Evaluating the Cost-Effectiveness of Proportional-Assist Ventilation Plus vs. Pressure Support Ventilation in the Intensive Care Unit in Two Countries

Background: Mechanical ventilation is an integral, but expensive, part of the intensive care unit (ICU). Optimal use of mechanical ventilation could save costs and improve patient outcomes. Here, the cost effectiveness of proportional assist ventilation (PAV™ ventilation by Medtronic) is estimated relative to pressure support ventilation (PSV). Methods: A cohort-level, clinical model was built using data from clinical trials. The model estimates patient-ventilator asynchrony >10%, tracheostomy, ventilator-associated pneumonia, other nosocomial infections, spontaneous breathing trial success, hypoxemia, and death. Cost and quality of life are associated with all events, with cost effectiveness defined as the cost per quality-adjusted life year (QALY) gained in the US and UK. Results: The mean cost of ICU care was lower with PAV™ than with PSV in the US and UK, but the total cost of care over 40 years was higher due to more patients surviving and incurring future care costs. Reduced time on mechanical ventilation, fewer nosocomial infections, and extended life expectancy with PAV™ drove QALY improvement. The cost per QALY gained with PAV™ was $8,628 and £2,985. Conclusion: PAV™ improves quality of life and reduces short-term costs. PAV™ is likely to be considered cost-effective over 40-years in the US and UK.


BACKGROUND
Mechanical ventilation via an endotracheal tube, is life-saving for patients with acute respiratory failure in the intensive care unit (ICU). As the most widely used supportive technique in the ICU (1), its patient benefit is generally accepted. It is, however, an invasive and expensive intervention. Uncomplicated mechanical ventilation in the US was found to have a mean cost of $59,770 per patient in 2009 USD (2). European studies have determined that the daily, direct costs of an ICU stay range between e1,168 to e2,025 (3,4), with UK costs reported at £1,738 in 2016 (5). The patient impact also cannot be ignored, with studies showing that patients in critical care have a negative (worse than death) quality of life (6). As it is expected that mechanical ventilation will remain an integral part of care for years to come (7), understanding the most beneficial modes of mechanical ventilation from both a patient and healthcare provider point of view is of increasing importance.
There are various modes of mechanical ventilation, which can be generally split into "controlled" and "assisted." With controlled modes the respiratory system is a passive structure and all breath characteristics depend on ventilator settings as determined by the caregiver and respiratory system mechanics (8). During assisted modes of support, the patient's control of breathing is under the influence of the ventilator pump and the ventilatory output is the final expression of the interaction between the ventilator and the patient's system of control of breathing (9). With common modes of assisted mechanical ventilation, such as volume-assist and pressure support ventilation (PSV), the algorithms used to deliver pressure and cycle off the ventilator are far from ideal and patientventilator asynchrony occurs regularly. This is not without consequences, since several studies have shown that increased asynchrony, particularly ineffective efforts, is associated with poor outcome (10)(11)(12)(13).
Proportional-assist ventilation with load-adjustable gain factors (PAV TM , Medtronic Inc.) is an assisted mode in which ventilator software measures, semi-continuously, elastance and resistance of the respiratory system and once triggered delivers pressure proportional to the instantaneous inspiratory flow and volume and, hence, to the inspiratory muscles' pressure (14). Several studies demonstrated that, compared to PSV, PAV TM improves patient-ventilator synchrony and unloads the respiratory muscles without the risk of over-assistance and periodic breathing (15)(16)(17). Whether or not PAV TM constitutes a cost-effective approach to care in the ICU remains, though, an open question.
The costs and patient outcomes, in particular mortality, related to mechanical ventilation are realized over years and decades (18,19). This makes an evaluation of cost-effectiveness FIGURE 1 | Impact of time horizon on model outcomes. The impact of the time horizon (X-axis, years) on model outcomes: total costs (primary Y-axis) and life expectancy (secondary Y-axis) is depicted for the US (A) and UK (costs converted to USD to allow comparison) (B). Life expectancy is shown in both life years and quality-adjusted life years (QALYs). All data are reported as PAV TM minus PSV. as part of a real-life study impractical. The analysis described here explores the question of cost-effectiveness using a computational model of the patient care pathway and is applied to the United States (US) and United Kingdom (UK) settings. The in-silico approach taken complies with good practice guidelines and the requirements of health technology assessment agencies (20,21).

