Pulmonary artery pressure technologies for remote monitoring of chronic heart failure

The committee discussed the evidence base for this evaluation, which consisted of:

• 3 randomised controlled trials (RCTs) comparing CardioMEMS with usual care (CHAMPION, GUIDE‑HF and MONITOR‑HF)

• 3 single-arm prospective studies for CardioMEMS, included for device-related outcomes only (COAST, MEMS‑UK and CardioMEMS‑PAS)

• 2 studies on patient experience of CardioMEMS (Assaad et al. 2018 and Haynes et al. 2020)

• 3 prospective single-arm studies for Cordella (SIRONA, SIRONA 2 and PROACTIVE‑HF).
  
  PROACTIVE‑HF was originally designed as an RCT and changed to a single-arm study part way through. Some data from the randomised phase of PROACTIVE‑HF was included in the evaluation. Two studies included patient survey results that contributed data on patient experience and satisfaction (PROACTIVE‑HF and SIRONA 2).

The sample sizes of the studies included for the main review of clinical effectiveness outcomes ranged from 15 people (in SIRONA) to 1,000 people (in GUIDE‑HF). The mean or median age of people in the studies ranged from 61 to 71 years across the studies. This is younger than the average age of people with first heart failure diagnosis in the UK, which is 77 years. There was a high proportion of men in the studies (for example, over 70% of people in CHAMPION were men). One of the clinical experts explained that there is no reason to believe that the technology would be less beneficial for women, trans and non-binary people.

The committee concluded that the study populations were less diverse than the real-world population of people with heart failure in terms of gender, age and ethnicity. The committee also noted that the technologies could have been more effective in the studies than they will be in routine clinical practice because of the younger age of the study participants.

One clinical expert explained that pharmacological management of chronic heart failure has changed since some of the studies were carried out, particularly for HFpEF. Previously, treatment for HFpEF relied on diuretics alone, but current practice includes mineralocorticoid receptor antagonists and sodium-glucose cotransporter 2 inhibitors in addition to diuretics. Most of the studies were done recently and reflect contemporary treatment of chronic heart failure. CHAMPION began enrolment in 2007 and reflected HFpEF treatment at the time. The committee concluded that this did not affect the interpretation of the results for this evaluation. This was because most people with HFpEF in the overall analysis of all the studies would have had treatment in line with contemporary treatment.

All-cause mortality was evaluated in the 3 RCTs on CardioMEMS and in the comparative and single-arm phases of PROACTIVE‑HF and SIRONA 2 on Cordella. The RCT results suggested a small decrease in mortality with CardioMEMS. For Cordella, data provided by its manufacturer for the comparative phase of PROACTIVE‑HF suggested a small decrease in all-cause mortality. Confidence intervals were wide and consistent with both an increased and decreased risk of death for CardioMEMS (HR 0.91, 95% CI 0.70 to 1.17) and for Cordella (HR 0.51, 95% CI 0.20 to 1.32). Indirect comparison of Cordella with CardioMEMS suggested no evidence of a difference in all-cause mortality between the 2 technologies. But the estimate was very imprecise. The committee concluded that it was uncertain whether either technology reduced all-cause mortality from the evidence available.

All the studies provided data on health-related quality of life (HRQoL). This was measured using the EQ‑5D‑5L visual analogue scale, the Kansas City Cardiomyopathy Questionnaire (KCCQ) or the Minnesota Living with Heart Failure questionnaire. The impact of PAP monitoring technologies on HRQoL was mixed. For CardioMEMS, the studies had results in the opposite direction, so the findings were inconsistent. MONITOR‑HF reported an increase in EQ‑5D‑5L score and GUIDE‑HF reported a decrease. The summary estimate from the meta-analysis was a mean non-statistically significant increase in EQ‑5D score of 1.75 EQ‑5D (95% CI -6.03 to 9.53; a higher score is better). So, the summary estimate was of limited value. The overall result of the meta-analysis for the KCCQ was a mean non-statistically significant increase in KCCQ of 3.63 (95% CI -2.24 to 9.47; a higher score is better). For Cordella, data on HRQoL was limited and lacked direct comparisons, making it difficult to draw conclusions. The committee thought that the evidence for HRQoL and mortality was inconclusive for both technologies. It concluded that it was uncertain whether either technology improved HRQoL.

Data from the quantitative studies included in this evaluation showed that:

• Implantation failure was rare. Implantation of the technology failed in 1.7% of people (95% CI 0.8% to 2.9%) with CardioMEMS and in 4.9% people (95% CI 3.1% to 7.0%) with Cordella.

• Device or system, or procedure-related complications were rare. The summary proportion of these complications across trials was 0.7% (95% CI 0.3 to 1.3%) with CardioMEMS and 0.1% (95% CI 0.0 to 0.09%) with Cordella.

