This is an aside and isn't to do with any of the PROMS authors here, however I just needed to get it out because I couldn't quite believe it (certainly given what we know/is acknowledged now). From the Crawley and White (2013) paper: Treatment outcome in adults with chronic fatigue syndrome: a prospective study in England based on the CFS/ME National Outcomes Database | QJM: An International Journal of Medicine | Oxford Academic (oup.com)
They have detailed below how they used a complex 'GEE model' in order to apparently adjust the scores where follow-up questionnaires were returned late - in order to be precise in working out what the 'fatigue level' would have been at 12 months. They say that some teams obtained data at additional follow-up points (6 and 24-months) so I guess they used these to back-fill the model. And I don't know either how many of the sample would have had this done nor how 'far off' they were 12months. But it seems a pretty astounding approach given they are using retrospective data, but on something that hasn't actually been researched before?? ie it isn't like they've got 25 trials worth of data covering all these time periods to show some curve.
Yet I can only assume they think they are being really diligent by doing this.
"Follow-up interval
Each team sent out follow-up questionnaires at 12 months. Variation in when the questionnaires were sent and delays in return of 12-month follow-up questionnaires led to variation in the exact time of follow-up. Also, some teams obtained data at additional follow-up points (e.g. 6 and 24 months). To maximize the availability of follow-up data for our analysis, we determined a margin of follow-up either side of 12 months. We did this by fitting fractional polynomial generalized estimating equation (GEE) regression models of fatigue against time.21,22 This method (implemented in Stata as fracpoly combined with xtgee) compares a linear GEE model with the best-fitting first and second degree models, each containing fractional polynomial terms (from a pre-defined set of integer, fractional and negative powers) for time. The differences in deviances between the linear and 1st-degree model and the first and second degree models are tested using a chi-squared test and the resulting P-values indicate whether the change in outcome over time is linear or whether it has a more complex shape. We inspected a plot of predicted values of fatigue against time from the model with the best fit to determine an appropriate follow-up interval in which observed fatigue scores could be assumed to be representative of the scores predicted at 12 months. In patients with more than one follow-up assessment, the closest to 12 months was used."
This post has been copied and discussion of this paper moved to:
Treatment outcome in adults with CFS: a prospective study in England based on the CFS/ME National Outcomes Database, 2013, Crawley et al
They have detailed below how they used a complex 'GEE model' in order to apparently adjust the scores where follow-up questionnaires were returned late - in order to be precise in working out what the 'fatigue level' would have been at 12 months. They say that some teams obtained data at additional follow-up points (6 and 24-months) so I guess they used these to back-fill the model. And I don't know either how many of the sample would have had this done nor how 'far off' they were 12months. But it seems a pretty astounding approach given they are using retrospective data, but on something that hasn't actually been researched before?? ie it isn't like they've got 25 trials worth of data covering all these time periods to show some curve.
Yet I can only assume they think they are being really diligent by doing this.
"Follow-up interval
Each team sent out follow-up questionnaires at 12 months. Variation in when the questionnaires were sent and delays in return of 12-month follow-up questionnaires led to variation in the exact time of follow-up. Also, some teams obtained data at additional follow-up points (e.g. 6 and 24 months). To maximize the availability of follow-up data for our analysis, we determined a margin of follow-up either side of 12 months. We did this by fitting fractional polynomial generalized estimating equation (GEE) regression models of fatigue against time.21,22 This method (implemented in Stata as fracpoly combined with xtgee) compares a linear GEE model with the best-fitting first and second degree models, each containing fractional polynomial terms (from a pre-defined set of integer, fractional and negative powers) for time. The differences in deviances between the linear and 1st-degree model and the first and second degree models are tested using a chi-squared test and the resulting P-values indicate whether the change in outcome over time is linear or whether it has a more complex shape. We inspected a plot of predicted values of fatigue against time from the model with the best fit to determine an appropriate follow-up interval in which observed fatigue scores could be assumed to be representative of the scores predicted at 12 months. In patients with more than one follow-up assessment, the closest to 12 months was used."
This post has been copied and discussion of this paper moved to:
Treatment outcome in adults with CFS: a prospective study in England based on the CFS/ME National Outcomes Database, 2013, Crawley et al
Last edited by a moderator: