Percentages versus patient years
Or: What are proper outcomes in clinical research?
Probably slightly beside the point:
1. Patients or events?
To my opinion patients with a specified outcome should be the unit
of analysis and of presentation, not the events itself. Particularely
because events can differ in severity and consequences. E.g.
fractures in osteoporosis studies: one or three (a)symptomatic
vertebral fractures is not as important an outcome of an intervention
as one or three patients with (a)symptomatic vertebral fractures.
Not speak of the vast difference between the perc. of patients with
osteoporotic fractures in general versus the perc. patients with a
hip fracture (mortality 25 %).
2. Problems with dichotomous scales < > ordinal & ratio scales
Dichotomous scales are easy to present in a sensible way.
Results on other scales are often presented in terms of 'mean
differences' (some years ago 'the certainty of the existence of a
difference with p-values). This is statistically a sensible method,
but not clinically. Clinically (or for teaching purposes) it seems
more appropriate to present something like 'The perc. of patients
who improve xx points on a 5-point Likert scale'; Or: 'The perc. of
patients who improve more than xx units being the clinically
relevant difference.
Nico van Duijn
Dr.N.P. van Duijn, General Practitioner
Department of General Practice
Division Public Health
Academic Medical Centre
University of Amsterdam
Meibergdreef 15
1105 AZ Amsterdam
& Primary Health Care Centre 'de Haak'
Almere
the Netherlands
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|