Document omkeR9RKR3o0eEKM66zyyaaYR

TEXAS DEPARTMENT- OF HEALTH Request for Personal Data All requests for personal data shall be in writing and shall include the following information and assurances: 1. Name and address of agency, institution, or firm sponsoring the project. 2. Name, degree(s), title, address, and phone number of person who will direct the project. 3. Name and address of agency. Institution, or firm funding the project (if other than that shown in Item 1). 4. Names, degree(s), and titles of other persons who will have supervisory responsibilities in the project. 5. Purpose of project. Be specific. 6. Data involved. Specify type of data needed and for what years, (e.g.. Mortality, 1975-1979). Specify if names or other Identifying data are needed. 7. Action planned. Be specific. 8. Results expected. Be specific. 9. The sponsoring agency, institution, or firm assures the Texas Department of Health that the release of the requested data is subject to the following conditions: (a) the data shall be treated as strictly confidential. (b) the data shall not be used for any purpose other than that specifically set forth above and shall not be used for any secondary purpose. (c) the data shall not be made available to any other individual, agency, institution, or firm. (d) appropriate controls shall be maintained to prevent unauthorized access to the data. (e) no follow back of any type shall be made to any individual, institution, or agency without written authorization by the Texas Department of Health. (f) any data released by a project shall be restricted to aggregate data and shall not identify any Individual or institution. (g) the Texas Department of Health shall be given credit as the source of the data. (h) a copy of the results of the project shall be furnished the Texas Department of Health. (i) If computer tapes are requested, such tapes, after serving the purpose set forth above, shall be returned to the Texas Department of Health unless purchase of the tapes is requested and approved in advance of their preparation. Further, retained work tapes shall be "scratched" unless specific authority is requested and granted for their retention and future use. 10. Name and address of person(s) to whom data and billings are to be sent must be provided. 11. The request oust be signed by the appropriate administrative officer of the sponsoring agency, institution, or firm. NOTE: If applicable, (1) data will be furnished on a cost reimbursement basis, and (*) \ navm^nf ntUQf Ka rard> 1nfirtF f rs FKm f +- H a *4 a* 4 UCC 096976 TEXAS DEPARTMENT OF-HEALTH Bureau of Vital Statistics Request for Copies of Vital Statistics Tapes All requests for copies of vital statistics computer tapes shall be in writing and shall include the following information and assurances: 1. Name and address of agency, institution, or firm sponsoring the project. 2. Name, degree(s), title, address, and phone number of person who will direct the project. 3. Name and address of agency funding the project (if other than that shown in Item 1). 4. Names, degree(s), and titles of other persons who will have supervisory responsibilities in the project. 5. Purpose of project. Be specific. 6. Vital records data involved. Specify type of data and the time period desired, (e.g.. Mortality, 1975-1979). Specify if names and other identi fying data are needed, and if cause-of-death data are needed from death files. 7. Action planned. Be specific. 8. Results expected. Be specific. 9. The sponsoring agency, institution, or firm assures the Texas Department of Health that the release of the requested vital statistics data is sub ject to the following conditions: (a) the data shall be treated as strictly confidential. (b) the data shall not be used for any purpose other than that speci fically set forth above and shall not be used for any secondary purpose. (c) the data shall not be made available to any other Individual, agency, institution, or firm. (d) appropriate controls shall be maintained to prevent unauthorized access to the data. (e) no follow back of any type shall be made to any individual or institution without written authorization by the Texas Department of Health. (f) any data released by a project shall be restricted to aggregate data and shall not identify any individual or institution. (g) the Texas Department of Health shall be given credit as the source of the data. (h) a copy of the results of the project shall be furnished the Texas Department of Health. (i) after serving the purpose set forth above, the tapes shall be returned to the Texas Department of Health unless purchase of the tapes is requested and approved in advance of their preparation. Retained work tapes shall be "scratched" unless specific authority is requested and granted for their retention and future use. 10. Name and address of person(s) to whom data and billings are to be sent. 11. The request must be signed by the appropriate administrative officer of the sponsoring agency, institution, or firm. NOTE: Data furnished shall be on a cost reimbursement basis. Payment must be received prior to the release of the data. TJCC 096977 Appendix Hi Accuracy of Cane r Death Certificat s and Its Effect on Cancer Mortality Statistics*1 Constance Perc y, MS, Edward StaNek, III. MS. and Lynn Gi.oeckler, MS Abstract: A study to determine the accuracy of cancer mortality data was done using cancer deaths occurring during 1970 and 1971 in eight of the nine areas included in the Third National Cancer Survey (TNCS). Death certificates with an underlying cause of death of cancer were compared to the hospital diagno sis for 48.826 resident cases of single primary cancers. The underlying cause of death as coded on the death certificate was found to be accurate for about 65 per cent of the cancer deaths in this study. Mis- classification problems occurred