Metamizol: een honderdjarige treurnis

Jerusalem, Boston, May 1987,

Our attention has recently been drawn to a commentary by Dr.Offerhaus (1987;479-81) on the International Agranulocytosis and Aplastic Anemia Study.1

1. Firstly, a matter of ethics: unless our translation is faulty, Dr. Offerhaus states that we ‘pretended (our emphasis) that all eligible hospitals in Berlin were involved in the study....’ This is a serious allegation implying dishonesty on our part, and we request that it be withdrawn.

2. As a matter of record, we repeat, without qualification, that in Berlin and elsewhere, all major hospitals were involved in the study as described,12 and we have detailed log books and other documentation to show this.

We turn next to matters of fact:

3. The Boston Collaborative Drug Surveillance Program is not associated with the Drug Epidemiology Unit (now the Slone Epidemiology Unit).

4. One region (Sao Paulo) was dropped, and one catchment population (Milan) was reduced in size early in the study for reasons of feasibility. None of the regions for which data were presented ‘failed to achieve the required number of patients’.

5. 221 cases of agranulocytosis, not 208, were compared with 1425 controls.

6. As already stated, we read to each patient a list of generic and trade names of drugs containing metamizol that accounted for over 90% of the sales. It was not necessary to mention ‘thousands of drugs’.

7. 21 patients, not 34, died before they could be interviewed.

8. It is not correct that numbers in ‘Milan and Sofia appeared to be too small to permit statistical analysis’. The data from these centers were analyzed and commented upon.

9. It is not correct that ‘only in the... regions of Barcelona, Ulm and West Berlin was there such homogeneity in terms of the figures as to justify’ pooling of the data. The data were also homogeneous for Israel and Budapest and these two regions were also pooled. In addition, we did not carry out ‘addition and averaging’.

10. We have not concluded at a congress in Stockholm, or anywhere else, that ‘the data collected in Israel are so contradictory and inconsistent that they had better be forgotten’.

11. The catchment population for the Ulm region was not ‘overestimated tenfold’. We used reliable census data to estimate the size of the population served by more than 80 hospitals in an extensive region in Southern Germany surrounding that city.

12. The deduction that ‘for the total material (Ulm, Berlin, Barcelona)... 1.25 patient(s) in 1,000,000 inhabitants per annum (our emphasis) have developed agranulocytosis following ingestion of metamizol’ is incorrect (the correct estimate is given below). But perhaps the most egregious error that Dr.Offerhaus makes is to take his (incorrect) figure (the assumed number of metamizol-induced cases occurring in one year) as a numerator, and to apply it to a denominator of 17,000 (the estimated number of persons using metamizol in one week).

13. The excess risk of ‘one in a million’ (or more accurately, 1.1 per million) does not ‘apparently relate’, either to the ‘total population’ or to a ‘single treatment’ (see below).

We turn lastly to matters of science:

14. If the extreme assumption is made that all 21 patients who died before they could be interviewed took metamizol, this would not materially change the excess risk estimate.

15. ‘Differential loss of memory’ between cases and controls would, as we have pointed out, tend to result in overestimation, not underestimation, of the risk due to metamizol.

16. We are not aware of any ‘rule of thumb for any multicenter study that analysis should be possible by center.’ To the contrary, it is the infeasibility of obtaining sufficient data in any individual center that is usually the main reason for performing a multicenter study. We agree, however, that data should be homogenous between centers before they are pooled, and we adhered to that principle. In doing so we estimated an excess risk for Ulm, Berlin and Barcelona that was the most adverse for metamizol. Elsewhere the risk appeared to be lower (Milan, Sofia) or perhaps not increased at all (Israel, Budapest).

17. Official sales figures for metamizol were determined by different survey instruments in Israel and in Europe,3 and cannot be compared. Earlier data ‘collected by the same team in Jerusalem’ were obtained by means of a different methodology, with a different questionnaire, and at a different time.4 Most importantly, the drug histories related to exposure in a one month interval, as against one week in our recent study.1 This also applies to data ‘collected by the same team in Berlin’.

18. The earlier Israeli studies of metamizol-induced agranulocytosis were based on an anecdotal evidence from case reports and uncontrolled studies,56 and one of the studies has been shown to be biased.7

19. Contrary to what is claimed, homogenous categories were compared, and Dr.Offerhaus offers no evidence in support of his claim that the ‘collected data... often are not representative of the general population’.

20. The number of persons taking metamizol cannot be inferred from the number of packages sold per year: doses vary between persons, and in the same person from time to time; it is common for the drug to be shared between family members or friends; preparations of metamizol, combined with other analgesics, are used in different ways from metamizol alone; and packages can be discarded wholly or partly unused.

