A real-world study comparing pre-post billed annualized bleed rates and total cost of care among non-inhibitor patients with hemophilia A switching from FVIII prophylaxis to emicizumab

Abstract

Objective

Factor VIII (FVIII) replacement and emicizumab have demonstrated efficacy for prevention of bleeds among patients with hemophilia A (PwHA) compared to on-demand (OD) use. Evidence investigating clinical outcomes and healthcare costs of non-inhibitor PwHA switching from prophylaxis with FVIII concentrates to emicizumab has not been well-established within large real-world datasets. This study aimed to investigate billed annualized bleed rates (ABR_b) and total cost of care (TCC) among non-inhibitor PwHA switching from FVIII-prophylaxis to emicizumab-prophylaxis.

Methods

This retrospective, observational study was conducted using IQVIA PharMetrics Plus, a US administrative claims database. The date of first claim for emicizumab was defined as the index date. OD patients and inhibitor patients were excluded. Bleeds were identified using a list of 535 diagnosis codes. Bayesian models were developed to estimate the probability ABR_b worsens and TCC increases after switching to emicizumab. Wilcoxon rank-sum tests were used to test statistical significance of changes in ABR_b and TCC after switch.

Results

Among the 121 identified patients, the difference in mean ABR_b between FVIII-prophylaxis (0.68 [SD = 1.28]) and emicizumab (0.55 [SD = 1.48]) was insignificant (p = .142). The mean annual TCC significantly increased for patients switching from FVIII-prophylaxis ($518,151 [SD = $289,934]) to emicizumab ($652,679 [SD = $340,126]; p < .0001). The Bayesian models estimated a 21.0% probability of the ABR_b worsening and a 99.9% probability of increasing TCC after switch.

Conclusions

This study found that in male non-inhibitor PwHA, switching from FVIII prophylaxis to emicizumab incurs substantial cost increase with no significant benefit in ABR_b. This evidence may help guide providers, payers, and patients in shared decision-making conversations around best treatment options.

Keywords:

• Hemophilia
• non-inhibitor
• real-world
• FVIII
• emicizumab
• comparative

Introduction

Hemophilia A (HA), a rare, recessive, X-linked bleeding disorder, is caused by a deficiency of clotting factor VIII (FVIII) occurring in approximately 1 in 5000 males born in the United States (US)¹. The national age-adjusted incidence for HA was 12.0/100,000 person-years during 2012–2018 among males born between 1995 and 2014². The severity of HA is categorized based on biochemical phenotypes as mild, moderate, or severe depending on the quantitative level of FVIII present, as well as phenotypically by the severity and frequency of bleeding patients experience³. Spontaneous bleeding episodes characteristically occur in joints, with ankles, knees, and elbows being the most frequent sites. Control and prevention of bleeds are paramount to minimizing acute and chronic damage and essential for reducing long-term sequelae, decreasing healthcare resource utilization, and improving quality of life^3,4.

FVIII products are approved for either prophylactic or episodic care (on-demand [OD] therapy)³. Typical FVIII prophylaxis regimens for patients with moderate-severe HA is ∼2–3 intravenous infusions weekly with the goal of increasing FVIII trough levels to eliminate spontaneous bleeding events and the subsequent acute and chronic joint damage. Current FVIII replacement therapies include plasma-derived and recombinant products. Among recombinant FVIII products, varying half-lives may have an impact on infusion frequency (standard half-life [SHL] or extended half-life [EHL]). The half-life of EHL products is approximately 1.5 times longer than SHL products, thereby potentially lowering treatment burden through fewer infusions and less time with sub-therapeutic FVIII levels^4,5. Approximately, 30% of patients with HA (PwHA) on prophylaxis will develop neutralizing alloantibodies or inhibitors to exogenous FVIII⁶, placing them at increased risk of fatal bleeds, prolonged bleeding episodes, mortality⁷, lower quality of life⁸, and compromised physical functioning⁹. The development of inhibitors introduces new levels of complexity to treatment options, such as bypassing agents and immune tolerance induction which requires more frequent administration and has decreased efficacy in bleed control¹⁰.

