The evidence provided by the company comprised an initial submission, an economic model (which is commercial in confidence) and the company’s response to the EAG’s clarification requests [3]. The EAG report comprised a summary and critical review of the clinical and cost-effectiveness evidence provided by the company.

3.1 Clinical Evidence

The primary source of clinical evidence presented by the company was the phase III, open-label, single-arm, multicentre SPR1NT trial, which assessed the efficacy of onasemnogene abeparvovec as a treatment for patients with pre-symptomatic SMA and two (n = 14) [17] or three (n = 15) [18] copies of the SMN2 gene. The SMN2 two-copy and SMN2 three-copy cohorts had different primary and secondary efficacy endpoints and lengths of follow-up in the trial.

In the SPR1NT trial two-copy SMN2 cohort [17], all 14 patients met the primary endpoint of functional independent sitting (as defined by the Bayley Scales of Infant and Toddler Development Gross Motor [BSID GM] subtest item number 26) at any visit up to age 18 months, and the secondary endpoint of event-free survival (i.e. survival without the need for permanent ventilation) at age 14 months. The majority (11/14, 78.6%) of patients achieved the primary endpoint within the normal development window (as defined by the World Health Organization Multicentre Growth Reference Study [WHO-MGRS]) [19]. All except one patient (13/14, 92.9%) met the secondary endpoint of weight maintenance at or above the third percentile (without non-oral/mechanical feeding support) up to age 18 months.

In the SPR1NT trial three-copy SMN2 cohort [18], all 15 patients met the primary endpoint of standing alone (as defined by the BSID GM subtest item number 40) at any visit up to age 24 months, and 14 patients (93.3%) met the secondary endpoint of walking alone (as defined by the BSID GM subtest item number 43) at any visit up to age 24 months. Most patients achieved these milestones within the normal development windows (as defined by WHO-MGRS) [19] (standing alone: 14/15, 93.3%; walking alone: 11/15, 73.3%).

In the company’s submission for the partial review of HST15, the company presented evidence for the comparison of providing onasemnogene abeparvovec pre-symptomatically to the pre-symptomatic patient versus BSC. As the SPR1NT trial [17, 18] did not have a control arm, the company presented data from the Pediatric Neuromuscular Clinical Research (PNCR) dataset [20] to provide evidence of clinical outcomes among patients receiving BSC only. The company compared data from the SPR1NT trial two-copy SMN2 cohort [17] with data from a cohort of patients in the PNCR dataset who had two copies of the SMN2 gene and type 1 SMA (n = 23), and compared data from the SPR1NT trial three-copy SMN2 cohort [18] with data from a cohort of patients in the PNCR dataset who had three copies of the SMN2 gene and any SMA type (n = 81).

In response to a clarification request, the company provided an updated model that included cost-effectiveness evidence to support the EAG’s preferred comparison. However, the company did not provide clinical-effectiveness evidence to support this comparison, other than the information included in the updated model.

Therefore, to inform the EAG’s preferred comparison, the EAG carried out simple naïve comparisons of data from the SPR1NT trial [17, 18] versus data from the START [21], STR1VE-US [22] and STR1VE-EU [23] trials, which assessed onasemnogene abeparvovec as a treatment for patients with type 1 (symptomatic) SMA. The EAG also compared data from the SPR1NT trial [17, 18] with data from the PNCR [20] three-copy SMN2 cohort. A subset of this cohort had type 2 or type 3 SMA, and therefore were relevant to the EAG’s preferred comparison.

Data from the SPR1NT [17, 18], START [21], STR1VE-US [22] and STR1VE-EU [23] trials and the PNCR [20] three-copy SMN2 gene cohort for the primary and secondary outcomes of the SPR1NT trial are presented in Table 4. For completeness, the EAG also presented data from the PNCR two-copy SMN2 cohort [20], as these data were used by the company to provide an external control arm for the SPR1NT trial [17].

Table 4 Comparison of key outcomes from the SPR1NT, STR1VE and START trials and the PNCR dataset

Generally, outcomes for patients treated pre-symptomatically with onasemnogene abeparvovec were better than outcomes for patients who received onasemnogene abeparvovec upon clinical diagnosis of type 1 SMA, and were better than outcomes for patients who received BSC only for any type of SMA.

