Researchers at IHME propose a method of estimating cause-specific mortality fractions (CSMFs), or the fraction of all deaths due to a specific cause. The study, Estimating population cause-specific mortality fractions from in-hospital mortality: validation of a new method, illustrates how this novel method can allow countries to utilize data already collected on in-hospital deaths to estimate the CSMFs in a population. This could be a valuable tool to help countries develop cause of death estimates, especially in countries without complete vital registration systems. The work was done in collaboration with scientists at the University of Queensland, the Harvard Initiative for Global Health, and the Ministry of Health in Mexico.
The researchers, using a method to estimate CSMFs, found that population CSMFs can be estimated with low average error in areas where accurate International Classification of Diseases (ICD)-coded cause of death data are available for deaths in hospital and for vital registration covering a subset of the population. For Mexico, this method substantially reduced error from biased hospital data, even when applied to areas with widely different levels of development. For countries with ICD-coded deaths in hospitals, this method potentially allows the use of existing data to inform health policy.
Within Mexico, the study found that perinatal infections represent the cause of death with the highest average proportion of deaths in a hospital, at 94%. For HIV/AIDS, diabetes mellitus, and cerebrovascular disease, the proportion dying in a hospital is lower in municipalities with lower socioeconomic status, as determined by literacy level. Overall, the proportion of in-hospital deaths was a distinct function of age, cause, and socioeconomic status.
Researchers utilized a method to estimate the proportion of people in an age, sex, and cause group that die in a hospital. The basis of the method is the use of observed proportions of in-hospital deaths by age, sex, and cause group to correct observed hospital CSMFs, yielding robust estimates of population CSMFs. The researchers then evaluated this method using nearly complete individual death records from Mexico from 1998 to 2005, which record whether a death occurred in a hospital. The method was validated using 45 disease categories both nationally and between communities. The estimation method was determined to have average relative errors between 20% and 31%, with lower error rates in the most developed states. Higher error rates occurred in the least-developed communities.
Cause of death data are not available for many developing countries. Information on deaths in hospital by cause is available in many low- and middle-income countries but is not a representative sample of deaths in the population, since deaths in hospital are not a random sample of deaths in the community. The researchers propose a method to estimate CSMFs using data that are already collected, but are rarely used, in many low- and middle-income nations: in-hospital death records. If these in-hospital deaths can be estimated, the proportion dying in the population using hospital death data can also be estimated. This research is part of ongoing work by IHME to understand what are the world’s greatest health problems and to provide timely, accurate, and comparable measurement of mortality and disease burden.
Recommendations for future work
The proportion of in-hospital deaths in other countries, as in Mexico, is likely to be a systematic function of age and sex covariates, cause of death, and community factors that influence physical, financial, and cultural access to hospital services. Future work could investigate how other community attributes that relate to the use of hospital services might improve estimates of population CSMFs. Further research may also yield important insights into the determinants of dying in a hospital that may strengthen the ability to predict the proportion of deaths by age, sex, and cause in various populations.
Because this study was validated for Mexico only, it is important that future research confirm the finding that the proportion of deaths by age, sex, and cause are predictable using partial data or data from other populations. These analyses should only be conducted in settings with nearly complete vital registration with good-quality ICD coding.