An alternative treatment regimen, emicizumab, has been recently approved to address the unmet need among PwHA. Emicizumab, a recombinant humanized bispecific monoclonal antibody that mimics FVIII by bringing together FIX and FX¹¹, was approved in 2018 and is currently indicated for routine prophylaxis for both inhibitor and non-inhibitor PwHA. Emicizumab is administered subcutaneously every week, every 2 weeks, or every 4 weeks. Long-term follow-up of patients enrolled in Phase III clinical trials (HAVEN 1-4), demonstrated efficacy of emicizumab for prevention of bleeding episodes while being well-tolerated^12–14.

HA is a clinically and economically burdensome disease with significant psychological and functional consequences, persistent treatment requirements, and lifelong healthcare needs and costs. New non-factor replacement therapies, such as emicizumab entering the market merit understanding the depth of clinical benefits relative to the economic burden placed on the healthcare system, payers, providers, and patients. While the benefits of emicizumab for inhibitor PwHA are well established, some uncertainty surrounds the clinical and economic outcomes among non-inhibitor PwHA in real-world settings¹⁵. With limited ability to compare therapeutic treatment options head-to-head, assessing real-world use of different treatment options is imperative. This study aimed to estimate and compare clinical outcomes, defined as billed annualized bleed rates (ABR_b) and total cost of care (TCC) for non-inhibitor PwHA, before and after switching prophylaxis from FVIII concentrates to emicizumab.

Methods

Data source

This retrospective, pre-post cohort study queried the IQVIA PharMetrics Plus database (a longitudinal US commercial health plan database with >190 million members) from 2015 to 2020. Because all patient data were de-identified, institutional review board approval was not required.

Patient population

Patients who met the inclusion criteria were followed from the beginning of the study period or start of continuous enrollment until the end of the study period or end of continuous enrollment (Figure 1). Male patients with ≥1 outpatient or pharmacy claim for emicizumab (first claim defined as the index date), ≥6 months of continuous enrollment before and after the index date, and prophylaxis with FVIII concentrates before switching to emicizumab met the inclusion criteria. Continuous enrollment was defined as no lapse in insurance coverage >60 days. Patients with a history of FVIII inhibitors or ≥2 claims at least 30 days apart for diagnoses of von Willebrand disease, hemophilia B, acquired HA or acquired coagulation factor deficiency were excluded.

Figure 1. Study design. *Continuous enrollment is defined as no lapse in insurance coverage for >60 days.

International Classification of Diseases (ICD) codes, National Drug Codes (NDCs), and Healthcare Common Procedure Coding System (HCPCS) codes were used to identify diagnoses, therapies, and procedures. Operational definitions were created to identify FVIII prophylaxis use, bleed events, and inhibitor status.

Identifying patients with FVIII prophylaxis

Patients with an annualized FVIII utilization of ≥2340 international units (IU)/kilogram (kg) (i.e. ≥45 IU/kg/week³) before the index date met the criteria for continuous FVIII prophylaxis. Annualized FVIII utilization was calculated as: (1) $$\text{Total annual dose}=\frac{\text{Total units of FVIII dispensed}}{\mathrm{Patien}{\mathrm{t}}^{\mathrm{’}}\text{s weight*Observation days}}\mathrm{\hspace{0.17em}*\hspace{0.17em}}365.25\mathrm{ }$$(1)

All FVIII claims associated with a bleed event (i.e. occurred up to 7-day prior to a bleed diagnosis) were eliminated from the calculation of the total annual dosage. This method identified bleed events that were evaluated or treated by the healthcare system and does not capture bleeds that were self-treated at home. Patient weight was not available in the data and was imputed using US national averages based on age^16–18.

Identifying patients with inhibitors

To ensure PwHA with active inhibitors were excluded from the study, they were operationally defined based on having at least one of the following:

• ≥1 claim(s) for bypassing agent (Feiba-Activated prothrombin complex concentrate [aPCC] or recombinant activated factor VII [rFVIIa]) or (Novoseven RT);

OR

• ≥1 claim(s) for rituximab

OR

• Immune tolerance induction therapy, identified using the following criteria:

•  ^ ≥1 claim(s) for Bethesda/Nijmegen assays in the study period;

  AND

•  ^ High dose of FVIII dispensed in the complete baseline period, was defined as >3 times of the median IU factors dispensed for patients in the same age-group for more than 5 consecutive 28-day intervals, allowing one interval gap¹⁹.

