A Method for Identifying Underlying Causes of Death in Epidemiological Study

To obtain the underlying causes of death in individuals of a cohort (male only), the items of date of birth, date of death and address code at the time of death were linked between the data of resident cards and the death tapes of National Vital Statistics supplied by the Ministry of Health and Welfare. As a result, the persons who have the same information for above mentioned three items between the resident cards and the death tapes accounted for 97.4%. If the persons who had the same information for three items except one item were considered to be identical, they accounted for 99.4%. It would be concluded that underlying cause of death can be obtained by record linkage of death tapes of National Vital Statistics with three informations in residence cards, dates of birth, dates of death and address codes at the time of death, even without names of the individual in the death tapes. The matched rate would be high enough for epidemiological studies. J Epidemiol, 2000 ; 10 : 362-365


INTRODUCTION
Since 1990, the Institute of Radiation Epidemiology of the Radiation Effects Association has been studying the relationship between exposure to radiation and causes of death in people who were engaged in radiation work at nuclear facilities.The study was entrusted to the Institute by the Science and Technology Agency to study the health effects of exposure to low-dose and low-dose rate radiation on human 1).As a study method, the copies of resident cards of the subjects under investigation were obtained to confirm whether they were alive or dead.For the persons who were dead, their underlying causes of death were identified by record linkage with the death tapes of National Vital Statistics (hereinafter referred to as "death tapes") supplied by the Ministry of Health and Welfare (M.H.W.) of Japan.In these death tapes, however, there were no descriptions of the names of the deceased.Therefore, three items of date of birth, date of death and address at the time of death were used for identification of individuals.Since the subjects under investigation were limited to males, only the death tapes of males were obtained and used.
The method for identifying underlying causes of death by mean of record linkage of resident cards and the death tapes was performed.
This paper reports the results of record linkage of the death information from the resident cards and the death tapes supplied by the M.H.W., discusses the problems of this method, and proposes this method as a useful procedure for epidemiological studies in Japan.

METHOD
The copies of the resident cards of the deceased contain the names, dates of birth, addresses at the time of death, and dates of death of the deceased, without information of the causes of death.To obtain the underlying causes of death, therefore, the death information in resident cards of the deceased were linked with the death tapes supplied by the M.H.W.The death tapes contain the information about dates of birth, dates of death, address codes at the time of death.In the present paper, it was adopted underlying causes of death coded by the ICD 9 (140-239) for about 3,200,000 of all the male deaths aged 18 or older from 1986 to 1992.Among radiation workers investigated, 1,758 deceased cases were confirmed by resident cards during the same periods.The identification work was performed by record linkage of 3 items of the dates of birth, dates of death and address codes at the time of death between the information of resident cards and that of the death tapes.
Identification method 1.The deceased who have the same information about the following three items as that of resident cards were retrieved from the death tapes: (1) Date of birth (seven figures including the name of an era) (2) Date of death (seven figures including the name of an era) (3) Address code at the time of death (five figures in each code assigned by the Ministry of Home Affairs) The deceased who had the same information about the above three items in the resident cards and magnetic tapes were regarded as identical.
2. The deceased who did not have the same information about these three items were subjected to record linkage again as follows: (1) The persons who have the same information about the date of death and address code at the time of death, except the date of birth.(2) The persons who have the same information about the date of birth and address code at the time of death, except the date of death.(3) The persons who have the same information about the date of birth and date of death, except the address code at the time of death.All the cases which did not completely match with each other in any of the above three items were retrieved as "a similar list."Among these retrieved cases, the persons who came under the following conditions were considered to be identical: For example, in (1) above, the persons with their dates of death and address codes at the time of death matching with each other were retrieved, and then the persons with their dates of birth most similar to each other (e.g., February 11 and February 17) were considered to be identical as shown in Table 1.In (2) above, the persons with their dates of birth and address codes at the time of death matching with each other retrieved likewise, and then the persons with their dates of death most similar to each other (e.g., March 10 and March 9) were considered to be identical as shown in Table 2.In (3) above, the persons with their dates of birth and dates of death matching with each other and only the address codes being different were retrieved likewise.Among those retrieved persons, those with the identical dates of birth and dates of death in the address codes of neighboring towns and villages (e.g., 14101 and 14102) were considered to be identical.In this record linkage, a COBOL program was used.

RESULTS
The persons whose dates of birth, dates of death and address codes at the time of death completely matched with each other were 1,713 (97.4%) of the 1,758 surveyed (Table 3).The per-  sons who had the same information about these three items were judged to be identical and given their underlying causes of death.For 45 persons who did not have the same information about these three items, the persons who have the same information for two of the three items as that in the resident cards were retrieved from the death tapes.Then, record linkage was tried again for the persons who have the same information for two of the three items, in the same way as for the identification method.
As shown in Table 3, the persons whose dates of death and address codes at the time of death matched with each other were 14 (0.8%), the persons whose dates of birth and address codes at the time of death matched with each other were 16 (0.9%), and the persons whose dates of birth and dates of death matched with each other were 5 (0.3%).These persons could be judged to be identical.On the other hand, the persons who were completely different from each other in this record linkage and could not be identified were 10 (0.6%).In other words, the underlying causes of death of 1,748 (99.4%) out of the 1,758 deaths were finally obtained by this method.
There was no duplication of deaths in this record linkage method.

DISCUSSION
Epidemiological studies have been carried out with people engaged in radiation work in order to know the effects of exposure to low-dose and low-dose rate radiation on the causes of their death, including cancers 1).The resident cards of subjects of investigation were obtained to confirm whether they were alive or dead.However, these resident cards did not contain the causes of death.
To know the causes of death , a method was used in which the underlying causes of death were confirmed by record linkage between the information of the resident cards and that of the death tapes supplied by the M.H.W.As a measure for verifying the identification of the deceased in this record linkage method, it will be necessary to confirm that the deceased is truely identical by perusing the death regis-tration form (kept only five years) at public health centers or death certificates kept at the Legal Affairs Bureaus, Ministry of Justice.
record linkage for related documents of atomic bomb survivors.According to their report, for example, the probability that in a paired sample of different persons the family name and present address of a person match with those of the counterpart is 0.52 X 10-5, and the probability that in a paired sample of different persons the family name and present address of a person do not match with those of the counterpart is 0.997.As shown in the above, the probability that the matched data of two persons are identical and they are actually different persons is very low.Therefore, we think this method shows high performance in record linkage.

Table 1 .
Cases identified by changing of date of birth.

Table 2 .
Cases identified by chaneine of date of death .

Table 3 .
Results of record linkage for underlying causes of death (Subjects: 1,758).high as 99.4%.It is considered that the reason for such a high correspondence rate is due to our study using the death record in resident cards and magnetic tapes, which are both the offi- This, however, only means the match of linkage between two records.It is impossible to confirm that they are definitely identical.Ohtaki et al. 6) have done a theoretical analysis of