METHODS
This analysis considers a cohort of patients receiving mechanical ventilation in the ICU and their progression through the care pathway to hospital discharge. Using clinical outcome data from randomized, controlled and large, prospective studies, an in-silico model of the care pathway was developed. Each clinical setting and event was associated with a cost and patient quality-of-life utility. Clinical outcome, cost, and quality-of-life data were sourced from a structured review of peer-reviewed literature. Searches of PubMed were performed on January 5, 2017 to identify recent data related to the critical care setting, healthcare costs, quality of life utilities, and efficacy and safety of PSV and PAV TM . The structured searches are available in the Supplementary Material. The developed Markov model ( Figure S1 in Supplementary Material) has patients ( Table 1) starting in the ICU and receiving either PSV or PAV TM . At initiation of mechanical ventilation, 38% of patients were found to have clusters of ineffective ventilatory efforts (or ventilator asynchronies), which were linked to a high asynchrony index (13). In 8.4% of cases, patients with clusters of ineffective efforts exhibited an asynchrony index >10% (13). In comparison, only 1.5% of patients without clusters of ineffective efforts had an asynchrony index >10% (13). Asynchrony >10% is associated with additional time in the ICU and increased risk of tracheotomy (11). At each time point in the model, patients are assessed for asynchrony ≤10%, asynchrony >10%, tracheostomy, ventilator-associated pneumonia (VAP), other nosocomial infection, and a spontaneous breathing trial (SBT). Patients who failed on a SBT remain on mechanical ventilation and may experience hypoxemia; whereas, if successful the patient may be weaned and is eligible for transfer to the general ward (17). Patients remain on the general ward until discharge home. Mortality rates in the ICU, in hospital, and after discharge home are taken from published literature specific to this patient population (18,19).
The relative efficacy of PAV TM compared with PSV was taken from published, randomized controlled trials (16,17). From their pilot study, Bosma et al. reported that PAV TM resulted in a significant reduction in time on mechanical ventilation and in the ICU, but was associated with an extended stay in hospital ( Table 2) (16). The authors also presented information on the number of patients experiencing asynchrony and receiving a tracheostomy tube, both of which were lower with PAV TM than with PSV (16). The study by Xirouchaki et al., determined that PAV TM was associated with increased probability on remaining on assisted modes (17). Other endpoints in this study were numerically lower with PAV TM but the difference did not reach significance (17). For each event, a cost (in 2015 currency units) and quality of life utility is applied ( Table 1). For example, VAP is known to increase the cost of care and the time on mechanical ventilation, adding about 11 days (2, 33), and likely resulting in reduced patient quality of life through increased time in the ICU. In addition, the interventions (PSV or PAV TM ) also have a cost, which were assumed to be equivalent at USD 27,000 or GBP 27,000 for purchase. Assuming a 5-year life cycle and usage of 80% of days, the cost per day for the each intervention was USD/GBP 18.48. The summation of costs and quality of life over the time horizon of the model allows for the cost effectiveness of mechanical ventilation methods to be assessed. The model was developed in line with ISPOR good practice guidelines and includes all relevant clinical aspects for which incidence data were available, irrespective of whether efficacy data for the comparators or any cost or quality of life data were available (34). The time line for analysis is 40 years, and provides a healthcare payer perspective of expected costs over the lifetime of the patient. All costs and utilities were discounted at 3.5% per annum after year 1. Results are reported as the cost per qualityadjusted life year (QALY) gained with PAV TM relative to PSV. To facilitate comparison, UK costs are also converted to 2015 USD using the mean annual exchange rate (1 USD = 0.68 GBP 1 ). The model outcome in the base case provides an estimate of the costs and QALYs associated with use of PSV and PAV TM . There is, though, inherent uncertainty about the parameters used as model inputs, which is often described via the standard deviation or 95% confidence interval. To understand how the uncertainty in model inputs influences outcomes, 2,000 patient cohorts were simulated using sampled age and gender distributions and run through the model. In each case, all model parameters (such as incidence of VAP and the relative risk of a successful SBT with PAV TM ) were also sampled from underlying distributions. Incidence data used normal, and relative risks log-normal, distributions. The 2,000 analyses were also used to estimate the cost-effectiveness plane and the likelihood of PAV TM being considered cost effective. The willingness-to-pay threshold, the maximum level at which a payer would consider an intervention to be cost effective, was taken to be the commonly accepted values of USD 50,000 and GBP 30,000 per QALY gained in the US and UK, respectively (35).
As model outcome data rarely follow a normal distribution, frequentist statistics and their calculated p-values are generally not appropriate measures of significance. As such, the 95% credible interval (CrI) for outcomes of the 2,000 simulations were also calculated. This is the outcome range, in which 95% of results fall (e.g., the bottom and top 2.5% of results are excluded). Although no direct relationship holds between the Bayesian CrI and the frequentist confidence interval, a 95% CrI that does not cross parity can be approximated to the same significance as a p-value of < 0.05.

RESULTS
Over 40 years in the base case compared with PSV, PAV TM was associated with increased life expectancy and quality-adjusted life expectancy ( Table 3). In the US setting, the increases were 1.58 years (10.53 vs. 12.11) and 0.79 QALYs (5.21 vs. 6.00), whereas in the UK setting PAV TM resulted in an extra 2.05 years (13.85 vs. 15.90) and 0.97 QALYs (6.52 vs. 7.48). The increase in QALYs with use of PAV TM will in part be due to fewer days on mechanical ventilation and a reduction in days with VAP (21.7 and 22.0% reduction in the US and UK, respectively). There were also fewer nosocomial infections and tracheostomies with PAV TM ( In varying the time horizon of the model, it was evident that results are sensitive to the time frame of the analysis (Figure 1). In both the US and the UK, PAV TM is associated with a lower cost of care in years 1 and 5. From year 10 onwards, PAV TM costs are higher than those with PSV. Given that all patients are in the home setting after 1 year, this indicates that it is the additional patient survival with PAV TM that is translating in to additional healthcare expenditure over the patient's life time. In terms of cost-effectiveness, PAV TM is dominant compared with PSV at a time horizon of 1 and 5 years, and would likely be considered cost-effective between time horizons of 10-40 years.

Probabilistic Sensitivity Analyses
In assessing the impact of parameter uncertainty on model findings, 2,000 simulations were performed with random input values sampled from underlying distributions. In the majority of cases, results of these simulations were aligned with the base case. In the US setting, PAV TM increased life expectancy and QALYs in 89. 40  Overall, PAV TM would be considered cost-effective in 82.95% of simulations at a willingness-to-pay threshold of USD 50,000 per QALY gained (Figure 2). For the UK setting, similar results were obtained. PAV TM would be considered cost effective in 88.65% of simulations (Figure 2) . As all CrIs at the 95%-level crossed zero, the differences are not considered significant. The trend in both countries is for improved patient outcomes with PAV TM , and this is reflected in the substantial percentage of simulations in both settings in which PAV TM would be considered cost effective.
The results of these analyses provide a consistent picture of PAV TM being considered a cost-effective option for mechanical ventilation in the ICU over a 40-year time horizon in both the US and UK. Input data and their associated uncertainties were taken from individual publications, and these could vary considerably between individual hospitals and ICUs. For example, VAP incidence varied from 7.3 to 20.7 events per 1,000 ventilator-days in the UK (23). To test the impact of such variation, a series of scenario analyses were performed. The results, presented in Table 4, demonstrate that in most scenarios the findings from the analysis remain unchanged. The items having the most impact were the age of the patient population and the relative risks (RRs) associated with use of PAV TM . If patients were aged 40 years on average, PAV TM remained cost-effective but the ICER increased. For patients aged 80 years on average, PAV TM dominated PSV in the UK setting and was very likely to be considered costeffective in the US setting; potentially because annual healthcare costs were accumulated for fewer years in this older population. Similarly, if the RR of events with PAV TM ( Table 2) was equal  to that with PSV (RR 1.0), then PAV TM dominated PSV. This outcome could also be effected by switching only the RR of mortality to 1.0, again indicating the impact of life expectancy and future annual healthcare costs on outcomes. This item is further highlighted by the results in which the future annual care costs themselves were halved or eliminated. In contrast, the costs of the interventions and the incidence of adverse events did not substantially alter the findings of this analysis.

