Previous studies have shown that the majority of costs associated with chronic fatigue are hidden or indirect, such as informal care and lost productivity
. Nevertheless, in the current context of cost containment and efficiency savings, a narrower focus on healthcare costs as the one adopted in the present study is still relevant.
The analysis of the outcome results at six months did not find statistically significant differences between the three treatments after adjusting by baselines characteristics. Therefore, the superiority of any treatment in clinical terms cannot be established. On the other hand, there were statistically significant differences between the costs for the three interventions.
The combination of outcomes and costs shows that counselling is dominated by the other two options. Therefore, it might be the case that the choice is reduced to that between GET and BUC. GET is more expensive and more effective and, therefore, it is a value judgement as to whether this represents value for money. The CEAC illustrates that the probability that GET is the most cost-effective intervention does not reach a convincing level even when a high level is placed on a clinically significant improvement. If society values such an outcome at £2500 then there is over a 60% likelihood that GET is cost-effective. This uncertainty has been found in previous studies that have analysed the cost-effectiveness of these or alternative interventions
[5, 6]. Chisholm et al. found that counselling was cheaper than cognitive behavioural therapy (CBT) while in the study by McCrone et al., CBT was also more expensive than the other active therapy (graded exercise). Nevertheless, a clear result as to which intervention was more cost-effective was not reported. These studies were also conducted in primary care with a design very similar to the one presented here although the number of participants in these studies were slightly lower.
A more comprehensive analysis of outcomes trial data
, including the analysis of effectiveness data at 12 months, found that there were no significant differences in change scores between the three groups at the 6- or 12-month assessments. This analysis also found that, compared to BUC, those treated with graded exercise or counselling therapies were more satisfied at 1 year. Patients with fatigue have reported low levels of satisfaction with diagnosis and management in primary care. Satisfaction is an important and remunerated outcome in the National Health Service, but it is not clear how this can be assessed against increased cost.
First, data collected under experimental conditions may be different from data collected under routine conditions. However, we believe that such bias is minimal in the present study given that patients were drawn from the same practices and the same instruments were used for assessment. Second, reliance on patient self-report might result in some inaccuracies in service use measures. While the schedule used is well developed and has been used in numerous other studies, it may still be the case that for some patients recall was difficult and this would have led to inaccuracies. A number of studies have though suggested that patient recall of service use can be acceptable
[23–25]. Another study noted a difference between self-report and administrative records, but pointed out that it was unclear which was more accurate
. Third, a unit change on the fatigue scale is difficult to interpret. This is addressed to some extent by using a four-point change to represent clinical significance. However, this is in itself somewhat arbitrary and still presents a challenge for interpretation by those not closely involved in the care of this patient group. Indeed, the self-reported nature of the Chalder Fatigue Scale could prove problematic as it requires respondents to compare themselves with how they were previously (termed as usual) and they may be unsure as to this. The duration of the follow-up period also imposes limitations to the significance of the results of the study. Availability of economic data for the follow-up of a year would have increased the validity and relevance of the results. Our analysis would also benefit from a wider perspective that includes informal care and productivity costs. It was not feasible to use quality-adjusted life years (QALYs) as an outcome measure, as planned in the trial protocol, and this is also a limitation. This was caused by unexpected unavailability of EQ-5D questionnaire data for the six months follow-up. Finally, there were incomplete data for 27 % of the individuals who were randomised. It was decided not to conduct imputations given that there were limited data to do this.