In the current era, “big data” is considered to be a way to find out things that studies run on small populations cannot derive. It has been used specifically to address comparative evaluation of new interventions, and assessment of cost versus quality in care. These approaches are very appropriate when 1) both the disease and the outcome are relatively straightforward, 2) there is no clear bias in where or how the patient presents and 3) the codes used to capture both diagnosis and outcomes are representative of what actually occurs. An example might be myocardial infarction, where the majority of patients come to the ER, they get a correct diagnosis, and their outcome may be recovery or death.
Unfortunately, both diagnosis and treatment of epilepsy pose huge issues in interpretation of “big data” sources. Epilepsy is an extremely heterogeneous disease, that may be benign or deadly. It is difficult to diagnose. It may be diagnosed in the ER, in an outpatient clinic, or during a hospitalization. Patients may come to a neurologist for assessment of symptoms, or may be referred because they are felt to need more comprehensive care. The diagnostic codes are not applied properly, and many associated diseases may be caught in the epilepsy net. A non-neurologist may apply an epilepsy diagnosis to cases that do not have epilepsy, or may use an intractability code inappropriately.
The outcomes, as noted by Hill and colleagues, are also not well captured. If people do not end up in the ER or with an injury, does that mean that they are “OK”? are they employed? Are they able to function and lead a normal life? Are they able to drive? Unfortunately, we have found time and time again that “big data” does not necessarily provide a clear picture of what happens with epilepsy care.
These very interesting results once again highlight that where epilepsy is concerned, “the devil is in the detail”. It would be fascinating to be able to dig more deeply into these cases, to find out how they are similar, and how they differ. Without such an ability, unfortunately, we will have to continue to ponder how to interpret these results, and others based on diagnosis codes alone.
- Hill, CE, Lin CC, Burke JF, et al. Claims Data Analyses Unable to Properly Characterize the Value of Neurologists in Epilepsy Care. Neurology Epub 2019 Jan 23