• coronary artery disease
• electronic medical records

Most readers of Heart will be no stranger to dealing with the comorbid patient. Cardiologists see increasing numbers of complex patients both in our clinics and on our ward rounds. They can be the most rewarding of cases, but they can also make your heart sink. Frequently, the most difficult case discussions are around higher risk intervention or operations in older patients with comorbidities. The heart might be easy, but it is the rest of the patient that is the problem. Often, individually, none of the comorbidities would be an absolute contraindication to a proposed procedure. However, we have to consider their impact in combination and in totality. What will the natural history of the patient in front of us be with or without our intervention, taking into account the comorbidities—difficult questions to answer? In this issue of Heart, Crowe provides us with a new perspective on comorbid phenotypes.1 No single article is going to give us the answer but this one provides food for thought. Do not let not knowing what a latent class analysis is put you off looking at it.

The study presents high-quality data on the risk of mortality in patients with ischaemic heart disease (IHD). It is drawn from The Health Improvement Network (THIN) in the UK.2 THIN extracts data from a widely used general practice management system. It provides a searchable database of demographics, symptoms and diagnoses. Approximately 25% of population of the UK are included meaning that the results can be readily extrapolated to the population as a whole.3 But how about for the specialist wanting to extrapolate to their secondary care patients? That is the first point for a cardiologist in considering this paper.

Any database is, of course, only as good as the data that are put into it. It is well recognised that IHD tends to be overdiagnosed in the General Practice environment. The typical scenario here may be the patient with a remote history of chest pain that was not fully investigated carrying forward an IHD diagnostic code for years. To ensure consistent quality recording of important clinical outcomes including coronary disease, many practitioners contributing the data to THIN did receive specific training. But nonetheless the authors concede that overcoding could have occurred. In the end, this was probably not to an extent that would impact on these results. We take reassurance from the data in supplement 5, which shows that when the analysis is restricted to myocardial infarction, a diagnostic code that should be more robust and less prone to overdiagnosis than IHD, the findings of this study are not materially altered. This suggests that the population in the analysis should be similar to a secondary care secondary prevention population and that the results could be extrapolated to our postmyocardial infarction patients.

So if we can apply it to our patients, then what does the data tell us as cardiologists? It does not tell us the particular impact of any of our interventions or their interaction with comorbidity, but it does suggest which clusters of comorbidities particularly influence outcome. To this end it is of note that of the 20 comorbidity diagnoses it was osteoarthritis, osteoporosis and rheumatoid arthritis grouped together as musculoskeletal conditions that emerged from the model when clustered with either other vascular conditions (atrial fibrillation, heart failure, hypertension, peripheral vascular disease or transient ischaemic attack) or respiratory conditions (asthma and chronic obstructive pulmonary disease). It will be apparent that within each of the three groups there are widely differing conditions, for instance osteoporosis and rheumatoid arthritis have very different aetiologies. But they can have similar effects in terms of contributing to frailty. Frailty has been shown to be a strong predictor of mortality in the British population.4 The authors rightly raise the possibility that what their data are showing is really just what a clinician might see as frailty but through the lens of a latent class analysis. Perhaps patients with lung or additional vascular disease in addition to musculoskeletal comorbidity are particularly prone to progressive frailty. Much more research is needed into comorbidity patterns and frailty.

Finally, most cardiologists will not be familiar with latent class analysis. It is an analysis in which the clusters are not prespecified by the investigators. The word ‘latent’ is a synonym for ‘hidden’. The analysis shows us what is hidden. For instance on the surface in the current study the effect of asthma was small, on its own the associated hazard ratio for mortality was an unimpressive 1.19. But in combination with one of the musculoskeletal diagnoses the hazard ratio jumps to 2.62. At the start of the analysis, this was a latent or hidden association. The hitherto hidden mortality impact of a musculoskeletal condition when associated with either further cardiovascular disease or respiratory disease that has emerged in this study is significant. Cardiologists should be alert to these combinations in their patients and factor the increased mortality risk into their decision-making. Meeting the comorbid patient should not be a heart sink moment.