In the last two posts I outlined some of the problems associated with late diagnosis of chronic liver disease and how colleagues and I brought together some of our ideas to tackle this.
Part of our proposal was to use a ‘case finding’ approach to diagnosing liver disease. In essence we would be using existing blood test results in our laboratory systems to attempt to identify patients likely to have significant liver disease. We could then check whether they were known to us and had already been seen and treated or whether they were in the group of undiagnosed patients. There was nothing that was spectacularly ground-breaking in this work – much research has gone before us, and the principle of ‘case finding’ was already established in a number of other disease areas including for Hepatitis C. Furthermore this approach is much cheaper than ‘population screening’, as no new tests are performed. It does of course have the downside that there will be other people missed simply because they have never had relevant blood tests performed.
There were two elements required. Firstly, we needed a complete overview of all our patients in the two hospitals in Somerset. We didn’t want to be chasing up patients we already knew about, it would confuse them, their GPs and just create extra work. Secondly, we needed to be able to look for patterns of blood tests in our laboratory computer systems to identify potential patients.
The second part of this is distinctly tricky. We are not trialling or inventing new ways of diagnosing liver disease, but simply trying to put into practice on a computer system what liver doctors routinely do in clinic. We look at a small number (often fewer than ten) blood tests, and look for patterns and changes over time, or look for collections of abnormalities that taken together indicate a particular diagnosis (for most liver disease there isn’t a ‘single diagnostic test’, although some come closer to this than others). Theoretically this should be quite manageable with current computer databases. Competent programmers or analysts would be able to analyse the database to provide an answer to the question – “which patients have the following [x, y, z] patterns within their previous blood tests”. Once we have the list, we can rule out those we already know about and start finding the patients in need of treatment.
There is nothing particularly difficult in terms of computer science, no complex ‘Artificial Intelligence’, and no incomprehensible mathematical algorithms.
In practice there are some key hurdles. One we found very early is that healthcare computer systems are not designed to be used in this way. Whilst they can report on all sorts of data, their prime use is to show a clinician all the information on a known patient, carefully identified (e.g. name, date of birth, NHS number etc). Because they are not designed to be used in reverse, finding a cohort of patients with particular blood test combinations becomes incredibly slow. In some of our simulations a desktop model was predicting a task time of many hours to analyse quite simple data. We could extrapolate some of these results showing it would take decades to answer the sort of questions we were looking at. Obviously, there were potential ways around this – more sophisticated analysis, more powerful computers etc, but realistically what we were looking for was not achievable. We wanted to make it possible for clinicians in hospitals to easily analyse data to look for ‘missing’ liver patients themselves, not rely on supercomputers and highly skilled analysts.
What we needed was someone who understood both computers and healthcare. My former colleague Neil Stevens (an experienced specialist in Information Management and IT systems, their implementation and the NHS). I had worked with Neil on a number of major projects and he and I were engaged on a small IT project. After one meeting I outlined the problems we were having with case-finding. After the usual surprise about how patients with potential liver problems can be overlooked, there followed more surprise, this time from me as to how simple this might be to solve.
We therefore started to write up our proposed potential solution, and started to look for funding for this exciting innovation.