example, England recentlyused cost-effectiveness modelling as amajor consideration in deciding tomove to using liquid-based cytology inits programme. This modelling beganwith the assumption that the effective-ness of new and conventional cytologywas the same (Payne et al., 2000).Theevaluation of the new technology wasthen based on costs derived locallyfrom pilot implementation of liquid-based cytology (Moss et al., 2003).The Payne et al. (2000) conclusion ofequivalence of effectiveness betweenconventional and liquid-based cytologywas based on a surrogate measure ofeffectiveness, the identification of cer-vical abnormalities, rather than long-term follow-up of outcome, that isreductions in incidence and mortalityfrom cervical cancer.

Data sourcesCost-effectiveness measures requiredata on natural history of cervical can-cer, the overall effectiveness of the pol-icy or intervention, survival rates asso-ciated with cancer, test characteristics,and quality of life.

Data sources could include: ran-domized trials, observational studies,meta-analyses; other published litera-ture, expert opinion and health systemsstatistics. However, as implied above,surrogate measures of efficacy mayhave to be used to support assump-tions relating to the likely effectivenessof new technology. To the extent thatthese assumptions are uncertain, theresult of the modelling will also beuncertain, even if sensitivity analysesare performed to attempt to encom-pass the extent of uncertainty.

Sources of cost data could includethe costs of:

• Training of staff• The screening test• Administration of the screening test• Laboratory procedures• Reporting and referral of

women with abnormalities• Diagnostic tests• Treatment of precursors• Treatment of clinically invasive

cancer• Patient time for all aspects

of screening• Transportation of specimens• Programme organization

These data must be collectedlocally or estimated according to localconditions. The eventual judgement asto whether a particular strategy is cost-effective or not will depend on localhealth circumstances.

Cost-effectiveness studiesAll models consistently support themessages that organized screening ismore cost-effective than opportunisticscreening and that the most pressingquestion in all settings is how to reachthe highest proportion of women atgreatest risk for cervical cancer.Increasing coverage is always morecost-effective than using resources inany other area of the programme. Inthe United Kingdom, altering the pay-ment system in 1990 as an incentive toprimary-care physicians to maximizecoverage rather than as a fee for ser-vices greatly facilitated the neededincrease in coverage, which rose fromaround 40% of women in 1989 to over80% of women by 1993 (NHS 2003a,b; Patnick, 2000).

Goldie et al. (2001) evaluated thecost-effectiveness of visual inspection,cytology and HPV testing, largelyusing surrogate measures of efficacy,within a developing country situation. Avery important determinant of cost-effectiveness in this setting was the

224


201-226 (VOIR IMAGES) 24/01/05 11:11 Page 224

requirement for three visits for womenwith abnormal cytological results,whereas for screening by visualinspection a one-visit strategy wasmodelled, and for HPV testing two vis-its. A substantial loss when women arerequired to return after the initialscreening visit is observed in manydeveloping countries, and this has animportant effect on the cost-effective-ness of cytology, and to a lesser extentof HPV testing.When the authors mod-elled the effect of a limited number oftests in a lifetime, three tests at five-year intervals from the age of 35 yearsproved to be more cost-effective than a10-year schedule commencing at the

same age. It was concluded that cervi-cal cancer screening strategies thatincorporate DVI or HPV DNA testingand eliminate colposcopy may offerattractive alternatives to cytology-based screening programs in low-resource settings.

Goldie et al. (2004) reported theresults of an analysis comparing thecost-effectiveness of HPV testing withthat of conventional cytology in womenaged 30 years or more. This was set inthe US context of annual conventionalcytological testing, which was com-pared with three-year screening usingliquid-based cytology and three-yearscreening using HPV testing.

[Although the latter strategy proved tobe more cost-effective in the analysis,it is unclear what the results wouldhave been if three-yearly conventionalcytology had been incorporated in theanalysis.] Goldie et al. (2004) consid-ered that for women aged 30 yearsand more, a strategy of screeningevery two or three years with eitherHPV DNA testing in combination withcytology for primary screening or cytol-ogy with reflex HPV DNA testing forequivocal results will provide a greaterreduction in cancer and be less costlythan annual conventional cytology.

225


201-226 (VOIR IMAGES) 24/01/05 11:11 Page 225

chapter 5 · 201 chapter 5 effectiveness of screening in populations this chapter deals with...

Documents