No 8785Saturday 11 January 1992

ORIGINAL ARTICLES

Systemic treatment of early breast cancer byhormonal, cytotoxic, or immune therapy

133 randomised trials involving 31 000 recurrences and 24 000deaths among 75 000 women

EARLY BREAST CANCER TRIALISTS’ COLLABORATIVE GROUP

Part 1, published last week (p 1-15), gave the general introduction and hormonal results. Part 2 nowgives the cytotoxic and immunotherapy results, and includes the general discussion of all results.

Results of chemotherapyLong-term polychemotherapy

Information is available on a total of 11 000 women with

early breast cancer in 31 randomised trials of long-termpolychemotherapy versus no chemotherapy. Information isalso available on a total of 8000 such women in 16 separaterandomised trials of preoperative, of perioperative, or ofprolonged single-agent chemotherapy versus no

chemotherapy. The long-term chemotherapy regimenslasted 2 to 36 months, the most common being 12 months,while the other regimens lasted less than 1 month. Anoverview of the results of all these trials, involving a total of18 000 women, provides statistically defmite evidence thatsome forms of chemotherapy can affect both recurrence andsurvival: in the top line of table IV, the odds reduction for allwomen of any age in any of these chemotherapy trials is 21 %SD 2 for recurrence (2p < 0-00001) and 11 % SD 2 formortality (2p<0’00001). However, the regimens are soheterogeneous, ranging from one week of single-agentchemotherapy to a year or more of multiple-agentchemotherapy, that they need to be subdivided before thesizes of their effects can usefully be described. A preliminarysubdivision is into those that involved prolongedpolychemotherapy (table IV, line 2), those that involved onlypreoperative or peri operative chemotherapy (table IV, line3), and those that involved prolonged monochemotherapy(table IV, line 4). The indirect comparison between thesethree groups of trials indicates significant heterogeneity ofeffect (2p < 0001 for recurrence-free survival and 2p < 0-05for overall survival: table IV, lines 2-4), with the greatesttherapeutic effect associated with prolongedpolychemotherapy.

In confirmation of that indirect comparison, directrandomised comparisons of long-term polychemotherapywith preoperative or perioperative chemotherapy (table IV,line 5, and fig 10, part a) or with long-term

monochemotherapy (table IV, line 6, and fig 10, part b)indicated significant differences, again with greater benefitobserved with long-term polychemotherapy. However,when long-term chemotherapy was compared with evenlonger-term chemotherapy, no significant differences wereseen between the shorter regimens (mean 6 months) and thelonger (mean 16 months): table IV, line 7 and fig 10, part c.Subsequent analyses are therefore restricted just to the trialsof prolonged polychemotherapy (table IV, line 2), which willsimply be abbreviated to "polychemotherapy".

Absolute reductions in long-term risk

Fig 11 gives the recurrence-free survival and overallsurvival during the first 10 years after randomisation for all11000 women in the randomised trials of adjuvantpolychemotherapy (ie, of chemotherapy with more than onedrug for more than one month) versus no chemotherapy.The shaded bars at the bottom of the figure indicate that themedian (and average) scheduled treatment duration in thesetrials was about 1 year. The number of recurrences anddeaths among these women was large and so the randomerrors in fig 11 are small (as are the selective errors that couldhave been produced by inadvertent omission of one or twotrials or by unduly data-dependent emphasis on particulartypes of treatment or of patient). Consequently, somestatistically definite patterns at 5 years and at 10 years afterrandomisation can be seen.

For recurrence-free survival (fig 1 1R) the main

divergence takes place during the first 5 years, and at the endof the fifth year 58-8% of polychemotherapy versus 49-6%of control patients remain alive and free of recurrence(absolute difference 9-2% SD 1-2). In subsequent years theannual recurrence rates in the polychemotherapy andcontrol groups do not differ significantly, so the gains inrecurrence-free survival during the first 5 years are notsignificantly increased or decreased after year 5.

72

TABLE IV-OVERALL EFFECTS OF CHEMOTHERAPY, AND VARIOUS DIRECT AND INDIRECT CHEMOTHERAPY COMPARISONS

a From age-standardised log odds ratio, calculated’ as h that for women under 50 plus 1/3 that for older women round parentheses () denote SD, and square parentheses []denote statistically unstable results with SD &ge; 9b Heterogeneity between three age-standardised odds reductions recurrence 2p < 0001, mortality 2p < 0 05c Heterogeneity between two age-standardised odds reductions not significant (2p < 0-05) for recurrence or for mortalityd Among node-negative patients, this 18% SD 8 mortality reduction yields 1 p < 002, but without age standardisavon the reduction is 16% S 8 8 (fig 12), 1 p = 0 03eage-standardised odds reductions for 1 to 3 involved nodes (by dissection) and for 4 + nodes recurrence 30% SD 6 and 28% SD 7, mortality 19% SD 7 and 8% SD 8

For survival (fig 11M) the pattern is different: theabsolute mortality difference at 5 years (3-2% SD 3-3) hasdoubled by 10 years (to 6-3% SD 1-4: 2p < 0-00001). Theadditional effect of polychemotherapy on the annual deathrates after year 5 was highly significant (annual death ratesduring years 5-9: 6-70% SD 0-32 polychemotherapy versus8-50% SD 0-38 control: 2p<0-001). So, the effects ofpolychemotherapy on survival are far more definite at 10years than at 5 years. Even after year 10 the survivaldifference continues to grow larger, but there is too littleextra information beyond year 10 for this additional benefit,taken on its own, to be statistically reliable.

Fig 12 subdivides these results by nodal status: at 10 yearsboth the absolute risks and the absolute benefits appeargreater (and, for survival, statistically more definite) fornode-positive than for node-negative women. Thus, forrecurrence-free survival the absolute improvement at 10years is 8-7% SD 1-5 for node-positive and 7-1% SD 2-7

(logrank test: lp<0-00001) for node-negative women(fig 12R). Likewise, for survival the absolute improvementat 10 years is 6-8% SD 1-6 for node-positive and 4-0% SD2-8 (logrank test: lp = 0-03) for node-negative women (fig12M).

Proportional reductions in annual risk

As for the tamoxifen results, analyses of proportional riskreductions were used to seek evidence of any heterogeneitybetween the effects of prolonged polychemotherapy indifferent trials, in different treatment schedules, and indifferent types of patient. The results for the 32 separatepolychemotherapy comparisons are given in fig 13 in thesame format as fig 4.

For recurrence-free survival polychemotherapyproduces, overall, a highly significant 28% SD 3 reduction(2p < 0-00001) in the annual hazard (fig 13R), this being anaverage of a large effect in the first 5 years and no significantadditional effect thereafter (fig 11R).

