Summary of results
0.5% of death records included an asthma-related cause. When the primary cause of death was asthma-related, ICD-10 code J45 (asthma) rather than J46 (asthma attack) was used 82% of the time. There were higher odds of infection (predominantly pneumonia) being reported as a contributing cause of death when J45 was the primary coded cause, compared to J46 (odds ratio = 7.51, 95% CI = 2.76 to 20.41). When asthma was only a secondary cause, 42% of deaths had primary cause related to diseases of the circulatory system.
Results in context
The process of completing a death certificate is liable to bias, particularly where there may be very co-morbid patients with chronic diseases. In the 2014 National Review of Asthma Deaths (NRAD), the MCCDs had been predominantly completed by junior doctors and therefore more likely to have inaccuracies than if they were completed by a senior doctor [20]. An expert panel review of the medical records of 900 cases classified with asthma as the underlying cause of death, they found that 10% had no evidence of asthma diagnosis at all, and 13% had asthma but did not die from it [21]. We consider such cross-referencing with primary care records to be the gold standard in reliable ascertainment of validated asthma deaths.
This bias can influence the accuracy of the final coded record in two different stages: the reporting of the death on the MCCD certificate by the attending physician, and the ICD coding of that certificate by the medical coder. A Japanese 2019 study reviewed 103 asthma ICD-10 coded deaths, and found that 16% were not asthma deaths, and that for 13% the cause could not be ascertained without further investigation [22]. A Swiss study compared the mortality data for in-hospital deaths to their terminal hospital discharge records, and found that for asthma mortality record cases (n = 50) there was only 24% agreement as primary cause [23]. Conversely, for the 20 cases with asthma as the principal terminal hospital discharge cause, the agreement to the mortality record was 60%. Unfortunately, we were not able to identify any UK data to directly compare these studies to.
The ontological coding from death certificate to ICD-10 value may also introduce some degree of bias, particularly when conducted by a human-coder rather than automated coding software. A study in the Netherlands compared the results of two independent ICD coders, and found that for respiratory deaths (n = 1145) there was agreement to the 4-digit level in 81% of cases, and to the 3-digit level in 84% of cases, but only to the chapter level in 88% of cases [24]. The chapter-level agreement was less than 70% for infectious, endocrine, and skin diseases, but over 95% for neoplasms. Although this study used manual coding, there may also be variation in automated coding between software and versions which affect outputs. An NRS review of the change to Iris in January 2017 from the previous Mortality Medical Data System (MMDS) system, which was an automated system in use from 2000, observed a decrease of 4.8% in the number of deaths allocated to respiratory causes in Scotland. This was mainly due to the switch of deaths from chest infections and aspiration pneumonia to dementia and diseases of the nervous system [25].
The typical standard for asthma death ascertainment from ICD-10 coded data is either J45 or J46 (and lower-level codes under these parents) as the primary cause of death. However, consideration must be paid to patients with a prior misdiagnosis of asthma [26], with an incorrect prior diagnosis of a different respiratory condition [27], with no reported history of asthma diagnosis [21], and with comorbid conditions [21], including infections and overlapping COPD and asthma [20, 21, 27]. As highlighted in our methods, the code J448 includes chronic obstructive asthmatic bronchitis however as it does not differentiate between asthma related and emphysematous related COPD, it was not used as an identification criterion in our study. As such our study may have underestimated any deaths associated with chronic asthmatic bronchitis. More complex rules and exclusions may be required to improve the accuracy of asthma mortality ascertainment, especially if such data were to be used for training disease prediction models.
We investigated changes in practices over the duration of the observed data (2000 to 2017). Variation in practice over time was observed, such as in the use of J45 versus J46 as the UCOD, but we failed to identify any clear trends. Further investigation is required to explore possible causes of temporal changes. Furthermore, the effect of the disruption to the healthcare system resulting from the CoVID-19 pandemic warrants further exploration [28].
The ALHS dataset contains records for a subset of the Scottish population: over half a million patients from 75 general practices in Scotland [19]. In this study population, Scottish people from areas with lower socioeconomic deprivation are slightly over-represented, however the population is otherwise fairly representative.