Introducing A New Diagnostic tool and Methodolgy for detecting outliers in aortic surgery early and reliably …
In summary, there are three points that probably explain the whole story:
- The medical tradition of checking performance has been in the form of audit-type method for ‘really’ long: get the notes; check your practice; compare to others; draw conclusions .. This has worked well in general; but has failed drastically in few occasions (Harold Shipman, Mid Staffordshorem, etc.)
- Industrial sector has revolutionised this approach over a decade ago: they moved into SPCs ..
- I am introducing an original methodology whereby one can apply SPCs effectively as a diagnostic tool in aortic surgery; and hence improve quality significantly and in time… I will show how to do this .. and will test it immediately and show the effectiveness of such a tool.
The basis of my research interests is a desire to improve the quality of monitoring outcome in aortic surgery using modern way of thinking.
The process for achieving this is by:
– first, identifying and understanding the problems faced by current policy makers in charge of monitoring aortic surgery outcome;
– second, formulating effective ways of addressing these problems
– thirdly evaluating the impact of these interventions;
– and finally, disseminating the data to enable policy change so that findings can be hard wired into clinical provision.
See prof Banjerjee on addressing dementia, university of Sussex
So in one sentence, my research question is: can vascular surgeons monitor and improve themselves better?
This project is probably bound together by one idea: is there a place for a new diagnostic test to detect outliers early and effectively in aortic surgery?
I was inspired by few things that really made me complete the project in full …
- I fall in love with Oxford .. since my 1st medical school year .. Oxford inspires me .. its art .. its colleges .. its busy scientific life .. and all the discussions running in there …
- Hence .. I connected to QRS team in Oxford extensively … to find out that they failed (recently and drastically at that time) in ‘knitting’ quality control methodologies (that we enjoyed talking and publishing about) into a proper tool …
- and I was lucky that I moved to Brighton … the lovely Brighton … the city that literally ‘knits’ everyone and everything together … and with an inspiring teacher at work, I proposed.. and they accepted my proposal …
This topic has in fact attracted many issues and debates recently –
- For example, will vascular surgeons be required eventually to publish their outcome for public scrutiny like their counterpart cardiothoracic surgeon? The answer came few years ago: YES
- What about all this variation in practices … how to control and improve such level of variation ..
- and what about the completeness and validity of NVD (now NVR)?
- what also will guarantee that we will not have another Harold Shipman or Mid Staffordshire drastic failing?
The problem I am talking about does worth tackling …
- We have to learn from recent inquiries … monitoring outcome is simply one of the major recommendations for the latest £13-million mid-staffordshire inquiry …
- We need to knit our service together while centralisation takes place .. failing to monitor outcome and act rapidly can result in another Bristol Inquiry …
- and we need to learn from better business in the field of quality control … the industrial sectors has identified this and tackle it while ago …
- and finally … we have limited resources .. and we can not offered losing any extra penny in a poorly organised poorly monitored service/surgeon who are not delivering a quality service that fits the purpose …
The team/person who had the strongest influence on the subject area I choose … is … Prof. Doug Altman and his team, Dr Gary Collins, at the Centre of Statistics in medicine, Oxford University .. Those guys are certainly exceptional … just google Doug Altman and see how much influence he has had on advancing statistics in the medical field .. and how much inventions he added to human knowledge … He, with the deep and inspiring discussions with Gary Collins and Peter mcCulloch, has certainly inspired me in full …
Then it was only when I moved to Brighton when the influence of another thinker, Waquar yusuf, came into my life … and with all, I managed to knit my ideas together and start testing my hypothesis under the umbrella of the University …
Which are the three most important papers that relate to your thesis?
- the unpublished paper by Gary Collins .. SELECTING THE CONTROL LIMIT IN CUSUM MONITORING OF A PROPORTION USING FRACTIONAL POLYNOMIALS – Gary Collins et al ..
- Wald, 1945, on: Sequential Tests of Statistical Hypotheses.
- Mohammed 2001: Bristol, Shipman, and clinical governance: Shewhart’s forgotten lessons
Any published work closest to me?
- Yes and No …
- as a methodology (a rather thorough one) – I have not found any similar work/ideas in literature whatsoever …
- but as using CUSUM for monitoring … few are there … Chen: on EMS system ..
What about the history in relation to your work?
- In the 1920s, Walter A Shewhart, an American physicist, engineer and statistician, was charged with improving the quality of telephone production in Bell Laboratories. He introduced and applied statistical process control methodology.
- In 1945, Abraham Wald from the Statistical Research Group in the US army was given permission to publish his paper on sequential probability ratio testing, following two years of restriction that was applied by the US National defense Research Committee on his work (Wald, 1945). Wald’s technique was considered a national security issue during the wartime due to the fact that SPRT was considered of special significance in allowing for substantial savings in the expected number of observations required.
- Bristol Heart Scandal – Set up in 1998 to investigate the deaths of 29 babies undergoing heart surgery at the Bristol Royal infirmary in the late 1980s and early 1990s, the vast 529-page report effectively provided a blueprint for wider reform of the NHS.
