Longer disease progression milestone-free time in people with amyotrophic lateral sclerosis treated versus not treated with intravenous edaravone: results from an administrative claims analysis

This was a retrospective, observational, propensity score-matched, comparative effectiveness cohort study using health claims data from the Optum Clinformatics^® Data Mart (CDM; Optum, Inc.). The CDM is statistically de-identified under the expert determination method consistent with the Health Insurance Portability and Accountability Act of 1996 and is managed according to Optum customer data use agreements. Therefore, it does not require an institutional review board assessment or approval for secondary analysis as it is not considered human patient research as defined in 45 CFR 46.102. The database includes approximately 17 to 19 million annual covered lives. The population is geographically diverse, spanning all 50 states, and is composed of members with commercial and Medicare Advantage plans. CDM administrative claims submitted for payment by providers and pharmacies are verified, adjudicated and de-identified prior to inclusion. These data, including patient-level enrollment information, are derived from claims submitted for all medical and pharmacy healthcare services with information related to healthcare costs and resource utilization.

Records from PALS who had previously been treated with IV edaravone (Radicava^® IV) were included in the IV edaravone-treated case group if they met the following criteria:

PALS may or may not have continued IV edaravone treatment until death or censoring. PALS not treated with IV edaravone were included in the control group if they met criteria 1, 2 and 4 above. For the control group, the index date was randomly assigned under the assumption that the index timepoint was uniformly distributed over the time interval between the diagnosis date and the last claim date.

Administrative claims for ALS-related symptoms and procedures identifying respiratory-, nerve-, limb- and bulbar-related symptoms in the pre-index period were tabulated for both individuals treated and not treated with IV edaravone and compared with the proportion of symptoms in the 2 groups [13]. These claims were extracted using ICD-9-CM and ICD-10-CM, Current Procedural Terminology (CPT), or HCPCS codes. HCPCS codes are a standardized coding system based on the American Medical Association's CPT that describe specific items and services.

Propensity score matching (PSM) estimates the effect of a treatment in nonrandomized, interventional studies by adjusting for potential confounding factors in a mixed population of treated or untreated individuals with varying characteristics to achieve a balanced covariate distribution between the treated and untreated groups [14]. After identifying PALS who had been treated versus not treated with IV edaravone as noted above and described in prior studies [13], a propensity score was estimated for each individual using extreme gradient boosting algorithm and accounted for the following covariates: age, sex, race, US region of residence, pre-index disease duration, insurance (Medicare Advantage vs commercial insurance), riluzole prescription; and pre-index claims for cardiovascular disease, artificial nutrition/gastrostomy tube placement, noninvasive ventilation and all-cause hospitalization. Pre-index disease duration was defined as the period between the date of first claim for ALS diagnosis and the first claim of IV edaravone for IV edaravone-treated PALS. For PALS who were not treated with IV edaravone, the pre-index disease duration was calculated between the date of first claim for ALS diagnosis and the randomly assigned index date. PALS in each group were then matched based on propensity score using the nearest-neighbor method with a caliper width equal to 0.1 of the SD of the logit scores [15] and PALS for whom no matching partner had been found were excluded. Each treated individual was allocated up to 1 untreated individual (in 1:1 matching) with the same (or a very similar) propensity score and treatment effect was evaluated in the matched groups. Standardized mean difference (SMD) was used to examine the balance of covariate distribution between groups treated versus not treated with IV edaravone [15]. SMDs of less than 10% were considered negligible imbalances.

Demographics and clinical variables were assessed descriptively using counts and percentages for categorical variables and measures of central tendency (mean/median/SD/interquartile range [IQR]) for continuous variables. Analysis of disease progression milestones and deaths used weighted estimates that considered the weights obtained from PSM and the time restriction, since weighted estimates may reduce bias and improve efficiency of the estimates, and thus provide better generalizability to the whole patient population [16].

Six disease progression milestones, including 1) use of gait assistive devices (canes/walkers/wheelchairs), 2) artificial nutrition, 3) noninvasive ventilation, 4) invasive ventilation, 5) speech-generating devices and 6) hospice, were analyzed and their CPT, HCPCS, ICD-10-CM and revenue codes are reported (Table 1). The milestones are not listed in the order that they occurred.

