| Japanese Journal of Clinical Oncology | Pages |
Introduction
Subjects And Methods
Study Population
Active Follow-up
Passive Follow-up
Analysis
Results
Discussion
Acknowledgments
References
Evaluation of the Death Certificate Follow-up Method for the Analysis of Survival Rate: Data from Aichi Prefecture, Japan
To evaluate the use, for studies on survival, of death certificates from population-based cancer registries in Japan, we compared 5-year survival rates by two different methods: passive, derived from death certificates, and active, using family register systems. Registered cancer cases from Aichi Prefectural Cancer Registry were used as a model. The study subjects comprised 9244 cancer patients (3830 males and 5414 females) newly diagnosed at the Aichi Cancer Center Hospital and recorded in the Aichi Prefectural Cancer Registry between 1983 and 1991. The passive follow-up method, using death certificates, identified 87-89% of deaths among the study subjects. The proportion of deaths which were not identified by the passive follow-up method did not vary greatly with age or gender, but was higher among patients in the earlier rather than the later stages of cancer. Overall, the absolute and relative effects of unregistered deaths on apparent survival rate vary with the absolute survival rate. The absolute and relative differences between the survival rates calculated by the two methods are explained more clearly when data are analyzed by cancer site. The results of the present study provide useful information for interpreting the survival rate following diagnosis of cancer estimated by the passive follow-up method, i.e. using death certificates from a population-based cancer registry.
INTRODUCTION
A 5-year survival rate is one of the most useful indices for cancer control measures. Many population-based cancer registries in Japan have tried to calculate 5-year survival rates for each cancer site by using their own follow-up methods, taking into account their available manpower and the size of their population base.
One of the most widely-used sources of follow-up in the population-based cancer registry is death certificates; they are also the source of information for `death certificate only' cases when we estimate their incidence. In Japan, every death certificate is compulsorily submitted to a local public health center operated by the local government, generally the Prefecture. The information from the death certificates is then compiled by the Ministry of Health and Welfare and transferred to the computer database for national vital statistics. However, due to restriction by the Management and Coordination Agency, each Prefectural government is allowed to obtain only its own local vital statistics death certificate file from the Ministry of Health and Welfare. Accordingly, each population-based (mostly Prefecture-based) cancer registry has no access to any death certificates of out-migrant registered cancer patients who died in other areas of registration (1). The proportion of out-migrant registered cancer patients varies with the area of registration. Moreover, it is not easy for an individual registry with a limited number of staff to conduct an active follow-up of registered patients on a routine basis by other methods, unless it has a relatively small population or few cases. Consequently, there is no way of calculating actual 5-year survival rates from most population-based cancer registries in Japan.
At the Aichi Cancer Center Hospital (ACCH), a hospital-based cancer registry was started in 1964 and a thorough follow-up of registered patients has been conducted by reference to family registers. In 1993, almost all registered cancer patients (99.8%) were traced, and we can thus regard this follow-up data as reliable. All registered patients living in Aichi Prefecture have been reported to the Aichi Cancer Registry, which is a population-based cancer registry in Aichi Prefecture, Japan. Aichi Prefecture has a population of 6.7 million (2), one of the largest in Japan. At the Aichi Cancer Registry, a follow-up of registered cancer patients has been conducted by using Aichi Prefecture death certificates since 1983. Consequently, for registered cancer patients diagnosed after 1983 in ACCH, two different kinds of death records are available: a hospital-based follow-up system and death certificates.
In the present study, we examined the 5-year survival rates of registered cancer patients at ACCH, comparing death certificates (passive follow-up) with death records which were systematically compiled at ACCH (active follow-up). The main purpose was to assess the 5-year survival rate difference between these two follow-up methods and to evaluate the validity of passive follow-up, using death certificates from population-based cancer registries in Japan in which only death certificates were available, for calculation of the 5-year survival rates.
SUBJECTS AND METHODS
Study Population
The subjects comprised 9244 cancer patients (3830 males and 5414 females), newly-diagnosed in ACCH between 1983 and 1991, who were residents of Aichi Prefecture at the time of diagnosis. All of the subjects were identified via the ACCH cancer registry database, and registration of the same subjects could also be confirmed at the Aichi Prefectural Cancer Registry (Aichi Cancer Registry). Two sources of survival and death records were thus available in these two different cancer registry systems.