Outcomes

Billed annualized bleed rates

Clinical outcomes were assessed as billed ABR and included bleeding episodes that resulted in a claim for evaluation, treatment, or procedure. A clinical review of diagnosis codes and previous literature¹⁶ resulted in a list of 535 ICD-10 codes used to identify HA-related bleeding episodes (e.g. hemarthrosis) in the primary diagnosis position. Bleeds treated at home or those that did not trigger any medical claims were not captured. ABR_b was categorized as either a spontaneous or traumatic bleed using specific diagnosis codes finalized during the clinical review. All bleed events caused by trauma, or a medical procedure was classified as traumatic bleeds and the remaining as spontaneous bleeds. A sensitivity analysis for ABR_b was conducted using additional ICD-9/10 bleed codes not included in the base case list. We also investigated the difference in ABR_b, irrespective of treatment before the COVID-19 pandemic (March 2020) and during the lockdown to identify potential confounding caused by systematic differences in reporting or delivery of care.

Total cost of care

Total costs included all hemophilia and non-hemophilia healthcare costs paid by the payer, expressed as the mean annualized cost per patient per year (PPPY) and adjusted to 2020 US dollars. TCC was further stratified by cost accrued from breakthrough/traumatic bleeds (FVIII costs accrued during ±7 day of a bleed event), prophylaxis treatment (remaining FVIII-related costs pre-switch and emicizumab-related costs post-switch), FVIII transitioning (first 4 weeks after switching), and concomitant use (after 4 weeks of switching). FVIII concomitant use after the expected transition period may represent FVIII dispenses for on-hand breakthrough/traumatic bleeds, additional prophylaxis, or maintaining chronic exposure to FVIII to minimize risk or complications of inhibitor development.

Subgroups

A sub-group analysis tested the impact of high FVIII utilization in the pre-switch period on mean change in ABR_b and TCC by stratifying patients by the top 10% FVIII utilization (high-utilizers) and the remaining 90% (low-utilizers). Another sub-group analysis stratified patients into age categories.

Statistical methods

The analysis generated descriptive statistics to characterize the cohort, as well as summarized pre- and post-switch ABR_b and TCC. Bayesian models were developed to generate inferences on the probability of a mean change in ABR_b and TCC before and after switching from FVIII prophylaxis to emicizumab. Gaussian model for ABR_b and gamma regression model for TCC difference were the best fitting models (Appendices). The models were run for 100,000 iterations to improve robustness. Inferences were conducted by computing posterior probabilities for hypotheses and summarized with 95% credible intervals (CrIs). The a-priori defined Bayesian inferences included: 1) what is the probability ABR_b worsens after switching to emicizumab? and 2) what is the probability the TCC increases after switching to emicizumab? Wilcoxon signed-rank tests (WSRT) were performed as secondary endpoints.

Results

A total of 121 male, non-inhibitor, PwHA met criteria for inclusion. The mean (median) age was 25.9 (25) years and ranged from 2 to 63 years. The mean observation time in the pre-switch and post-switch periods was 2.5 and 1.1 years, respectively. The mean (standard deviation [SD]) baseline Charlson Comorbidity Index score was 1.08 (2.49). Prevalence of joint pain and soft tissue disorders in the pre-switch period were 46 and 28%, respectively. Prevalence of blood-borne viral infections was 14.1% for hepatitis B/C and 9.1% for Human Immunodeficiency Virus (HIV). During the pre-switch period, the mean (median) FVIII replacement dose was 82.8 (70.2) IU/kg/week (Table 1).

Table 1. Distribution of demographic and clinical characteristics among non-inhibitor hemophilia A patients.

Characteristic	Overall cohort [N = 121]
Age at index date, years
Mean [SD] (min–max)	25.88 [14.23] (2.00–63.00)
Median [interquartile range]	25.00 [13.00–38.00]
Age-groups (in years) at index date, N (%)^a
0–2	1 (0.83)
3–6	7 (5.79)
7–12	20 (16.53)
13–18	20 (16.53)
>18	73 (60.33)
Stratification of age-group (in years) >18 years, N (%)
19–26	17 (14.05)
>26	56 (46.28)
FVIII utilization IU/kg/week
Mean [SD] (min–max)	82.85 [39.26] (45.45, 251.81)
Median [interquartile range]	70.18 [56.01–98.00]
Baseline period, years
Mean [SD]	2.50 [1.53]
Follow-up period, years
Mean [SD]	1.11 [0.44]
Index year, N (%)^a
2018	9 (7.44)
2019	84 (69.42)
2020	28 (23.14)
Charlson comorbidity index (CCI) in complete baseline^b
Mean [SD] (min–max)	1.08 [2.49] (0.00–11.00)
Frequency of CCI in complete baseline, N (%)^a,b
CCI = 0	79 (65.29)
CCI = 1	28 (23.14)
CCI = 2	1 (0.83)
CCI = 3+	13 (10.74)
Comorbidities, N (%)^a
Musculoskeletal pain	59 (48.76)
Joint health related^c	55 (45.45)
Soft tissue disorder (muscles, synovium, and tendons)	34 (28.10)
Hepatitis B/C (acute and chronic)	17 (14.05)
HIV	11 (9.09)