• Data from all quantitative studies except GUIDE‑HF showed that the proportion of people with a sensor implanted in whom the sensor subsequently failed was low (0 to 1.2% overall). The summary estimate was 0.1% (95% CI 0.0 to 0.6%) for CardioMEMS and 0% (95% CI 0.0 to 0.1%) for Cordella. GUIDE‑HF did not report sensor failure.
  
  The clinical experts explained that failure after implantation was rare and that, when implantation failure happens, it does not lead to products being discarded. The committee concluded that the technologies are safe to use and have an acceptable failure rate.

To benefit from using a PAP technology, people need to be comfortable with:

• having a general anaesthetic and a right heart catheterisation procedure

• taking routine measurements at home

• acting on any changes to their medication that are needed because of PAP changes.
  
  Qualitative data included in this evaluation suggested that people were positive about using the technologies and that the technologies improved their understanding of their condition. The clinical experts said that people generally feel comfortable living with an implanted sensor and some people may feel reassured by knowing that their heart failure is being continuously monitored. But other people could feel uncomfortable with having a sensor implanted, living with it and find the monitoring requirements to be a burden. The person's ability to adhere to using the technology and medication changes, and their comfort with living with the implanted sensor, would need to be considered as part of shared decision-making.
  
  Other options for remote monitoring of chronic heart failure are available, including algorithm-based remote monitoring systems, which use data from cardiac implantable devices, and virtual wards. Patient selection would also need to consider which methods of remote monitoring are available and would be most suitable for the patient. The committee concluded that patient selection would need to be carefully considered.

The EAG developed a multistate Markov model to estimate the cost effectiveness of remote PAP monitoring technologies compared with current monitoring practice (standard care). People with NYHA class 3 heart failure entered the model in a stable state, reflecting the health state for people whose condition has stabilised following an index admission. The model consisted of 8 mutually exclusive states (stable heart failure 1, first recurrent heart failure hospitalisation, stable heart failure 2, second recurrent heart failure hospitalisation, stable heart failure 3, subsequent recurrent heart failure hospitalisation, stable heart failure 4, and death). The model used a cycle length of 1 month and a lifetime time horizon.

The key clinical input in the model was heart failure hospitalisation rates. In the model, the risk of having a subsequent heart failure hospitalisation increased with the number of previous heart failure hospitalisations. The effect of PAP monitoring was to reduce the rate of heart failure hospitalisations, and this effect was assumed to continue regardless of how many previous heart failure hospitalisations a patient had had. The risk of death increased with each stable state, and the utility associated with each stable state declined as heart failure hospitalisations increase. The committee agreed that the design of the model was appropriate.

The committee discussed the key drivers of the model. One way sensitivity analysis showed that the effectiveness of CardioMEMS at reducing heart failure hospitalisations had the largest impact on the incremental cost-effectiveness ratio (ICER). At the lower bound of the CardioMEMS hazard ratio, the ICER was reduced by over £17,000 per quality-adjusted life year (QALY). At the upper bound, the ICER was increased by over £40,000. All other variables included in the deterministic sensitivity analysis had a less than £10,000 impact on the ICER. The implant failure rate had the next biggest impact on the ICER, followed by the costs for third or subsequent recurrent heart failure hospitalisations. The committee concluded that heart failure hospitalisations were a key driver of the model.

Heart failure hospitalisation rates were calculated using data from 2 studies done in the UK. The Lahoz et al. (2020) study reported data from 8,603 people with heart failure who had already had an index heart failure hospitalisation from the UK Clinical Practice Research Datalink. Lahoz et al. reported the median number of days to the next heart failure hospitalisation. From this, the EAG calculated the hazard ratio for a recurrent event compared with having no recurrent heart failure hospitalisations after the index heart failure hospitalisation. The EAG used data from COAST to estimate the recurrent heart failure hospitalisation rate in the first stable heart failure state in the Markov model. It did this using data from Lahoz et al. to adjust for the fact that the COAST data included people with multiple previous heart failure hospitalisations. Data in Lahoz et al. was not reported according to NYHA class. People with NYHA class 3 heart failure were more likely to be admitted to hospital than people with NYHA class 1 or 2. Lahoz et al. data was used to adjust the COAST data to account for the number of previous heart failure hospitalisations. So, the committee concluded that it was possible that heart failure hospitalisation rate may have been underestimated or overestimated. But it also noted that this was uncertain based on the data available.

In the EAG's base case, the utility value for the initial stable heart failure state was based on a published meta-analysis of utilities for people with heart failure (Santos et al. 2024). Reductions in utility were made for the second, third and fourth stable heart failure states using data from Gohler et al. (2009). As people progressed through the subsequent stable heart failure states, their utility declined. The clinical experts agreed that utility declines with each hospital admission. Utility data from the EAG's meta-analysis of trials of the technologies included in this evaluation were not used in the base case. This was because of a high degree of uncertainty in the results.

