Jump to comment:

• Improving reproducibility
  Darrel P Francis
  Published on: 28 November 2002

• Published on: 28 November 2002

  Improving reproducibility

  • Darrel P Francis, Cardiology Specialist Registrar

  Dear Editor

  Dr Otterstad makes some incisive points about the reproducibility of left ventricular volumes and ejection fractions. In general, it is remarkable how silly we are from a quantitative point of view, when we plan research. We make extensive arrangements to design a study, consent subjects, transport them to hospital, arrange them on an echo table, set up our machinery, measure our variable of interest jus...

  Show More

  Dear Editor

  Dr Otterstad makes some incisive points about the reproducibility of left ventricular volumes and ejection fractions. In general, it is remarkable how silly we are from a quantitative point of view, when we plan research. We make extensive arrangements to design a study, consent subjects, transport them to hospital, arrange them on an echo table, set up our machinery, measure our variable of interest just once, and then do all the previous steps in reverse. The only scientifically valuable step is the measurement of the variable with as little random error as possible, yet (usually) we spend no time on minimising this error.

  The consequences are not only the problem that Dr Otterstad points out, but also failure of the measured variable to show its true correlation with any other variable.[1]

  There are many strategies we can take to minimise random error, but the only universally effective technique is to measure the variable more than once and take the average. As a (remarkable) consequence of the central limits theorem of statistics, regardless of the distribution of the measurements we make (i.e. even if they are skewed), the mean of several measurements takes on a normal distribution whose standard deviation falls with the square root of the number of measurements we have meaned.

  Thus measuring left ventricular volumes 4 times (i.e. from 4 separate echo recordings), rather than once, and then taking the mean, may be expected to halve the reproducibility range. Our instincts may be to laugh at the prospect of repeating the study so many times, but actually it is probably the most effective use of the time of staff and patients. It only seems time-wasting if one ignores the consequences of a study whose reproducibility is twice as wide -- four times as many patients are needed!

  Which is more wasteful - quadrupling the scanning time or quadrupling the study size overall?

  Darrel Francis
  Specialist Registrar in Cardiology
  St Mary's Hospital

  Reference

  (1) Francis DP, Gibson DG, Coats AJS. How high can a correlation coefficient be? Effects of limited reproducibility of common cardiological measures. Int J Cardiol 1999, 65:185-189.

  Show Less

  Conflict of Interest:
  None declared.

  • Back to top