For survival polychemotherapy produces, overall, a

highly significant 16% SD 3 reduction (2p < 0-00001) in theannual hazard (fig 13M), representing a persistentdifference both in the first 5 years and in the next 5 years (fig11M). This mortality difference is not large, so these trialscan assess it reliably only if information from many of themis combined, since no single one of them is sufficient on itsown to achieve a result that is both significant and plausible.The only regimen tested in many trials is CMF

(cyclophosphamide, methotrexate, 5-fluorouracil), usuallyfor about 12 months, and separate subtotals are given for(a) the CMF trials (4000 women), (b) the trials of CMFplus other drugs (3000 women), and (c) the other

polychemotherapy trials (4000 women), but there is no

significant heterogeneity between these three subtotals

(=3-0 for recurrence and 35 for mortality). Indeed,

73

taking the 32 polychemotherapy comparisons as a wholethere is, despite the heterogeneity of their designs, onlylimited heterogeneity between their 32 results (p = 0-06 and0.04, respectively, for heterogeneity in figs 13R and 13M).Moreover, the only separately significant sources of thisheterogeneity were the unusually favourable results in theGuy’s/Manchester CMF trial (79E)13 and, particularly, inthe small Milan 8004 trial (80F)14 of CMF for node-negativepatients, neither of which is reinforced by the other 9 CMFtrials. Apart from this, virtually no heterogeneity remains.So, the best estimate of the real effect of CMF alone may begiven by the overview of all 32 polychemotherapy trials, notjust the 11 CMF trials. However, even though these resultsprovide no clear evidence of it, common sense suggests thatsome of these chemotherapy regimens must be better thanthe overall average, and that some must be worse.

Indirect comparisons between different treatment

durations are not reliable, because other aspects of thechemotherapy regimens may be confounded with duration;nevertheless, there is in fig 13 no tendency for the apparentefficacy of the 13 treatment regimens that lasted less than ayear (median 6 months) to differ from the overall average,reinforcing the directly randomised comparisons in fig 10.

Like the treatments, the patients themselves can besubdivided in various ways. Fig 14 describes the effects ofsubdivision of the polychemotherapy results by age and bymenopausal status before randomisation. These trials

provide information about the effects of polychemotherapythroughout middle age, but provide virtually no informationabout women aged over 70.For recurrence-free survival (fig 14R) there is a

statistically significant (2p = 0-0006) trend towards theeffects of polychemotherapy being greater at younger than atolder ages; but even among women aged 60-69

polychemotherapy still produces a 20% SD 5 delay ofrecurrence, which is highly significant (lp<0’0001). So,any difference in response between these women and

younger women is merely a "quantitative" difference in thesize of the response, not a "qualitative" difference betweenno response and some response. Age is related to

menopausal status, but the relationship is so close that it isnot possible to disentangle the effects of these two factors onthe response to polychemotherapy. Among women under50 only 225 were known to be postmenopausal, while amongwomen over 50 only 911 were known to be premenopausal.These numbers with atypical menopausal status for theirage are too small to determine reliably whether, amongwomen of similar age, menopausal status is correlated withthe response to chemotherapy (fig 14R).For survival (fig 14M) the general pattern resembles that

for recurrence-free survival, and the trend towards greatereffect among younger women is again significant(2p=0’01), though less than for recurrence. The effect ofpolychemotherapy on mortality is statistically significantamong women aged >50 (13% SD 4, 1p< 0-002), butappears to be only about half as big as among youngerwomen (24% SD 5, Ip < 0-00001: table iv, line 2).Age-standardised analyses of subgroups of the

polychemotherapy trial results with respect to concurrenttamoxifen and with respect to nodal status are given in thebottom part of table iv. These and the other subgroupanalyses in table iv are further subdivided at age 50,although this makes most of the numbers too small to bestatistically reliable.For the question of whether tamoxifen prevents

chemotherapy from working properly, trials are needed not

only of chemotherapy versus control in the absence oftamoxifen, but also of chemotherapy versus control in thepresence of tamoxifen. The latter trials include only a fewhundred women under age 50, so this comparison has to berestricted to women aged over 50, among whom tamoxifendid not seem to interfere with the efficacy of chemotherapy.In the absence of tamoxifen and in the presence of

tamoxifen, the effects of polychemotherapy among theseolder women involved odds reductions of 22% SD 4 and26% SD 5, respectively, in recurrence-free survival and of14% SD 5 and 10% SD 7, respectively, for survival (tableIV). For nodal status, the age-standardised results indicatethat the proportional reductions in annual risk are similar fornode-negative and for node-positive women, so that (as infig 12) the 10-year absolute risk reductions are larger fornode-positive than for node-negative patients. If,artificially, attention is limited just to women with node-negative disease then there is still a 26% SD 7 (lp < 0.0001)improvement in recurrence-free survival, and an 18% SD 8(lp = 0-03) improvement in overall survival (table IV, fig 12).

Other outcomes

Information on contralateral breast cancer was availablefrom about half the patients in the polychemotherapy trials.Contralateral breast cancer was reported in 42/2290 (1-8%)of the control patients compared with 40/2351 (1-7%) of thepolychemotherapy patients (logrank O-E -3-9 andvariance 19-8: odds reduction 18% SD 20, NS).

In three-quarters of the trials, confirmation was generallyavailable that any deaths before relapse were not due tobreast cancer (total: 297 non-breast-cancer deaths beforefirst relapse), but specific causes were often not available.There was, however, no apparent effect of

polychemotherapy on these deaths (logrank O-E 2-6 andvariance 71-7: odds reduction -4% SD 12, NS).

Results of immunotherapy6300 women were randomised into a total of 24 trials of

immunotherapy (8 of BCG, 9 of levamisole, and 7 of otheragents). Neither in total, nor in any of these three subtotals,nor in any of the 24 separate trials was there any significantlyfavourable difference between treatment and control inrecurrence-free survival or in overall survival (fig 15: theodds reduction was - 3% SD 4 for both of these two

endpoints). Indeed, for BCG the subtotal was significantlyadverse, though not strongly so (2p = 002). These resultsapply directly only to the specific types of immunotherapytested, and not to newer types. It is, however, of somemethodological interest to note that an overview can yield astrongly null result, and in particular to contrast the

apparently adverse effects of BCG (20% SD 8 greatermortality, 2p = 0-02) in the present overview of randomisedtrials with the claims of benefit15 for it in "historicallycontrolled" comparisons.

Discussion

This second cycle of the worldwide collaboration betweenearly breast cancer trialists involves the largest amount ofproperly randomised evidence ever brought together forreview of the effects of any forms of cancer therapy onlong-term outcome. It can therefore assess reliably thehighly significant effects of polychemotherapy and ofendocrine therapy (by tamoxifen and, for younger women,by ovarian ablation) on 10-year outcome.

74

Fig 10-Prolonged polychemotherapy versus three other types of chemotherapy.

Women in each of the three trials (77A, 80H, 81 F) comparing prolonged polychemotherapy with pre/perioperative chemotherapy were almost

Comparison of benefits during years 0-5 and years 5-10

For ovarian ablation the trials include fewer than 2000

younger women, which is not enough for the time-

dependence of the main benefits of such treatment to beassessed reliably (fig 7). For tamoxifen, however, there are30 000 women randomised between treatment and controland for polychemotherapy there are 11 000, and in bothcases the time-dependence of the main benefits is rathersimilar: for recurrence-free survival (figs 2R and 11R) themain divergence between treatment and control takes placeonly during the first 5 years, whereas for overall survival(figs 2M and 11M) a fairly steady divergence takes placethroughout years 1-10. In particular, although thetreatments are both relatively brief-median only 2 years fortamoxifen, median only 1 year for polychemotherapy-theadditional divergences in mortality after the first 5 years thattook place many years after the treatments had ended areboth highly significant (each 2p<0-0001). This was not

anticipated by most of the collaborators, who were askedbeforehand to give a written prediction of the extra survivalgains during years 5-10: 78 trialists replied, and the means of

their predictions were zero for polychemotherapy and zerofor tamoxifen, with some predicting small extra gains, somepredicting slight losses, but not one predicting additionalsurvival gains during years 5-10 as large as those in figs 2Mand 11M. The present overview does include a greaterpercentage of good-prognosis patients (particularly fromJapan, and from trials in node-negative disease) than the1985 overview did,l but this does not bias the size of theproportional risk reduction either in years 0-5 or in years5-9.

