We’re a dynamic bunch and involved in lots of projects and collaborations. Please get in contact if you want to know more.
GlobalSurg/ NIHR Global Health Unit on Global Surgery
We co-founded GlobalSurg and this work sits at the centre of what we do. We collaborate with our friends at the Universities of Birmingham, Warwick and in many countries across the world. We are grateful to the NIHR for funding this work. Please visit the GlobalSurg/NIHR website which is maintained by Katie for details of the details of the many studies and initiatives we are involved in. The GlobalSurg 3 study is currently running and causing many sleepless nights. Have a look and tell us that it is far too ambitious.
GlobalSurg Training Centre
Training.globalsurg.org is a Moodle run by Katie. This is the centre of operations for the NIHR GSU Education and Training Centre. Here we provide a number of training courses which are stand-alone or related to the trials the group are running. We all help run Edinburgh Surgery Online and support global scholarships to these programmes.
GlobalSurg Data Centre
Data.globalsurg.org is a growing space run by Riinu. Here we run auto-updating apps giving minute-to-minute reporting of where our various studies are in terms of recruitment. We publish all our open data here as well. Check out ssi.globalsurg.org, a cool shiny app in which you can explore You will find a dedicated blog about all things data.
We are promiscuous in our approach to data science, but collaborate with amazing experts across the world. Here are some themes that we are working in.
Mobile technology for direct patient outcomes
The Tracking Wound Infection with Mobile Technology (TWIST) Randomised Controlled trial is led by Kenny and has recruited over 250 patients. We use SMS text messaging after emergency surgery to retrieve data and photographs on surgical wounds to try and reduce the time to the diagnosis of wound infection. Stephen also works on mobile technology in global settings, including GPS tracking to measure post-operative recovery.
Mobile data collection platforms
We use REDCap platform from Vanderbilt University like no one else. Riinu is the expert, but we all no it very well. We have tens of thousands of users, use the API until it smokes, and pull data directly into RStudio for analysis. We’ve trialled the stand-alone app and like it. Most of what we do has REDCap lurking in it some where.
Well who isn’t using deep learning? We have focussed on Tensorflow and specifically Keras and specifically in RStudio, because we love RStudio. But this is going really well. Tom is going prediction models for outcomes after surgery that are showing great AUCs. And we have some great recurrent ANNs (LSTM) for processing radiology text to identify surgical diagnoses. Well gallstones for now, but with plans to expand. Cameron is leading on this.
Natural language processing
But we started with rules-based approaches rather than ML. We use the R API to the Stanford NLP algorithms to get dependencies, then apply conditional logic to determine negation. And have got some awesome results with it, almost (but not quite) as good as the ML.
Ewen keeps going on about wearables and everyone else looks sceptical. Thanasis is a believer and we are now expanding this line of work. Clinical Surgery in Edinburgh has a lot of data supporting accelerometer data in accurately reflecting recovery from surgery, so we’re getting on it. GPS or LoRraWAN for something big in Edinburgh? If you know about the latter, get in touch. We haven’t got much response from the Internet of Things people.
Decision modelling and Bayesian
Kenny did a great paper on decision modelling in transplant surgery using a frequentist Markov-model. We have used Stan for most of our recent studies and have moved a lot of the analysis into a Bayesian hierarchical framework. We will likely continue to do so and will write more about Stan in the future.
The stuff that is easy to measure rarely reflects what is actually important to patients in the recovery from surgery. A patient may not mind having a complication if they are cured of cancer, so does quantifying 30-day complication rates mean much? We have worked hard on patient-reported outcomes in the past – particularly Catrin, Kenny, and Alesandro. We are now building PCOs into our reporting apps and are expanding this line of work.
Administrative Data Analysis
So much data is already collected and not used. We work with ISD predominately on hospital episode data and collaborate closely with our anaesthetic colleagues – Mike Gilles, Naz Lone and Tim Walsh.