for colorectal cancer, the second leading cause of death from cancer. Colon cancer was overreported and rectal cancer was under reported on death certificates. Other misclassification problems were found for cancers of the uterus, brain, and buccal cavity including most of its sub-sites. Physicians tended to report a non-specific site of can cer on the death certificate rather than the specific site identified by the hospital diagnosis. (Am J Public Health 1981: 71:242-230.) Cancer mortality statistics have been part of many epi demiological investigations including a variety of etiological studies of cancer. Deaths from cancer have been used in geo graphic studies, studies of time trends, correlation studies, and therapy evaluation. They have also been used to identify cases for retrospective evaluation of possible etiological fac tors. Because of the varied uses of mortality data, it is impor tant that they be reliable and accurate. In the past*, a number of authors have pointed out the inaccuracies of cancer death certificates1 1 by comparing the specified underlying cause of death to autopsy diagnosis and to more specific hospital and pathologic information. Most of these studies have been of limited scope and dealt only with a small series of cases. To date, the only large-scale studies on the accuracy of cancer death certification are Dorn and Horn's 1941 study4 based on the First National Cancer Sur vey: Dorn and Cutler s 1938 study' based on the Second Na tional Cancer Survey: the Pan American Health Association study in 1967*-. and studies by the Atomic Bomb Casualty Commission7 " in Japan. No national assessment of the ac curacy of cancer mortality data in the United States has been undertaken in (he last 20 years. The study to be reported compares the underlying cause of death from cancer with the hospital cancer diagnosis of persons in specific areas of the United States. In order to be included in the analyses to be presented, a patient must have had a medical record which indicated that a diagnosis of one and only one cancer had been made and must have died with cancer coded as (he un derlying cause of death on the death certificate. Further more. the patient must have been a resident of the areas in volved m the study. From the Biometry Branch. DCCP. NCI. NIH. Address repniM requests to Constance Percy. Biometry Branch. National Cancer Insinute. National Institutes of Health. Bcthesda. MD 20203. This pa per. submilled lo the Journal October 26. 1979. was revised and ac cepted for publication October 2S. 1900 Editor's Note: See related editorial p. 2)1. this issue. 1 This article was first published in tho American Journal of Public Health 71:242-250.1981. It is reproduced here with Use per mission of the publisher. Materials and Methods The data on deaths used in this study were drawn from the Third National Cancer Survey (TNCS), which was con ducted in two states and seven metropolitan areas during the years 1969-1971. The survey reviewed hospital records, in cluding autopsy and surgical pathology reports, for all active cancers, both incident and prevalent, in all area hospitals. Over 90 per cent of the TNCS cases were microscopically proven. This data set differs from that reported in Mono graph 41* which dealt mainly with incident cases. Prevalent cases were included in this study in addition to the subset of the incident cases reported in this monograph.* Death certificates with any mention of cancer were rou tinely collected from vital statistics sections of state health departments during the study period. However, not all cop ies of certificates received by TNCS included an Inter national Classification of Diseases Adapted (ICDA-8)'* code for the underlying cause of death. Sometimes copies of the certificates were reguested by TNCS staff before the coding was done in the state health department or copies were sent without the underlying cause code appearing on the copies of the certificate. It would have been possible for the authors to code the underlying cause of death directly from the death certificate when the ICDA-8 code was missing. However, the coding rules for 1CD-8*1 arc complicated and codes are usually assigned by a trained nostotogist. To prevent possi ble bias, no coding of the death certificates was attempted by (he authors. As a result, deaths whose certificates lacked the ICDA-8 codes for cause of death had to be excluded from the analysis. `Prevalent cases were defined in the TNCS study as having ac tive cancer during 1969-1971 but having been first diagnosed pre vious to 1969 UCC 096978 467 PERCY. ET Al in 1970. TNCS staff began abstracting from the death certificates the underlying cause of death coded according to the ICDA-8.1" Death certificates must be obtained from the state health departments because the National Center for Health Statistics INCHS) receives certificates without any personal identification. NCHS recodes all death certificates sent from the states but both the state and NCHS use the same classification and coding schemes, so the results of this study should be applicable to both U.S. mortality data (NCHS) and state mortality data. If a hospital abstract was not on file for a death certifi cate with an underlying cause of cancer, an abstract of the medical record was obtained. Only 2.1 per cent of the cases in TNCS were "death certificate only" cases, i.e.. the death certificate was the only evidence that the person had cancer. These cases were not included in this study since there was no corresponding hospital record. When an autopsy protocol indicated cancer, the death certificate was obtained and the autopsy diagnosis was considered as the hospital diagnosis. Hospital diagnoses of cancer were abstracted and coded for site and histologic type according to the Manual of Tumor Nomenclature and Coding (MOTNAC).1-' To facilitate