The company presented adverse event (AE) data from the SPR1NT trial [17, 18]. All patients (29/29, 100%) experienced at least one treatment-emergent AE (TEAE), most frequently pyrexia (18/29, 62.1%) and upper respiratory tract infection (14/29, 48.3%). Eighteen patients (62.1%) experienced at least one TEAE that was considered by the investigator to be related to treatment with onasemnogene abeparvovec. No patient experienced a TEAE that resulted in death or trial discontinuation.

Neither patient or carer HRQoL data were collected as part of the SPR1NT [17, 18], START [21], STR1VE-US [22] or STR1VE-EU [23] trials.

3.2 Critique of the Clinical Evidence and Interpretation

The EAG considered that SPR1NT trial results suggested that onasemnogene abeparvovec is a clinically effective treatment for babies with pre-symptomatic SMA and two or three copies of the SMN2 gene. However, clinical advice to the EAG was that uncertainty remains about the long-term efficacy of onasemnogene abeparvovec in clinical practice, particularly whether deterioration would occur. The company presented interim efficacy and safety data from a long-term study (LT-002) of patients with SMA (follow-up to age 15 years) treated with onasemnogene abeparvovec in clinical trials, but final results will not be available until the study’s completion in December 2035.

The EAG cautioned that the naïve comparisons of data from the SPR1NT, START [21], STR1VE-US [22] and STR1VE-EU [23] trials and the PNCR [20] dataset were not robust, as differences between trial and patient characteristics were not accounted for. For example, the mean age at symptom onset for patients in the PNCR [20] dataset (3.0 months) was greater than for patients in the START [21] (1.4 months), STR1VE-US [22] (1.9 months) and STR1VE-EU [23] (1.6 months) trials. Furthermore, sample sizes of the included trials and the PNCR dataset were all relatively small; this was expected given the rarity of SMA.

Evidence to inform the EAG’s preferred comparison was also limited as there was no evidence for the effectiveness of onasemnogene abeparvovec as a treatment for patients with type 1 SMA and three copies of the SMN2 gene, as the START [21], STR1VE-US [22] and STR1VE-EU [23] trials only included patients with type 1 SMA and two copies of the SMN2 gene. Furthermore, in the PNCR [20] dataset, the cohort of patients with three copies of the SMN2 gene included some patients with type 1 SMA; in NHS clinical practice, patients with type 1 SMA may be eligible for [25] and receive treatment with onasemnogene abeparvovec, rather than BSC.

Clinical advice to the EAG was that safety data from all onasemnogene abeparvovec trials provides more comprehensive information than safety data collected from patients with pre-symptomatic SMA only. The EAG noted that safety data for 99 patients who received onasemnogene abeparvovec as a treatment for pre-symptomatic or symptomatic SMA at the recommended dose were reported in the European Medicines Agency (EMA) European Public Assessment Report (EPAR) [26]. The AEs most frequently reported from five open-label trials (SPR1NT, START [21], STR1VE-US [22], STR1VE-EU [23] and STR1VE-AP [27]), and described as very common (> 10%) or common (> 1%), were increased hepatic enzyme (24/99, 24.2%), hepatotoxicity (9/99, 9.1%), vomiting (8/99, 8.1%), thrombocytopenia (6/99, 6.1%), increased troponin (5/99, 5.1%), and pyrexia (5/99, 5.1%). The EPAR highlighted that outside clinical studies, including in the postmarketing setting, there had been reports of children experiencing thrombotic microangiopathy and developing signs and symptoms of acute liver failure.

The EAG also highlighted that more recently (11 August 2022), it was reported that two children out of more than 2300 patients worldwide who had been treated with onasemnogene abeparvovec experienced acute liver failure resulting in death [28]. These deaths were reported to occur between 5 and 6 weeks after onasemnogene abeparvovec infusion, and between 1 and 10 days after corticosteroid tapering occurred.

3.3 Cost-Effectiveness Evidence

The company’s economic evaluation compared the cost effectiveness of onasemnogene abeparvovec versus BSC for treating patients with pre-symptomatic SMA and two or three copies of the SMN2 gene. The company considered the population as a whole, with results weighted by number of copies of the SMN2 gene. Subgroup analyses for the patient cohorts with two or three copies of the SMN2 gene were carried out.

The company developed a two-part model (short- and long-term components) using Microsoft Excel (Microsoft Corporation, Redmond, WA, USA). The experience of patients receiving onasemnogene abeparvovec was modelled using the short-term component (61 months) and the long-term component (lifetime), while the experience of patients treated with BSC was modelled using only the long-term component.