Validity of main findings, and of some subgroup analysesThe general validity of systematic overviews in general,

and of these methods in particular, is more widely acceptednow than when this collaboration first began." Because themethods are valid and efficient, the previous estimates’2 ofthe effects of treatment during the first 5 years are notmaterially altered (see years 0-5 in figs 2M and 11M) by thepresent threefold increase in information. What is new,however, is that there is now substantial evidence on 10-yearoutcome, and that more information is now available for

75

all "node-positive". Whereas about three-quarters of the women in 81 F were aged under 50, only about one-third of those in 77A and 80H wereunder 50 Greater benefit was observed with prolonged polychemotherapy in these trials among women aged under 50 (48% SD 9 recurrencereduction, 37% SD 12 mortality reduction) and 50 or over (33% SD 11 recurrence reduction; 27% SD 13 mortality reduction).

tentative exploration of subgroup analyses by age, stage, orother characteristics. The overall results still involve only amoderate-sized benefit, however, so there may still be

false-negative or exaggeratedly positive results if analyses ofparticular trials or particular subgroups are interpretedwithout reference to the overall results, or without somecross reference between subgroup analyses of recurrence-free survival and those of overall survival. (If the differencebetween treatment and control in some particular subgroupis not conventionally significant, then there are still threepossibilities to investigate-there may be good evidence ofno benefit, there may be no good evidence of benefit, orthere may, when a wide range of evidence is reviewed, begood, but indirect, evidence of benefit.)For helping determine patient treatment, there are two

very different types of subgroup analysis. First, there aresubgroup analyses that directly assess the effects of mutuallyexclusive treatments (eg, tamoxifen duration: 1 year, 2 years,or over 2 years), where a choice has to be made between thesesubgroups if any such treatment is to be given. Second, thereare all the other subgroup analyses (eg, d and e in table III),which do not directly assess mutually exclusive treatments.

For mutually exclusive treatment regimens, the choice ofwhich to adopt may (until better evidence becomes

available) have to depend on non-significant outcomedifferences between the apparent effects in particularsubgroups, on biological plausibility, or on convenience.For other subgroup analyses diametrically oppositeconsiderations may apply, since appropriate emphasis onthe overall results may mean that even clearly significant"quantitative" differences between the sizes of the effects ofa treatment in different patient subgroups do not imply"qualitative" differences between those subgroups inwhether or not that treatment would be of benefit: seeIntroduction. In the previous overview, subgroup analyseswith respect to nodal status, hormone receptors, and agewere of particular concern.

Nodal status: effects in node-negative patientsFor recurrence, the evidence is clear that nodal status has

little effect on the proportional risk reductions produced bytamoxifen and by polychemotherapy, both of which arehighly significant even for node-negative patients (each

76

Fig 11-Ten-year outcome in polychemotherapy trials: overall results for all 11 000 randomised women.

Because the rates after year 10 are based on small numbers, combination of the data for years 5-9 and 10+ yields annual rates in years 5+ thatfavour active treatment (recurrence 6-05% SD 032 vs 6-72% SD 038, NS mortality 6-46% SD 0 30 vs 8-30% SD 0 36, difference 1-84% SD 0-46,2p < 00001 ) The shaded bars show, just for those with a first event in each year, the proportions who, if allocated active treatment, would have beenscheduled still to be on polychemotherapy the median (and the mean) scheduled polychemotherapy duration was 1 year

lp < 0-0001). For mortality, the age-standardisedproportional reductions among node-negative patients intables III and IV are also both significant (tamoxifen 17% SD5, lp < 0.001; polychemotherapy 18% SD 8, lp < 0-02), andare the same size as the reliably known mortality reductionsproduced by these treatments among node-positivepatients. Hence, the pattern that was definitely true forrecurrence (ie, the approximate irrelevance of nodal status tothe age-standardised proportional risk reduction) appearsalso to be true for mortality. This may somewhat reinforcethe estimated sizes of the proportional mortality reductionsamong node-negative women. Still, however, the absolutereduction in 10-year mortality will be considerably less innode-negative than in node-positive women, and it is theabsolute risk reduction that helps determine whethertreatments are worthwhile in human terms.Both for tamoxifen and, particularly, for chemotherapy

the node-negative subgroup already involves so few deathsthat further subdivision of it by any other characteristic (eg,age: see the irregularities in table IV) may well yieldstatistically uninformative results. The ovarian ablationresults for women aged under 50 are likewise too sparse forsubdivision by nodal status to be statistically reliable, so theplay of chance may have inflated the large benefit thatablation seems to produce in node-negative women agedunder 50 (see legend to fig 8).

Oestrogen receptor measurements

The effects of tamoxifen on long-term outcome werelargest if the excised primary tumour was oestrogen-

receptor (ER) positive, but were still significantlyfavourable, especially above age 50, if it was classified as ERpoor (table III). One-quarter were classified as ER poor, butthe measurements were of variable quality (including,inevitably, some false negatives) and the cut-off was

generally 10 fmol/mg. Detailed receptor studies, with alower cut-off, could presumably identify a smaller

proportion of women with somewhat17 less benefit fromtreatment-but, even among truly receptor-negativewomen tamoxifen might also reduce contralateral9 and/orvascular’8 disease.

Age-specific proportional reductions in annual risk fromendocrine therapy, chemotherapy, and indirectly, chemo-endocrine therapy

In countries such as Britain about 40% of breast cancer

patients are over 70, about 40% are aged 50-69, and onlyabout 20% are under 50 when diagnosed.8 For women over70 only one treatment, tamoxifen, has been extensivelytested, and it produced very definite risk reductions (fig 5:recurrence 28% SD 5, mortality 21% SD 6, or perhapsslightly more, eg, 33% and 25%, if restricted to trials of atleast 2 years of tamoxifen).

For women aged 50-69, two treatments each produced ahighly significant (lp<0’00001) percentage reduction inrecurrence (fig 5, fig 14, and table v, sections la and 2a:tamoxifen alone 30% SD 2, chemotherapy alone 22% SD4). The evidence, however, is still reasonablystraightforward, for only two treatments (and hence onlyone "interaction") are to be studied, large numbers of

77

Fig 12-10-year outcome in polychemotherapy trials, subdivided by nodal status.

Among the node-negative women, the balanced totals were 273/1411 polychemotherapy vs 283/1360 control deaths (logrank 0- E = - 20 7, variance121.2,1 1p=0 03), of which 18 vs 23 (1 3%vs 1 1 ’7%) were deaths before relapse attributed to causes other than breast cancer and 228 vs260 (16,9% vs191%) were other deaths, a proper overview of which yielded logrank O-E= -18-5, variance 111-9, odds reduction 15% SD 9,1p=0’04.

women have been randomised, and in this age range both forrecurrence and for survival the effects of the two treatments

appear approximately independent of each other. Hence, areasonably reliable estimate of the effects of combinedchemo-endocrine therapy can be constructed by combiningresults from trials of chemotherapy with results from trials oftamoxifen (see footnote to table v for details). In this age rangethe effects of tamoxifen and the effects of chemotherapyappear to be largely independent of each other (table v,sections 1 and 2). Hence, different methods of combiningtheir effects can give similar results (table v, section 3). Onepossible method (table v, section 3a) is to combine the effectsof tamoxifen alone (from trials of Tam vs nil) with the effectsof adding cytotoxics to tamoxifen (from trials of CTX + Tamvs Tam). Another method (table v, section 3b) is to combinethe effects of chemotherapy alone with those of addingtamoxifen to chemotherapy, while a third method (table v, 3c)is to ignore completely any possible interactions. These threemethods all give virtually identical results, with the last resultintermediate between the first two. Hence, allocation to acombination of polychemotherapy and tamoxifen wouldproduce risk reductions of about 45 % SD 3 in recurrence and30% SD 4 in mortality (table v, section 3c)&mdash;or, perhaps,slightly more, eg, 50% and 33%, if restricted to trials of over 2years of tamoxifen. Among women over 50, tamoxifenproduced significant benefit even among ER-poor women(recurrence reduction 16% SD 5, lp = 0-001, and mortalityreduction 16% SD 6, 1p=0.004).Below 50 years of age, three treatments each produce a

highly significant (lp< 0-001) percentage reduction in

recurrence (table VI, sections la, 2a, and 3a): tamoxifen alone27% SD 7, chemotherapy alone 37% SD 5, and ovarianablation alone 30% SD 9. Among premenopausal or,

particularly,13 perimenopausal women chemotherapy canhave a double action, involving not only some cytotoxicitybut also, in most women.,19 ovarian ablation. The

therapeutic effects of such ovarian ablation may well besubstantial20 but must depend on what endocrine treatmentsare also to be given. Among such women tamoxifen mightalso have a double action, but of an opposite type, for itsantioestrogenic effects induce a compensatory increase incirculating oestrogen ’2122 unless the ovaries are preventedfrom producing this. With three significantly effectivetreatments there are several "interactions" to be studied, butunfortunately, although the number of questions is largerthan at older ages the number of randomised patients issmaller (table VI). Hence, the combined effects of more thanone of these three treatments among women aged under 50cannot yet be estimated reliably.

If, among women under 50, attention is restricted tomortality then at present only two of these three treatmentshave clearly significant effects (table VI, sections 2a and 3a:chemotherapy alone 27% SD 6, and ovarian ablation alone28% SD 9). Even for just these two treatments the trialresults do not allow a reliable assessment of any"interaction" between their main therapeutic effects. Thisinteraction may well be substantial since chemotherapyoften produces permanent cessation of menses: for instance,in one trial19 two-thirds of all premenopausal women ceasedmenstruation permanently after polychemotherapy, while

78

Fig 13-Separate results for all 32 polychemotherapy comparisons, subdivided by the cytotoxic drug regimen.

Only 31 tnals are represented since tnal 76H generates two comparisons.

in another13 97% of those aged over 40 (but only 37% ofyounger women) ceased menstruating, so these two

treatment effects may well be much less than additive.

However, since chemotherapy does not always cause

cessation of menses, the sequence of chemotherapy thenablation may well be somewhat more effective than

chemotherapy alone, at least in women with persistentovarian activity.

For the third treatment, tamoxifen, qualitativeextrapolation from its effects on recurrences and from itseffects in other circumstances indicates that women under50 given ovarian ablation alone, given chemotherapy alone,or given ablation plus chemotherapy should on average gain

some additional survival benefit from tamoxifen (at least ifstarted after completion of any other treatments, to avoid anydirect interference between cytotoxic and hormonal effects,and continued for some years). Quantitatively, however, it isnot yet possible to predict reliably the size of the survivalbenefits of adding long-term tamoxifen to the treatment ofyounger women: they could be small or (as is suggested bythe premenopausal ECOG 5181 trial,23 fig 6, and by theER-positive trial of 5 years of tamoxifen3O) substantial.

Correction for non-complianceTrial analyses by allocated treatment provide an unbiased

test of whether or not treatment has any real effect, but

79

provide a biased underestimate of the size of that realeffect-that is, of the difference in outcome that would havebeen seen with full compliance among those allocated activetreatment and with no such adjuvant therapy at all amongthe controls. (For endocrine therapy correction of this biasmight perhaps change the 19% proportional mortalityreduction to an average reduction of at least 20-25%.)Hence, the expected effects of actual use of the trialtreatments are likely to be somewhat better than in theanalyses of allocated treatment throughout the results, andin table v and VI.

Absolute improvements in 10- year survival

The effect of treatment on the average duration of survivalwill not be known for decades, and even estimates of median

survival may be unreliable (see, for example, figs 7M and8M). Instead, therefore, the effects of treatment on survivalare described either by proportional reductions in annualmortality rates, or by absolute differences in long-termsurvival. If, as seems generally to be the case, poor-prognosisand good-prognosis patients derive fairly similar

proportional reductions in risk, then the absolute reductionin 10-year risk is greater for the poor-prognosis than for thegood-prognosis women. (For instance, a 30% proportionalreduction in the annual mortality rate might improve the10-year survival by 12 per 100 stage n patients, but by onlyabout half as many stage I patients: fig 1, table I.) But, withmuch longer follow-up the absolute benefits for stage I

disease, especially in younger women, may become

substantial (see, for example, the 30-year follow-up of the

80

Fig 14-Subtotals by age and menopausal status of polychemotherapy trial results.

Among women aged under 60, 11 % had unknown menopausal status (10% were from trials 76K and 81 E, 1 % from other trials), but exclusion of suchwomen from the analyses never alters the last digit of the percentage risk reduction by more than one

first randomised trial of ovarian irradiation in

premenopasual stage I breast cancer4).For stage n disease, in which half or more of the untreated

women die within 10 years, the corresponding differences in

TABLE V-AGE >50: INDIRECT ESTIMATION OF THE EFFECTS OFCONCURRENT PROLONGED POLYCHEMOTHERAPY AND ABOUT

2 YEARS OF TAMOXIFEN

I I I

Few in the chemotherapy comparisons were aged over 70, so all but the first of theseestimates are of directly demonstrated relevance only at 50-69 years of agea Mean scheduled tamoxifen durations for those dying in the trials of tamoxifen (1 a)2 1, (1 b) 1 -8, and (1 c) 2 0 yrb The process of estimation simply mvolves letting one percentage reduction followanother for example, in (3a) a 30% reduction in recurrence (from 1 0 down to 0-7)followed by a 25% reduction (from 0 7 down to 0 52) combine to yield a 48%reduction (from 1 0 down to 0 52) Likewise, prevention of ’/-- and then preventionof 1/3 would combine to prevent ’12

10-year survival would be about 8-12 per 100 women givenendocrine therapy alone (ie, ablation in women under 50 andtamoxifen in those aged 50 or over), and about 12 or moreper 100 middle-aged women treated with combined chemo-endocrine therapy (table VII). The effects for women under50 are uncertain, but are likely to be at least as great as forolder women (see notes to table vil). Among stage II

patients, fig 1 shows that the sizes of the absolute differencesin 10-year survival may well be largely independent ofimportant prognostic factors such as the actual number ofaxillary lymph nodes in which cancer is detectable. (Thissurprisingly simple conclusion applies if, as in tables III andIV, the proportional risk reduction is similar for good-prognosis and for poor-prognosis stage II disease.)For stage I disease in which only about a quarter of the

untreated women die within 10 years, the correspondingdifferences in 10-year survival would be only about half asgreat as in table VII. For stage I breast cancer with a

particularly good prognosis (such as, for example, might befound by mammographic screening) the absolute benefitsmight well be even smaller (see table I). For instance, if thesame proportional risk reductions apply to patients of whomonly 10% will eventually die of their breast cancer, then theabsolute differences in 10-year survival might be only about2 per 100 women treated with endocrine therapy alone, and3 per 100 treated with combined chemo-endocrine therapy.