DISCUSSION
Mechanical ventilation via endotracheal tube is recognized as an important but costly intervention that helps maintain life in the critical care setting. There are several modes of mechanical ventilation, but one of the most common is PSV (19). Recent randomized, controlled trials have demonstrated that PAV TM has advantages over PSV (16,17), but the question of costeffectiveness has never been addressed. Our analysis determined that in both the US and UK settings, PAV TM is likely to be a cost-effective mode of mechanical ventilation compared with PSV. In both countries, PAV TM was associated with increased patient life expectancy and QALYs, as well as increased total costs of care over a 40-year time horizon. If a shorter time perspective was taken, then PAV TM may also be considered a cost saving intervention. Our analyses identified that differences in future costs of healthcare between the two interventions, driven by increased life expectancy with PAV TM , was a key determinant of model outcomes.
As future annual care costs may be offset by insurance premium payments in the US and general taxation and national insurance payments in the UK, there is a reasonable question as to how these should be applied from a payer perspective. Guidelines from the National Institute for Health and Care Excellence (NICE) and Academy of Managed Care Pharmacy (AMCP) do not specify a time horizon to use (36,37), but note that the time horizon implemented should be sufficient to capture "the period over which the main differences in health effects and use of healthcare resources between interventions are expected to be experienced" (36). As all patients are off mechanical ventilation and out of the hospital setting at 1 year, a short time horizon may be appropriate for decision makers. In fact, guidelines for economic analyses in the Netherlands do note that a lifelong time horizon is often inappropriate for evaluation of medical devices (38). In this setting, the minimum time horizon for evaluation of cost impact is specified at 3 years (38). Given uncertainty in the most appropriate time horizon to use for evaluation of medical devices, results are presented here after differing periods of time (Figure 1). If shorter time horizons are most appropriate to medical device decision making then it is important to note that in both countries at 1 and 5 years, PAV TM dominates PSV. Furthermore, in scenario analyses running for 40 years, halving or eliminating these future annual care costs resulted in PAV TM being considered highly cost-effective or dominant, respectively. The results of the analyses were robust to changes in input parameters. The model is, however, only a representation of real life and based on average patients and mean outcome data. Clinical practice and even guidelines vary by country and hospital, the model cannot account for all these variations but aims to account for varying rates of events and intervention and levels of efficacy through sensitivity and scenario analyses. Results in individual hospitals would likely vary from those presented here because care practices, patient populations, and costs vary by institution. Furthermore, there is likely a learning curve to become familiar with the operation of PAV+ that has not and cannot (due to a lack of data) be explored in the model. The randomized, controlled trials informing this model were run by groups familiar with PAV+ and this may bias results relative to users new to the technology. Still, over a set of hospitals we would expect that results would be in line with those presented here. Probabilistic sensitivity analyses have shown that a payer would have a 95% chance of seeing a cost difference of between USD −17,813 and USD 47,194 in the US and GBP −7,709 [- $11,320] and GBP 18,985 [$27,878] in the UK, when a 40-year time horizon is used. The actual healthcare cost burden for a particular hospital using PAV TM will be dependent on local event rates and costs for devices, interventions, and personnel.
To fully validate the results of a model would require an extended, prospective cost collection study. This is beyond the realms of feasibility, and so model outputs are compared with previously reported values. In 2015, Hjelmgren et al. considered the cost consequence of neutrally-adjusted ventilatory assist (NAVA) mechanical ventilation relative to PSV in the Swedish setting (39). Their model varied substantially from that presented here, not considering future healthcare costs or any items outside of the ICU. Where comparable model outcomes were reported these were, however, in line with our analysis. Hjelmgren et al. reported a mean time on mechanical ventilation of 6.23-7.93 days (39), which is consistent with 6.77-6.97 days estimated here with use of PSV.
