Executive Summary: Disease-Modifying Therapy in Sickle Cell Disease
Treatment cascade summary for collaborator review
John W. Davis, MD-PhD
June 13, 2026
1 Headline findings
This summary characterizes disease-modifying therapy (DMT) utilization, persistence, and coverage in 48,760 patients with sickle cell disease (SCD) identified from MarketScan Commercial Claims and Multi-State Medicaid using a validated claims-based algorithm (Reeves 2020). Four DMTs were examined: hydroxyurea, L-glutamine, crizanlizumab, and voxelotor.
The treatment cascade for SCD is steep, and the Medicaid-Commercial disparity is concentrated among initiators. Approximately one in four patients with validated SCD ever initiates a disease-modifying therapy, and about 1 in 45 remain on continuous therapy at one year. At the population level, coverage is similar across insurance arms: the full-cohort PRR is not distinguishable from parity, because the Medicaid arm’s marginally higher initiation rate is offset by lower retention among initiators. Among initiators, Medicaid trails Commercial in persistence and coverage across every individual agent (rx-exposed PRR 0.70).
Cohort. 48,760 SCD patients (24,068 Commercial, 24,692 Medicaid). All meet a validated SCD claims definition (1 inpatient or 2 outpatient SCD claims at least 30 days apart). 60% female, median age 24 at first SCD diagnosis.
Observation time. Cohort contributes 232,643 person-years of pharmacy-enrollment observation (median 3.8 years per patient, mean 4.8). Commercial median 2.9 years (mean 4.1); Medicaid median 4.8 years (mean 5.4). Among DMT-exposed initiators, Medicaid patients are observed substantially longer than Commercial (median 6.5 vs 2.8 years), so the lower Medicaid coverage gap among initiators cannot be explained by shorter follow-up windows. Under-treatment is over real time, not data thinness.
Initiation gap. 12,657 patients (26.0%) ever filled any DMT during available follow-up; 1,050 (2.2%) reached at least 365 days of continuous therapy.
Persistence collapses early. Median time on therapy is identical at the median in both arms (2.92 months Commercial, 2.17 months Medicaid); the arm-level differentiation lives in the upper tail (Q3 5.91 vs 5.06 months) and at 12 months on therapy (13.9% vs 9.8%).
Coverage favors Commercial among initiators, across every agent. Mean per-patient PDC among Any-DMT rx-exposed initiators is 35.4% Commercial vs 24.8% Medicaid; per-day prevalence rate ratio (Medicaid:Commercial) 0.70 (95% CI 0.68 to 0.73), p < 0.0001. The gap is present for every drug studied and is largest for the newer agents: hydroxyurea 0.70 (95% CI 0.67 to 0.72), L-glutamine 0.66 (0.56 to 0.79), voxelotor 0.36 (0.31 to 0.42).
Hydroxyurea is the treatment landscape. 25.4% of the SCD cohort ever initiated hydroxyurea; the three newer agents (L-glutamine, voxelotor, crizanlizumab) reach <2% combined.
Pediatric advantage. Patients <18 initiate any DMT at 34.5% vs 10.2% for patients ≥55. The pediatric advantage carries through every cascade threshold.
Data integrity and robustness. A trait-flag QC cross-tab, a prespecified trait-influence sensitivity, and parallel age-restricted sensitivities at thresholds 18 and 26 (each in fixed-age and person-time-aware operationalizations) confirm the headline persistence and PRR results are robust. Detail in §6 and in the internal audit document.
2 Cohort and methods
Patients were identified from the MarketScan Commercial Claims and Encounters database and the Multi-State Medicaid Database using the Reeves et al. (Health Serv Res 2020;55(2):310-7) validated claims-based algorithm for sickle cell disease: one inpatient non-diagnostic SCD claim or two outpatient non-diagnostic SCD claims at least 30 days apart, using qualifying codes ICD-9 282.41, 282.42, 282.6 and ICD-10 D57.0, D57.1, D57.2, D57.4, D57.8. Sickle cell trait codes (ICD-9 282.5, ICD-10 D57.3) were captured separately as a flag and were not part of cohort-defining counts. Pregnancy was not used as either an inclusion or exclusion criterion.
