Label without a cause

To talk informally about a concept, we need only give it a recognisable name. For example, we use the label "medical error" and we all know what is meant - or at least we think we do. However, there are clearly large differences between mis-diagnosing a condition, prescribing an incorrect dosage or removing the wrong internal organ, so our informal certainty is of limited practical use. We know that to analyse or measure a problem robustly over time (and ultimately to improve it), we need more than a high-level label. We need a rigorous classification, universally adhered to.

Rigorous classification is what the World Health Organisation's ICD system seeks to provide for analysts working with mortality data. We've talked before about challenges when working with mortality data disaggregated by cause-of-death, and we've discussed specifically the issue of dealing with revisions in the ICD classification system through time. However, a research article published in the BMJ earlier this month suggests that while fretting over changes to existing classifications, we ought to be even more concerned with classifications that have never existed at all. The article proposes that the third largest cause-of-death in the USA may be deaths arising from medical error, an area the ICD system does not explicitly code for.

This conclusion has particular relevance to those interested in using extrapolative mortality projections to inform or benchmark a proposed mortality basis. While with all-cause projections we can be reasonably sure our data will accurately capture the fact of a death, we have a potentially intractable problem for by-cause projections if one of the largest mortality-causes is not reflected in our data. Part of the answer will lie in better classification and reporting, however research suggests achieving this won't be easy. There are obvious legal sensitivities around disclosing medical error, and barriers to reporting and disclosure exist at many levels.

Of course, the authors might well be described as bold or even be considered guilty of overreaching. They attempt to draw a hard (and sensational) numerical conclusion from uncertain data, culled from a small number of mismatched studies. This approach seems unlikely to pass rigorous statistical scrutiny. However, that very fact underlines the problem the research seeks to highlight. If we are to have any hope of applying lessons learned from the science of safety to medical systems and processes, then we urgently require better data, but currently have no means of gathering it.

This latest research doesn't represent the only issue around mortality cause classification. We have previously discussed how the fall in autopsy rates may have left our assigned causes vulnerable to a high rate of error. In addition we've noted the impact of losing vital information on age at death. Whilst the desire to extrapolate by cause of death remains understandable, the list of problems ranged against being able to credibly accomplish it seems only to be growing. Clearly our health systems must work harder to ensure that by-cause data will one day be fit for some of the purposes that mortality analysts want to put it to. Alas, that day seems as far away as ever.