Richard Stevens, Course Director: M.Sc. in EBHC Medical Statistics
Richard Bright’s paper of 1833, a series of case reports from his clinical practice, is thought to be the first publication to make or posit a connection between diabetes and cancer. When we first meet the undergraduate medical students, for their first statistics workshop of their undergraduate career, we begin by asking them to consider this example of the science of 1833:
“The Renal capsules were both involved like the pancreas, so that it was difficult to say whether the disease actually occupied their structure or was superadded; but they appeared to me to be pervaded so completely by the disease, that in one of them a very small portion only retained a natural texture …”
…and compare it to a more recent publication on diabetes and cancer, which produced this forest plot:
Figure 1. Forest plot showing cancer incidence or mortality in randomised trials of metformin. Reproduced from Stevens et al., Diabetologia, 2012.
The point we’re making is that in the early 19th Century, medical science could proceed by subjective, qualitative and speculative assessment (all of which are present in Bright’s comprehensive paper), but in the 21st Century, medical science is highly objective, quantitative and statistical.
We do this to explain to undergraduates in our medical school why we expect them to learn statistics. Undergraduates who’ve enrolled to study medicine arrive expecting to study physiology, pharmacology, biochemistry, not to mention neurology, pathology and perhaps even social science, but it’s often not clear to them why they also have workshops in a subject that they perceive, from their schooling, as mathematics.
Our postgraduate students are quite different. These students, in our Graduate School of Evidence Based Medicine and Research Methods, are already healthcare professionals, or professionals in a related field, such as nursing, pharmacy or other allied subjects. These veterans of the real medical world arrive well aware of the importance of medical statistics, and highly motivated to learn it. Some have little statistical knowledge and want to grasp the basics: it’s a pleasure to teach these students, especially when we can make something beautfiully clear that has eluded them with other teachers on other courses. Others have quite advanced statistical knowledge and a hunger for more detail and rigour. For these students we have more advanced statistical modules, which give us the chance to teach some of the subtler points and more interesting techniques.
For example, one of my favourite things to teach on our advanced courses is the difference between fixed effect and random effects meta-analysis. The fixed effect method takes a kind of average (a “weighted mean”) of the effect estimates from all the studies in the meta-analysis. The random effects method does the same, but with an extra mathematical term that expects to find differences (heterogeneity) between different study protocols in different populations and settings. I greatly respect the mathematical statisticians who found the way to embed this in the algebra and the computer code; but, much as I love algebra, and computer code all the more, that’s not where the real beauty lies. (In fact, I find I can get a remarkably long way teaching this topic before touching on algebra at all.) The beauty of the subject is the way the “mathematics” reflects the different clinical assumptions (are these trials fundamentally the same, or fundamentally varied?), and the way the slightly different results from different mathematical approaches map back to different clinical conclusions. It turns out that the decision to use a fixed effect approach or a random effects approach isn’t a mathematical question at all: it’s a question for the clinician/scientist.
From next academic year the students who enjoy our advanced courses will have the option of an entire M.Sc. in medical statistics. The new course, M.Sc. in EBHC Medical Statistics, is a part of the same Graduate School and includes several of the same modules. But the advanced statistics modules on the existing programme will become the core modules of the new course; and new modules will give the students the chance to go deeper into specialist statistical areas. The module in Big Data Epidemiology for example will capitalize on our team’s expertise in database research (see, for example, the Clinical Practice Research Datalink, or Ben Goldacre’s Open Prescribing project). Meanwhile, another new course, the M.Sc. in EBHC Systematic Reviews will similarly expand our existing teaching on systematic review and meta-analysis into a specialist course in every form of quantitative and qualitative review.
For the first time, the University of Oxford is offering a M.Sc. in statistics that is aimed at health care professionals, rather than mathematicians and statisticians. We’re excited about this new direction and we’re looking forward to meeting our first students in Autumn, 2017.
Find out more about the MSc in EBHC Medical Statistics here.