Active Follow-up
In the ACCH cancer registry, survival and death checks of registered cancer patients have been conducted by reference to family registers with the permission of the Ministry of Justice. In Japan, which has no system of unique identification such as a social security number, a family register is one of the most thorough and accurate registry systems within the national administration. Although the information is accurate, strict confidentiality prevents our obtaining data in computerized form from family registers. Original family registers are kept by each local government office.
After getting permission from the Ministry of Justice we sent lists, including information on self-identification of all patients registered in the ACCH cancer registry still recorded as surviving, to each local government office in charge of the family register, and obtained the latest information on survival. With regard to patients whose survival or death was omitted from a family register (approximately 0.4% per year), further inquiries were made at the office at which they were registered to ascertain whether they were alive or dead.
Active follow-up has been conducted annually for every registered patient and has been continued for 10 years after diagnosis. As of 1993, 99.8% of all subjects were traced by this active follow-up method. Fifteen untracked patients (0.2%), most of whom had emigrated to other countries, were considered as surviving.
At the Aichi Cancer Registry, death has been determined for registered patients diagnosed after 1983 from Aichi Prefecture death certificates only. The Aichi Cancer Registry could obtain computerized data files of the deaths of Aichi Prefecture residents only. In this database, indices such as the national index number and name were not available for matching, for reasons of confidentiality. Accordingly, we used gender, date of birth and city code in matching with the cancer registry file, and verified each matched case by making reference to the original death certificates which were kept by the local public health center. Passive follow-up has been conducted annually because the data file is provided on a yearly basis. In this passive follow-up, deaths were checked up to 1991, and any registered patient whose death had not been noted in the registry was considered as surviving.
The end-point of this study was determined by death, regardless of cause. With every case used in this study, we confirmed that a death noted in the death certificates was also in the family registers.
Five-year observed survival rates were calculated by gender, age group ( <= 49, 50-59, 60-69, >= 70 years) and site of cancer, using two different follow-up methods: passive follow-up from death certificates and active follow-up from the family register. Age was defined as the age at which cancer was first diagnosed. The date of this first diagnosis was also defined as the starting date of the follow-up period. The cancer site of each patient was classified according to the ninth revision of the International Classification of Diseases (ICD-9) (3).
To clarify other factors which caused distorted survival rates in passive follow-up, 5-year survival rates were also assessed according to clinical stage for the most common cancers in Aichi Prefecture, i.e., stomach and lung cancer in males and stomach and breast cancer in females. In the Aichi Cancer Registry, the reported cancer patients were divided into four groups according to their clinical stage, i.e., localized, extension to regional lymph nodes, direct extension to adjacent organs or tissues, and distant metastasis. For analysis in the present study, these clinical stages were regrouped into only three stages: Stage 1 (localized), Stage 2 (extension to regional lymph nodes) and Stage 3 (direct extension to adjacent organs or tissues, and distant metastasis).
The life-table method was used for calculating these survival rates in registered cancer patients in all sites having >= 50 patients. The Logrank test was used to test the equality of the survivor function across the different follow-up methods. The LIFETEST procedure from the SAS (4) computer system was used to perform the calculation.
Table 1 shows the 5-year survival rates based on the two different follow-up methods among registered cancer patients at ACCH, by gender and age group. Average follow-up periods were 3.0 years by active follow-up and 3.2 years by passive follow-up. In males, 1800 of the 2012 deaths (89%) were assessed from death certificates, and the 5-year survival rate for the total cases was 46.6% by passive follow-up and 40.7% by active follow-up. In females, 1388 deaths were recorded through use of death certificates compared with 1604 deaths by active follow-up (87%). The 5-year survival rate was 65.9% by active follow-up but 70.1% by passive follow-up. In each age group, 84-90% of patient deaths determined by active follow-up were recorded through reference to death certificates. The absolute difference between 5-year survival rates was greater in older age groups than younger, and greater in males than females except the group aged >= 70 years. The relative difference in 5-year survival rates (ratio of survival rate by passive follow-up to active follow-up) was lower and closer to 1.0 in females than males, for all age groups.
