Introduction Accurate information from routinely collected cause of death information is essential for epidemiological studies and for health systems planning. Given the relatively low sensitivity of clinical examination for the detection of valvular heart disease (VHD), we hypothesised that cause of death data based on clinical information would routinely underestimate deaths due to VHD. We therefore compared clinical information available pre-mortem with autopsy findings.

Methods The OxVALVE Accuracy of Death certificates for Valvular heart disease as a Cause of Death (OxVALVE-ADVanCeD) study examined post-mortem reports for adults from 2004 to 2012. We calculated the sensitivity and specificity of the documented pre-mortem clinical information for predicting cause of death determined after the result of an autopsy (and any other supplementary investigations). Autopsies performed both on behalf of HM Coroner and on behalf of the hospital were included.

Results There were 7324 post-mortem reports for adults aged 18 years and older from 1^st Jan 2004 until 31^st December 2012. Of these, 583 had any mention of VHD pre- or post-mortem (mean age 75 years, range 19–98 years, 48.2% female). 140 (1.9% of total cohort) had VHD as the primary cause of death, 273 (3.7%) had VHD as any cause of death, and 289 (3.9%) had VHD as an incidental rather than causal finding. Presence of VHD in the pre-mortem clinical information had a sensitivity of 73.6% and a specificity of 97.0% for VHD being listed as the primary cause of death after autopsy, and a sensitivity of 70.3% and specificity 98.2% for VHD being listed as any cause of death. A small number (21, 0.2%) of reports contained pre-mortem descriptions of VHD that were not found post-mortem, with the majority of these being regurgitant lesions that would be less reliably detected post-mortem.

Conclusions Clinical information available before death is highly specific for cause of death established after autopsy, but has limited sensitivity. It is therefore likely that population-based cause of death statistics, which rely most commonly on clinical information, routinely underestimate the mortality burden of VHD.