Variation between practices in high-risk prescribing: multilevel modelling study of six indicators over five years in 190 practices

Talk Code: 

The problem

Prescribed drugs have large benefits but also cause significant amounts of harm, being responsible for 5-6% of UK emergency hospital admissions. There are a number of consensus-validated indicators of high-risk primary care prescribing, where there is reasonable evidence that the risk of harm to patients is significant. In previous work, we have shown that high-risk prescribing measured using a composite of 15 indicators is both common and varies four-fold between practices. The aims of this study were to examine whether (1) Practices which had above or below average on one high-risk prescribing indicator were above or below average on five other high-risk prescribing indicators; (2) Practices were consistently above or below average on all six high-risk prescribing indicators over time.

The approach

We defined six indicators of high-risk prescribing that could be consistently measured across the period of analysis 2005-2009: drugs to avoid in heart failure, high-risk NSAID use, methotrexate in mixed strengths, antipsychotics in dementia, beta-blockers in asthma, and long-acting beta-agonists (LABA) without inhaled corticosteroids (ICS) in asthma. Multilevel modelling was used to reliably estimate variation between practices for each of these measures. For objective 1, we estimated correlations between rates of all six high-risk prescribing indicators at practice level in quarter 1 2009. For objective 2, for each individual indicator we estimated correlations between rates in quarter 1 2005 and quarter 1 2009.


Objective 1: In quarter 1 2009 there were no consistent or strong correlations between practice rates on any one indicator and practice rates on any other (Pearson correlation coefficient range -0.140 to 0.212). Objective 2: Between 2005 and 2009, high-risk prescribing significantly reduced for high-risk NSAIDs (from 7.0% to 4.6%), methotrexate in mixed strengths (14.9% to 11.0%) and LABA without ICS (4.4% to 2.5%) and was unchanged for the other three indicators. For all six indicators, there were uniformly positive and statistically significant correlations between practice rates in 2005 and practice rates in 2009 (from a weak-to-moderate 0.414 for methotrexate in mixed strengths, to a strong 0.851 for high-risk NSAID use). For all indicators, practices were fairly consistent in the quartile of high-risk prescribing over time, irrespective of whether the mean rate of high-risk prescribing was falling or static.Consequence Practices in this study were not consistently higher or lower risk across indicators, but were remarkably stable in their ranking relative to other practices over time, even where average high-risk prescribing was steadily reducing. This suggests that there is no small group of particularly risky practices that could be targeted by prescribing safety interventions, but that such interventions should rather seek to influence prescribing behaviour across multiple indicators in all practices, since most will have room for improvement on some indicators.


  • Bruce Guthrie, NHS Tayside, Dundee, UK
  • Peter Donnan, NHS Tayside, Dundee, UK
  • Douglas Murphy, NHS Tayside, Dundee, UK
  • Tobias Dreischulte, IMS Health, London, UK
  • Ning Yu