Abbreviations. CCI, Charlson comorbidity index; HIV, Human Immunodeficiency Virus; HMO, Health Maintenance Organization; IU/kg, International units per kilogram; Min, Minimum; Max, Maximum; N, Number of patients; PPO, Preferred provider organization; SD, Standard deviation.

Notes:

^aDenominator for each sub-category percentage is N, i.e. the total cohort size.

^bCCI was captured using Quan et al.²⁰

^cList of ICD-10 joint diagnoses included is based on Osooli et al.⁹

The Bayesian model estimated a mean change in ABR_b of −0.128 [95% CrI: −0.441–0.184] and predicted the ABR_b will worsen 21.0% of the times after switching to emicizumab (Table 2). The mean change in ABR_b was not statistically significant after switching (pre-switch: 0.68 [SD = 1.28] vs. post-switch: 0.55 [SD = 1.48]; WSRT p = .142). The Bayesian model estimated a less than 1% probability of ≥0.5 reduction in ABR_b after switching (Table 3).

Table 2. Bayesian posterior summaries of difference in ABR_b and TCC before and after switch from FVIII prophylaxis to emicizumab.

Outcome	FVIII prophylaxis	Emicizumab prophylaxis	Mean change after switch (95% CrI)	Bayesian model posterior probabilities
ABRb^a	0.68	0.55	−0.128	21.0%
(Per patient)			(−0.441–0.184)	(Probability of ABRb worsening on emicizumab)
TCC^a	$519,854	$679,535	$159,680	99.9%
(Per patient per year)			($74,842–$247,841)	(Probability of emicizumab being more expensive)

Abbreviations. ABR_b, Billed annualized bleed rate; CrI, Credible intervals; TCC, Total cost of care.

^aThe mean values reported above may slightly vary from the actual mean values as reported previously due to variability added by multiple iterations used in the Bayesian model.

Table 3. Bayesian probability distribution of range of difference in ABR_b and TCC before and after switch from FVIII prophylaxis to emicizumab.

Outcomes	Intervals	Probability (%)
ABRb	ABRb worsens (μ > 0)	21.04
	No change or improvement in ABRb by ≤ 0.1	22.03
	(−0.10 ≤ μ ≤ 0)
	Improvement in ABRb between 0.1 and 0.2	24.51
	(−0.20 ≤ μ < −0.10)
	Improvement in ABRb between 0.2 and 0.5	31.49
	(−0.50 ≤ μ < −0.20)
	Improvement in ABRb between 0.5 and 1.0	0.94
	(−1.00 ≤ μ < −0.50)
	Improvement in ABRb by 1.0	0.0
	(μ < −1.00)
TCC	TCC decreases (μ^C < 0)	0.01
	TCC increases by 0–20%	10.10
	(0 ≤ μ^C ≤ $103,970)
	TCC increases by > 20–40%	76.30
	($103,970 < μ^C ≤ $207,940)
	TCC increases by >40%	13.58
	(μ^C > $207,940)

Abbreviations. ABR_b, Billed annualized bleed rate; TCC, Total cost of care; μ, Difference in ABR_b; μ^c, Difference in TCC.

Spontaneous bleeds comprised the majority of ABR_b. Pediatric patients (0–12 years) had the greatest reduction in bleeds, however, adolescents aged 13–18 years experienced increased bleeds. Adults had a 0.05 reduction in ABR_b. For patients aged >26 years, ABR_b changed from 0.71 (SD = 1.12) to 0.68 (SD = 1.89). High-utilizers of FVIII had a 0.03 increase in ABR_b after switching (Table 4). Sensitivity analyses of ABR_b using additional diagnosis-based definitions for bleeds yielded similar results.

