This study showed that using a remifentanil-based sedation reduces costs compared to conventional sedation. The current economic evaluation is based on a model, allowing extrapolation beyond the moment of censoring (at 10 days MV) as it was defined in the clinical study. In that study, after three days still 59% of the patients in the CS group were intubated and 42% in the RS group, even though the intention was to include patients with an expected duration of ventilation of 24 to 72 hours. For this reason, also a subgroup analysis was included for patients that started weaning within the first 72 hours. The reduction of LOS and costs with remifentanil for the subgroup was similar to that for the whole population.
In the clinical study, after 10 days 21% of patients were still intubated in the CS group versus 8% in the RS group. It is clear that the lack of observation of time of extubation leads to uncertainty. We have, therefore, included an extensive probabilistic sensitivity analysis to address this uncertainty. From this we found that there is 79% certainty that using remifentanil-based sedation is indeed cost-saving when considering all patients, and 90% when limiting the analysis to the subgroup.
Commonly in drug trials, a 95% certainty is required before we allow a statement that a new treatment is more effective than the comparator. However, in health economics such strong risk aversion is not common. Some health economists have even argued that it is most rational to base the decision about introduction solely on the expected value [15
], but even for risk averse decision makers, 79% probability of being actually cost-saving will be judged as favourable. For comparison, a technology with a 50% probability that its total incremental cost-effectiveness ratio (ICER) is below £20,000 to £30,000/QALY is seen as cost-effective in the UK [16
]. Often an ICER of $50,000 US/QALY is also cited by health economics as threshold value - again, at a 50% likelihood [17
In our model we included the transition to death, based on the deaths that occurred in the clinical study. However, data on deaths were scarce, so, consequently, the transition probabilities to death are surrounded by much uncertainty. But since the death rate was the same in the two treatment groups, any changes in these probabilities will have a very limited effect on the outcome.
To date, no similar economic evaluations have been published. Most studies looking at the costs of remifentanil have focussed on the costs of sedation and analgesics of remifentanil and its alternatives. Only a few have addressed all relevant costs, for example, an open-label randomized study in Germany comparing remifentanil/propofol (n
= 39) versus midazolam/fentanyl (n
= 33) [18
]. The total costs for these groups were €1,712 versus €1,729. The additional costs of the remifentanil regime were compensated for by lower costs for physicians and nurses.
The perspective used for the cost calculations was that of the hospital. The question may be raised to what extend our results would change had we adopted a societal perspective. In this study, such a perspective would only have changed the drug costs. In the hospital perspective, the costs per mg are based on the prices hospitals pay after discounts. In the societal perspective, the prices before discounts should have been used. Since there was no discount for remifentanil versus large discounts for the other analgesics and sedatives, adopting a societal perspective would have been in favour of the RS arm.
Of course, the model outcomes can only be as good as the input. Since most of our input was derived from the UltiSAFE study, it is important to discuss its findings in relation to our results. The main result of UltiSAFE was that the treatment effect of remifentanil was time-dependent, that is, on Days 1 to 3 patients in the RS group were 1.86 times more likely to be extubated than in the CS group (P
= 0.018), while no difference was observed during Days 4 to 10 (rate ratio 0.98; P
= 0.951) [11
]. Though our study used Weibull time-to-event curves instead of Cox's proportional-hazards models as in the clinical study, we can still recognize this time-dependency in Figure . Here, the transition probability to start weaning is much higher for RS than CS during the first two days, while slightly lower after three days.
Additionally, the strengths and limitations of UltiSAFE should be mentioned as they also apply to the current study. UltiSAFE was not blinded, which may have caused biases. However, the purpose of the study was not to compare two treatments, but two rather different sedation regimens applied in 'real life'. While an inclusion criterion of the study was an anticipated two to three days of mechanical ventilation, after three more days 60% of the patients were intubated. This caused the study to be underpowered, which also impacts the degree of uncertainty around our model estimates of costs and LOS.
It is important to realize that the current cost-consequence study was limited to the parameters evaluated in the main study UltiSAFE. For example, ventilator associated pneumonia and other ICU acquired infections have not been taken into account in the current study.
From the literature it is clear that patients on mechanical ventilation are at an increased risk of developing pneumonia (VAP, ventilator-associated pneumonia). Due to the sample size of UltiSAFE, data on this adverse event were not collected as part of the clinical study on which we based our model input. VAP is the most prevalent infection acquired on ICU; the VAP frequency reported in various studies ranges from 8 to 28% [19
These studies also show varying results regarding the risk per day on MV. While one study showed a constant risk of 1% per day [20
], another study showed a decreasing hazard, going from 3% risk per day at Day 5, to 2% at Day 10 and 1% at Day 15 on MV [21
]. The differences in results can mostly be explained by differences in populations being studied [19
]. Due to this uncertainty, we opted for not considering these adverse events.