Alternative scenarios for capturing utilities in the model were considered, in which utility outcome data from the trials was directly applied in the model. In scenario 6a, data was applied from CHAMPION. The EAG had some concerns about using CHAMPION data because of how old it was and because the trial was done in the US. In scenario 6b, data was applied from MONITOR‑HF. MONITOR-HF was done in the Netherlands, so was preferred by the EAG because the data was from a European population. But the EAG had concerns about using data from MONITOR‑HF because it took place during the COVID‑19 pandemic lockdown. This could have affected the generalisability of the results to usual UK practice. In scenario 6c, utilities were used from MONITOR‑HF for the first 12 months, and health-state based utilities were used for extrapolation beyond 12 months. This scenario analysis resulted in an ICER of £36,000 per QALY gained for CardioMEMS. The committee concluded that scenario 6c was the most plausible of the analyses considered.

The clinical experts explained that the frequency of monitoring is likely to reduce after the initial implant. So, the frequency of 3 times per week used in the model over the lifetime was likely an overestimate. The technologies can be set with parameters specific to the people using them. Also, they can alert people using them and their healthcare professionals when measurements fall outside of range. The model was based on a band 5 nurse monitoring PAP data. But a band 5 nurse would not be able to prescribe medications if the data shows that a change in medication is needed to avoid decompensation of heart failure. A doctor or nurse prescriber would need to prescribe the medication. Five minutes per month of a medical consultant was included in the base case. This may have been insufficient if a doctor or nurse prescriber is needed to oversee medication changes and prescribe medication. So, the committee thought that the band 5 nurse costs in the model were likely to have been an underestimate.

The committee agreed that the appropriate monitoring schedule would be to calibrate the technology after sensor implantation, then monitor weekly for the weeks 1, 2 and 3, then monitor every 3 months. The monitoring schedule would be repeated from the start if heart failure worsens. The committee thought that adjusting the monitoring costs in the model would be unlikely to reduce the ICER sufficiently for CardioMEMS to fall within the range that NICE considers a cost-effective use of NHS resources.

Primary care resource use was not included in the model because there was no data on it. The clinical experts advised that, in the real world, people using the technology would remain under secondary care. The committee agreed that the omission of primary care resource use was not a limitation.

The results of the economic model suggested that the ICER for CardioMEMS in the probabilistic base-case analysis was £41,878 per QALY gained. This is above the range that NICE considers an acceptable use of NHS resource. All iterations of the probabilistic sensitivity analysis resulted in more health at a higher cost for CardioMEMS. The committee concluded that CardioMEMS was not cost effective at its current price. There were also some areas of uncertainty in the model for CardioMEMS, including that:

• monitoring of PAP data was more frequent in the trials than it would be in clinical practice

• reducing the frequency of monitoring in the model may reduce costs.
  
  The committee agreed that there was evidence that CardioMEMS reduced heart failure hospitalisations but that there was uncertainty about the size of the effect. The committee agreed that the clinical effectiveness was likely greater in the trials than in the real-world clinical setting. This was because:

• of the difference in the mean age of the trial population and that of the real-world heart failure population

• there have been changes in the clinical management of heart failure since some of the trials were done

• the intensive monitoring in the trials may have increased the size of the effect.
  
  The committee concluded that, given the cost of CardioMEMS, accounting for these uncertainties in the model would be insufficient to make it cost effective.

The ICER for Cordella could not be estimated because its cost was unknown. In the model, the cost of Cordella was assumed to be the same as the cost of CardioMEMS. This assumption was made to enable Cordella to be included in the model and was for illustrative purposes only. The committee concluded that it was not possible to estimate the cost effectiveness of Cordella.

The committee concluded that the base case in this evaluation must:

• adopt the approach to utilities used in scenario 6c (that is, use utilities from MONITOR‑HF for the first 12 months and use health-state based utilities for extrapolation of utilities beyond 12 months)

• use the cost of monitoring based on the monitoring schedule that was considered plausible by the committee (see section 3.22).

• use the cost of monitoring based on the salary for a healthcare professional who can prescribe medication when needed, in response to changes in PAP measurement.

The committee was aware that a model was developed for the Health Technology Wales appraisal of percutaneous implantation of PAP sensors for monitoring treatment of chronic heart failure. The ICER in the base case was lower than the ICER in the EAG's model. The committee discussed the key differences between the models and agreed that it preferred the model developed by the EAG. In the EAG's model, the disutility associated with hospitalisation for heart failure stopped after 1 month, but the disutility lasted for 1 year in the Health Technology Wales model. The committee agreed that 1 month was a more plausible timeframe for disutility because of hospitalisation and immediate recovery. In the EAG's model, the utility associated with each stable heart failure state decreased with each subsequent stable state. The committee agreed that this was clinically appropriate. Also, in the Health Technology Wales model, monitoring costs were based on a band 5 nurse monitoring PAP data. The committee recalled its earlier conclusion that a healthcare professional who can prescribe would need to do the monitoring.