Publications.Thesis Index.donnelly_mph

MPH Thesis - Donnelly, N

Analysis issues in a general practitioner based randomised controlled trial
Donnelly N (1993)

Abstract:
General practice based interventions can provide data on both practitioner and patient (encounter) variables. Both these units of analysis were used in the evaluation of a general practitioner based RCT aimed at modifying benzodiazepine prescribing for anxiety and insomnia.

With the GP as the unit of analysis, the outcome of interest was the rate of benzodiazepine prescriptions. Two alternative approaches for the analysis of change in prescription rates via linear regression were compared  (i) the change score approach in which the outcome was the difference between the pre and post rates and (ii) the regressor variable method in which the post rate was regressed on the pre rate and other covariates. The two approaches gave consistent results, though in this example use of the change score better satisfied distributional assumptions.

The patient encounter was the more appropriate unit of analysis when assessing change in insomnia and anxiety specific benzodiazepine prescriptions. However here adjustment may need to be made for the cluster randomisaton were the assumption of independence violated. A method proposed by Rao and Scott (1992) was used to provide estimates of intra-cluster dependency. Resulting variance inflation factors were fond to vary by both outcome and diagnosis subgroup.

The results of logistic regression (ignoring the clustering) were compared with those of binomial generalised estimating equations (GEE) which took the clustering into account. Overall there were no major differences between the logistic regression and GEE models once important covariates had been adjusted for , though the standard errors for the GEE models tended to be larger than for the logistic regressions. Adjustment for important covariates reduced the estimated intra-class correlations, though this was more pronounced among the insomnia diagnosis subgroup.

<back to thesis index>