Direct and indirect comparisons

Questions such as the relative merits of several yearsversus only a few years of tamoxifen may be addressed eitherby a direct randomised comparison (as in fig 6) of oneduration versus another, or by an indirect randomisedcomparison (as in fig 4) between the results of the varioustrials of tamoxifen versus no tamoxifen. Indirect

comparisons do not compare the treatment group in one trialdirectly with that in another. Instead, each treatment group

81

* For balance, controls from 84C count twice just in deaths/patients

Test for age trend = 2-5 SD; 2p = 0.01

Test for menop. trend = 1 6 SD; NS

is first compared with its own concurrent control, and onlythen are these results from different trials compared witheach other. (This procedure assesses whether the advantageof the tamoxifen group over its concurrently randomisedcontrols is greater in trials of several years of tamoxifen than

TABLE VI-AGE <50 INDIRECT ESTIMATION OF THE EFFECTS OFCONCURRENT PROLONGED POLYCHEMOTHERAPY AND

OVARIAN ABLATION

I I

Separate emphasis on those under 50 is a data-derived subgroup analysis based onlimited numbers of women, so the estimates m this table may be of limited reliabilityunless taken in conjunction with those m table V

a Square parentheses [] denote statistically unstable results with SD &ge;9b Mean scheduled tamoxrfen durations for those dying in thetrials oftamoxifen. (1 a)2 6, (1 b) 1 6, and (1 c) 1 8 yrC Chemotherapy produces ovarian ablation, leading to cessation of menses in most,but not in all, premenopausal women This suggests the combined effects ofchemotherapy then ovarian ablation will be less than from ignoring any interactionsand treating each as independent, but more than from giVing either separately (ie,indicating a mortality reduction definitely less than 45% and definitely more than25%)

Treatment effect 2p < 0.00001

in trials of only a few years: fig 4). The weight that can begiven to such indirect randomised comparisons (or, indeed,to extrapolation of any trial results to other circumstances) isa matter of judgment on which opinions differ and on whichthe present report takes no definite stand. Physicians whoare convinced by particular such comparisons may wish totreat their patients accordingly, while those who remainsubstantially uncertain may wish to collaborate in directlyrandomised trials that provide further evidence.

TABLE VII-PROPORTIONAL AND ABSOLUTE IMPROVEMENTS IN

10-YEAR SURVIVAL FROM TREATMENT OF 100 MIDDLE-AGED

WOMEN WITH STAGE II BREAST CANCER

-

Absolute benefit might be about half as great for stage Ia Results cited are for analysis (by allocated treatment) of women m all tamoxifentrials, so the median tamoxifen duration IS only 2 yr but more prolonged tamoxifenappears better (table III).b Estimates of the effects of one or other treatment alone for age < 50 yr are

somewhat unreliable (1a, 2a, or 3a in table VI), owing to small numbers anddata-dependent emphasis on this particular age-range The overall results for thesetreatments (1 c, 2c, or 3c in tablevi) are subject to smaller random errors, but may welltend to underestimate the effects of these treatments The proportional reductions mannual mortality produced by polychemotherapy alone and by ovarian ablation aloneshould therefore be at least as great as in (2c) and (3c) m table V) that is, (i) at least25% SD 5 for polychemotherapy alone, (II) at least 25% SD 7 for ovarian ablationalone, and (ui) at least 30% for the combination (the lower limit of the range in 4a mtable VI) This would yield absolute 10-year mortality reductions for women aged< 50 yr with stage I I disease of at least (I) 10 deaths (SD 2) per 100 women treatedwith polychemotherapy alone, (II) 10 deaths (SD 3) per 100 women treated withovarian ablation alone, and (m) about 12 deaths per 100 women treated with thechemo-endocrine combination.

82

Fig 15-Separate results for all 20 immunotherapy trials, subdivided by the immunotherapeutic agents tested.

Implications for treatmentTreatment regimens evolve, often for the better, yet the

10-year results now available may be from regimensdesigned about 20 years ago, such as CMF 25,26 (given lessintensively than might nowadays be possible), or an averageof just 2 years2’as of tamoxifen (whereas 5 or more yearsmight nowadays be considered routinely:2329.30 table III).The definite effects of these old regimens on 10-yearmortality therefore provide only lower limits as to whatadjuvant treatment might now offer.For tamoxifen, the most widely tested regimen is

20 mg/day for 2 years. Indirect randomised comparisons(standardised, where appropriate, for age and ER status)suggest that this daily dose may be sufficient, but thatlonger-term tamoxifen may well be better (fig 4, table III: seealso the few directly randomised comparisons of differenttamoxifen durations in fig 6). Trials of a few years oftamoxifen versus 5 or more years of tamoxifen are currentlyin progress, but many began randomising after Jan 1, 1985,and long-term follow-up data on substantial numbers maynot be available until the mid-1990s. Meanwhile, it may notbe appropriate to equate the effects of 5 or more years of

tamoxifen with those of the present overview of an averageof only 2 years of treatment.

For prolonged polychemotherapy, the most widely testeddrug combination is CMF, so this is the only one with aseparately demonstrated survival advantage in the presentoverview (even though others appear at least as promising,such as the B-1631 trial of only a few months of doxorubicin-based chemotherapy: fig 13). Both direct and indirect

comparisons show long-term treatment (eg, 12 months) tobe no better than shorter-term treatment (eg, 6 months) withthese types of chemotherapy.

Overviews of many trials are neither necessary nor even

particularly helpful for assessing the common acute side-effects of chemotherapy, or those of endocrine therapy forpremenopausal women (by surgical ovarian ablation,"radiotherapeutic ovarian ablation,24,32 or tamoxifen33). Atpresent, the only apparent effect in any of the overviews onnon-breast-cancer death was that tamoxifen (which reducesblood cholesterol concentrations substantially18) was

associated with a marginally significant 25% SD’ 13reduction in vascular mortality. When detailed causes ofdeath become available these overviews will help assess any

83

rare, or delayed, side-effects but even then they may notsuffice to detect small risks of inducing or preventing somespecific types of cancer by chemotherapy34 or by ovarianirradiation.35Treatment recommendations depend on a wide range of

considerations, of which trial results are only one part.Hence, the present report makes no recommendations as towhat treatments should be used. Trial results-or, better,overviews of them-provide information, not instruction, tothose concerned with treatment. Where individual trials oroverviews do produce very definite answers-as, for

example, the effects on 10-year survival of tamoxifen, ofpolychemotherapy, and of ovarian ablation (table VII),which are far more definite now than at the time of the recentUS36,37 and UK 38 consensus conferences-those treatingearly breast cancer should merely make themselves aware ofthose answers, and of the reliability of the methods thatproduced them. Uncertainties as to the exact value ofcombined chemo-endocrine therapy for younger womenshould not divert attention from the very definite effects ofthe separate components of this combination.