The short-term component of the model was populated with data from the START [21], STR1VE-US [22] and STR1VE-EU [23] trials for patients with type 1 SMA and two copies of the SMN2 gene. In the absence of data for patients with type 1 SMA and three copies of the SMN2 gene, the company assumed that the efficacy of onasemnogene abeparvovec for these patients was the same as for patients with type 1 SMA and two copies of the SMN2 gene.

The long-term component of the model had a time horizon of 100 years and a cycle length of 1 month. This Markov state-transition model comprised five mutually exclusive health states that reflected the highest motor function milestones achieved by patients. These health states were referred to by the company as: ‘HS1 (non-sitter, permanent assisted ventilation [PAV])’; ‘HS1 (non-sitter, no PAV)’; ‘HS2 (sitter)’; ‘HS3a (delayed walker)’; ‘HS3b (experiences later-onset SMA)’. Data from patients with type 1 (HS1), type 2 (HS2) and type 3 (HS3) SMA were used to populate these health states.

The time point at which patients entered a health state was estimated using the WHO [19] thresholds for sitting and walking. Patients who did not meet developmental milestones were moved to lower functioning health states. Patients could progress to death from any health state. Estimates for the proportions of untreated non-sitter patients requiring PAV were derived from Wijngaarde et al. [29] and from the NeuroNext [20] study. Milestone losses were estimated using data published by Wadman et al. [30].

Patients treated with onasemnogene abeparvovec entered the long-term model component in the same health state that was assigned to them in the short-term model component (based on motor function milestones achieved at the end of the SPR1NT trial and interim data from the LT-002 study [31]), where they remained until death. In the BSC arm, the distribution of patients between initial health states was informed by the distribution of patients across SMA type reported by Calucho et al. [6] (n = 3459), based on the proxy relationship between SMA type and motor milestone achievement.

Survival data sources used to populate the short-term and long-term model components were sourced from the SPR1NT trial and the LT-002 study [31], UK National Life tables (2018–2020) [32], an Italian natural history study [33], the NeuroNext study [5, 20] and Wijngaarde et al. [29].

Standard methods were used to fit parametric distributions to available data. To avoid clinically implausible survival estimates, curves were terminated based on observed life expectancy, clinical expert opinion, or assumptions that were preferred by the Evidence Review Group (ERG) in HST15 [2].

The utility values used to populate the model were sourced from the literature (Table 5).

Table 5 Company model utility values

The cost of treatment with onasemnogene abeparvovec was estimated based on the confidential discounted Patient Access Scheme (PAS) price. Health state costs were sourced from NHS Reference Costs 2019–2020 [38], the NHS Business Services Authority prescription cost analysis 2021/2022 [39] and the literature; where necessary, costs were inflated to 2021 prices using the National Health Service Cost Inflation Index (NHSCII) [40]. Costs associated with AEs were not included in the company model due to difficulties distinguishing between AEs due to treatment and SMA complications.

In response to a clarification request from the EAG, the company provided cost-effectiveness results to inform the EAG’s preferred comparison [providing onasemnogene abeparvovec pre-symptomatically versus providing (1) onasemnogene abeparvovec at symptom onset if the patient develops type 1 SMA, and (2) BSC at symptom onset for all other SMA types].

For the comparison of pre-symptomatic onasemnogene abeparvovec versus BSC, the company’s cost-effectiveness results suggested that the ICER per quality-adjusted life-year (QALY) gained was likely to be less than £100,000. For the comparison of pre-symptomatic onasemnogene abeparvovec versus onasemnogene abeparvovec on development of symptoms of type 1 SMA and BSC for all other types of SMA, the results suggested that pre-symptomatic treatment with onasemnogene abeparvovec was likely to be dominant.

3.4 Critique of the Cost-Effectiveness Evidence and Interpretation

The EAG carried out a comprehensive check of the company model data inputs and algorithm and was satisfied that the model algorithms were accurate. The EAG was satisfied that the cost-effectiveness results generated by the company’s model were robust and suitable for decision making for both the company’s preferred comparison (providing onasemnogene abeparvovec pre-symptomatically versus BSC) and for the EAG’s preferred comparison [providing onasemnogene abeparvovec pre-symptomatically versus providing (1) onasemnogene abeparvovec only at symptom onset if the patient develops type 1 SMA, and (2) BSC at symptom onset for all other SMA types].

