614
Correspondence
their method, while almost 11 per cent chose the intrauterine device. A similar shift was seen in women with varicose veins or coronary artery disease. In order to assess the reasonableness of this explanation for the difference in risk between idiopathic and nonidiopathic cases, a sample of the cases of thromboembolism was reinterviewed a considerable period after their discharge from the hospital for the purpose of learning whether their physicians had reinstituted oral contraception. Had they done so, it would have been in disregard of the advice on the package insert that the method not be used by those who have sustained a prior thrombosis attack. Actually, of 31 previous oral contraceptive users among thrombosis patients who were queried after discharge, only one was found to have resumed this method. .4mong 65 women with thrombosis who had not used oral contraceptives within one month before hospitalization, again, only one was an oral contraceptive user after discharge from the hospital. These findings support the view that the non-idiopathic cases as a group are less likely to have the method prescribed for them than healthy women; hence, they would have a spuriously low relative risk compared to healthy women when studied by the casecontrol method, even though their ‘true’ risk might be as high as that of the idiopathic cases. They also provide heartening evidence that physicians are adhering to the current advice with respect to the contraindications for these agents.’ Also, on page 906, the authors stated that “no dose-related effect was demonstrated by this epidemiologic study.” We actually concluded the opposite, i.e., that the risk of occurrence of thrombotic disease was related to oral contraceptive dosage. Finally, their reference to our paper was incorrect and is as below.’ Paul D. Stolley, M.D. Associate Projessor Jamrs A. Tonasria, Ph.D. Associate Prqfrssor Department of Epidemiology The Johns Hopkins University School Hygiene and Public Health 615 North Wolfp Strret Baltimore, Maryland 21205
qf
REFERENCES
1. Stolley, P. D., Tonascia, J. A., Tockman, M. Thrombosis with low-estrogen oral contraceptives, Epidemiol. 102: 197, 1975.
S., et al.: Am. J.
To thr Editors: Among the
many errors and misrepresentations in the article by Goldzieher and Dozier (December 15, 1975, pages 378 to 914), three of those referring to our work are gross. On page 894, in discussing two of the retrospective studies of the relationship between oral contraceptives and venous thrombosis which we” ’ carried out in the 1960’s, Goldzieher and Dozier made the following statement: “The incidence of oral contraceptive use in the control subjects second samples
is signzjicantly
(italics theirs):
dz;fferent
between
the$rst
25 per cent of control
and
sub-
Table I. Recurrence of thromboembolic disease in III o groups of women in relation to previous history (I[ thromboembolic disease and contraceptive choice (from Badaracco and Vessey”)
.history ojtiwombormbolic dtieasr
No. without previous history No. with subsequent recurrences No. with previous history No. witd subsequent recurrences
30 3 (10%) 12 2 (17%‘)
25 10 (40%)
17 9 (53%)
>(I. 1
jects in the first study were users, against 8.6 per cent in the second control sample, a difference that is statistically significant (P = 0.01). The use of oral contraceptives certainly did not &crQnsr (italics theirs) threefold in the time intervals examined.” In fact, in the first study, which covered the years 1964 to 1966, we found that 10 of 116 control women (8.6 per cent) used oral contraceptives while in the second study, which covered the year 1967. we found that 13 of 52 control women (25 per cent) did so. In view of the small size of both samples and the fact that the sales of oral contraceptives more than trebled3 in the United Kingdom from 1964 to 1967, these data are eminentl! reasonable. Goldzieher and Dozier have evidently interchanged the control data from the two studies. On page 901, Goldzieher and Dozier present the data shown in Table I, but without the percentages, which have been added by us. In their discussion of this table, they stated, “In the subjects uith C[~WZ&ZL.\ histo? (italics theirs) of thromboembolic disease, the frequency of recurrence was the same in the former users andin the nonusers. . Quite another situation was observed in those uith WI pr?zriouJ kistoq (italics theirs) of thrombophlebitis . . those who were WIVU.WI\ (italics theirs) at the time of their first attack had a significantly higher (italics theirs) recurrence rate than those who had been users at the time of their first attack .” Inspection of the percentages in the table indicates ho\\ misleading these comments by Goldzieher and Dozier are. The fact that one of the differences is “statistically significant” while the other is not does not detract from the over-all conclusion that the data for those with and without a past history of thromboembolic disease are remarkably similar. Also on page 901, Goldzieher and Dozier start a discussion of “precipitating factors” hv stating: “Since thromboembolism is basically a recurrent disease, one would suspect that a precipitating or augmenting factor (such as the oral contraceptives are hypothesized to
Correspondence
be) would increase the incidence of recurrence. Data available in the British studies fail to support this notion (Table XI).” Where do the numbers in their Table X1 come from and what do they mean? Even as the authors of the two reports cited in Table XI, it took us some time to find out. In fact, of the 58 women with venous throtnboetnl~olism included in the first report,’ 21 ga\‘e a historv of l)ro~ri~~ thromboembolism. Of these :! I. nine (43 per cent) were users of oral contraceptives and 12 (37 per cent) were not. Similarly, of the 26 women with vrnous thromboembolism included in the second report.” eight gave a history of previo2cs thrombocmbolism. of whom three (37 per cent) were users of oral contraceptives and five (63 per cent) were not. Therefore. the data refer to the women’s past history and provide no information whatsoever about the “rccurrcnce of thromboembolism as related to oral contra< eptive use..” An interitn report of the findings in our large prospective stud! of women using oral contraceptives, which has been in progress since 1968,5 will be published shortly.’ So far as throtnboembolic disease is concerned, the results conform closely with those obtained in the retrospective studies. Al. P. I’~w~~, MA., M.D., F.F.C.M. Profrwr oj Social and Community .?w Rirhnrd Doll, D.M., F.R.C.P., Regius O.$rfi 8 &hle
Prqfessor
Medicine F.R.S.