Implications for research

The present overview should help generate many newtrials, for even among older women the optimum duration

of tamoxifen remains uncertain (as, among youngwomen, does the effect of adding tamoxifen to ablation)and for chemotherapy the range of unanswered questionsabout drugs, duration, and treatment dose intensity isalmost limitless. Finally, endocrine therapy and

chemotherapy are not mutually exclusive, so may notneed to be randomly compared, but the combination(given concurrently or consecutively in a wide range ofpatients) may need further comparison against each

separately.The most important research implications, however, are

not about what questions to ask but about how to answerthem, both in breast cancer and in many other diseases.Reliable assessment of moderate survival differencesrequires not only proper randomisation of large numbers(appropriately governed by the "uncertainty principle"1&deg;39)but also collaboration between those who randomise.Non-randomised methods are, with the growth of

computerised databases, inevitably going to be elaborated,often at considerable expense, but cannot reliably excludemoderate biases. Randomised methods can avoid biases andare, with the growth of large overviews, now capable ofavoiding not only false-positive but also false-negativeresults. Every decade millions of women are treated for earlybreast cancer, and in the treatment of just one millionwomen mortality reductions such as those now

84

demonstrated could well prevent or substantially delay100 000 deaths.

This report is dedicated to David Byar (1938-1991) and Sir AustinBradford Hill (1897-1991), two wise and humane clinical trial statisticians.The continuing collaboration of breast cancer trialists is funded by a specialgrant from the Imperial Cancer Research Fund to the ICRF/MRC ClinicalTrial Service Unit in the Nuffield Department of Clinical Medicine,University of Oxford. The chief acknowledgement is to the many thousandsof women who took part in these trials and so helped determine how best totreat breast cancer, and to the many medical, statistical, and administrativetrial investigators who carefully checked any queries and provided details ontrials. The trial groups and trialists who collaborated in the 1990 overview

were, in alphabetical order of group, of institute, or of location:

AGETBC, Japan: O. Abe, R. Abe, K. Asaishl, K. Enomoto, T. Hattori, Y. Ino, K.Kikuchi, H. Koyama, K. Sawa, J. Uchino, M. Yoshida. Addenbrooke’s Hospital,Cambridge: J. L. Haybittle. Amsterdam, Integraal Kankercentrum: T. van de Velde, J. B.Vermorken. Auckland Breast Cancer Study Group: V. J. Harvey, T. M. Holdaway, R. G.Kay, B. H. Mason. Australaan-New Zealand Breast Cancer Tnals Group: J. F. Forbes.BMFT Germany: G. Bastert, W. Sauerbrei, H. Scheurlen, M. Schumacher. BelgzanAdjuvant Breast Cancer Project: C Focan, J P. Lobelle. Berlin-buch, Akademie derWissenschaften: U. Peek. Birmingham General Hospital: G D Oates, J. Powell. Bordeaux,Fondation Bergoni&eacute;: M. Durand, L. Maunac. Boston, Dana-Farber Cancer Insntute: R. S.Gelman, 1. C. Henderson, C. L. Shapiro. Bradford Royal /M/:rMMry.’ A. K. Hancock.British Columbia Cancer Agency: S. Jackson, J. Ragaz. Cancer & Leukenna Group B: I. C.Henderson*, A. Korzun, W. C. Wood. Cancer Institute Hospital, Tokyo: M. Yoshimoto.Cancer Research Campazgn: M. Baum,J. Houghton. Cardiff Surgery Tnalists: K. Horgan,L. Hughes, H. J. Stewart. Case Western Reserve Umversaty: N. H. Gordon. CentralOncology Group, U.S.A.: H. L. Davis. Centre Regional Fran&ccedil;ois Baclesse, Caen: T.,Delozier, J. Mace-Lesec’h. Centre Rene Huguenin, St Cloud: P. Rambert. CharlesUniversity, Prague: O. Andrysek, . Barkmanova. Cheltenham General Hospital: J. R.Owen. Chrism HospItal & Holt Radium Institute: A. Howell, G. C. Ribeiro, R. Swmdell.Coittibra Instituto de Oncologia: C F. de Oliveira. Copenhagen, Damsh Cancer Registry : B.Carstensen, T. Palshof. Copenhagen Radium Centre: H. Johansen Cracow Institute ofOncology : S. Korzmiowski, J. Skolyszewski. Danuk Breast Cancer Cooperative Group: K.W. Andersen, P. Dombernowsky, H. T. Mouridsen, C. Rose. Dublin St Luke’s Hospital:N- Corcoran. Dusseldorf University: H. J. Trampisch. Eastern Cooperative OncologyGroup. M. D. Abeloff, P. C. Carbone, J Glick, R. Gray, D. C. Tormey. EuropeanOrganization for Research and Treatment of Cancer M Buyse, F. Mignolet, R. Paridaens,O. J. Repelaer van Dnel, R. J. Sylvester, C J. H. van de Velde, J A. van Dongen, K.Welvaart. Evanston Hospital, U S A.: E. F. Scanlon, S. Schurman. Fox Chase CancerCenter, Phaladelphia: R. Catalano, R. H. Creech. Ghent University Hospital: A de

Schryver Glasgow Beatson Oncology Centre: K. McGregor, H. M. A. Yosef. GlasgowVictorca Infirmary C. S McArdle, D. C. Smith Granada University Hospztal’ P. C. Lara.Groote Schuur Hospztal, Cape Town: D. M. Dent, C A. Gudgeon, A. Hackmg. GruppoRicerca Ormono Chemio Terapia Adiuvante, Italy F. Boccardo. Gunma University,_7apanM. Izuo Guy’s Hospital, London. A. Bentley, Z. Doran, I. S. Fentiman, J. L Hayward, R.D. Rubens. Gynecologzcal Adjuvant Breast Group, Germany: M. Kaufmann, W. Jonat.Headelberg Urtiverszty I: H. Scheurlen Heidelberg University II D. von Foutmer, MKaufmann. Helsinki, Deaconess Medical Centre. P. Klefstrom. ICRF, London: J Cuzick.Innsbruck University: R. Margreiter. Institut Curze, Paris: B. Asselain, P. Pouillard.Instztut Salah Azazz, Tunis: J Bahi. Instituto Policlinico, Barcelona: A Milla, F SanchizInternational (Ludwig) Breast Cancer Study Group: M Castiglione, F. Cavalh, J. Collins,R D Gelber, A. Goldhirsch, M. R Isley, J. Lindtner, K. N Pnce, C. M. Rudenstam, H.J. Senn. Internauonal Collaborative Cancer Group, Charing Cross Hospital, London: J M.Bliss, C. E. D. Chilvers, R. C. Coombes, M Marty. Israel Technion, Rambam MedicalCenter. R. Borovik, G Brufinan, E. Robinson. Italzan Cooperative Chemo-Radio-SurgicalGroup: F. Patmun. Japanese Vatwnal Hospitals Group Breast Cancer Study Group. STakashrma, T. Yasutorm. Karohnska Hospital, Stockholm: L. E. Holm. Kawasakz MedicalSchool H. Sonoo. Kumamoto University Group J YamashIta. Louvam, AcademischZiekenhuzs St Rafael : J. Bonte M D. Anderson Cancer Center: A. U. Buzdar, T SmithMarsezlle, Laboratoire de Cancirologie Bzologique APM: P. Martin, S Romam. MayoClinic, U.S.A.: D Ahmann, D. J Schard Memonal Sloan-Kettenng Cancer Center : THakes, L. Norton, R. Wlttes. Metaxas Memonal Cancer Hospital, Athens: P. Foroglou, B.Llssaios Mzlan Isntuto Nazumale per lo Studio e la Cura dei Tumon: G. Bonadonna, M.del Vecchio, P. Valagussa, U Veronesi Montpelher Centre Paul Lamarque. J. B Dubois.NSABC, Israel: G. Brufman, H Hayat. Naples Uraverstty: A. R. Blanco National CancerInstitute, U.S.A M. E. Lippman, L. J. Pierce, R. Simon, S. M. Stemberg. NationalKyushu Cancer Center, Japan Y Nomura. National Medical Centre, Mexzco: R. de laHuerta, M. G. Samz. National Surgzcal Adjuvant Project for Breast & Bowel Cancers,U S.A.. A Brown, B Fisher, C. Redmond, N. Wolmark. Nolvadex Adjuvant TrialOrganuatzon M. Baum, I. M. Jackson, M. K. Palmer North Central Cancer TreatmentGroup, USA.: J N Ingle, D. J. Schard North Sweden Breast Cancer Group. N. O.Bengtsson, L. G Larsson North-Western British Surgeons J. P. Lvthgoe, R SwmdellNorthunck Park Hospital, London. M Kzssm. Nottingham Cuy Hospztal: R. W Blamey,A.K.Mitchell,J F F R. Robertson OSAKO & Swus Group for Clinical Cancer Research:M Castighone, H. Fluckager, H. J Senn. Otta Prefectural Hospttal Y Nakamura.Oncofrance. G. Mathe, J L Misset Oncology Hospital A Busimco, Sardima: N.Deshpande, L- d, Martmo. Ontario Cancer Treatment & Research Foundation E A.

