This study examined the cost-effectiveness of a primary care intervention to prevent T2DM and CVD from a societal perspective, compared with provision of health brochures. Small, statistically non-significant differences in risk scores and QALYs gained were found between the intervention and comparison group. The mean difference in costs was €-866 (95% CI -2372 to 370). The probability of cost-effectiveness for health risk modification was 90% at a ceiling ratio of €0. Increasing the ceiling ratio resulted in a slight drop in the probability of cost-effectiveness for risk of CVD, but severely for risk of T2DM. The probability of cost-utility of the intervention, compared with the brochures group, was around 90% at all ceiling ratios. The results were confirmed in the sensitivity analyses, except in the complete case analysis. This showed that the intervention was not more likely to be cost-effective for reducing health risks than provision with health brochures. Furthermore, the probability of cost-utility reduced to a maximum of 77%.
The results indicate that there were no statistically significant differences between the intervention and control group with regard to the clinical outcomes. Baseline health utilities were already high, which implied that there was little room for improvement in QALYs gained. High baseline utilities and lack of improvement might be related to the use of the EQ-5D-3L to measure health status. This instrument is known for its ceiling effects in relatively healthy populations. Other instruments, such as the recently developed EQ-5D-5L might resolve this problem, but this is still under research . Lack of differential effect on risk profile could be related to the low attendance to the counseling sessions. On average only 2.4 of the 6 available face-to-face counseling sessions had taken place, whereas the mean use of the 3-monthly phone-calls was 2.3. These results underscore the difficulties in translating efficacious methods to interventions that are feasible and effective in real-world settings.
Cost-differences were in favor of the intervention albeit statistically not significant. Due to the skewed nature of the cost data, the study may have been underpowered to reach statistical significance for cost differences. The difference in total costs was mostly explained by differences in costs of productivity losses. To explore the effect of outliers on these costs, a post-hoc analysis was done. In this, the number of sick leave days was truncated at 30 days, as it is improbable that sick leave over 30 days would have been influenced by the intervention. The cost difference reduced to €-179 (95% CI -725 to 311). However, a reduction in societal costs remained.
To our knowledge, no other studies of lifestyle interventions to prevent T2DM or CVD have found immediate cost reductions as a consequence of the program, in the absence of health effects . Some researchers have suggested that health promotion programs have non-health benefits that are currently not measured, such as increased health literacy . These benefits may have a direct influence on health care use and sickness absenteeism . The reduction in costs of personal expenses, mainly consisting of sports costs, is puzzling. Because the intervention participants were stimulated to be more physically active, higher costs were expected. All in all, the finding of possible cost savings in favor of the intervention cannot be easily explained.
The main aim of an economic evaluation is to decide whether the treatment under scrutiny offers value for money. Ceiling ratios for reductions in risk of CVD or T2DM are not determined. The ceiling ratio for QALYs gained in the Netherlands is also not established, but it has been proposed to be set at €8000 for diseases with a low burden and at €80,000 for diseases with a high burden . The immediate burden of disease of an elevated risk for T2DM and CVD is unknown, but likely lies at the lower end of the range. At a ceiling ratio of €8,000 per QALY gained, the probability of cost-utility of the intervention was 93%. At the higher end, probability of cost-utility was 89% at a ceiling ratio of €80,000 per QALY gained. The intervention therefore has a high probability of cost-utility at all acceptable ceiling ratios. However, there is methodological uncertainty regarding this probability. The complete case analyses showed that the probability that the true cost-utility ratio falls below €80,000 is 75%. The analysis from the health care perspective also showed a lower probability of cost-effectiveness. The post-hoc analysis in which the number of sick leave days was truncated at 30 days, showed a maximum probability of 70% at a ceiling ratio of €0. Thus, in these sensitivity analyses, cost-utility results were less positive. Furthermore, people generally demand larger compensations for losses compared with how much they are willing to pay for gains (willingness to pay < willingness to accept). If this aversion to loss is applicable to losses in health, a lower probability of cost-utility would have been found . Finally, as explained before, questions remain about the cause of the cost-differences. In light of these uncertainties, it is unsure if the possible benefits outweigh the efforts involved in the implementation of this new intervention.
Limitations and strengths
A limitation of the study is the amount of incomplete cost data. Intervention costs were missing for 140/314 (45%) participants, mainly because the practice nurses failed to report the number of phone sessions to the research team. However, a more complete report may be difficult to achieve in real-world settings. In addition, 36% of the self-reported data on health care utilization, personal spending and sick leave was missing. This is comparable to other RCTs in the Netherlands with a follow-up of one year or longer [32, 33]. Studies in the Netherlands that need individual data on health care use have to rely on self-report because it is not feasible to collect these data from health insurers. Nevertheless, methods to improve data-completeness in studies with many self-reported cost-measurements should be devised to increase the internal validity of these studies.
Over one-third of all cost data were imputed, using multiple imputation techniques. In multiple imputation it is assumed that unobserved data are (in part) dependant on the observed data (e.g. available costs). However, this assumption cannot be fully tested. Although methodological studies show that multiple imputation is preferred over complete case analyses and simple imputation methods results from this study should be treated with some caution.
The use of ‘projected’ risk scores for T2DM and CVD was useful to accentuate absolute risk, but should not be considered as current absolute risk. For both risk scores and for each participant, age was extrapolated to 60 years while all other variables in the risk scores (e.g. blood pressure or cholesterol) were absolute scores. Thus, the scores estimated the risk of participants as if they were 60 years old, but with current values.
Lastly, the time horizon of the study was too short to observe the development of T2DM or CVDs, and to measure their associated changes in QALYs achieved. Decision modeling could be used to extend the time horizon of the study. This would give more insight in the longer term cost-effectiveness, and would increase the comparability of the results to those from other studies. Decision modeling also has the advantage that the results of similar interventions could be included, thereby broadening the evidence base . This would however have to be done in a separate study.
Strengths of the study include the relatively long-term follow-up in terms of intermediate outcomes, the use of multiple imputation to address the large amount of missing cost data, the use of both the friction cost approach and the human capital approach to value productivity losses, and the application of a randomized controlled design in a real-world setting.