21. Since Dr.Offerhaus has misunderstood the nature of the excess risk estimate (see above), it is worth describing how it was calculated: Metamizol-induced agranulocytosis is an acute immunological reaction and therefore can only occur when the drug (or its metabolites) is still present in the body, or, at most, within a matter of days thereafter. The median duration of use was two days by the cases, and three days by the controls, and the drug is metabolized rapidly within a few days. Based on these considerations we defined etiologically relevant exposures as those that occurred within one week of the onset of the illness. In Ulm, Berlin and Barcelona we estimated that 27% of all cases of agranulocytosis were due to such exposure. In those regions the age and sex-adjusted incidence rate of agranulocytosis that arose in the community and resulted in hospital admission was 3.4 per million per year, or 0.065 per million per week (3.4/52). Therefore, in any week the incidence among persons not exposed to metamizol was: (1 – 0.27) 0.065 = 0.05/106 (A).

In Ulm, Berlin and Barcelona the estimated rate ratio for exposure to metamizol was 23.7. Therefore, in any week the incidence among persons exposed to metamizol was: 23.7 (1 – 0.27) 0.065 = 1.12/106 (B).

Thus the excess risk (risk difference) that could be attributed to the use of metamizol for up to one week was: 1.12 – 0.05 = 1.07/106, which rounds off to 1.1/106 (B)-(A).

Elsewhere, of course, where the rate ratio estimate was lower, the excess risk was correspondingly reduced.

Rijswijk, August 1987,

I have to apologize for the delay in answering Dr.Levy and Dr.Shapiro's letter. This was caused on the one hand by waiting for the results of the discussion on the same subject in the Journal of the American Medical Association,1-4 and by the subsequent untimely death of my colleague Dr.van Dijke, and on the other by the time needed to collect data necessary to give a proper answer. I will try to answer their queries in the order given in their letter.

1. I do not know in what way the authors obtained an English translation of my article, though I suppose that it was made and sent to them by the sponsor of the study. Editorials do not have an English summary, and online abstracts are the responsibility of the database vendor. Moreover, even in the best of translations subtle nuances tend to get lost as the translation of the title of my Editorial clearly shows.

2. I could not find any protocol requirements as to the recruitment of hospitals in the two quoted articles, and neither is the number of participating hospitals nor the number of hospital beds involved exactly mentioned. An esteemed clinical pharmacologist, who was working in Berlin during the onset of the study asked – purely out of scientific curiosity – a number of chiefs of large medical services in the region whether they had been approached to participate in the study, and two of them were not aware of any trial of that kind going on. Unfortunately, in view of the legal harassment to which the sponsor is apt to subject its German opponents (as during the period preceding the withdrawal of their antidepressant drug nomifensin) I am not at liberty to reveal his identity. I do not question his honesty, nor do I question the scientific honesty of Dr.Levy and Dr.Shapiro in any way, but they cannot have been present in person at every study centre, and it must in retrospect have been impossible to verify every single data.

Turning to the matters of fact:

3. Though the Drug Epidemiology Unit is not anymore associated with Dr.Jick's program it took its origin from that centre. To be correct, they should at the present time be seen as totally separate units.

4. The number of patients observed in Milan was insufficient to arrive at any valid statistical conclusions for this particular centre in contrast to other regions where sufficient numbers were collected to do the same. The reasons to drop the largest and most important region of Sao Paulo were never officially revealed.

5. I left out the cases collected in Sweden, because dipyrone is not available in that country and these data are therefore not of any use in calculating the excess risk of dipyrone use for that particular region.

6. By restricting the drug list to those with supposedly the highest known turnover some sort of bias must have been introduced; the ‘at least 90% of the sales turnover’ condition was not mentioned in the protocol. In the Federal Republic of Germany dipyrone is present in literally thousands of different combination preparations (WHO data). Earlier experience in the Ulm university hospital has been that it is often difficult to pinpoint one particular drug, because patients with agranulocytosis had taken several drugs in the recent past which could have been causally implicated;5 polypharmacy in German hospital patients is the rule rather than the exception.

Moreover, the fact that dipyrone is marketed under at least 11 different generic names adds to the general confusion:6 ‘The reasons for this multiplicity of names are not clear. There is at least the possibility that they merely confused some physicians, leading them to believe that dipyrone and the now-feared aminopyrine were quite different’.7 Moreover, Shapiro has recently admitted that problems of possible information bias in ascertaining exposure are quite common in case-control studies.8

7. (7% of 300) + (8% of 160) = 21 + 13 = 34 (aplastic anaemia and agranulocytosis).

8. Vide paragraph 4.

9. True, but the rate ratio estimate in Israel was double that in Hungary: Can a difference of one hundred percent be called ‘homogeneous’?