of Medicine
C~rriwrsit~ Road
REFERENCES
1. Vessey, M. P.. and Doll, R.: Investigation of relation between use of oral contraceptives and thromboembolic disease, Br. Med.J. 2: 199. 1968. 2. Vessey. M. P., and Doll, R.: Investigation of relation between use of oral contraceptives and thromboembolic disease. A further report, Br. Med. J. 2: 651, 1969. 3. Vessey, M. P., and Inman, W. H. W.: Speculations about mortality trends from venous thromboembolic disease in England and Wales and their relation to the pattern of oral contraceptive usage, J. Obstet. Gynaecol. Br. Common. SO: 562. 1973. 4. Badaracco, M. A., and Vessey, M. P.: Recurrence of venous thromboembolic disease and use of oral contraceptives. Br. Med. J. 1: 215, 1974. 5. Vessey, M. P.. Johnson. B., and Donnelly, J.: Reliability of reporting by women taking part in a prospective contraceptive studs, Br. 1. Prev. Sot. Med. 28: 104, 1974. 6. Vessey. M. P., Doll, <. Peto. R., Johnson, B., and Wiggins, P.: A long-term follow-up study of women using different methods of contraception. An interim report, J. Biosoc. Sci. 8: 1976. In press.
Reply thY Editon: Comment on letter by Dr. C. R. Kay. We respectfully suggest that Dr. Kay contain his outrage, for his categorical denial that “there is no arithmetical error in any of our tabulations” is simply not true. Kay confuses To
615
the situation by describing his use 01 more rigorous definitions of predisposing causes. This iz not the issue. Kay’s original denominators comaill exposures of women with conditions that prechltic their being classifiable as cases of first attack of- idiopathic thrombosis. This is a basic impropriety in epidemiology. a> discussed, for example, by B. Macblahon and ?I‘. F. Pugh (Epidemiology. Principles and ,2ltathods, Boston, 1970, Little, Brown & C:otnpa!l)~, p. 70). Simply put, wotnen with predisposing condlttons c‘artnot, bv definition, contract idiopathic disease. The recalculations published by Ka!, in 1975, were not, in fact. figures to correct the error\ in the original 1974 report. They apparently pertair! to expanded data with the exclusion of predisposed and pregnanr women from both the nutneratol and the denotninator. We still have no idea what the original rat.es or ratios of rates would have been whctt i.orrec-tl>- calculated. For comparison, we do not kno\\ what the updated calculations would have been if done the old way. What Kay has given us insteacl is a comparison of original data calculated incorrectly with expancled data calculated correctly. This is hardlv the resoiurion of doubt that Kay purports to have provided. I’t~rtraps this will be revealed in the “full report” he has twicr promised (Lancet 1: 138 1, 1975: ,J. K. Coil. <&n f%ac.t. 25: 904, 1975). The use of “standardized” rates (ft of age/raceadjusted values (as in the Heyman stud\; I)I~ strokes) is a proper way of trying to compensate for differences in populations subjected to epidemiologic surveys. The magnitude of the changes (or the lack tht%reof, as in the present instance) which can be brought about bv such adjustments is disquieting. In the letter oi‘O’Fallon and associates, we find that such an adjustment can lower a relative risk from 13.6 to 2.6; the;, had previously adjusted certain published relative disks of 8.8 and 9.5 down to 4.1 and 4.4. Such flexibilitJt. a( hieved entirely within allowable epidemiologicmanipulation, does not increase confidence in the realit\. of‘ ihc computed values. Comment on letter by Dr. A. T. Ma.& Dr. Masi’s comments regarding the Sartwell and Z’csscy-Doll papers are taken up elsewhere in our commentary. His discussion regarding sample size in prospective studies is quite confused. First, he caref’ully obfuscates the fundamental difference between prospecrive clinical observations (such as the studies of the Koyal College of General Practitioners and the .1tlstralian College of General Practitioners) where users and nonusers are self-selected (with the attendant bias) and clinical exprimrn/.~ (such as the Fuertes-de la Ilaha study) where users and nonusers are assigned h\: property ratidomized design. (The statement that a 10,000 patient exfipuriment is “small-scale” is rather ;tmu+ing.) Then he proposes zero occurrences in a theot,t*tic al control group and proceeds to calculate the clti-square value. That this is improper is widely recogniT/eti by statisti-