com parison. a computer program was written converting the MOTNAC codes to ICDA-8 categories.1' Since this con version consists of converting a detailed code involving both site and histologic type to the less specific categories of ICDA-8 which is based mainly on site, it is highly reliable for the site involved. In this study, malignant neoplasms are de fined as ICDA-8 categories 140-207. For a death to be included in this study, the copy of the death certificate received at the National Cancer Institute (NCI) must have an ICDA-8 cancer category coded for the underlying cause of death and it must have a hospital diagno sis of cancer. When a person had two or more cancers, it was difficult to determine which cancer caused the death. There fore. patients with multiple primaries were excluded. There were 48.826 deaths among residents during 1970 and 1971 in the TNCS areas for persons with only a single primary cancer and an underlying cause of death category of cancer coded on the death certificate. Excluded from these 48.826 deaths were those from the Minneapolis-St. Paul area which did not start recording the underlying cause of deal categories until 1971. Also excluded were less than 100 deaths from malignant carcinoids (category 258) that were not considered to be malignant neoplasms by ICDA-8. Of these 48.826 patients. 60 per cent were incident (diagnosed m 1970 or 1971) and 40 per cent prevalent (diagnosed before 1970) cases. To assess the representativeness of the cancer deaths in this study, a comparison of these 48.826 cancer deaths w,(s made to those deaths repuned by the NCHS for the survey areas. The 48.826 cancer death certificates from TNCS com prised 80.6 per cent of those reported as cancer deaths in the TNCS areas by NCHS in 1970 and 1971. In 1970. 70 per cent of the NCHS deaths were in our study and in 1971. 90 pet cent were included. The discrepancy results from excluding patients with multiple primaries, death certificate only cases, and certificates received without an ICDA-8 category for the underlying cause of death. Overall, the most common cause of exclusion was the lack of an ICDA-8 cause of death cate gory coded on the certificate. After the first year of the study, a special effort was made to get this item from the health departments (hence the difference between the 1970 and 1971 percentages). Cross classifications of hospital diag noses with underlying cause of death hy site were examined for the individual years. 1970 and 1971. and found to be very similar. Therefore, it was decided to combine the data for these years. The survey deaths were cross-classified by sex. race (White. non-White), and age (0-49. 50-69. 70+), and subsequently compared to NCHS deaths (Table I). Although there were slight differences in percentages, none of ttfl comparisons suggested serious biases in the data. The accuracy of the death certificate code was assessed by comparing the primary cancer site reported on the hospi tal diagnosis with the cancer site coded as the underlying cause of death on the death certificate (Table 2). This was done at a 3-digit level of the ICDA-8 classification10 for most sites. Most mortality data are reported by similar groupings The basis of the analysis is a cross-tabulation of the hospital diagnosis (vertical axis) with the underlying cause of death (horizontal axis). The diagonal of the cross-tabulation repre sents cases whose hospital diagnosis and underlying cause of 46b TABLE 1--Number of Cancer Oaatlw Reported to tho Notional Cantor tor HooMh StattaOca (NCHS) tor TNCS Arooo* and Naroontogo of Such Oaotha Inctudod In thta Study by Ago, Sax, and Raco (1976 + 1971) No ot NCHS Cancar DaaOW by Aga % at NCHS Canear Onn mcmm m *xa Study by *oa Raea anoSai White Mates Females Non-white Mates Fsmales Total 0-49 6.552 3.142 3.410 1.269 614 655 7.821 50-60 23.983 13.400 10.583 3.696 2.202 1.494 27.679 70- 23.022 12.462 10.560 2.060 1.241 828 25.091 AM AOM 53,557 29.004 24.553 7,034 4.057 2.977 60.591 0-40 77.6 799 75.5 805 80.6 802 78.1 50-60 79 5 800 787 80.4 81 6 786 79 6 70* 81.8 801 840 89 0 87 6 91.1 825 AS 803 80.0 80.6 62.9 83.3 82.4 806 `Mnnaapofta-Sl Paul *>* andudad NATIONAL CANCER INSTITUTE MIWV-.uaou mq 59 UCC 096979  TA8LE M*P,UI W^naM* wtth Uml#rly`n9 C.UM ol Oeath toy Cancer Site* (ICOA-B: 140-207) tor Cancer Deathe Iron, the Third National Cancer Sur- ACCURACY OF CANCER DEATH CERTIFICATES tBaa ICOA-e (rat. to} (or i PERCY. ET AL death were identical. This agreement was measured by two different rates, detection and confirmation: The detection rate for a specific site was defined as the number of cases diagnosed as cancer of that site in the hospi tal and having cancer of the same site on the death certificate divided by the total number of persons diagnosed with that specific site of cancer in the hospital and dying of cancer. It is. therefore, the proportion of hospital diagnoses with can cer of a certain site in which the cause of death reflects the same hospital diagnosis. The confirmation rate is the same numerator divided by the number of persons who died with this particular site as the underlying cause of death and had had previous diag nosis of cancer. It is therefore the proportion of cancer deaths in which the specified underlying cause is confirmed by the hospital diagnosis. Selected sites of cancer were classified into four groups according to the relative and absolute values of their detec tion and confirmation rales as follows: Group 1: Both high detection and confirmation rates at about the same level lover SO per cent). This means that there was good agreement between the primary site diag nosed in the hospital and that recorded on the death certifi cate. Mortality rates for these sites should be fairly accurate. The majority of the deaths were found in this group. Group 2: Both low detection and confirmation rates