Cancer deaths in India: is the model-based approach valid?

Brief communication Cancer deaths in India: is the model-based approach valid? P.C. Gupta,1 R. Sankaranarayanan,2 & J. Ferlay3 The model-based cancer mortality estimates for India (775 800) are nearly double the data-based esti- mates (433 000), and are higher than even the incidence estimates (612 300). The model-based approach is therefore, at least in the case of India, incorrect. Established practice is to use real data to validate theoretical models, not to reject the data if apparently the model does not fit. The problem In their recent article on global and regional cause- of-death pattems in 1990, Murray & Lopez estimated the total cancer deaths for India as 775 800 (1). Based on the World Bank estimates of age/sex- specific total deaths and linear regression modelling procedures, they divided the total number first into three broad categories of causes using adjustment procedures for arriving at corrected totals, and then into specific disease groups, one of which was can- cer. For India, they used the Survey of Causes of Death (8) to check the validity of their division into the three broad categories, adjusting if the estimates differed by more than 10%. They then used cancer registry data to generate site-specific numbers from total cancer deaths, but preferred indirect estimates of cancer mortality, suggested by the models, rather than estimates based on cancer incidence because "the level of underreporting by registries in some areas (e.g., India) seemed... to be sufficiently high to invalidate this approach" (1, p. 456). In India there exists a network of several well- established population-based cancer registries pro- viding reasonable incidence data for many years. The Cancer Registry in Bombay, for example, has operat- ed for 30 years and the registries in other metropoli- tan areas for over 10 years. Their data have been included in the well-known volumes on Cancer inci- dence in five continents. Regular reports have been I Visiting Scientist, International Agency for Research on Can- cer, Lyon, France. 2 Scientist, International Agency for Research on Cancer, 150 Cours Albert-Thomas, 69372 Lyon Cedex 08, France. Requests for reprints should be sent to this author. 3 Assistant (Statistics), International Agency for Research on Cancer, Lyon, France. Reprint No. 5552 published by the individual registries, with details of registration methods and reliability indices including two research papers (2, 3). Combined reports have been published by the Indian Council of Medical Research. Murray & Lopez have not provided any reason for their impression of incompleteness, nor any quantification of it. We have compared the Indian data with that from other countries and have not found any evidence of serious underreporting. Discussion Cancer registry data were used for estimating the worldwide cancer incidence (4) and mortality (5) in 1985. In order to produce a national estimate, the age-sex-site-specific cancer incidence rates from the three longest-functioning cancer registries in metro- politan areas in India (Bombay, Bangalore, Madras) and the only rural registry (Barshi) were used with equal weightage (0.25) for each. This may have resulted in some overestimation, as the rates for Bar- shi were the lowest and 70% of the Indian population is rural. These estimates can easily be updated to 1990 by taking into account the 5-year trends in inci- dence for each site and age group from Coleman et al. (6) and population changes. The results for all sites combined and some selected sites among men and women are displayed in the Table along with the percentage of "underreporting" if the estimates from the models are taken as correct. It can be seen that the model-based mortality estimates are about twice as high as the data-based estimates, and are higher than even the incidence estimates. Looking at some specific sites the differ- ences seem even more inexplicable; for example, the incidence and mortality estimates for mouth/pharynx cancers in India are already one of the highest in the world. They are comparatively easily diagnosed can- cers and it would be rather ludicrous to suggest that Bulletin of the World Health Organization, 1994, 72 (6): 943-944 © World Health Organization 1994 943 P.C. Gupta et al. Table 1: Incidence and mortality estimates (in thousands) for cancer in India during 1990 Incidence Mortalit Percentage (Gupta (Gupta (Murray "under- et al.) et al.) & Lopez) reporting" All sites 612.3 433.0 775.8 44 Male 290.6 229.5 450.2 49 Female 321.7 203.5 325.6 38 Mouth/pharynx 90.1 66.5 157.1 58 Male 61.3 44.4 108.0 59 Female 28.8 22.1 49.1 55 Oesophagus 42.1 40.6 83.2 51 Male 24.4 23.7 51.5 54 Female 17.7 16.9 31.7 47 Colorectal 27.1 18.5 39.5 53 Male 15.1 10.4 22.7 54 Female 12.0 8.1 16.8 52 Lung 30.8 27.4 55.3 50 Male 26.0 23.2 46.4 50 Female 4.8 4.2 7.9 47 Breast 58.3 30.0 40.7 26 Cervix 95.5 43.5 63.0 31 there is nearly 60% underreporting. The model pre- dicts consistently higher underreporting for men rather than women and the smallest underreporting for female breast and cervix cancers. If there were some underreporting, one would expect it to be the other way around. Using different procedures, the National Cancer Registry Programme of India esti- mated the total number of cases (incidence) in 1990 as 629 000 (7), about 2.7% higher than our estimate (612 300). It seems clear that the model-based approach is, at least in the case of India, incorrect. The justifica- tion for using it is presumably to ensure that the pro- portionate distribution of deaths by cause is similar to that observed in other countries with a similar total (all causes) mortality. Yet there is no reason to suppose that this is necessarily so; it might be, for example, that mortality from cardiovascular disease is more common in India than in other countries at an equivalent level of socioeconomic development. In any case, established practice is to use real data to validate theoretical models, not to reject the data if apparently the model does not fit. Acknowledgements We are grateful to D.M. Parkin and P. Pisani for their help in preparing the estimates. Resume Deces par cancer en Inde: I'approche fond6e sur les modbles est-elle valable? En Inde, les estimations de la mortalite par cancer fondees sur les modeles (775 800) sont pres du double des estimations fondees sur les donn6es (433 000), et depassent meme les estimations de l'incidence (612 300). Par cons6quent, du moins en ce qui concerne l'inde, I'approche fondee sur les modeles est incorrecte. La pratique usuelle est d'utiliser les donnees reelles pour valider les mo- deles th6oriques, et non de rejeter les donn6es si le modele ne semble pas s'y adapter. References 1. Murray CJL, Lopez AD. Global and regional cause- of-death patterns in 1990. Bulletin of the World Health Organization, 1994, 72: 447-480. 2. Shanta V et al. Cancer registration methods in Madras Metropolitan Tumour Registry in India. European journal of cancer, 1994, 30A: 974-978. 3. Yeole BB, Jussawalla DJ. An assessment of reli- ability and completeness of Bombay Cancer Registry data (1963-1985). Indian journal of cancer, 1988, 25: 177-190. 4. Parkin DM, Pisani P, Ferlay J. Estimates of the world-wide incidence of eighteen major cancers in 1985. International journal of cancer, 1993, 54: 594- 606. 5. Pisani P, Parkin DM, Ferlay J. Estimates of the world-wide mortality of eighteen major cancers in 1985. International journal of cancer, 1993, 55: 891-903. 6. Coleman MP et al. Trends in cancer incidence and mortality. Lyon, International Agency for Research on Cancer, 1993 (IARC Scientific Publication No. 121). 7. National Cancer Registry Programme. Annual report 1987. New Delhi, Indian Council of Medical Research, 1990. 8. Registrar-General, Government of India. Survey of causes of death (rural). Annual Report 1988. New Delhi, 1990. 944 WHO Bulletin OMS. Vol 72 1994