Table 4. Billed annualized bleed rate (ABR_b) during treatment with FVIII prophylaxis (baseline) and treatment with emicizumab (follow-up).

Stratum	N	Baseline period (pre-switch)				Follow-up period (post-switch)
		Mean	SD	Median	(Min, max)	Mean	SD	Median	(Min, max)
Overall	121	0.68	1.28	0.00	(0.00, 7.38)	0.55	1.48	0.00	(0.00, 9.65)
Spontaneous	121	0.63	1.17	0.00	(0.00, 5.71)	0.54	1.48	0.00	(0.00, 9.65)
Traumatic	121	0.06	0.23	0.00	(0.00, 1.84)	0.03	0.15	0.00	(0.00, 1.00)
Age-group (years)
0–2	1	7.38	–	7.38	(7.38, 7.38)	0.00	–	0.00	(0.00, 0.00)
3–6	7	0.41	1.08	0.00	(0.00, 2.86)	0.14	0.36	0.00	(0.00, 0.95)
7–12	20	0.40	1.18	0.00	(0.00, 5.09)	0.16	0.34	0.00	(0.00, 1.26)
13–18	20	0.85	1.27	0.49	(0.00, 5.71)	0.96	1.59	0.00	(0.00, 5.75)
>18	73	0.65	1.08	0.00	(0.00, 4.71)	0.60	1.68	0.00	(0.00, 9.65)
Stratification ≥19 years
>19–26	17	0.44	0.97	0.00	(0.00, 4.00)	0.32	0.46	0.00	(0.00, 1.15)
>26	56	0.71	1.12	0.00	(0.00, 4.71)	0.68	1.89	0.00	(0.00, 9.65)
FVIII utilization
High-utilizers	13	1.31	1.75	0.59	(0.00, 5.09)	1.34	2.77	0.00	(0.00, 9.65)
Low-utilizers	108	0.61	1.20	0.00	(0.00, 7.38)	0.46	1.23	0.00	(0.00, 8.12)

Abbreviations. ABR_b, Billed annualized bleed rate; SD, Standard deviation; Min, Minimum; Max, Maximum.

The Bayesian model estimated an increase in TCC of $159,680 [95% CrI: $74,842–$247,841] PPPY after switching and predicted a 99.9% probability the TCC will increase after switch (Table 2). The mean annual TCC increased significantly from $518,151 [SD = $289,934] to $652,679 [SD = $340,126] PPPY after switching (WSRT p < .0001). The Bayesian model estimated an 89.9% probability TCC would increase by >$103,970 (>20%) PPPY (Table 3). Higher cost in the follow-up was accounted by increase in prophylaxis cost after starting emicizumab, followed by added costs of FVIII transitioning and concomitant use (Figure 2).

Figure 2. Total cost of care (TCC) per patient per year (PPPY) before and after switch from FVIII prophylaxis to emicizumab. PPPY, Per patient per year; TCC, Total cost of care; USD, United Stated Dollars. Notes: 1. The total of all the cost components may not add-up exactly to the overall costs as some of the claim lines were captured under both FVIII-specific and emicizumab-specific costs while rolling up costs at visit-level. 2. All costs are adjusted to 2020 USD. 3. †Non-FVIII and non-emicizumab related costs. 4. ‡Mean cost for first 4 weeks after switching. 5. §PPPY cost accrued after 4 weeks of switching. 6. ¶FVIII-specific cost related to breakthrough or traumatic bleeds. 7. ¤Non-bleed related FVIII costs pre-switch and emicizumab-related cost post-switch.

HA medication costs accounted for >94% of TCC in both pre- and post-switch periods, of which prophylaxis costs composed >90% of TCC in both periods. The mean TCC increased after the switch for all ages, with a larger increase among older age groups. For example, for the seven patients aged 3–6 years, the TCC increased from $121,227 to $172,084, whereas for 56 patients aged >26 years it increased from $684,072 to $853,187. The only stratum where the TCC decreased after switching was for the 13 high FVIII utilizers (from $1,033,222 to $885,533) (Table 5).

Table 5. Total cost of care (TCC) during treatment with FVIII prophylaxis (baseline) and treatment with emicizumab (follow-up).

