Regional and educational disparities in inaccurately coded deaths in Sweden, 1997-2023: a systematic analysis

Accurate underlying cause of death (CoD) data is critical for informing public health policy, but inaccurate CoD assignment, here called garbage code (GC) deaths, compromise CoD research and monitoring. Since 1997, GCs have consistently made up over 20% of all underlying CoDs in Sweden, but the distribution of GC deaths by sociodemographic status of the deceased remains poorly understood.

We used the Swedish Cause of Death Register containing 2.50 million death records from 1997 to 2023. We mapped each record to the Global Burden of Disease (GBD) project cause list and categorized GC deaths by disease groups. We calculated the fraction of deaths that were GCs by individual age, sex, region of death, and highest educational attainment. We performed redistribution of GCs onto well-defined CoDs and assessed the odds of GC assignment with a binomial logistic regression.

Since 1997, Sweden has coded at least 23% of deaths to GCs each year with 25.5% coded to GCs in 2023. The lowest educated consistently received more GC deaths, with 45.8% more GC deaths relative to non-GC deaths between ages 20 and 39 compared to the highest educated, and there were more GC deaths in (1) infections, (2) blood and endocrine diseases, (3) injuries, (4) cancers, and (5) maternal, neonatal, and congenital (MNC) diseases in 2023. GC deaths among the highest educated have continued to increase in infections, injuries, cardiovascular, digestive, and MNC diseases. After redistribution, well-defined death counts among the lowest educated increased by over 20% in 13 of the leading 20 CoDs in Sweden. Our model suggested low education increased the likelihood of having a GC by 12.8% (11.4%-14.2%) compared to the highest educated. This was second to point estimates of standardized age at death (25.2% [24.8%-25.6%]) and exceeded sex (12.1% [11.4%-12.8%] increase for males) and region (at most 7.3% [6.6%-8.1%] decrease for death outside of Stockholm).

We found consistent trends of high GC level in Sweden with doctors assigning more GCs to the lowest educated. Our results reveal stark sociodemographic disparities in CoD coding in Sweden and it is probable that similar disparities would be found elsewhere. This underscores the need for improving procedures and national guidelines in CoD assignment to correctly represent all social groups in research.