Restricted mean time lost (RMTL) was measured and is defined as the increase/decrease in lost life expectancy due to a specific cause of death, or in the case of a non-fatal outcome, the difference in disease/milestone-free time, considering adjusting mortality as the competing risk [16]. RMTL is reported up to 24 months since the index date for each milestone, and differences in RMTL between groups are presented for controls not treated with IV edaravone and cases treated with IV edaravone. In both groups, PALS were analyzed as a group and also subdivided by whether they were covered by commercial or Medicare Advantage insurance.

A sensitivity analysis strategy proposed by Lin et al. was used to examine the robustness of calculated treatment effects for RMTL analyses to unmeasured confounding [17] using a strategy adapted from the E-value method [18,19]. For unmeasured binary confounder U, two sensitivity analysis parameters are defined: relative risk (RR) of the treatment on a level of U with a given propensity score value, and mean ratio (MR) of U on the RMTL outcome with a given propensity score value. For a pair of specified sensitivity parameters, corresponding 95% CI bounds can be calculated. The behavior of the 95% CI bounds shows the robustness of estimated treatment effects with respect to any unmeasured confounder.

All analyses were conducted using R (version 4.0.3, R Core Team, 2020). Propensity score estimating and matching were done using the MatchIt (version: 4.1.0, 2011) package.

Between 8 August 2017 and 31 December 2021, 15,099 adults with ALS in the Optum CDM had initiated IV edaravone, riluzole, or had no treatment (Figure 1). During 1:1 PSM, 7 out of 423 cases did not report sex information and were excluded. Additionally, there were no matches found for 21 patients. Therefore, the matching process resulted in a study population that included 395 PALS with at least 1 claim for IV edaravone and 395 controls not treated with IV edaravone. Demographic and clinical characteristics for both groups after PSM are presented in Table 2. Covariates were balanced between groups as indicated by an SMD of ≤0.1 for all variables. Median (IQR) IV edaravone treatment duration in the IV edaravone-treated group was 7.47 months (3.08, 14.55).

Claims for ALS-related symptoms and procedures were identified for both groups (Figure 2). These claims were not considered confounding factors and were evenly balanced between cases and controls – individuals in both groups reported a similar proportion of claims for ALS-related symptoms, including bulbar-, limb-, nerve- and respiratory-related symptoms, in the pre-index period. The most common symptoms for individuals in both groups were limb related. IV edaravone-treated PALS had more claims for the insertion of an IV catheter than PALS not treated with IV edaravone.

RMTL was calculated for 6 disease progression milestones, including new use of canes/walkers/wheelchairs, artificial nutrition, noninvasive ventilation, invasive ventilation, speech-generating devices and hospice (Table 3). For all milestones, IV edaravone-treated cases had a smaller amount of time lost due to reaching disease progression milestones than controls not treated with IV edaravone based on RMTL for PALS overall, and when grouping PALS based on commercial or Medicare Advantage insurance coverage.

Disease progression milestones and deaths are reported from 0 to 12 months, 12 to 24 months and 0 to 24 months after the index date (Table 4). Weights were estimated from the propensity score model and the resulting estimated percentages may be generalized to the patient population. Overall, more IV edaravone-treated cases (n = 129) reported an absence of milestones than controls not treated with IV edaravone (n = 103) from 0 to 24 months after the index date. Fewer deaths occurred in IV edaravone-treated cases (n = 131) than in controls not treated with IV edaravone (n = 232) from 0 to 24 months after the index date.

In Figure 3, contour plots show the lower bounds of 95% CI for use of canes/walkers/wheelchairs, artificial nutrition, noninvasive ventilation and speech-generating devices, which had relatively large estimated treatment effects. The gray scales represent different lower bound values of the lower 95% confidence bound of the treatment effect, with the lighter shades indicating large effect values and darker shades indicating negative values. The solid curve in the middle of the plots represents when the lower bound of the lower 95% confidence bound of the treatment effect is zero. For the disease progression milestones including use of canes/walkers/wheelchairs and artificial nutrition, the benefit of IV edaravone treatment still holds under RR <2 and MR <2, as the lower bound of the 95% confidence bounds of the treatment effect is above the solid zero curves. This indicates the estimated treatment effects for these 2 milestones are tolerable to relatively high deviations from the ignorability assumption, i.e., no unmeasured confounder. The RR <1.6 and MR <2 for noninvasive ventilation to hold the above-zero treatment effect also indicates a moderate robustness to unmeasured confounding.