Ecological studies of relation between hospital policies on neonatal vitamin K administration and subsequent occurrence of childhood cancerBMJ 1998; 316 doi: https://doi.org/10.1136/bmj.316.7126.184 (Published 17 January 1998) Cite this as: BMJ 1998;316:184
- S Jane Passmore, research officera,
- Gerald Draper, director ()a,
- Pat Brownbill, research programmera,
- Mary Kroll, statisticiana
- Correspondence to: Dr Draper
- Accepted 31 October 1997
Objective: To investigate the possible link between neonatal administration of intramuscular vitamin K and childhood cancer.
Design: Ecological studies comparing incidence of cancer in groups of children classified by the vitamin K policy in operation at their hospital of birth.
Setting: Selected large maternity units in England, Scotland, and Wales.
Subjects: Children born in these units in varying periods between 1966 and 1991.
Main outcome measures: Cancer occurring among these children before age 15 years identified by using the National Registry of Childhood Tumours. Ratios of observed to expected numbers of these conditions calculated for hospitals where the policy was to give all babies intramuscular vitamin K (non-selective) and where the policy was to use this treatment only for a selected minority of babies at increased risk of vitamin K deficiency bleeding (selective).
Results: These ratios were calculated for children born in 94 hospitals with varying vitamin K policies. A raised risk was occasionally associated with vitamin K, but the overall results were not significant, and there was no evidence to support the previously suggested doubling of the risk of childhood cancer.
Conclusions: On the basis of the results reported here it is unlikely that there is a greatly increased risk of childhood cancer attributable to intramuscular vitamin K given to newborns, if indeed there is any.
Intramuscular vitamin K given to babies is known to be effective in the prevention of vitamin K deficiency bleeding but it has been suggested that these preparations, or one of their constituents, may increase the risk of childhood cancer
Most studies have not shown a significant association between childhood cancer and vitamin K but are unable to exclude the possibility that its use increases the risk of childhood cancer by up to 10%
Intramuscular vitamin K has been given to “high risk” babies as part of all the various prophylaxis policies in the United Kingdom; this should continue
As a small risk cannot at present be excluded it seems prudent to recommend a policy of giving intramuscular vitamin K only to those babies at particularly high risk and giving it orally to others
It is essential that a record should be made of whether or not vitamin K is given and of the preparation, route of administration, and dose
Research carried out after the publication of papers by Golding et al, which reported a possible doubling of risk of childhood cancer after the administration of intramuscular vitamin K,1 2 has suggested that the risk, if any, is not as large as this. In an accompanying paper (p 178) we describe the background to the present study and our own case-control study.3 We also summarise previous studies and give details of policies on the use of vitamin K in newborns.
In this paper we present the results obtained with an “ecological study.” We use this term rather than “cohort study” as only the average level of exposure for each group in the study was (approximately) known. The analysis is based on knowledge of the policies on vitamin K administration for births in selected hospitals. We do not know for individual births whether the policy was carried out. We assume, however, that it was in most cases; any increase in incidence of cancer for the total group of cases will depend on the carcinogenic risk, if any, of vitamin K and also on the proportion of babies receiving it.
Ascertainment of cases
For some hospitals and time periods it has been the policy to give intramuscular vitamin K to all babies; for other hospitals and times it has been given only to “high risk” babies-usually about 25-30% of births. These are referred to in this paper as “non-selective” and “selective” policies. The rates of cancer after these two policies should differ if there is indeed an effect of vitamin K, and the rates for the two groups can be compared even if the individuals receiving vitamin K are not identified. To carry out such an analysis it is necessary to know, for each hospital and birth year included, the total number of births, the policy concerning vitamin K administration, and the date of diagnosis of each case of malignant disease occurring among these children by a specified date.
We identified several groups of hospitals for which it was possible to obtain the necessary information. By using the National Registry of Childhood Tumours, which includes birth records for children born and diagnosed in certain periods, we identified cases of malignant disease occurring among these children by the latest date for which, firstly, the registry was complete and, secondly, cases diagnosed up to that time had been linked to place of birth. This is, in effect, a record based, follow up study. We excluded retinoblastoma, cases associated with Down's syndrome and neurofibromatosis, and cases occurring in children aged under 1 year.
Information on hospitals to be included in this study was obtained from four sources.
(1) Study of selected maternity units: For 16 hospitals visited in carrying out the case-control study and for those in the Cardiff births survey, described by Passmore et al,3 we also obtained the information required for the ecological study.