- Harold Shipman – On 31 January 2000, a jury found Shipman guilty of 15 murders. He was sentenced to life imprisonment and the judge recommended that he never be released. After his trial, The Shipman Inquiry, chaired by Dame Janet Smith, began on 1 September 2000. Lasting almost two years, it was an investigation into all deaths certified by Shipman. About 80% of his victims were women. His youngest victim was a 41-year-old man.[3] Much of Britain’s legal structure concerning health care and medicine was reviewed and modified as a direct and indirect result of Shipman’s crimes.
What are the most recent major developments in your area of research?
Nothing much.. maybe the most important is the move of NVR into an online portal .. there remain the major problem of capturing the data in time and in full.
The research question emerged as follows & in this chronological consequence –
- CUSUM is a good tool to consider for application in clinical databases (the CUSUM bubble era). But, it doesn’t fit into any statistical model we use in clinical medicine (kay square, t-test, correlation, etc.) .. there is no clear p values!!
- if we can apply CUSUM on large database; how can we test its accuracy?
- what can be a gold standard to measure CUSUM against?
- What does industrial sector use?
- How can we acquire NVD and use it?
15. What were the crucial research decisions you made?
· Moving from Upper GI database to aortic aneurysm (NVD) & HES
· Moving from observational (feasibility) study into experimental design
· Validating the ‘already validated’ industrial/statistical CUSUM using NVD – defining key concepts and definitions
Putting in writing a real-time online system analysis!
16. Why did you use this research methodology? What did you gain from it?
· CUSUM(SPRT) as a diagnostic test requires no complicated mathematics. The standard approach of sensitivity/specificity analysis is all what’s required. The difficulty is therefore not statistical, but rather judgemental: deciding how good a diagnostic test should be to be clinically valuable.
· Other methodologies: RCT with Feedback from CUSUM (intervention) vs no feedback (controls). But with no solid background to CUSUM in practice, it is doubtful that such RCT will have a go ahead.
· Retrospective case-control study .. can be done; but, will lack the prospective and real time ability to adjust CUSUM parameters .. and infringe the concept as a whole.
17. What were the alternatives to this methodology?
18. What would you have gained by using another approach?
19. How did you deal with the ethical implications of your work?
No challenges. Ethical approval (generic) is already agreed at VSGBI on a third party requesting access to anonymised NVD.
20. How has your view of your research topic changed?
from simple retrospective CUSUM application and few case studies-applied on SAGOCS; into CUSUM application on NVD; into CUSUM application in real time; into CUSUM validation using simulation and FP; into CUSUM validation using gold standard audit; into CUSUM validation using ROC.
21. How have you evaluated your work?
22. How do you know that your findings are correct?
Simple clear methodology
Peer review
Working with authoritive pioneers in the field
Acceptance of program by r engine
23. What are the strongest/weakest parts of your work?
Strongest – methodology and automation
Weakest- no comparing methodology/approach ever before; apart from retrospective publications
24. What would have improved your work?
Funding …
Migration to MAC
Availability of R programmer on site
More time availability (I did this while progressing in my career to a consultant surgeon post, and while working in a very heavy tertiary centre)
25. To what extent do your contributions generalise?
Can be applied on any major surgical database and on other vascular surgeries with enough numbers
26. Who will be most interested in your work?
The quality improvement committee of the VSGBI and our hospital managers
27. What is the relevance of your work to other researchers?
I link to QRS team on finding better methods for quality improvement
I link to the centre of statistics in medicine
I link to quality improvement committee at VSGBI
28. What is the relevance of your work to practitioners?
As above
But also, I am providing a tool that allows for better judgement
29. Which aspects of your work do you intend to publish – and where?
Five papers
30. Summarise your key findings.
31. Which of these findings are the most interesting to you? Why?
32. How do your findings relate to literature in your field?
33. What are the contributions to knowledge of your thesis?
34. How long-term are these contributions?
35. What are the main achievements of your research?
36. What have you learned from the process of doing your PhD?
37. What advice would you give to a research student entering this area?
38. You propose future research. How would you start this?
39. What would be the difficulties?
40. And, finally… What have you done that merits a PhD?
see this one for p values –
http://statpages.org/ctab2x2.html
http://ktclearinghouse.ca/cebm/practise/ca/calculators/statscalc
https://books.google.co.uk/books?id=hPuEpp9dxbEC&pg=PA223&lpg=PA223&dq=calculating+p+values+in+diagnostic+tests&source=bl&ots=TWyVWZtjbM&sig=BU_JCwY70WHhwyNYtZY4t1rTbFU&hl=en&sa=X&ei=0bvmVJPSI8HnUrSyhKAK&ved=0CEwQ6AEwBjgK#v=onepage&q=calculating%20p%20values%20in%20diagnostic%20tests&f=false
this gives the equations for calculations
http://www.cct.cuhk.edu.hk/stat/confidence%20interval/Diagnostic%20Statistic.htm
http://en.wikipedia.org/wiki/Diagnostic_odds_ratio
http://stats.stackexchange.com/questions/61349/how-to-calculate-the-confidence-intervals-for-likelihood-ratios-from-a-2×2-table
with R code:
http://stats.stackexchange.com/questions/61349/how-to-calculate-the-confidence-intervals-for-likelihood-ratios-from-a-2×2-table