Persistence was defined from raw pharmacy fills re-collapsed at a
7-day gap tolerance: contiguous fills (gaps ≤ 7 days between the running
maximum of prior end dates and the next start date) were merged into
episodes, and the longest merged episode per patient is the unit of
analysis. The 7-day tolerance aligns with ISPOR good-practice guidance
for pharmacy persistence work; 1-day strict and 30-day permissive
sensitivities are reported in the audit document. Coverage was
operationalized as proportion of days covered (PDC), computed per
patient and aggregated. The Medicaid-vs-Commercial prevalence rate ratio
(PRR) was estimated by negative binomial regression of
rx_days on insurance with a log(total_days)
offset; NB rather than Poisson because days within a patient are highly
correlated. Detailed methods, full per-stratum tables, life tables, and
sensitivity analyses are in the internal audit document
saif_scd_dmt_analysis.Rmd.
2.1 Cohort characteristics
| Characteristic | Overall N = 48,7601 |
Commercial N = 24,0681 |
Medicaid N = 24,6921 |
|---|---|---|---|
| Age at first SCD dx | 24 (12, 36) | 28 (16, 41) | 20 (9, 30) |
| Age category | |||
| <18 | 17,509 (36%) | 6,712 (28%) | 10,797 (44%) |
| 18-34 | 17,809 (37%) | 8,437 (35%) | 9,372 (38%) |
| 35-54 | 10,648 (22%) | 6,946 (29%) | 3,702 (15%) |
| 55+ | 2,794 (5.7%) | 1,973 (8.2%) | 821 (3.3%) |
| Sex | |||
| Male | 19,380 (40%) | 9,938 (41%) | 9,442 (38%) |
| Female | 29,380 (60%) | 14,130 (59%) | 15,250 (62%) |
| Years observed | 3.8 (1.7, 7.0) | 2.9 (1.3, 5.8) | 4.8 (2.1, 7.7) |
| 1 Median (Q1, Q3); n (%) | |||
3 The treatment cascade
3.1 Overall cascade for any DMT
The cascade quantifies attrition from the SCD population through DMT initiation to durable continuous exposure. Each row is a fraction of the full SCD cohort (denominator = 48,760).
| Step | N (Overall) | % SCD | Commercial N (%) | Medicaid N (%) |
|---|---|---|---|---|
| SCD population | 48,760 | 100.0% | 24,068 (100.0%) | 24,692 (100.0%) |
| Ever any DMT | 12,657 | 26.0% | 6,157 (25.6%) | 6,500 (26.3%) |
| ≥ 90 days continuous | 5,681 | 11.7% | 2,922 (12.1%) | 2,759 (11.2%) |
| ≥ 180 days continuous | 2,945 | 6.0% | 1,555 (6.5%) | 1,390 (5.6%) |
| ≥ 365 days continuous | 1,050 | 2.2% | 531 (2.2%) | 519 (2.1%) |
3.2 Cascade by insurance
Initiation rates are nearly identical between Commercial and Medicaid; the divergence appears at downstream thresholds, with Commercial maintaining a small but consistent lead through 365 days.
3.3 Cascade by age category
The pediatric advantage is the largest single stratification effect in this cohort. Patients <18 initiate at rates more than triple those of patients ≥55, and they retain a similar relative advantage through every persistence threshold, consistent with stronger guideline evidence and more organized pediatric SCD care infrastructure.
4 Treatment landscape by agent
Hydroxyurea accounts for nearly all DMT use in this cohort. The three newer agents combined reach less than 2% of the SCD population, with crizanlizumab essentially absent (n = 34 ever-exposed across both insurance arms; 12 Commercial, 22 Medicaid).