Qarke, J R. McLaughlm Ontano Clinical Oncology Group: R. M Clark, M. Levme.Osaka CIty Medical School: K Monmoto. Oslo Radium HospItal. S Gundersen, M.Hauer-Jensen, H Hast Oxford Churchzll Hospital: E. Crossley, K Durrant, A HarrisOxford, ICRF;MRC Clinical Tnal Serrncz C’ntt Secretariat) M Clarke, R Collms*,1- J.Godwin, R. Gray*, E. Greaves, C Harvood, G Mead, R Peto*, K Wheatley. PartsInslZtut Gustave-Roussy’ - C. Hill, J Lacour, A Laplanche,M Le, D Sarrazin Petro-v

Research Institute of Oncology. V. Semtglazov. Ptedrnrntt Oncology Assoczanon J.Brockschmidt, M R Cooper Prencess Margaret Hospital, Torrnuo: J. W. Meakin, TPanzarella, K I. Pritchard Rorterdam Radio-Therapetic Institute A D. Treurniet-Donker, W L J. van Putten Royal Marsden Hospital, Institute of Cancer Research,London D. Easton, T. J. Powles St George’s Hosptral, London: J. C Gazet. SASIBInternational Trialists- P Douglas, A. Hackmg, H liost, A Lmdmer, G Notter.

Saskatchewan Cancer Foundatzon: A. J. S. Bryant, G. H. Ewing, J. L KrushenScandinavian Adjuvant Chemotherapy Study Group: R Nissen-Meyer. Scottish CancerTnals Office: A. P. M. Forrest, C. McDonald, H. J. Stewart. South Swedish Breast CancerGroup: T. R. Moiler, S. Ryden. South-East Sweden Breast Cancer Group: J. Carstensen,T. Hatschek, M. Soderberg. Southeastern Cancer Study Group & Alabama Breast Cancer

Project: J. T. Carpenter. Southwest Oncology Group: J. Crawley, S. Green, C. K. OsborneStockholm Breast Cancer Study Group: L. E. Rutqvist, A. Wallgren. Tel Aviv University -

H. Brenner, A. Hercbergs. Toronto-Edmonton Breast Cancer Study Group: G DeBoer,A. H. G. Paterson, K. I. Pntchard. Toulouse Centre Claudius Regaud. A. Naja. UKMultlcentre Cancer Chemotherapy Study Group: M. Reid, M. Spittle. UK AsiaCollaborative Breast Cancer Group: F. Senanayake. University of Chicago P P MeierUniversity of Lund: I. Tengrup, L. Tennvall-Nittby. University of Wurzburg: H. Caf’fierUniversity of the Witwatersrand: W. R. Bezwoda. Uppsala-Orebro Cancer Study Group : LHolmberg. Vienna University Hospital 1st. Department of Gynaecology P. Sevelda, C C.Zielinsky. Vienna University Hospital, 1st. Department of Surgery: R Jakesz Wessex

Radiotherapy Centre: R. B. Buchanan, M. Cross. West Midlands Oncology AssociationJ. A. Dunn, Gillespie, K. Kelly, J. M. Morrison. West of Scotland Breast Trial: A. Litton.Western Cancer Study Group: R T. Chlebowski.*Writmg committee.

REFERENCES

1. Early Breast Cancer Trialists’ Collaborative Group. Treatment of earlybreast cancer, vol 1: worldwide evidence 1985-1990. Oxford: OxfordUniversity Press, 1990.

2. Early Breast Cancer Trialists’ Collaborative Group. Effects of adjuvanttamoxifen and of cytotoxic therapy on mortality in early breast cancer:an overview of 61 randomized trials among 28,896 women. N Engl JMed 1988; 319: 1681-92.

3. World Health Organization. World Health Statistics Annual 1990.Geneva: WHO, 1991.

4. ISIS-2 (Second International Study of Infarct Survival) CollaborativeGroup. Randomised trial of intravenous streptokinase, oral aspirin,both, or neither among 17,187 cases of suspected acute myocardialinfarction: ISIS-2. Lancet 1988; ii: 349-60.

5. Anti-Platelet Trialists’ Collaboration. Secondary prevention of vasculardisease by prolonged antiplatelet treatment. Br Med J 1988; 296:320-31.

6. Peto R, Pike MC, Armitage P, et al. Design and analysis of randomizedclinical trials requiring prolonged observation of each patient. Part I:Introduction and design. Br J Cancer 1976; 34: 585-612.

7. Peto R, Pike MC, Armitage P, et al. Design and analysis of randomizedclinical trials requiring prolonged observation of each patient. Part II:analysis and examples. Br J Cancer 1977; 35: 1-39.

8. Office of Population Censuses & Surveys. Cancer statistics registrations:cases of diagnosed cancer registered in England and Wales, 1985 (SeriesMB1, no. 18). London: HM Stationery Office, 1990.

9. Fornander T, Rutqvist LE, Cedermark B, et al. Adjuvant tamoxifen inearly breast cancer: occurrence of new primary cancers. Lancet 1989; i:117-20.

10. Nevinny HB, Nevinny D, Rosoff CB, Hall TC, Muench H. Prophylacticoophorectomy in breast cancer therapy. Am J Surg 1969; 117: 531-36.

11. Ravdin RG, Lewison EF, Slack NH, Dao TL, Gardner B, State D,Fisher B. Results of a clinical trial concerning the worth of prophylacticoophorectomy for breast carcinoma. Surgery Gynecol Obstet 1970; 31:1055-64.

12. Matthews KA, Meilahn E, Kuller LH, Kelsey SF, Caggiula AW, WingRR. Menopause and risk factors for coronary heart disease. N Engl JMed 1989; 321: 641-46.

13. Richards MA, O’Reilly SM, Howell A, et al. Adjuvantcyclophosphamide, methotrexate and fluorouracil in patients withaxillary node-positive breast cancer: an update of the Guy’s/Manchester Trial. J Clin Oncol 1990; 8: 2032-39.

14. Zambetti M, Valagussa P, Brambilla C, Di Fronzo G, Silvestrini R,Bonadonna G. Adjuvant CMF in node-negative (N-) and estrogenreceptor negative (ER-) breast cancer: updated results. Proc Am SocClin Oncol 1989; 8: 21 (abstr).