The company provided results for the whole population and independently for patients with two or three copies of the SMN2 gene. The EAG considered that cost-effectiveness decisions should be made based on number of copies of the SMN2 gene due to the following.

Model results showed that QALYs and BSC costs differed substantially by number of copies of the SMN2 gene. Patients with two copies of the SMN2 gene have a higher likelihood of having type 1 SMA than patients with three copies of the SMN2 gene. Furthermore, patients with type 1 SMA with three copies of the SMN2 gene tend to have longer survival than those with two copies of the SMN2 gene.

Patients with two copies of the SMN2 gene and those with three copies of the SMN2 gene are identified at the time of diagnosis of SMA.

Approximately 85% of patients with three copies of the SMN2 gene have type 2 SMA (54.3%) or type 3 SMA (30.9%), not type 1 SMA (14.7%) [6], and therefore are not eligible for treatment with onasemnogene abeparvovec following the development of symptoms based on the recommendations made by NICE in HST15 [2].

The EAG therefore generated scenario results independently for patients with two copies of the SMN2 gene and patients with three copies of the SMN2 gene. The EAG scenario analyses explored two areas of uncertainty, namely loss of milestones achieved and social care costs.

In the company model, patients in the onasemnogene abeparvovec arm were modelled to maintain the best milestone they achieved, while, over time, patients in the BSC arm could lose milestones previously achieved. Milestone data were available from the SPR1NT trial for a maximum follow-up of 24 months, and from the START [21] trial for 6.2 years. This means that there is still uncertainty whether, over a lifetime, patients treated with onasemnogene abeparvovec would lose previously achieved milestones. To explore the impact of this uncertainty on company cost-effectiveness results, the EAG ran a scenario analysis applying the company base-case loss of milestone assumptions for the BSC arm of the long-term model to patients in the onasemnogene abeparvovec arm of the long-term model.

In the company model, social care costs accounted for the largest proportion of total costs after hospitalisations. The EAG highlighted there was uncertainty to how the company calculated social care costs as the value in the model did not match the costs presented in the publication by Noyes et al. [41]. To test the impact of these costs on company cost-effectiveness results, the EAG carried out a scenario in which the costs of social care were set to zero. The EAG considered that patients with SMA were likely to rely heavily on social care and that setting social care costs to zero is an extreme scenario; however, this scenario was undertaken to explore whether reducing social care costs would change the conclusions that can be drawn from the company’s cost-effectiveness results.

The EAG’s scenario cost-effectiveness results were generated for both the company’s preferred comparison and the EAG’s preferred comparison. All EAG scenario analyses generated ICERs for pre-symptomatic treatment with onasemnogene abeparvovec that were less than £100,000 per QALY gained.

3.5 Conclusions of the EAG Report

The EAG considered that results from the SPR1NT trial supported the company’s conclusion that onasemnogene abeparvovec is a clinically effective treatment for babies with pre-symptomatic SMA and two or three copies of the SMN2 gene. Naïve comparisons of data from the SPR1NT trial, the PNCR [20] dataset, and other trials [21,22,23] that evaluated onasemnogene abeparvovec as a treatment for patients with symptomatic SMA, suggested that outcomes for patients treated pre-symptomatically with onasemnogene abeparvovec are better than outcomes for patients who receive (1) onasemnogene abeparvovec upon a clinical diagnosis of type 1 SMA; and (2) BSC only for any type of SMA. However, these naïve comparisons were not robust and there remained some uncertainty about the long-term efficacy of onasemnogene abeparvovec in clinical practice given it is unclear whether some deterioration may occur.

The EAG also concluded that it is important to consider patients with two copies of the SMN2 gene and patients with three copies of the SMN2 gene separately as outcomes for these two groups differ substantially.

For the comparison of pre-symptomatic onasemnogene abeparvovec versus onasemnogene abeparvovec on development of symptoms of type 1 SMA and BSC for all other types of SMA, the EAG considered that pre-symptomatic treatment with onasemnogene abeparvovec is likely to be dominant. For the comparison of pre-symptomatic onasemnogene abeparvovec versus BSC, the EAG considered that the ICER per QALY gained is likely to be less than £100,000. Although the long-term efficacy of onasemnogene abeparvovec and costs associated with social care provision to children with SMA remained uncertain, these uncertainties were considered unlikely to change the conclusions that could be drawn on the cost effectiveness of onasemnogene abeparvovec given pre-symptomatically.