Stratum	N	Baseline period Mean ± SD [median] PPPY	Follow-up period Mean ± SD [median] PPPY
Overall	121	518,151 ± 289,934 [480,610]	652,679 ± 340,126 [669,065]
Age (years)
0–2	1	370,823 ± 0 [370,823]	566,397 ± 0 [566,397]
3–6	7	121,227 ± 16,855 [127,568]	172,084 ± 11,510 [174,439]
7–12	20	283,548 ± 172,868 [249,745]	308,726 ± 117,202 [297,613]
13–18	20	407,060 ± 177,789 [362,949]	600,825 ± 158,285 [568,641]
>18	73	634,651 ± 281,645 [565,366]	829,082 ± 304,341 [796,807]
Age-group ≥19 years
≥19–26	17	483,816 ± 180,754 [423,769]	744,189 ± 154,906 [704,778]
>26	56	684,072 ± 291,055 [621,666]	853,187 ± 330,883 [830,916]
FVIII utilization
High-utilizers	13	1,033,222 ± 359,834 [902,815]	885,533 ± 562,970 [834,054]
Low-utilizers	108	461,048 ± 215,300 [430,050]	624,044 ± 288,876 [645,716]

Abbreviations. PPPY, Per patient per year; SD, Standard deviation; TCC, Total cost of care; USD, United States Dollar.

All costs are adjusted to 2020 USD.

Among the 104 patients with available data, the mean ABR_b changed from 0.64 (SD = 1.79) prior to the COVID-19 pandemic (March 2020) to 0.53 (SD = 1.81) during the pandemic (results not shown in tables), irrespective of treatment (FVIII or emicizumab).

Discussion

This study assessed the comparative effectiveness of billed ABR and costs associated with FVIII prophylaxis as compared to emicizumab prophylaxis among male non-inhibitor PwHA in the US. The analysis used a national claims database that is geographically diverse with a broad age range. The study also applied algorithms and clinical practice definitions to identify prophylaxis FVIII use, bleed events, and inhibitor status^3,16,19. In addition to traditional statistical testing, the study also applied Bayesian statistical approaches, that provide a powerful framework to calculate the probability of an event or change in event over time for a given exposure²¹. Treatment approach for bleed prevention and management differs between the inhibitor and non-inhibitor PwHA therefore, by removing the inhibitor population, the clinical characteristics of a non-inhibitor population become apparent and are not confounded by the more severe disease characteristics of an inhibitor population, or by the different treatment efficacies and costs in the inhibitor population. Our results demonstrate a significant increase in costs after switching therapies for an already expensive and clinically burdened disease, without a discernible improvement in ABR_b among non-inhibitor PwHA. The Bayesian analysis suggests that the 21.0% potential worsening of the disease with emicizumab needs to be carefully weighed, especially for patients who are well-controlled on FVIII prophylaxis. The 99.9% probability of cost increase after switching to emicizumab is important to understand amidst the increasingly complex treatment landscape of this already very costly disease with restrictive payer environment. The substantial cost increase seen with switching to emicizumab is not offset by a clinical benefit.

Some studies investigating emicizumab with real-world data have generally followed patients to observe clinical outcomes, such as safety and efficacy of longitudinal use; correlation between laboratory monitoring and emicizumab’s hemostatic effect; association of spontaneous and traumatic bleeds with age; inhibitor status; and other outcomes^22–27. Other studies have assessed use of additional hemostatic agents for surgery²⁸, treatment patterns, clinical, and economic outcomes among FVIII users by age²⁹, or are simply descriptive analyses of bleed outcomes²⁷. There are two relevant studies^30,31 (to date) conducted by McCary et al. and Samelson-Jones et al. that pursued similar research questions using a similar study design to our study. Both studies extracted data from three hemophilia treatment centers in the US and followed a cohort of 93 patients. McCary et al. reported that the annualized rate of treated bleed dropped significantly from 1.6 to 0.4 (p = .0025) among non-inhibitor PwHA after switching to emicizumab³¹. However, 86% of the patients were on prophylaxis before the switch, with a potential confounding impact from those patients who were on demand, and the population was 93.3% pediatric patients, limiting its generalizability. Additionally, this study suffered from threats to external validity namely, sampling and self-selection bias because the included patients switched to emicizumab of their own volition and may represent preferred candidates for a switch of therapy. Unlike this study, Samelson-Jones et al. estimated that the mean 6-month total costs before ($204,987) and after the switch ($180,571) did not statistically significantly change³⁰. Differences in data sources and patient populations may explain divergent results. The study also calculated costs using wholesale acquisition costs (WAC) whereas, our study reports actual payer costs thereby providing superior estimates because WAC differs from the actual drug price due to discounting practices, particularly for products on market for a longer time³². Additionally, the differences between prescribed amount and actual consumed amount (adherence), are not captured.