10. The authors did not conclude that the Israeli data were contradictory and inconsistent, I did. My arguments were mentioned in the Editorial.

11. Ulm is a small provincial city in a thinly populated and wooded area in the heart of Southern Germany (population on 1.1.1983 99,725). To the region of Ulm belong the adjoining regions of Alb-Donau, Biberach and Heidenheim (total population on 1.1.1983 536,021; on 31.12.1986 538,205). The region has nine hospitals with services for internal medicine with 673 beds reserved for that branch of medicine (either in specialist wards or in general hospitals) out of a total of 3,510 hospital and sanatorium beds. In 1982 in the regions Ulm and Alb-Donau a total of 136,448 patients was hospitalized including 3,100 from the Biberach and Heidenheim regions, a substantial portion (10.355) from adjoining Bavaria and 3.221 from elsewhere. To the ‘medical’ region Ulm may be added app. 30% Bavarian patients, mainly from the regions Neu-Ulm and Günzburg in adjoining Schwaben (total population on 31.12.1986 250,181), bringing the population covered by the hospitals of the Ulm region to 536,021 + 250,181 = 786,202 persons, of whom app. 100,000 are estimated to go to hospitals outside the Ulm region.9 Large and renowned university hospitals in Heidelberg, Tübingen and Stuttgart cater for the majority of the other inhabitants of Baden-Württemberg, and in adjoining Bavaria Augsburg and Munich. Moreover, the very regionalised German sick-fund system, just as in Holland, does not easily allow insured patients to consult doctors outside their own region except with special permission.

Even after leaving all city regions to their local medical centres the number of 5,300,000 inhabitants, quoted to live in the Ulm region, has not been adequately explained. The Baden-Württemberg census authority has called this number grossly exaggerated (‘weit überhöht’). If the academic centres in the neighbourhood, Stuttgart and Munich (the Klinikum Grosshadern is reportedly the largest hospital in Europe), had cooperated in the study, nobody would have spoken about the ‘region Ulm’; it would be as if the University of Worcester, Mass. had organized a study involving Greater Boston and would have used the term ‘region Worcester’. The official regional census data tally remarkably with the number of app. 490,000 shown by Gaus on a slide at the Berlin hearing.

12. My rough estimate of 1:14,000 would indeed be totally wrong (certainly by a factor 52) if dipyrone is being used continuously, but such chronic use would only marginally increase the chances of developing agranulocytosis. If every German user of dipyrone-containing drugs follows 2-3 treatment courses per year (a more realistic figure) the risk in users could be around 1:50,000. As the authors rightly stress that the data obtained from the patients and the control groups only apply to the one week preceding the interview, any extrapolation over a whole year is loaded with many uncertainties and assumptions. As matters stand at the moment, we simply do not know. This has been pointed out by Miettinen and by others, including the statistician involved with the study in the Ulm region.210 There is considerable uncertainty whether the control patients, who were mainly admitted to hospital for conditions associated with pain and fever are representative of the population as a whole, including children, who rarely are given analgesics: No case controls were sampled from a healthy population.

Moreover, it may be questioned whether a period of one week is long enough, because after a first exposure to the drug the interval may, even in the ‘immune type’ of agranulocytosis be considerably longer.1112 We do not know whether these were first or repeated exposures, and we do not know whether at the first signs of the disease the drug was stopped or continued; when dipyrone is used as an antipyretic and continued, the symptoms of agranulocytosis might be masked,5 with disastrous consequences. Such a sequence might easily occur at home and go unnoticed until death.

13. My criticism, that the data obtained in the control group were extrapolated to the population as a whole to arrive at the disputed incidence of 1.1 to 1,000,000, thereby assuming equal drug use habits, age and sex distribution, has not been answered; what really matters is the chances of contracting a horrible disease and perhaps dying of it after taking a particular drug, and not disengage this chance from drug use and compare it with a traffic accident or a thunderbolt. The risk of developing cisplatinum-induced bone marrow aplasia might be well in the order of one in one billion for the population as a whole; nevertheless cisplatinum rightly is regarded as a dangerous drug, which should only be used exceptionally.

Turning to the matters of science:

14. The argument is not so much that adding all patients who died before the interview to the total would not materially have affected the statistical outcome (which argument happens to be true), but that an unknown number of cases might have fallen ill and recovered or died undiagnosed either at home or even in the hospital. The data collected in the patients who contracted the disease during their hospital stay might have been helpful, but they were inconclusive. Coulter counters do often not function at night or during the weekend, and counting leukocytes under the microscope and colouring blood smears is an art which junior doctors generally do not master anymore. Moreover, I doubt whether all Bulgarian, Hungarian or Spanish patients have routine blood counts at admittance. The protocol does not mention any requirements as to routine laboratory investigations in all patients.