at about the same level (under SO per cent). In this group, there was considerable disagreement between hospital diagnoses and the corresponding underlying causes reported on the death certificates. Nevertheless, since both the detection and confirmation rates were equally low, the number diag nosed in the hospital and the number of deaths reported on the death certificates were about the same (columns I and 2 of Table 3). Such rates were observed usually for sites of low frequency. For cancer sites in this group the proportionate mortality would remain the same. Group 3: Detection rate higher than confirmation rate. The sites in this group are characterized by a greater number of deaths reported on death certificates for the specific site than reported by the hospital diagnosis; this means that these sites will be overreported in the subsequent mortality statis tics. This occurs for a site such as booe where metastasis from other primary sites frequently occurs. Unless the certi fier specifies bone as a metastatic site, it will be coded as the primary site. Group 4: Confirmation rate higher than detection rate. In this group more cases of a site were diagnosed in the hos pital than were actually reported on the death certificates. This resulted in an underreporting on death certificates. Results Frequencies and detection and confirmation rates are presented in the Tables for the 48,826 cancer deaths. The detection and confirmation rates are given with their stan dard errors for each site to aid researchers in evaluating the accuracy of cancer mortality statistics for these sites. 470 As mentioned above, Table 2 cross-classifies the hospi tal diagnoses of the 48,826 cancer deaths with the underlying cause of death on the certificate. Overall, when the 30-sue groupings shown in this Table were used. 86,7 per cent had the same site reported on the hospital diagnosis as that in dicated as the underlying cause of death on the certificate. However, if the number of groups (3-digit categories) is in creased to 49 as in Table 3. the overall agreement decreases to 82.7 per cent. Naturally, if 4-digit categories are used (Ta bles 5-7), agreement decreases even more. Table 3 presents the data of Table 2 in a different format and includes the detection and confirmation rates and their standard errors. The figures and rates are given for each 3digit primary site and a few combinations of sites. Table 4 identifies the principal sites which fell into the four groups described earlier. In this study 65 per cent of the total cancer deaths belong in Group I. A similar percentage (64) of United States cancer deaths in 1970 were attributed to cancers of the sites included in Group 1. Extremely high de tection and confirmation rates of over 93 per cent were found for cancers of the lung and bronchus, breast, prostate, and multiple myeloma. Connective tissue tumors are a good example of Group 2. Although there was much misdassification between the hospital diagnosis and the underlying cause of death, about the same number (259 cases) were reported in the hospital as were found for the underlying cause on the death certificate^ (252 cases). However, as seen in Table 3, column 3. only I4^P connective tissue cancers were diagnosed and confirmed on death certificates, a detection rate of 54.8 per cent and a con firmation rate of 56.3 per cent. Very few cancer deaths are in Group 2. The category "malignant neoplasm of the bone" is a good example' of Group 5 (detection rate higher than con firmation rate). Table 3 shows that 160 bone cancers were diagnosed in the hospital but many more (252) were stated as the underlying cause of death. Since only 125 of these were confirmed in the hospital, the confirmation rate was only 49.6 per cent but the detection rate was 78.1 per cent. An examination of those deaths coded to primary cancer of bone on the death certificate showed diagnoses in the hospital of primary cancers of many other sites. No doubt bone was actually a secondary site of these cases and thus they were misclassified. Therefore mortality figures for bone cancer are considerably overreported in vital statistics data. Colon can cer, one of the principal sites of cancer, falls into Group 3 Group 4 sites have higher confirmation rates than detec tion rates. For example, more buccal cavity cancers were diagnosed in the hospital (1,397 cases), than were reported on death certificates (1.187 cases). Since 1.098 deaths were confirmed by the hospital diagnosis, the confirmation rate (92.5 per cent) was much higher than the detection rate (78 6 per cent). Malignancies of the rectum, a frequent site of can cer. fall into this group as well. For many sites, a specific diagnosis was made in the hospital but only a non-specific site was stated on the death certificate. Table 5 shows a detailed comparison of cancer of the cervix uteri, corpus uteri and uterus. NOS (not otherwise specified). Many more deaths were reported for cancer ucc 096981 ACCURACY OF CANCER DEATH CERTIFICATES TABLE 3--Number of Cape* Diagnosed in Hospital and Humber with Underlying Causa of Death by Sit* and Number with Sam* Sit* on both. Cancer Deaths from Third National Cancer Survey, 1970-1971 Detection and Confirmation Rate* ICDA-8 Category Primary Sice No. diagnosed No. *Uh this in hospital site 4s the with this site cause of death 140-207 All Sites 46,826 48,826 140-149 140 141 142 143 144 145 146 147 148 149 Buccal cavity Lip Tongue Salivary gland Gum Floor of mouth Mouth. NOS Qropnaryn* Nasopharynx Hypopharynx Pharynx, NOS 1.397 32 324 90 54 135 176 220 123 174 69 1,187 14 267 67 23 79 177 149 104 111 196 150 151 152 153-154 153 154 Esophagus Stomach Small intestine Colon & Rectum Colon Dec tun 921 2.34S 113 6.644 4.546 2,098 997 2,321 109 6.498 5.131 1.367 155 Liver 6 Intrahep. bile duct 536 156 oal1 bladder 727 157 Pancreas 2.496 158 