(2) Questionnaire study: A further 80 hospitals were selected for possible inclusion by searching the National Registry of Childhood Tumours for hospitals where more than 25 children who subsequently developed cancer had been born in the period 1968-85. Paediatricians in each of these hospitals were asked to complete a form recording annual numbers of births and vitamin K policies. One follow up request was sent to those who had not returned a form within 3 months, and any unclear responses were clarified by post or telephone. (Some did not respond, and for many the forms had incomplete information for the relevant years.) No attempt was made to verify the information on vitamin K policies for the 47 hospitals included from this source; for six of these, more accurate birth numbers and more years of known vitamin K policy were subsequently obtained from the Scottish Neonatal Network (see below). Thus 41 hospitals were included in this part of the study.
(3) Study by Ansell et al: We ascertained total numbers of births and subsequent cases of cancer among these for four hospitals included in the study reported by Ansell et al.4 These authors provided us with information on the vitamin K policies in these hospitals.
(4) Scottish Neonatal Network: Dr Edmund Hey generously provided us with information obtained by members of the Scottish Neonatal Network on numbers of births and vitamin K policies for 30 hospitals in Scotland.
Births from a further three hospitals in the Cardiff births survey3 were also included, so that the whole study covered 94 hospitals, with a total of 2.3 million births occurring in periods when intramuscular vitamin K was routinely administered and 1.4 million births occurring when a selective policy was in operation. Most of the births occurred between 1968 and 1985, and cases were diagnosed by the end of 1986 (England and Wales) or 1994 (Scotland).
If complete information for the 15 years after birth was available the analysis would consist in comparing cumulative rates of cancer among groups of children born in hospitals and periods when there was a selective policy with those when there was a non-selective policy. For most of the births the effective follow up period (that is, the period between birth and the latest date for which we were able to link childhood cancer registrations to the hospital of birth) was less than the full 15 years for which rates of childhood cancer are calculated. We used the following standardisation procedure to allow for this.
For each hospital/policy and for seven diagnosis/age groups we calculated the expected number, E, of cases by using the number of births in each hospital in each year, the effective follow up period, and the cumulative disease rate for that year and period for cases in the National Registry of Childhood Tumours. The observed number of cases, O, in any such group will in general differ from E because of chance fluctuations and possibly because of incomplete ascertainment of cases. If there is no effect of vitamin K and we assume that the completeness of ascertainment is not related to the policy of vitamin K administration, the ratio O:E should on average be the same for hospitals/periods with selective and non-selective policies, and the hypothesis that there is no effect of vitamin K can be tested by comparing the O:E ratios; if vitamin K has an effect this ratio will be higher for the non-selective policy. If the cases of cancer could be regarded as statistically independent observations then each O would follow a Poisson distribution with mean equal to E. While it is probably a reasonable assumption that cases occur independently they may not be ascertained independently-for instance, for children born and remaining in a particular area the likelihood of ascertainment will depend, among other things, on the local cancer registry. Because of this the analyses are based on the assumption that the statistically independent quantities to be analysed are not the individual cases but the total numbers of cases observed among children born in each hospital and period.
Two sets of analyses were carried out. Firstly, for each group of hospitals (other than that reported by Ansell et al,4 where the number was insufficient) and for the total group, we compared those having only a non-selective policy during the period covered by the study with those having only a selective policy by ranking the O:E ratios and using the Wilcoxon two sample rank sum test5 to determine whether these ratios differ between the non-selective and selective policies. Secondly, for hospitals for which data had been obtained for periods with both selective and non-selective policies a “score” was allocated to each hospital by subtracting the O:E ratio for the selective period from that for the non-selective period. If vitamin K does not affect the cancer rates these scores should be symmetrically distributed with mean zero. For three groups of hospitals it was possible to test this hypothesis by ranking the absolute values of the scores for hospitals within a group and using the Wilcoxon paired rank test5 to determine whether the scores showed any tendency to be positive or negative, reflecting carcinogenic or protective effects of vitamin K. This test was also carried out for the total group of such hospitals with two policies.
Although the rank tests are statistically valid whereas those based on the assumption that the observed numbers follow a Poisson distribution may not be, the former lack statistical power, and we therefore also carried out analyses based on the Poisson assumption but with the proviso that there are circumstances, which we cannot in practice always identify, when they may be invalid.