In addition to testing efficacy, studies using HAVEN trial data have provided insight into long-term outcomes^14,33. Klamroth et al.’s indirect comparison showed that individualized recombinant factor VIII-Fc fusion protein (rFVIIIFc) was more efficacious than emicizumab once every 4 weeks and reported similar efficacy when compared with more frequent emicizumab regimens³³. In Reyes et al., a network meta-analysis (NMA) of emicizumab use across varying dosages, reported a greater reduction in bleed rates compared to FVIII prophylaxis¹⁴. However, due to stringent inclusion criteria, such as the requirement to have a measure of “treated bleeds”, which was not a common end-point before emicizumab, this NMA included only a limited number of FVIII trials which had notable heterogeneity in treated ABR measurement methods and could therefore be subject to information bias. Beyond clinical efficacy, existing literature in cost-effectiveness and other models were limited to patients with inhibitors, did not isolate non-inhibitor patients, or did not assess prophylactic and OD FVIII regimens separately, biasing the results toward the outlier inhibitor patients and leading to invalid comparisons^34–38. On the contrary, the current analysis appropriately studied non-inhibitor PwHA switching from FVIII prophylaxis to emicizumab.

While future evidence is still needed, our study adds important real-world, comparative observations for non-inhibitor PwHA. The use of claims data presents an accurate estimation of costs to the healthcare system that is more informative than models or registries that require assumptions on price, adherence, and dose. The incorporation of billed bleeds allows for the comparison of an important clinical outcome for patients and providers. Our results further suggest that the discussion of treatment options for non-inhibitor PwHA needs to be more nuanced as clearly, every patient cannot be treated with emicizumab given potential for disease worsening and ongoing need for FVIII. Although some patients benefit by initiating emicizumab both clinically and economically, others will benefit more from the current standard of care. Personalization of care and identification of patients who receive the greatest clinical benefit for each therapy can optimize patient care. Payers and decision-makers should understand their patient population and the future treatment landscape to prepare for changes in costs as well as clinical outcomes. Treatment decisions should consider inhibitor status and the evidence from this study should only be used to inform disease management strategies among non-inhibitor PwHA. As additional real-world data becomes available for non-inhibitor PwHA, future research should compare different prophylaxis treatment options among diverse patient populations.

This study possesses limitations resulting from the inherent nature of claims databases, including lack of information about specific patient characteristics, medications administered during inpatient admissions, and variation in coding and billing practices. Furthermore, due to the absence of required valid information in administrative claims, some of the potential sub-group analyses, such as outcome assessments by frequency of emicizumab injection (e.g. biweekly vs. monthly) or by long-acting vs. short-acting FVIII replacement products were not able to be conducted. Moreover, because the sample was sourced from the commercially insured population, the cohort may not be representative of uninsured, Medicare, or Medicaid populations. Our study could not capture bleeds treated entirely at home or those that did not trigger any claim. Misclassification could arise from lack of disease-specific and clinical parameters in administrative claims, such as inhibitor status, HA severity, FVIII trough target, weight, or differentiation between prophylaxis and OD treatment. Although our study investigated direct costs and benefits in terms of bleed rates, other tangible benefits to patients or caregivers, such as convenience of subcutaneous infusion over the costs of maintaining venous lines, treatment schedules/frequencies, and other quality of life-related aspects were not studied. The study is susceptible to history and maturation bias because it included patients with prevalent HA, and changes in age or internal states between switching therapies may have affected the outcomes, which may have been further amplified due to large difference between the assessment time periods. Last, because the study period for this investigation coincided with the onset of the COVID-19 pandemic, it may have reduced the bleeding risk or willingness to seek bleed treatment, more so toward the end of the inclusion period.