15. The term ‘differential loss of memory’ has not been clearly explained by the Steering Committee. It could mean that the controls remember more clearly what they have taken than the patients or vice versa. In all probability the patients were pressed harder to provide a proper drug history because such data are important for their treatment and outcome (resulting in an underestimation of the risk). If there is a general loss of memory in both groups estimations of previous dipyrone use could be very variable. Much depends on the availability of the drug to the public: Data sheets of OTC products are often absent or discarded, but doctors issuing prescriptions for a PO product might warn patients against certain side effects. The general experience in taking drug histories is that patients tend to ‘forget’ having taken OTC drugs more readily because they are not really regarded as ‘medicines’. Considerable differences between countries might exist; moreover, the chance that such ‘information bias’ might have resulted in a false estimation of the risk in the German and the Spanish regions is admitted by the authors in their report, but it is not clear why this should only apply to the controls, and not to the patients.

16. Though agreeing that the main reason for performing multicentre studies is to obtain numbers which might never be reached in a single centre, any multicentre study which shows discrepancies between centres of the order observed in the IAAAS seems at least disquieting. In therapeutic trials such differences might lead to disqualification of the study as a whole. Disregarding Israel, which has an excellent standard of medical care, much of the difference could be explained by the obviously different standards and traditions in medical (and nursing) care between countries – something which any Northwestern European who ever had the misfortune to end his holiday in a Spanish or Italian hospital can confirm. That is an additional reason to restrict scrutiny of the results to the German data.

17. It is a great pity that the sponsor as the leading producer of dipyrone has not from the start of the trial opened his sales figures and put them at the disposal of the trial committee. We now have to deal with incomplete and conflicting data which leave indeed much to be desired.13 There is, even though the methods used were slightly different, no obvious reason why the results of earlier studies in Jerusalem and Berlin should have given such completely different results, the more so since the Berlin results were essentially in agreement with the German data collected by the editors of the Arzneimittelbrief, as mentioned in my Editorial.14

18. I cannot enter into the discussion on the subject held in Israel. Dipyrone is undoubtedly an extremely popular drug in that country, as shown by Levy,15 and I cannot explain why one local authority thinks that the drug causes agranulocytosis and the other thinks it does not. Moreover, I cannot explain the low incidence in the present study, nor can Dr.Levy.

19. Examples of inconsistencies are: In the region Ulm 6 out of 28 patients with agranulocytosis admitted previous use of dipyrone = 21%. In 1983 36 out of 63 patients in Southern Germany (57%) did so, and 21 out of 34 cases in Eastern Switzerland (62%).14 The consumption data for dipyrone in Western Germany are (expressed in defined daily doses) four times higher than in Israel, but the reverse is the case in the patients and controls in the present study: Nearly five times as much in Israel than in Western Germany. It is unclear which data are representative of the population as a whole.

20. I agree that there is generally no strict correlation between sales figures and actual drug consumption, but this cannot explain extreme variations: There is no reason why some populations should throw away or hoard more than others.

21. The argument concerning the calculation of the excess risk has been the subject of previous discussions;1 Dr.van Dijke has patiently tried to explain why he thought the calculation was wrong, but the answer is very brief and beside the point he tried to make.4 Moreover, the calculation of the excess risk made by the authors is in obvious contrast with a recently voiced opinion that in case-control studies only the relative risk, and not the excess risk, can be estimated,8 and we are left in doubt about the confidence limits of the relative risk, which should always be attached to indicate the statistical reliability.1617

Finally, the study proved beyond all doubt the causative role of dipyrone in the development of agranulocytosis, a syndrome which after use of this particular drug has an overall mortality of app. 25%.5 Many have, not unfairly, criticized the way in which the incidence of the risk was calculated, the calculation methods used and the data on which the calculations were based. In retrospect it seems questionable whether the study plan was at all suitable to compute incidence and risk data relating to single drugs. The IAAAS study is in itself eminent and very useful, but would have been better off if the main reason for performing it, i.e., disculpating dipyrone, had been less forcefully stressed. But then it would, for obvious financial reasons, never have been done. The answers provided by the authors have thus far seemed less than satisfactory. The fact remains that dipyrone is a drug for which there is no obvious need (certainly not as an OTC analgesic), and for the restricted indications for which it might be useful, less dangerous alternatives are available. Though Swedish data show that the total incidence of agranulocytosis did not change after the drug was banned, no cases of dipyrone-induced agranulocytosis were reported anymore from Finland and Denmark after the drug was taken off the market.

Finally, I would never have written the disputed Editorial if there had not been a clear inducement in the report, published in the same issue, on the umpteenth unnecessary case of dipyrone-induced agranulocytosis where the drug was prescribed for a trivial indication and had nearly caused the death of a young woman.18