As promised last week, here is another 1 000 000+ data study.
Sudden infant death dropped fast and early in affluent groups, but slower and longer in deprived areas. So an inequality opened up and later closed.
It is quite a long write-up but worth a careful read. Notice the use of record linkage to relate the infant death to characteristics at birth; also note that in Scotland all stillbirths and neonatal deaths are classified with a version of the Wigglesworth classification. Few (no?) other jurisdictions do this, so the study is only possible in Scotland. The Wigglesworth is simple but crude; various sophistications have been proposed if you ask a 'single medically qualified individual' im to code over 50 000 records per annum, simple is probably good.
With over 1 000 000 events in a database, you can answer some questions with certainty. It isn't a randomised controlled trial or a meta-analysis, but this study showed beyond doubt that stemmed metal-on-metal prostheses fail sooner than other types. Note also that the size of the database enabled the researchers to prove that this was a general design problem not a feature of one notoriously poor performing brand - there was still ample data to analyse after excluding that brand.
And finally: there was a good theoretical reasons for thinking that large, stemmed prostheses would have lower failure ares. Nature thought otherwise. Theory must always be tested by data from the real world.
Carpenters need to understand woods - how teak differs from cherry, or oak from balsa. We need to understand routine data - where its strengths lie, where the weaknesses and traps are, what we can use it for and what we can't.
So here is an interesting story about pneumonia. Pneumonia is a particularly squishy disease, the balsa wood of epidemiology (maybe I'm pushing this analogy too far...). You don't get coding problems with diseases as distinctive as, say, leukaemia. You also wouldn't get coding problems with pneumonia in a young person. But in older people with multiple morbidities, it can be very difficult to work out what is going on, and hence to code it. Heart failure likewise.