Tables 2 and 3 show the 5-year survival rates by two different follow-up methods, gender and cancer site. The proportion of patient registrations lost to passive follow-up varied by site. In males (Table 2), the proportion of deaths identified using passive follow-up was high in cancer of the colon (0.94) and lung (0.93), and low in liver cancer (0.83). The absolute difference between the 5-year survival rates was remarkably high in cancer of the liver (16.5%) and oesophagus (9.4%). The excess ratio of survival rate calculated by passive follow-up was <5% (1.05) in 5 years in cancer of the tongue (1.04), colon (1.04) and larynx (1.03). On the other hand, the ratio was high in cancer of the liver (3.45), pancreas (3.46), oesophagus (1.49) and lung (1.28). In females (Table 3), the proportion of deaths identified using passive follow-up was high in biliary tract cancer (0.95) and low in cancers of the thyroid (0.58), corpus uteri (0.77), colon (0.79) and rectum (0.79). The absolute difference between the 5-year survival rates was remarkably high in cancer of the ovary (11.3%) and lung (9.9%). The excess ratio of survival rate calculated by passive follow-up was low in cancer of the breast (1.02) and cervix uteri (1.04). On the other hand, the ratio was high in cancer of the biliary tract (1.68), lung (1.34) and ovary (1.28). In both genders, the relative difference in 5-year survival rates of these cancers tended to be higher in the cancer sites with poor prognosis.
Table 1
Table 2
Table 3
Table 4
Gender and age
No. of patients
>Active follow-up
Passive follow-up
p-value (Logrank)
Comparisons
Deaths (a)
% 5-year OSR (c)
(SE)
Deaths (b)
% 5-year OSR (d)
(SE)
(b)/(a)[dagger]
(d) - (c)[Dagger]
(d)/(c)§
Male
All ages
3830
2012
40.7
(0.9)
1800
46.6
(0.9)
<0.01
0.89
6.0
1.15
Age
<= 49
669
275
53.0
(2.2)
246
58.4
(2.1)
NS
0.89
5.4
1.10
50-59
1073
520
45.5
(1.7)
463
51.3
(1.7)
0.01
0.89
5.8
1.13
60-69
1183
624
39.5
(1.7)
564
45.2
(1.7)
0.02
0.90
5.7
1.15
>= 70
905
593
27.5
(1.7)
527
34.1
(1.8)
0.01
0.89
6.6
1.24
Female
All ages
5414
1604
65.9
(0.7)
1388
70.1
(0.7)
0.01
0.87
4.2
1.06
Age
<= 49
2079
457
73.6
(1.1)
406
76.3
(1.1)
0.05
0.89
2.7
1.04
50-59
1405
394
68.8
(1.4)
340
72.9
(1.3)
0.02
0.86
4.1
1.06
60-69
1213
390
62.1
(1.6)
337
66.6
(1.6)
0.02
0.86
4.6
1.07
>= 70
717
363
45.1
(2.1)
305
52.8
(2.1)
<0.01
0.84
7.8
1.17
Site
(ICD-9)
No. of patients
Active follow-up
Passive follow-up
p-value (Logrank)
Comparisons
Deaths (a)
% 5-year OSR (c)
(SE)
Deaths (b)
% 5-year OSR (d)
(SE)
(b)/(a)[dagger]
(d) - (c)[Dagger]
(d)/(c)§
Larynx
(161)
113
36
63.1
(5.3)
32
65.1
(5.2)
NS
0.89
1.9
1.03
Colon
(153)
292
100
57.3
(3.5)
94
59.8
(3.4)
NS
0.94
2.5
1.04