Conclusion

Our findings illustrate the high probability of a substantial increase in cost when patients switch to emicizumab and the low probability of a clinically meaningful change in ABR_b. Increased use of emicizumab may cause an increase in economic burden for payers without providing a substantial clinical benefit, substantively impacting their ability to manage the health of their patients. This evidence can aid providers and payers to better understand the economic impact of switching to emicizumab and meaningfully engage in shared decision-making process to improve patient care.

Transparency

Declaration of funding

Funding was provided by Takeda Pharmaceuticals U.S.A., Inc., Lexington, MA.

Declaration of financial/other relationships

KB is a consultant to Complete HEOR Solutions, North Wales, PA, USA, and current holder of equity in publicly-traded companies. BGS, JC, and MB are employees of Takeda Pharmaceuticals U.S.A., Inc., Lexington, MA, USA, and current holders of individual stock/stock options. CSH is a consultant to Complete HEOR Solutions. NA and SC are employees of Complete HEOR Solutions, which has received financial compensation for conducting the study analysis. The authors report no other relevant financial relationships or otherwise to disclose. Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Author contributions

KB, BGS, JC, and MB contributed to the conceptualization and design of the study. CSH, NA, and SC contributed to the study conduct, data analysis, and interpretation of results. All authors contributed to revising the article critically for intellectual content. All authors approved the final version to be published and agree to be accountable for all aspects of the work.

Acknowledgements

The authors thank Irene Doherty, PhD for medical writing assistance, and Niranjan Kathe, PhD for his supervision throughout the project. ID is a consultant to, and NK is an employee of Complete HEOR Solutions, North Wales. Funding for the manuscript writing was provided by Takeda Pharmaceuticals U.S.A., Inc., Lexington, MA.

Data availability statement

The data that support the findings of this study are available from the IQVIA repository, https://www.iqvia.com/solutions/real-world-evidence, but restrictions apply to the availability of these data, which were used under license for this study and so are not publicly available.

Appendices

Development of Bayesian models for billed annualized bleed rates (ABR_b) and total cost of care (TCC)

Bayesian models for ABR_b

Several Bayesian models, such as Gaussian model of ABR_b difference, binomial model of risk of bleed after switching, and independence of ABR_b were developed. For the mean difference in ABR_b, the Gaussian model was the best-fitted model, which estimated the probability of change in ABR_b after a switch. The likelihood function was Gaussian for this model and the priors were Gaussian and inverse Gaussian. The data was modeled as follows: $$y_i^a-y_i^b={\delta }_i\sim N\left(\mu ,\mathrm{ }\tau \right)$$ $$\mu \sim N\left(0,0.0001\right)$$ $$\tau \sim G(0.0001, 0.0001)$$

Model parameters are μ and t. Test the hypothesis that μ > 0.

The sampler was extremely well-behaved with almost immediate convergence and no evidence of auto-correlations.

Bayesian models for TCC

Several Bayesian models, such as Gaussian model of annualized cost difference, gamma model of increase in annualized cost after switching, and independence annualized cost were developed. Bayesian gamma regression model was the best-fitted model. A gamma distribution is continuous and limited to the positive real line and can also be skewed, which is consistent with the properties of costs. We next estimated a gamma model of costs, where the shape and scale parameters were modeled hierarchically to build in a linear function of switching. The data was modeled as gamma distributed as follows: $$c_i\sim \text{Gamma}\left(k,{\theta }_i\right)$$ $${\theta }_i=k/\exp ({\beta }_0+{\beta }_1{t}_i)$$ $${\beta }_0\sim N\left(0,0.0001\right)$$ $${\beta }_1\sim N\left(0,0.0001\right)$$ $$k\sim \text{Unif}\left(0,1000\right)$$ where i = 1…, N indexes patients, c_i is the annualized cost of patient i, and t_i is an indicator of switching. The gamma distribution has a shape parameter, k, and a scale parameter, theta. The scale parameter contains a linear predictor. Inference centers on exp(β1) – exp(β0), which shows the incremental cost of switching to emicizumab from FVIII prophylaxis. The model was fit in WinBUGS, and posteriors obtained for model parameters a Metropolis–Hastings algorithm was used to simulate the posterior. Markov chain initially was not well behaved, and exhibited drift and trend, as well as high auto-correlations after 1000 iterations. However, after running the chain for a longer burn-in period and thinning every tenth draw, an excellent sample was obtained.