In 2012, Kollef et al. indicated that the cost of uncomplicated mechanical ventilation was USD 59,770 per patient (2). Over a single year in our model, PSV was at a cost of USD 60,008. This is closely aligned to the published value, and no cost data from Kollef (3). Converting this to GBP using the mean annual exchange rate for 2008 (1.259 EUR = 1 GBP 2 ), and then inflating it to 2015 pricing [using CPIH INDEX 06 (L528)] gave a value of GBP 1,940. The consistency between costing studies and our model results may help to validate the model structure and costing items included and provide more confidence in longer-term outcomes estimated.
Optimizing provision of mechanical ventilation is of high importance to healthcare providers. For payers, the costs of ICU care and mechanical ventilation in particular are a substantial burden on budgets. In general, an ICU accounts for fewer than 10% of hospital beds but over 20% of hospital costs (3). In a prospective study, ICU costs accounted for 64.7% of a patient's total hospital costs (40). The authors found, though, that per patient ICU costs reduced by 40% between 2008 and 2011, although the daily ICU cost decreased by only 3.3% in this time (40). Reducing the cost burden of ICU care is therefore possible without reducing direct costs, but rather through optimization of care. Multivariate analysis determined that fewer nosocomial infection and reduced ICU length of stay were two of six significant factors reducing ICU costs (40). Our analysis has demonstrated that use of PAV TM , relative to PSV, can reduce patient-ventilator asynchrony and decrease time on mechanical ventilation and in the ICU. Aligned with previous studies, this resulted in a lower cost burden for ICU care. Over the long term, general healthcare (non-ICU) costs may increase due to improved life expectancy.
From a provider perspective, use of PAV TM could result in lower resource use and improved patient comfort and quality of life. A key aspect of the PAV TM technology is the reduction in patient-ventilator asynchrony. Scenario analyses changing the relative level of asynchrony did not, however, have a large impact on cost-effectiveness outcomes. More pertinent were fewer days on mechanical ventilation and in the ICU, as reported by Bosma et al. (16) This was linked to other model results, such as lower incidence of nosocomial infections and reduced use of tracheostomy with PAV TM . As in the randomized, controlled trials reporting on these outcomes, these differences in the model did not reach significance. To confirm the potential cost and health benefits of PAV TM identified in this analysis would require large scale clinical and cost collection studies. To date, the two randomized, controlled trials of PAV TM have enrolled 262 patients and no costs have been reported (16,17). The implication for improved care they provide is important, particularly as larger studies may impinge on clinician duties and may not be feasible on the scale required. To this end, computational models are one option available to combine data from multiple sources and inform the debate on best practice in the ICU. The presented clinical model finds that PAV TM is likely to be considered cost-effective in the US and UK settings. The validity of the data is supported by the fact that the results are in line with published cost and outcome data not used to inform the model development.

CONCLUSION
The published reductions in asynchrony and length of stay in the ICU with PAV TM (16,17). led to shorter time on ventilation, and reduced incidence of VAP and tracheostomy in this decision analytic model. Increased patient survival with PAV TM resulted in annual healthcare costs being accumulated over a longer period. This made it most likely that PAV TM is cost-saving in the short term and cost-effective over the long term. For payers, PAV+ is likely a cost-effective option for mechanical ventilation when compared with PSV. For hospitals, PAV TM is expected to reduce costs and resource use.

AUTHOR CONTRIBUTIONS
DG provided information on the patient care pathway and clinical and economics outcomes to include. RS performed the literature search and review and created the initial model design. The model was finalized after review and updates from DG. RS ran analyses and sensitivity analyses and drafted the manuscript. Manuscript review and updates were performed by DG. Both authors met to approve the final manuscript for submission.

FUNDING
This study was funded by Medtronic.