Projects

Posted 29.05.16

 

In November 2014 data was presented to suggest that there had been a significant increase in the number of cases of infective endocarditis – more than would have been expected – since the change in guidelines for the use of antibiotic prophylaxis by dentists to prevent infective endocarditis.

 

It caused something of a storm, and has resulted in an ongoing battle between those who believe and those who do not believe in antibiotic prophylaxis.

 

It has become apparent that some patients are coded as having had multiple discharges with infective endocarditis when having OPAT – outpatient antimicrobial chemotherapy – and this may be accounting for some of the change seen.

 

Furthermore, there was concern that demography had not been adequately controlled for, and that the statistical analysis was not thorough enough. In particular, fitting two straight lines before and after March 2008 may not have been the most appropriate way of modelling the data, and that forcing a breakpoint of March 2008 may have been artificial.

 

An analysis was also undertaken by Craig Ramsay, as part of the NICE process.

 

This study is an attempt to rectify those issues.

 

These are the steps that will be taken, clearly there are challenges to completing this, not least access to the data and my abilities.

 

  1. Obtain accurate details of demographic change in the UK
  2. Obtain data from DrFoster of all discharges with a primary diagnosis of I33.0 from January 1st 2000 onwards, excluding patients admitted on successive days (to exclude cases of OPAT) and also those readmitted within 6 months; It is generally believed that patients who re-present within 6 months have a recurrence of their original disease (although this is not invariable). Nonetheless, there is concern that there is pressure to shorten length of stay and that this pressure may result in an increased rate of recurrence.
  3. Update prescribing data.
  4. Re-analyse the data:
    1. Correct the raw data for variations in month length and changes in demography.
    2. Replicate Lancet methodology.
    3. Replicate Craig Ramsay Methodology.
    4. Fit polynomials / exponential functions to the data pre and post March 2008 – selecting the two curves with the least residuals.
    5. Calculate the average slope before and after and the standard deviation.
    6. Perform a T-test to compare the two sets of data.
    7. Fit first order polynomials not specifying the break point and select the cut-off at which the lines fit best. Consider using higher order polynomials too. Potential lines to fit, assuming order of polynomials up to 17 = 17 x 17 x n-2, where n is the number of months of data, but possibly to January 2016, which would be 192 months, giving a total of 54,910 curves.
  5. Present all data and code online for others to analyse.

 

MJD