Retroperitpneum 108 159 Peritoneum 134 3*7 674 2.531 102 79 ICO Nose, ear 2 sinuses 161 Larynx 162 . Trachea, lung 1 bronchus 163 Pleura, med. 6 other resp. 93 395 10.059 76 71 433 10.178 102 170 Gone 171 Connective tissue 172 Melanoma of skin 173 Other skin 174 breast 160 259 502 51 4.734 252 252 486 67 4,583 180 182 183 121*134 lcS 186 137 138 139 190 191-192 191 192 193-194 193 194 195-199 Cervix Corpus 6 uterus, NOS Ovary, f. tube, etc. Other female genital Prostate Testis Other male genital Bladder Kidney f ye Brain i other new. Brain Other nervous irltM Thyroid i other eneo. thyroid Other endocrine III defined 1 ueknoun 995 674 1,497 232 2,621 136 so 1,211 984 71 1.U74 965 109 206 I4t 57 2.320 869 768 1.499 138 2.579 125 46 1.179 930 43 1.171 919 252 205 142 61 2.761 200*202 201 203 204-207 204 205 206 207 Non-Hodgktn'i lympnom* Hodgkin's disease Multiple eyelome Leukeeias Lymphocytic Myeloid Monocytic Other 6 unspecified 1,562 572 69t 2,152 743 1,107 96 204 1.470 516 688 2,140 188 914 104 *34 same site on both 40,379 1.098 1* 219 56 14 51 83 115 82 74 45 G58 2,109 75 6,171 4.062 1,180 266 583 2.252 45 20 53 317 9.560 36 125 142 *40 18 4.498 786 549 1,322 117 2.483 113 41 1.103 865 35 1.044 03 79 169 IX X 1.231 1.300 496 675 2.069 594 43 58 149 Oetection Rate Standard % error 82.7 0.17 78.6 43.8 67.6 62.2 25.9 37.8 47.2 52,3 66.7 42.5 65.2 1.10 8.77 2.60 5.11 5.96 4.17 3.76 3.37 4.25 3.75 5.73 93.2 89.2 66.4 92.9 89.4 56.2 0.83 0.64 4.44 0.32 0.46 1.08 *9.6 80.2 90.2 41.7 14.9 2.16 1.48 0.59 4.74 3.08 57.0 00.3 95.0 47.4 5.13 2.00 0.22 5.73 78.1 54.8 87.6 35.3 95.0 3.77 1.09 1.47 6.69 0.32 79.0 81.5 88.3 50.4 94.7 81.1 82.0 91.1 87.9 49.1 97.2 11.2 72.5 2.0 87.2 68.4 51.1 1.29 1.50 0.03 3.X 0.44 3.21 5.4] 0.82 1.04 5.93 0.50 1.20 4.28 2.70 2.73 6.18 1.00 83.2 6.7 96.6 96.1 79.9 76.2 57.1 71.0 0.90 1.42 0.89 0.40 1.47 1.28 $.00 3.11 Confirmation Rate Standard t error 82.7 0.17 92.5 100.0 82.0 83.6 60.9 64.6 46.9 77.2 78.8 66.7 23.0 0.76 2.35 4.53 10.18 5,38 3.75 3.44 4.01 4.47 3.00 86.1 90.9 68.8 95.0 79.2 86.3 1.10 0.60 4.44 0.27 0.57 0.93 76.7 86.5 89.0 44.1 25.3 2.27 1.32 0.62 4.92 4.89 74.6 73.2 93.9 X.3 5.16 2.13 0.2*. 4.73 *9.6 56.3 90.5 26.9 98.1 3.15 3.12 1.33 5.42 0.20 90.4 71.5 88.2 4.8 96.3 90.4 9.1 93.6 93.0 1.00 1.63 0.83 3.10 0.73 2.64 4.5* 0.72 0.8* 1.4 9.2 7.4 11.1 82.4 91.5 61.9 44.6 W.4 92.5 98.1 $.93 0.90 1.10 2.rz 2.70 2.33 6.12 0.90 0.20 1.14 0.52 96.7 18.1 92.2 53.8 14.1 0.40 1.11 0.88 4.89 2.28 of (he uterus. NOS (cuiegory 182.9) on (he death certificates than were so diagnosed in the hospital. Of the 470 cases hav ing cancer of the uterus. NOS on the death certificate. 103 were diagnosed m the hospital as cervical cancer and 184 as corpus cancer. In contrast, only 89 were diagnosed cancer of the uterus. NOS in the hospital, a confirmation rate of only 18.9 per cent. Table 6 shows a comparison of the 4-digit ICDA-8 cate gories for cancer of the colon and rectum. The non-specific site of cancer of the colon. NOS (category 133.8) was found CANCER MORTALITY IN THE U.S., 1990-1977 471 UCC 096982 PERCY. Et AL TABLE 4--Groupings of Certain Cancer Silas According to the Absolute and Ratetiva Values of the Detection and Confirmation Rates* (Cancer Deaths from Third National Cancer Survey, 1970-1971) Group 1 Both rales high Stomach (151) Pancreas (157) Bronchus and Lung (162) Melanoma ol skin (172) Breast (174) Ovary (183 0) Prostate (185) Bladder (188) Thyroid (193) Multiple myeloma (203) Group 2 Both rales low Mouth, NOS (145) Small intestine (152) Connective tissue (171) Group 3 Deledion rate higher then Confirmation Rate (overreporting on death certificates) Colon (153) Larynx (161) Bone (170) Uterus, NOS <182 9) Pharynx, NOS (149) III defined and Unknown sites (195-199) Group 4 Ccnhrmeson Rate higher then Detection Rate (underreporting on death cervficatet) Buccal cavity (140-149) Rectum (154) Cervix (180) Corpus (182.0) Eye (190) Myeloid leukemia (205) Transverse colon (153.1) Sigmoa] colon (153.3) `See Table 3 for rates. NOTE ICDA-8 (ref. 10) categories given in perenthesee. as the underlying cause of death on 3,299 certificates, but only 354 cases were classified in this category from the hos pital diagnosis, a detection rate of 76.0 per cent but a con firmation rate of only 12.8 per cent (Group 3). Of these 3,299 colon cancers (category 133.8), 468 were diagnosed in the hospital as cancer of the transverse colon. 749 as sigmoid colon. 266 as recto-sigmoid, etc. As can be seen from this Table, a specific site was diagnosed in the hospital, but the physician did not record this specific site as (he underlying cause on the death certificate. When ail the sites of the colon (category 153) are combined, the detection rate is 89 per cent and the confirmation rate is 79 per cent, putting this site in Group 3. Cancer of the rectum, on the other hand, falls into Group 4 because the confirmation rate (86 per cent) was much higher than the detection rate (36 per cent). Thus can cer of the colon was overreported on death certificates and cancer of the rectum was underreported. If cancer of the co lon and rectum are combined, the detection rate is 93 per cent and the confirmation rate is 93 per cent, placing the combination in Group 1. ICDA-8 has several categories for coding cancers of the liver: 155.0-- primary liver cancers, such as hepatocellular carcinoma: 197.8--malignant neoplasms ot tnc liver, not specified primary or secondary; and 197.7--liver cancer specified as secondary. Table 7 gives the comparison and rates for the various liver categories. Of the 536 cases diag nosed as primary liver cancer in the hospital, only 347 cases had an underlying cause of death category of 155.0 (primary malignant neoplasm of liver) on the death certificate, a detec tion rate of 49.6 per cent. However, in addition there were 375 cases with a cause of death category of 197.8 (malignanM^^ neoplasm of liver, unspecified). By adding the deaths frort^^F category 155.0 and category 197.8 together, the detection rate rose to 79.1 per cent as shown in Table 7. If only the category 155.0 were used, there would be gross underre porting of malignant neoplasms of liver on death certificates. Therefore, a more accurate total of liver malignancy deaths was obtained by combining ICDA-8 categories 155.0 and 197.8 together. Myeloid leukemia and lymphocytic leukemia were un derreported on death certificates (Table 3). Again the diagno sis was specific in the hospital but the physician just wrote "leukemia" (category 207) on the death certificate. Over twice as many cases were in this unspecified category on the TABLE 5--Compaireon of Hospital Otegnosla with Underlying Cauee of Death: Cancor Daatha from tha Third National Cancor Survey, 1970-1971, Dataction and Confirmation Rates lor Malignant Nsopissms of Specific sues of the Uterus Hospital Diagnosis icoA-a Category Malignant Neoplasm ot Underlymg Cause ol Death _____________ _____________ Cams 180 Corpus 1820 Ut*ru* NOSt 1829 Total no (fcagnosed m hosp ot tna site' 180 182.0 182 9 Cervix Ulan Corpus Uteri Uterus, NOSf Total no with this underlyi ng causa* 786 23 ~1 2 8 103 2M___ 184 to 1_____ s_ 869 296 470 995 547 127 `Figures lor Other sites are netudad *i tha total but not shtewi m the Table TNot Otherwise Specified Deieelon Rate Per Cent Standard Error 79 0 48 4 70 1 1 29 2 14 406 Confirmation Rate Per Cera Standard Error 904 895 189 1 00 1 78 1 81 UCC 09690 472 NATIONAL CANCER INSTITUTE MONOGRAPH NO 59 ACCURACY OF CANCER DEATH CERTIFICATES TABLE 6--Comparison of Hospital Diagnosis with Underlying Causa Of Doath: Cancar Oaaths from ths Third National Cwicar Survay, 1970-1971, Dataction and Confirmation Ratos for Malignant Nsopiasms of Colon and Rsctum ICDA-8 Malignant Catagoty Neoplasm pi Underlying Causa ol Death 1530 153 1 153.2 153 3 153.8 153.9 154 0 154 1 154 2 Total no.* diagnosed n hospital ol this sits Detection Rate Far Standard Cent Error Confirmation Rate Per Standard Cant Error 1530 153.1 153.2 1533 Cecum, ascending colon & append.* Tianaverse colon Descending colon Sigmoid colon 153 8 NOSt 153.9 154.0 154.1 154 2 Intestate, NOSt sigmoid Redum Anus. sic. Total no. with this underlying cates' 551 8 18 144 42 13 1 13 3 4 13 16 8 51 23 33 473 8 13 610 25 468 22 280 5 749 47 421 19 1 2 2 37 1 8 2 8 62 15 1,311 723 394 1.519 554 --- 2 10 25 - - 48 2. 111 80 288 16 j 127 827 1 Li__2 1 7 39 334 25 1 41 a ?T?T 5 1,438 622 184 133 868 X399 248 188 1.189 10 42 0 19.9 12.9 31.1 76.0 1.36 1.49 1.69 MO 1.61 886 87 8 383 71.1 12.8 1.28 2.56 375 1.76 0.58 58.6 13.7 83.1 7.41 1.37 1.27 10.1 51.2 76.2 80.0 1.91 3.88 1.24 12.68 `Figures lor other sees are nduded in total but not shown in the table. fNot Otherwise Specified death certificate as were diagnosed in the hospital. A final observation concerns deaths and diagnoses in ICDA-8 categories 195-199, cancers of ill-defined and un known sites. The detection rate was 53.1 per cent and the confirmation rate was 44.6 per cent. As seen in the last row of Table 2, many of the cases diagnosed as "unknown" in the hospital were coded to a specific site on the death certifi cate. Large frequencies were found for common metastatic sites such as lung, liver, and bone. Conversely, when we look in the other direction (at the column) for this category in Table 2, we note that over one-half of the cases with un known or ill-defined site on the death certificate had a hospi tal diagnosis of cancer of a specific site. Discussion and Summary Variability and biases in cancer mortality data can dis tort, limit, or inhibit the value of this important resource in epidemiologic studies. The data in this study have been used to pinpoint some of these problems. The reasons the absolute numbers of cancer deaths in TNCS did not correspond to those reported by NCHS have been previously presented. To ensure that this discrepancy in numbers did not bias this study, a comparison was made of the site distribution of cancers (ICDA-8 categories 140207) reported by NCHS14 for the total U.S. in 1970-71 to the TNCS cancer deaths in 1970-71 used in this study. Table 8 TABLE 7 --Comparison or Hospital Diagnosis with Underlying Cause at Death: Cancer Oenlhe horn the Third HeMonel Cancer Survey, 1970-1971, Detection and Confirmation Helen of Mefignent Neoplasms eI the User Hospital Diagnos* UndwVng Cause of Dean Prtmory Uwr Uwr. NOSt anal Stosndvy Tout Leer User Unknown 899 Tow no. Aaptoatd to hoipm pi Wtete* Data (dun Raw ICDA-8 Catagoty 155-0 Uvet, primary 155 0 229 197.8 14* 155.0 f 197.8 377 197.7 14 195-199 29 462 156.0 155.0 4 197.8 Standard Standard Par Cant Error Per Cant Error 49.8 2.33 81.8 1.80 Total no. with this underlying cause* Confirmation Rate 315 72.7 ^ 2.51 375 39 5 2.52 722 52.2 1.88 213 `Figures for other met are inducted n tin* total but not shown m tite Tatte. "Eckjdmg Mas 197 7 and 197 8 tNot Otitenmaa Spooled. CANCER MORTALITY IN THE U.S., 1950-1977 473 ucc 096984 TABLE 8--Comparison o( Percentage* of U.S. Cancar Oaatht and TNCS Cancer Deaths by Death Certificate and Hospital Diagnosis for Selected Sites of Cancer, 1970-1971 ICOA Category 140-149 150 151 153.1 153.2 153.3 153.8 154 160-163 174 180 183 165 Site Term Buccal cavity Esophagus Stomach Transverse coton Descending colon Sigmoid colon Colon, NOS Rectum Lung and other respiratory Breast Cervix Ovary Prostate total cancer death* mUS (14) 2.3 18 4.8 0.4 03 1.5 6.7 3.1 21.4 9.0 2-0 3.0 5.3 v % of TNCS Cancer deaths by cause of death on certificates 24 2.0 4.8 0.3 0.3 1.4 6.8 2.8 22.1 9.3 1.8 3.1 5.3 % ot TNCS Cancer deaths by hospital diagnosis 29 19 48 15 08 3.1 11 4.3 21 8 94 2.0 31 5,4 displays these data for selected sites comprising over 60 per cent of the cancer deaths, showing that there is very good agreement between the distribution