Tongue
(141)
75
25
57.2
(7.3)
23
59.2
(7.1)
NS
0.92
2.0
1.04
Stomach
(151)
1191
493
53.9
(1.6)
431
59.2
(1.6)
0.01
0.87
5.3
1.10
Rectum
(154)
288
112
52.7
(3.5)
100
58.1
(3.4)
NS
0.89
5.4
1.10
NHL
(202)
155
71
50.0
(4.4)
62
56.9
(4.3)
NS
0.87
6.9
1.14
Lung
(162)
729
518
20.8
(1.7)
480
26.7
(1.9)
0.04
0.93
5.9
1.28
Oesophagus
(150)
220
156
19.2
(3.1)
138
28.6
(3.5)
NS
0.88
9.4
1.49
Liver
(155)
188
150
6.7
(2.7)
125
23.3
(3.7)
<0.01
0.83
16.5
3.45
Pancreas
(157)
78
74
2.6
(2.3)
68
9.1
(3.8)
NS
0.92
6.5
3.46
Site
(ICD-9)
No. of patients
Active follow-up
Passive follow-up
p-value (Logrank)
Comparisons
Deaths (a)
% 5-year OSR (c)
(SE)
Deaths (b)
% 5-year OSR (d)
(SE)
(b)/(a)[dagger]
(d) - (c)[Dagger]
(d)/(c)§
Thyroid
(193)
109
12
86.7
(4.0)
7
93.0
(2.8)
NS
0.58
6.3
1.07
Breast
(174)
2181
403
78.9
(1.0)
356
80.8
(9.8)
NS
0.88
1.8
1.02
Corpus uteri
(182)
166
31
77.6
(3.9)
24
82.5
(3.3)
NS
0.77
4.8
1.06
Cervix uteri
(180)
850
182
75.5
(1.7)
154
78.7
(1.6)
NS
0.85
3.3
1.04
Tongue
(141)
62
22
62.3
(7.0)
19
65.8
(6.9)
NS
0.86
1.6
1.06
Colon
(153)
257
80
62.3
(3.6)
63
69.2
(3.4)
NS
0.79
6.9
1.11
Rectum
(154)
164
58
58.3
(4.5)
46
65.5
(4.4)
NS
0.79
7.2
1.12
NHL
(202)
97
35
57.5
(5.9)
28
65.7
(5.5)
NS
0.80
8.2
1.14
Stomach
(151)
717
300
52.0
(2.1)
272
56.7
(2.1)
NS
0.91
4.8
1.09
Ovary
(183)
116
53
39.8
(6.1)
44
51.0
(5.9)
NS
0.83
11.3
1.28
Lung
(162)
247
151
29.3
(3.5)
131
39.2
(3.7)
NS
0.87
9.9
1.34
Biliary tract
(156)
69
59
7.6
(3.5)
56
12.8
(4.4)
NS
0.95
5.2
1.68
Site (ICD-9) and stage
No. of patients
Active follow-up
Passive follow-up
p-value (Logrank)
Comparisons
Deaths (a)
% 5-year OSR (c)
(SE)
Deaths (b)
% 5-year OSR (d)
(SE)
(b)/(a)[dagger]
(d) - (c)[Dagger]
(d)/(c)§
Male
Stomach (151)/TD>
Stage 1
547
82
82.6
(2.0)
64
85.8
(1.8)
NS
0.78
3.2
1.04
Stage 2
352
196
34.6
(3.1)
174
41.6
(3.1)
NS
0.89
7.1
1.20
Stage 3
132
123
3.0
(1.9)
116
8.4
(2.8)
NS
0.94
5.5
2.84
Lung (162)
Stage 1
246
123
41.2
(3.6)
109
47.5
(3.7)
NS
0.89
6.3
1.15
Stage 2
175
128
18.0
(3.5)
121
22.9
(3.7)
NS
0.95
4.8
1.27
Stage 3
166
149
1.9
(1.3)
141
7.7
(2.3)
NS
0.95
5.8
4.04
Female
Stomach (151)
Stage 1
303
35
84.7
(2.6)
29
87.6
(2.3)
NS
0.83
3.0
1.04
Stage 2
235
144
30.1
(3.5)
132
35.5
(3.6)
NS
0.92
5.4
1.18
Stage 3
81
73
5.2
(3.0)
68
13.0
(4.0)
NS
0.93
7.8
2.51
Breast (174)
Stage 1
966
48
94.3
(9.4)
36
94.8
(9.1)
NS
0.75
0.5
1.00
Stage 2
683
183
65.9
(2.2)
168
68.4
(2.2)
NS
0.92
2.5
1.04
Stage 3
118
79
26.7
(4.6)
73
31.6
(4.8)
NS
0.92
4.9
1.18
From the 5-year survival rates for selected sites of cancer (stomach and lung in males and stomach and breast in females; Table 4), we see that excess rate ratios were very small in Stage 1 but increased sharply with stage progression.