The method of comparison for two Poisson distributions with unequal means with a χ2 test is given in Breslow and Day.6
If all cases were ascertained and the expected numbers calculated from rates that are based on data from cancer rates among children not given vitamin K, the O:E ratio for a group of babies all given vitamin K would be an estimate of the relative risk associated with such administration. In practice it may not be; firstly, because of the problems of ensuring that all cases have been ascertained and, secondly, because the expected numbers are calculated by using baseline rates derived from data collected during periods of vitamin K administration and which would be affected by any risk associated with vitamin K. Moreover, if the relative risk varies with age the estimate will depend on the distribution of follow up ages. If these factors can be ignored for a group where only some babies are given vitamin K the ratio O:E, which may be thought of as a group relative risk, gives an indirect estimate of the risk of vitamin K. For instance, if 30% of babies receive vitamin K a true relative risk of 1.5 would give on average a group relative risk of 1.15, using the O:E ratio; if 90% receive vitamin K the group relative risk would be 1.45. The ratio of these two group relative risks, which should be largely independent of the problems of ascertainment, etc, referred to above, is a measure of any vitamin K effect as it depends on the difference in the proportion of babies receiving vitamin K in the two groups. In this instance this ratio is 1.26, which is substantially less that the relative risk for an individual given vitamin K.
We obtained information on the vitamin K policies and numbers of births for 94 hospitals. A total of 3266 childhood cancers were identified among the 3.7 million children born in these hospitals in the periods covered by this study. The results are summarised in tables 1, 2, and 3.
In table 1, for each group of hospitals and for all taken together, we present for each age/diagnosis group the observed and expected numbers of cases and the O:E ratio, firstly, for hospitals with non-selective policies and, secondly, for those with selective policies. This is followed by the result of the Wilcoxon rank sum test, which tests the hypothesis that these ratios, and hence the rates of malignant disease, are the same for the two policies. The only significant (P<0.05, two tailed test) comparison suggests a protective effect of vitamin K, and the general pattern of results suggests that there is little, if any, difference in the rates associated with the two policies. The final column gives the results of the χ2 test on the basis of the, possibly invalid, assumption that the observed numbers each follow a Poisson distribution. The results are identical with those given by the rank test.
In table 2 we summarise the results of comparing the two policies within each hospital where both policies had been used by calculating the differences in the O:E values for the two policies and using Wilcoxon's paired rank test as described above to test for differences in malignant disease rates associated with the two policies. None of the differences was found to be significant at the 5% level with a two tailed test. Again, the results of the χ2 test on the basis of the assumption that the observed numbers each follow a Poisson distribution are given in the last column. The only significant values occur in the small group of seven “selected maternity units” (that is, selected for inclusion in our case-control study.) These four significant values are based on overlapping sets of data and seem to depend largely on the low O:E ratios found in the selective policy periods. In view of all the caveats to be attached to this analysis it would clearly be impossible to attach very much weight to these particular results, though, as explained below, we cannot exclude the possibility of an association between vitamin K and childhood leukaemia.
The overall results, separately for the one policy and two policy hospitals, and then for all hospitals together, are summarised in table 3. In this table we also give in the last column the ratio of each pair of O:E values, referred to here as the “risk ratio.” This ratio provides for each age/diagnosis group a comparison between the risk for children born in hospitals in which all, or nearly all, babies are given vitamin K and for those in hospitals in which only some (usually about 25-30%) are so treated. It can be seen that there is a slight tendency for the use of vitamin K to be associated with an increase in rates of leukaemia. (The χ2 value (3.79, not shown) for leukaemia comparing non-selective and selective policies for all hospitals is not quite significant at the 5% level.) As with ecological studies in general this may be due to some confounding effect-that is, some factor associated with use of vitamin K that is itself causally related to childhood cancer. In particular, there are time trends in the use of different policies, and the standardisation procedures used may not completely allow for this.
The cases analysed here include some from our case-control study reported in the accompanying paper3 and some from the study by Ansell et al4 so that the results are not independent. Those for the present paper, however, are based on more cases, and the method of analysis is entirely different and, in particular, does not use the controls from the other studies. Although (with some minor and possibly questionable exceptions) the results of the present analyses are not significant and there is no suggestion of a doubling of the risk of malignant disease arising from the use of intramuscular vitamin K, the findings for childhood leukaemia are compatible with an increased risk of around 20-30%, as, by using the same argument as at the end of the section on statistical methods, it can be shown that an individual relative risk of 1.25 gives a risk ratio of 1.14, and such values occur in table 3
Various colleagues are thanked in the accompanying paper. We are very grateful to Dr Hey and the members of the Scottish Neonatal Network for providing carefully validated information on numbers of births and vitamin K prophylaxis policies in Scottish hospitals.
Funding: This work was funded by the Department of Health; the views expressed in this publication are those of the authors and not necessarily those of the Department of Health.
Conflict of interest: None.
Contributors: SJP and GD initiated the study and designed the protocol. PB was responsible for setting up and manipulating computer databases. Statistical analyses were carried out by GD and MK. The paper was written jointly by SJP and GD, who are guarantors for the paper.