of U.S. cancer deaths (column 1) and TNCS cancer deaths (column 2) by site. What would happen to cancer mortality statistics if the death certificates were reviewed and the underlying causes changed to correspond to the site of cancer as reported in the hospital? The absolute value of the mortality rate would not change much for any particular site since the number of deaths is small in comparison to the total population of the areas. However, the percentage change is dependent on the site. Sites which fall into Groups 3 and 4 would have sub stantial percentage changes in the mortality rate. This would be important in any time trend analysis. Similarly, the pro portionate cancer mortality ratio due to a particular site of cancer would show little change (Table 8, columns 2 and 3) but the percentage change based on the number of deaths of a particular site would be large for sites in Group 3 and 4. For example, if the hospital diagnosis was used instead of the cause recorded on the death certificate, the number of trans verse colon deaths would increase from 164 to 723 (see Table 6), an increase of 340.9 per cent (723-164VI64. Similar calcu lations for descending colon, sigmoid colon, colon, NOS, rectum, and buccal cavity yield changes of 196.2 per cent, 128.4 per cent. - 81.9 per cent, 33.3 per cent, and 17.7 per cent, respectively. This illustrates the fact that mortality trends may show a per cent change due to an artifact caused by physicians changing their method of recording the diag noses when they fill out the death certificate. Death certificates are frequently used as a means of case ascertainment for retrospective studies. Cases identified for sites with low confirmation rates would be subject to pos sible bias, as many cases identified by the death certificates for a particular site would not have that site as the hospital diagnosis. For example, if bone cancer cases were selected from the death certificates, over one-half would not be bone cancers according to the hospital record. Another bias is that if the detection rate is low. cases pulled from the death cer tificates for a particular site will not represent all the cases diagnosed of that site. For example, over one-half the deaths due to malignant neoplasms of eye (category 190) would be missed if only death certificates were used because the cer tificate often says only "melanoma" with no mention of eye ; these certificates are coded to category 172 (melanoma of skin) and not to eye. There are a few additional factors besides the site of the neoplasm that may affect the accuracy of death certificates, such as the presence of an autopsy, evidence of microscopic confirmation, age at death, sex, race and geographic area. Although mechanisms exist for amending the original death certificates after autopsies are completed, this is rarely done. As a result, the performance of autopsies does not routinely result in improvements in death certificate accuracy. There was a slight variation in overall accuracy from area to area but no substantial differences were observed except for a few sites of low frequency. The overall agreement of the oth er factors mentioned, i.e. age, sex. and race, did not cause a difference of more than one or two per cent from the 86.7 per cent overall agreement in Table 2. The place of death may influence the accuracy of the recorded cause of death. In this study 73 per cent of the deaths occurred in the hospital, 12 per cent in nursing homes, and 13 per cent at home. Certifiers are not always familiar with the indexing of ICDA-8 or the guidelines and instructions1' for filling out death certificates and do not realize how the order of entry of the terms which they record ultimately determines (he selec tion of the underlying cause of death by the coders. Even though, in some cases, the NCHS had expanded the rales1* of ICD-8, it was still difficult for coders to apply the rules in a uniform and consistent manner. The ICD-9,1' which went into effect in January 1979. has further expanded the rules and improved the index. Word choice is critical for determining the correct code. For example, when malignant brain tumors such as astrocytoma or glioma were diagnosed in the hospital, the physician frequently signed out the death certificate as "brain tumor." This cause of death is coded to category 238.1. neoplasm of unspecified nature of the bra and is therefore not recorded as a cancer death. Difference? NATIONAL CANCER INST,T1 "re ""wv-.bsPh NO 59 UCC 096985 ACCURACY OF CANCER DEATH CERTIFICATES in understanding between certifiers and coders of the mean ing of a term may introduce systematic biases to the cause of death assignments. The term "metastasis-" is often difficult to interpret on death certificates. For instance "metastatic lung cancer" can mean metastasis from a primary lung can cer or metastasis to the lung."* The combination of the detection and confirmation rates for each site should help determine the reliability of the fre quency of each particular site. Among the ten leading sites of cancer deaths.1' seven sites--lung, breast, prostate, pan creas. urinary bladder, ovary, and leukemia--fell into Group I. Thus, with both a high detection and confirmation rate, the mortality rates for these sites can be considered reliable. Furthermore these sites represent 65 per cent of all cancer deaths in ihis study. However, statistics on the remaining one-third of the cancer sites are not as accurate: Cancer of the colon and cancer of the rectum, together the second leading cause of cancer deaths (IK per cent), are different entities with separate characteristics. Colon cancer is overreported on death certificates and rectal cancer is un derreported. This crossover forces some statisticians to combine mortality figures of colon and rectal cancer in order to get a more accurate count of colorectal cancer. Unfortu nately. this obliterates the individual characteristics of each disease. Cancer of the uterus, the fifth leading cause of cancer deaths among females, has the same type of problem as colo rectal cancer. Cancers of the corpus uteri and cervix uteri are not the same disease and have different etiologies. They should be identified specifically on the death certificate and not be called simply "uterus". Buccal cavity cancer, the tenth leading cause of cancer deaths, and its components--cancers of lip, tongue, salivary gland, gum. etc.