The present study shows that about 10% of deaths of all registered cancer cases over a 5-year period were overlooked using the death certificate method. We identified two major reasons for persons being lost to follow-up; one is collation error between the registry and the death certificate files, and the other is out-migration of registered patients to other Prefectures or other jurisdictional areas. Unfortunately, we could not quantify the individual contributions of the two reasons because the family register sometimes prevents the active follow-up process. We refer to family registers as the first step in the follow-up process. In practice, we were able to obtain information on survival for 99% of registered cases by reference to family registers only, and therefore it was unnecessary to make further inquiries at other information sources. However, family registers do not give a residential address and, in most cases therefore, it is not possible to get information by active follow-up on the change of residential address. Consequently, in the present study, we could not assess in a quantitative way which reason contributed more to those being lost to follow-up.
Collation error of the death certificate file and the registry file is one possible reason for persons being lost to follow-up. This error is considered to be mainly due to matching process error. As mentioned before, we are able to obtain the computerized death certificate file of Aichi Prefecture residents only. This file lacks name and unique identification number, so we use gender, date of birth and address code in matching each case in the cancer registry file to its reference on the original death certificate. One reason for matching error is inaccuracy of the reported gender, date of birth and other index items on the cancer registry files. ACCH reports its cancer cases to Aichi Cancer Registry by computer file index items which come from the hospital administrative data file. The ACCH administrative data are first loaded into the computer database at the time of the patient's first visit to the hospital and, at the same time, patients receive their computerized hospital ID cards showing index items such as hospital ID, name, gender and date of birth. Correction of these index items can be conducted at any time by either the patients or their families, or by the hospital staff. Therefore it is likely that errors in ACCH administrative data are few; nevertheless, we cannot discount the possibility of this sort of error altogether.
The complete loss by out-migration of registered cancer patients is critical (7). However, it is suggested that most of the discrepant survival ratio is due to the loss of out-migrants. The proportion of those lost to follow-up, 10%, matches the average annual proportion (2-3%) of out-migrants for Aichi Prefecture in the study period (8). One would expect that there are more out-migrants among patients with cancers in the early stages, or with good prognosis, than among those in later stages, or with poor prognosis. Indeed, the proportion of deaths not identified by passive follow-up was high among cancers with good prognosis, e.g., cancers of thyroid and corpus uteri in females, and among cancers with less advanced stages. We could not clarify which factor correlated with out-migration, due to the process of active follow-up mentioned above, but there may be two reasons for this; one is that patients in earlier stages or with good prognosis have a greater chance to move because of longer survival time, and the other is that the rate of out-migration per unit time is higher because of higher daily activity.
Assuming the active follow-up of ACCH to be reliable, we calculated three indicators for the assessment of survival differences: (i) the proportions of deaths identified using passive follow-up, (ii) the absolute difference in 5-year survival rates between the two follow-up methods, and (iii) the relative difference in 5-year survival rates (ratio of survival rates comparing passive follow-up to active follow-up). Interpretation of these three indicators is not simple. For example, the proportion of deaths identified by passive follow-up for males was no different from that for females (89% vs 87%), but the 5-year survival rate was more overestimated for males than for females in terms of both absolute (6% vs 4.2%) and relative (1.15 vs. 1.04) difference. For cancers with poor prognosis, the survival ratio exaggerated the difference. Using passive follow-up, 5-year survival following cancer of the pancreas in males was about 3.5 times the survival rate calculated by active follow-up, but the absolute difference was only 6.5% (9.1% vs 2.6%). For cancer of the corpus uteri, for which the prognosis is relatively good, the rate ratio was only 1.04 even though the absolute difference in 5-year survival rate is similar to that for cancer of the pancreas (4.9%). Which indicator is the better depends on the situation, the type of information one needs, and the cancer site.
One would expect a greater effect on both absolute and relative differences in estimated survival following cancers with poor prognosis, since in this situation a larger proportion of cases will have incorrect survival times assigned by the passive follow-up method. For cancers with good prognosis, unmatched cases are more likely to be those of surviving patients, even if such patients move out of the Prefecture. Consequently, survival rate estimated by passive follow-up is expected to be close to that of active follow-up. On the other hand, even if the proportion of out-migrants was much less among patients with cancers with poor prognosis than among those with good prognosis, overestimate of the actual survival rate would be higher in the former than the latter. One reason may be that the relative differences between the survival rates obtained by the two follow-up methods become larger, due to the relatively small value of estimated survival rates following cancers with poor prognosis. Also, all unmatched deaths were wrongly considered to be surviving and that made a larger absolute difference between the two follow-up methods.
One methodological issue in this study is whether cancer patients at ACCH are representative of Aichi Prefecture cancer patients as a whole. ACCH is a Prefecture-based local hospital, and about 80% of all cancer patients are domiciled in Aichi Prefecture (5). In the study period, 9% of all cancer patients registered in Aichi Prefecture were from ACCH (6). It is unlikely that cancer patients treated at ACCH are more likely to leave Aichi Prefecture after reporting than are cancer patients at other hospitals. All newly diagnosed cancer patients are registered with the ACCH Cancer Registry, and we used all the registered patients living in Aichi Prefecture. Therefore, it is unlikely that the patients who were used in this study were selected from some biased population.
Another methodological issue in this study is completeness of the reporting system for death certificates. All death certificates are compulsorily filled out by a medical doctor licensed by the national government, and are submitted to a local public health center operated by the government. Reporting death is compulsory in Japan, and essential for insurance, inheritances and other legalities. Therefore, accurate documentation of deaths is considered to be totally reliable in Japan. Osaka Prefectural Cancer Registry has also mentioned this point (1). Furthermore, we did not focus on the actual cause of death, because survival and death information is more important in the present study.
In Japan, the population-based cancer registries are organized by local governments and not the national government. Therefore, it is possible to overcome matching error if out-migrant follow-up is conducted in cooperation with neighbourhood and other population-based cancer registries, or if comprehensively identifying death certificate files were to become available on a national basis.
In summary, 5-year survival rates were compared between two different follow-up methods, passive follow-up by death certificates and active follow-up by family registers. Passive follow-up by death certificates identified 87-89% of deaths. The proportion of deaths which could not be identified by the passive follow-up method did not differ greatly with age or gender, but it was higher among patients in earlier stages of cancer than in later stages. However, the absolute and relative effects of missing death registrations on apparent survival rate vary with the absolute survival rate. The absolute and relative differences between the survival rates calculated by the two methods are explained more clearly when data are analyzed by cancer site. When we estimate the survival rates using passive follow-up methods, careful interpretation of these data is needed.
The authors are grateful to Mss Takako Kuno, Masae Torii and Sumie Yasui and Mr Kiyoshi Goshima for their work on the Aichi Prefectural Cancer Registry, and to Messrs Masatsugu Ando and Yasushi Mori for their work on the Aichi Cancer Center Hospital Cancer Registry. This study was supported in part by the Grant-in-Aid for Cancer Research (8-2) from the Ministry of Health and Welfare, Japan.
References
For reprints and all correspondence: Manami Inoue, Division of Epidemiology, Aichi Cancer Center Research Institute, 1-1, Kanokoden, Chikusa-ku, Nagoya 464-8681, Japan
Abbreviations: ACCH, Aichi Cancer Center Hospital; ICD-9, the ninth revision of the International Classification of Diseases
This page is run by Oxford University Press, Great Clarendon Street, Oxford OX2 6DP, as part of the OUP Journals
Comments and feedback: www-admin{at}oup.co.uk
Last modification: 19 May 1998
Copyright© Japanese Journal of Clinical Oncology, 1998.
CiteULike 
Connotea 
Del.icio.us What's this?
This Article 
Abstract
Alert me when this article is cited
Alert me if a correction is posted
Services
Email this article to a friend
Similar articles in this journal
Similar articles in ISI Web of Science
Similar articles in PubMed
Alert me to new issues of the journal
Add to My Personal Archive
Download to citation manager
Request Permissions
Google Scholar
Articles by Inoue, M
Articles by Inuzuka, K
Search for Related Content
PubMed
PubMed Citation
Articles by Inoue, M
Articles by Inuzuka, K
Social Bookmarking
What's this?