It is possible that inclusion of these might lead to averted cases of VAP in the RS group. Since various studies have reported that VAP increases LOS on MV and on the ICU [22
], excluding VAP in our model may have led to a conservative estimate of the cost savings due to remifentanil. However, it is possible that some patients in the UltiSAFE as well as in the micro-costing study suffered from VAP, and in that case, some of the increased costs and LOS due to VAP may have been implicitly included in our results.
Conversely, inclusion of ICU acquired infections could also have a negative impact on the cost-consequences of remifentanil. Recently a retrospective case-control study was published that showed that remifentanil discontinuation is an independent predictor of ICU acquired infections [24
]. This mechanism has been found in animal studies where morphine withdrawal caused immunosuppression resulting in an increased risk of infection [25
]. However, it is not clear to what extent this is remifentanil-specific. Given that the animal studies involved morphine it seems likely that ICU patients receiving morphine are also at risk for post-discontinuation infections. Whether this is also true for patients receiving fentanyl is unknown at this time.
Overall we can conclude that more data are needed in order to incorporate ICU acquired infections (including VAP) into a cost-consequence analysis of remifentanil.
Another parameter that was not explicitly studied in the UltiSAFE is the occurrence of acute withdrawal syndrome after opioid discontinuation. In the literature the occurrence of withdrawal syndrome has been reported, though little is known about the frequency of withdrawal syndrome [27
]. In one retrospective study, patients with an ICU stay of more than seven days were studied for the occurrence of withdrawal syndrome [27
]. Of the 28 patients included, 32% developed this syndrome. There was no difference between patients receiving fentanyl or morphine. No studies have been done in patients with shorter ICU stays. In the UltiSAFE study, withdrawal syndrome was reported in only one patient, in the RS group. Clearly larger studies are required to come to a meaningful conclusion on what the probability of withdrawal syndrome is after opioid-discontinuation and whether this probability differs between remifentanil and other opioids.
Additionally, pain resulting from remifentanil discontinuation has been reported [29
]. However, this was in the context of a double blind controlled trial of remifentanil versus fentanyl for analgesia based sedation in the ICU. In that study remifentanil patients who experienced pain did so for significantly longer during extubation, post-extubation and post-treatment. This is explained by the rapid offset of the analgesic. However, the authors suggest that in clinical practice, where the clinician is aware of this issue, proactive pain management can avoid this problem. In the UltiSAFE study, three patients in the RS group received morphine or fentanyl during the weaning phase, so it is likely that the costs associated with pain after remifentanil discontinuation are, at least to some extent, already incorporated in our data.
One of the issues with any clinical and cost-effectiveness study is that of generalisability of results to other countries. The UltiSAFE clinical study was performed in Dutch hospitals, and the control group was treated according to Dutch guidelines. As a result, a variety of drugs was used in this group, which may not all be used in other countries. For example, the proportion of patients treated with fentanyl versus morphine can vary. If the reduction in MV found in the UltiSAFE study is explained by a relatively large proportion of patients in the control group using morphine (59%), then the projected cost-savings may not be achieved in a setting where the current analgesic of choice is fentanyl.
However, in other countries studies have been performed with remifentanil in ICU patients where the control group was treated differently than the UltiSAFE.
In a randomized, open-label study remifentanil plus midazolam (n
= 57) was compared to midazolam plus fentanyl or morphine (n
= 48) in ICU patients expected to require MV for at least four days [30
]. In the control group, 62% of patients received midazolam with fentanyl, 15% received midazolam with morphine and 23% received midazolam alone. In this study the time on MV was, on average, 147 hours in the comparator group versus 94 hours in the remifentanil group, that is, a reduction of 53 hours (36%, P
Another randomized, open label study compared remifentanil plus propofol (n
= 39) to midazolam plus fentanyl (n
= 33) in post-operative ICU patients expected to require MV for 12 to 72 hours [18
]. The time on MV was, on average, 24.2 hours in the control group versus 20.7 hours in the remifentanil group, that is, a reduction of 3.5 hours (14%, P
While the patient populations in these studies and the UltiSAFE study are not fully comparable, especially with regards to the expected duration of MV at the time of inclusion, all studies show a clear reduction in MV time in the remifentanil group, both when only fentanyl was used in the control group and when both fentanyl and morphine could be used. Thus, it seems that the impact of different sedation/analgesic regimes on the reduction of MV and thus potential cost-savings is limited.
Finally, we would like to point out that the estimated savings of remifentanil-based sedation represent potential savings: Only if the hospital can use the freed resources (staff and increased ICU-capacity) it can exploit this potential. Furthermore, this analysis was performed in The Netherlands and its results cannot be directly transferred to other countries without necessary adjustments, for example, for country specific relative costs.