--are underreported on death certificates. If physicians understand how their reporting affects the classification of the underlying cause of death, they are likely to be more accurate in completing the death certificate. State vital statistics departments can and must play a role in mon itoring and encouraging accurate certification. Ensuring that autopsy diagnoses are reflected on death certificates would be a step in the right direction. The quality of cancer death certificate data will continue to be an important issue for cancer researchers, not simply to acknowledge but also to address. When using cancer mor tality data in studies of a specific cancer site, attention must be paid to the limitations as well as to the potentials of this data source. REFERENCES 1. Steer A. Land CE, Moriyama IM.c/ /: Accuracy of diagnosis of cancer among autopsy cases: JNIH-ABCC Population for Hiroshima and Nagasaki. Gann 1976: 67:625-612. 2. Monyama IM, Baum WS. Hacnszcl WM. rt ai. Inquiry into di agnostic evidence supporting medical certifications of death. Am J Public Health 1958. 48 1376-087. 3. James G. Patton Rh and Heslin AS: Accuracy of cause-of-dealh statements on death certificates. Pub Health Rep 1935; 70 39SI. 4. Dom HF and Horn JL: The reliability of certificates of deaths from cancer. Am J of Hygiene 1941: 34:12-23. 5. Dom HF and Cutler SJ: Morbidity from cancer in the United Stales Comparison of death certificates and case reports. Pub Health Mono 1958; 56:117-124. 206-207. 6. Puffer RR and Griffith GW Patterns of urban mortality. Pan American Health Organization. WHO, 1967. pg. 90. 7. Steer A. Moriyama IM and Shimizu K: The autopsy program and the life span study, Jan. 1951-Dec. 1970 ABCC-JNIH Pa thology Studies. Hiroshima and Nagasaki. Report 3. Technical Report 16*73. 8. Yamamoto T. Monyama IM. Asano M and Guralnick L: The autopsy program and tl)e life span study. Jan. 1961-Dec. 1975. Radiation Effects Research Foundation. Report 4. Technical Report RERF TR 18-78. 9. Cutler SJ and Young JL Jr teds): Third National Cancer Survey Incidence Data. NCI Monograph 41. DHEW Publication No. (NIH) 75-787. 1975. 10. U.S. Dept, of Health. Education, and Welfare. International Classification of Diseases. Adapted for use in the United States. Eighth Revision, PHS Publication No. 1693, 1967. 11. World Health Organization: International Classification of Dis eases. 1965 revision: Geneva: WHO, 1967. 12. Percy C. Berg JW and Thomas LB (eds): Manual of Tumor No menclature and Coding. 1968 Edition. New York: American Cancer Society. 1968. 13. American Cancer Society: Conversion Table No. 3. Conversion of Neoplasms by site and histology from Manual of Tumor No menclature and Coding, 1968 Edition to 8th Revision of the In ternational Classification of Diseases Adapted. ACS No. 0803. 1973. 14. Vital Statistics of United States, U-MorUlity, Part A. National Center for Health Statistics. US DHEW. 1970, 1971. 15. U.S. Dept, of Health, Education, and Welfare: Physicians' handbook on medical certification: death, birth, fetal death. Pub. No. (PHS) 78-1108. 16. Vital statistics: Instructions for classifying the underlying cause of death. 1976-77. National Center for Health Statistics. Part 2a. DHEW. 1975. 17. World Health Organization: International Classification of Dis eases. 1975 revision: Geneva: WHO. 1977. 18. Percy C and Oolman A: Comparison of the coding of death cer tificates related to cancer in seven countries. Pub Health Rep 1978; 93:335-350. 19. American Cancer Society: Cancer Facts and Figures, New York. ACS. 1980. ACKNOWLEDGMENTS The authors wish to thank Valeric Van Holten for the extensive programming of the MOTNAOICDA-8 conversion. Without this program, this study would not have been possible. Also the authors thank William McKay of the National Cancer Institute for supplying the necessary figures for Table I that he adapted from U S. mortality tapes supplied by NCHS. The authors appreciate the comments of Dr. Susan Dcvcsa. and the technical assistance of Carol Kalb and other members of the Demographic Analysis Section of the Biome try Branch of the National Cancer Institute. An earlier version of this paper was presented at the American Public Health Association Annual Meeting on November 7. 1979 in New York City. CANCER MORTALITY IN THE U S., 1960-1977 UCC 096986 . u.s, xvCTMprr ruiFCPG wmer i-h? : - jw-ns TEXASHealth Service Areas for the State of i pt Amarillo Trmm 1 1 1"an >L/ Lul >bo<:k* 4 U r* r * Abilene -pffjr Dallas* 'Tyler Odessa 4 r"`Waco 10 3 *** ' an Ai itonio"> 11 Port Arthur \*.Galv st n | `.4 JJJI'I. U El Paso ( mi e TtlAj Ucc 096987 Texas Department of Health 1100 West 49th Street Austin, Texas 78756 Robert Bernstein, M.D. Commissioner of Health 512-458-7375 Robert MacLean, M.D. Deputy Commissioner for Professional Services 512-458-7378 W. D. Carroll Bureau Chief Bureau of Vital Statistics 512-458-7692 William Barrington Director of Statistical Services Bureau of Vital Statistics 512-458-7362 Charles Alexander, M.D. Bureau Chief Bureau of Epidemiology 512-458-7207 Amelia Menchetti Director, Statewide Cancer Program Registry Bureau of Epidemiology 512-458-7265 Jeanne Martin, Ph.D. Data Management Bureau of Health Planning and Resource Development 512-458-6580 Lucina Suarez, M.S. Leslie Boss, Ph.D. Bureau of Epidemiology 512-458-7207