Modelling heterogeneous data: challenges from the cost effectiveness of acupuncture for chronic pain case study

Abstract: There is lack of clarity about the effectiveness and cost-effectiveness of acupuncture for the treatment of chronic non-cancer pain in primary care. Individual patient level data (IPD) from 28 high quality RCTs were available which evaluated acupuncture and relevant comparators in reducing pain in 3 different conditions comprising head, musculoskeletal and knee osteoarthritis (OAK). The trials used a variety of health-related quality of life (HRQoL) instruments including SF-36, SF-12, VAS pain, CMS and WOMAC. Additionally, a systematic review was undertaken, specifically for OAK treatments (inc. acupuncture), which found numerous relevant studies of heterogeneous quality and heterogeneous pain measurement reporting.

This on-going project aims to generate homogenous HRQoL scores and pain outcomes, to undertake the synthesis of effectiveness evidence using both IPD and of AD + IPD (in a continuous outcome context) and to undertake a cost-effectiveness analysis of acupuncture in each of the three pain conditions.

In fulfilling these aims a number of methodological challenges have been encountered, some of which will be the focus of the seminar discussions. First, to deal with multiple HRQoL evidence this study makes use of a variety of published and unpublished approaches to mapping to predict the target preference-based measure (EQ-5D) for all IPD studies. Second, to make best use of the available data, a mixed treatment comparison (MTC) synthesis model is developed to combine continuous EQ-5D outcome data using IPD. Third, to consider the inclusion of baseline characteristics and to assess the impact of potential treatment effect modifiers. Fourth, to check the
consistency of the evidence networks. Fifth, to incorporate SMD summary evidence in the synthesis modelling for OAK. Sixth, to incorporate the very limited resource use data available in the modelling. Seventh, to utilise an appropriate time horizon of the analysis and eighth, to consider the placebo effect, making use of the results of the sham intervention.