15. Hortobagyi GN, Gutterman JU, Blumenschein GR, et al.

Chemoimmunotherapy of advanced breast cancer with BCG. In:

Terry WD, Windhurst D, eds. Immunotherapy of cancer: presentstatus of trials in man. New York: Raven, 1978: 155-68.

16. Anonymous. Review of mortality results in randomized trials in earlybreast cancer. Lancet 1984; ii: 1205.

17. Jordon VC. Estrogen receptor-mediated direct and indirect antitumoreffects of tamoxifen. J Natl Cancer Inst 1990; 82: 1662-63.

18. Love RR, Newcomb PA, Wiebe DA, et al. Effects of tamoxifen therapyon lipid and lipoprotein levels in postmenopausal women withnode-negative breast cancer. J Natl Cancer Inst 1990; 82: 1327-32.

19. Ludwig Breast Cancer Study Group. Chemotherapy with or withoutoophorectomy in high-nsk premenopausal patients with operablebreast cancer. J Clin Oncol 1985; 3: 1059-67.

20. Bianco AR, Del Mastro L, Gallo C, et al. Prognostic role of amenorrheainduced by adjuvant chemotherapy in premenopausal patients withearly breast cancer. Br J Cancer 1991; 63: 799-803.

85

21. Love RR. Tamoxifen therapy in primary breast cancer: biology, efficacyand side effects. J Clin Oncol 1989; 7: 803-15.

22. Powles TJ, Smith IE. Adjuvant therapy-interpreting current results.In: Powles TJ, Smith IE, eds. Medical management of breast cancer.London: Martin Dunitz, 1991.

23. Carbone PP. Breast cancer adjuvant therapy. Cancer 1990; 66: 1378-86.

24. Nissen-Meyer R. Primary breast cancer: the effect of primary ovarianirradiation. Ann Oncol 1991; 2: 343-46.

25. Bonadonna G, Brusamolino E, Valagussa P, et al. Combination

chemotherapy as an adjuvant treatment in operable breast cancer. NEngl J Med 1976; 294: 405-10.

26. Bonadonna G, Valagussa P, Rossi A, et al. Ten-year experience withCMF-based adjuvant chemotherapy in resectable breast cancer. BreastCancer Res Treatment 1985; 5: 95-115.

27. Nolvadex Adjuvant Tnal Organisation (NATO). Controlled trial oftamoxifen as a single adjuvant agent in the management of early breastcancer: analysis at eight years. Br J Cancer 1988; 57: 608-11.

28. Fisher B, Redmond C, Brown A, et al. Adjuvant chemotherapy with andwithout tamoxifen in the treatment of primary breast cancer: 5-yearresults from the National Surgical Adjuvant Breast and Bowel ProjectTrial. J Clin Oncol 1986; 4: 459-71.

29. Medical Research Council Scottish Cancer Trials Office (MRC SCTO)Breast Cancer Trials Committee. Adjuvant tamoxifen in the

managmeent of operable breast cancer: the Scottish trial. Lancet 1987;ii: 171-75.

30. Fisher B, Costantino J, Redmond C, et al. A randomized clinical trial

evaluating tamoxifen in the treatment of patients with node-negativebreast cancer who have estrogen-receptor-positive tumors. N Engl JMed 1989; 320: 479-84.

31. Fisher B, Redmond C, Legault-Poisson S, et al. Postoperativechemotherapy and tamoxifen compared with tamoxifen alone in thetreatment of positive-node breast cancer patients aged 50 years andolder with tumors responsive to tamoxifen: results from the NSABPproject B-16. J Clin Oncol 1990; 8: 1005-18.

32. Meakin JW, Allt WEC, Beale FA, et al. Ovarian irradiation and

prednisone therapy following surgery and radiotherapy for carcinomaof the breast. Can Med Assoc J 1979; 120: 1221-29.

33. Legha SS. Tamoxifen in the treatment of breast cancer. Ann Intern Med1988; 109: 219-28.

34. International Agency for Research on Cancer. Carcinogenicity of

alkylating cytostatic drugs. Lyon: IARC, 1986.35. Smith PG, Doll R. Late effects of X-irradiation on patients treated for

metropathia haemorrhagica. Br J Radiol 1976; 49: 224-32.36. Anonymous. Consensus conference. Adjuvant chemotherapy for breast

cancer. JAMA 1985; 254: 3461-63.37. Anonymous. NIH Consensus conference. Treatment of early-stage

breast cancer. JAMA 1991; 265: 391-95.38. Anonymous. Consensus development conference: treatment of primary

breast cancer. Br Med J 1986; 293: 946-47.39. European Carotid Surgery Trialists’ Collaborative Group. MRC

European Carotid Surgery Trial: interim results for symptomaticpatients with severe (70-99%) or with mild (0-29%) carotid stenosis.Lancet 1991; 337: 1235-43.

Routine ultrasonography in utero and schoolperformance at age 8-9 years

Most fetuses in developed countries are exposedin utero to diagnostic ultrasound examination. Manypregnant women express concern about whether theprocedure harms the fetus. Since most routineultrasound examinations are done at weeks 16-22,when the fetal brain is developing rapidly, effects onneuronal migration are possible. We have sought anassociation between routine ultrasonography inutero and reading and writing skills among childrenin primary school.At the age of 8 or 9 years, children of women who

had taken part in two randomised, controlled trials ofroutine ultrasonography during pregnancy were

followed-up. The women had attended the clinics of60 general practitioners in central Norway during1979-81. The analysis of outcome was by intentionto treat: 92% of the "screened" group had beenexposed to ultrasound screening at weeks 16-22,and 95% of controls had not been so exposed, butthere was some overlap. 2428 singletons wereeligible for follow-up, and the school performance of2011 children (83%) was assessed by their teacherson a scale of 1-7; the teachers were unaware ofultrasound exposure status. A subgroup of 603children underwent specific tests for dyslexia. Therewere no statistically significant differences betweenchildren screened with ultrasound and controls in the

teacher-reported school performance (scores forreading, spelling, arithmetic, or overall performance).Results from the dyslexia test sample showed no

differences between screened children and controlsin reading, spelling, and intelligence scores, or indiscrepancy scores between intelligence and readingor spelling. The test results classified 21 of the 309screened children (7% [95% confidence interval

3-10%]) and 26 of the 294 controls (9% [4-12%])as dyslexic.The risk of having poor skills in reading and writing

was no greater for children whose mothers had beenoffered routine ultrasonography than for thosewhose mothers had not been offered the procedure.

Introduction

Because the indications for diagnostic ultrasound in

pregnancy are common, and because ultrasonographicscreening is routine in many countries, most pregnantwomen in developed countries are exposed to the procedure.During ultrasonographic examinations, pregnant womencommonly express concern that the procedure may harmthe fetus. Ultrasound is used therapeutically to disintegraterenal calculi and in surgery of the brainy which implies that

ADDRESSES: Department of Community Medicine and GeneralPractice (K A. Salvesen, MD) and Department of Psychology (ProfJ. O Undheim, PhD), University of Trondheim; Department ofEpidemiology, National Institute of Public Health, Oslo (Prof L. S.Bakketeig, MD); National Centre for Fetal Diagnosis and Therapy,Trondheim University Hospital (Prof S. H Eik-Nes, MD); andDepartment of Pediatrics, &Aring;lesund Central Hospital, Norway(O &Oslash;kland, MD). Correspondence to Dr K &Aring;. Salvesen, University ofTrondheim, Department of Community Medicine and General Practice,Medical Technical Centre, N-7005 Trondheim, Norway.