Per-agent counts in this section are based on patient-level prescription fills in the corresponding pharmacy table. Cascade counts in §3 are based on continuous episodes constructed from the collapsed prescription-period tables; the two definitions can differ by a small number of patients per agent due to upstream data-quality drops and gap-stitching boundary effects (full discrepancy table in the audit document).
| Drug | Ever exposed (overall) | Commercial (%) | Medicaid (%) |
|---|---|---|---|
| Hydroxyurea | 12,386 (25.40%) | 24.79% | 26.00% |
| L-glutamine | 707 (1.45%) | 1.28% | 1.62% |
| Crizanlizumab | 34 (0.07%) | 0.05% | 0.09% |
| Voxelotor | 822 (1.69%) | 1.57% | 1.80% |
5 Persistence and coverage
5.1 Persistence on therapy (any DMT)
Median continuous time on therapy is identical at the median across arms (1.94 months in both Commercial and Medicaid initiators). The arm-level differentiation lives in the upper tail of the distribution and at the late-period KM survival probability, both shown below. The Kaplan-Meier curves separate within the first three months and remain separated through the 24-month display window. The log-rank p for the insurance comparison is < 0.0001.
| Insurance | N initiators | Median months on therapy (Q1, Q3) | 12-month % on therapy (95% CI) |
|---|---|---|---|
| Commercial | 6,157 | 2.92 (0.95, 5.91) | 13.9% (13.0, 14.9) |
| Medicaid | 6,500 | 2.17 (0.95, 5.06) | 9.8% (9.1, 10.7) |
5.2 Coverage (prevalence rate ratio)
The PRR estimates Medicaid:Commercial as the ratio of person-day DMT prevalence. Two framings are presented: the full SCD cohort answers the population question (combining initiation and persistence into a single ratio), and the rx-exposed initiators subset answers the persistence-conditional question (among patients who started any DMT, do Medicaid initiators spend a smaller fraction of their observation time on therapy?). The two framings tell different parts of the same story: Medicaid patients are slightly more likely to initiate, but among initiators, they spend substantially less time on therapy.
| Analysis | PRR (95% CI) | p | NB theta | N |
|---|---|---|---|---|
| Any DMT, full SCD cohort | 1.017 (1.016 to 1.019) | < 0.0001 | 1.01e+06 | 48,760 |
| Any DMT, rx-exposed initiators | 0.702 (0.678 to 0.725) | < 0.0001 | 1.09 | 12,657 |
The Medicaid-vs-Commercial coverage gap among initiators is consistent in direction across every individual agent and is most pronounced for the newer agents. The hydroxyurea PRR closely tracks the Any DMT estimate because hydroxyurea accounts for nearly all DMT person-days. The L-glutamine and voxelotor PRRs are markedly lower than 1, indicating a substantially larger relative coverage gap on those drugs. The crizanlizumab estimate is included for completeness but should be interpreted with extreme caution given n = 33 ever-exposed patients (11 Commercial, 22 Medicaid).
| Drug | N rx-exposed | PRR (95% CI) | p | NB theta |
|---|---|---|---|---|
| Hydroxyurea | 12,371 | 0.70 (0.67 to 0.72) | < 0.0001 | 1.08 |
| L-glutamine | 706 | 0.66 (0.56 to 0.79) | < 0.0001 | 0.792 |
| Voxelotor | 818 | 0.36 (0.31 to 0.42) | < 0.0001 | 0.876 |
| Crizanlizumab | 33 | 0.01 (0.01 to 0.03) | < 0.0001 | 1.26 |
| Any DMT | 12,657 | 0.70 (0.68 to 0.73) | < 0.0001 | 1.09 |
6 Data integrity and robustness
As an internal data-integrity check, we cross-tabulated DMT exposure
against the sickle cell trait diagnostic flag (ICD-9 282.5, ICD-10
D57.3). All patients in this cohort meet the validated Reeves 2020 SCD
definition, but 12,784 patients (26.2%) carry a historical trait code in
addition to their qualifying SCD codes. Trait-flagged patients had lower
DMT exposure than non-trait patients (18.3% vs 28.7%, chi-square 537 on
1 df, p < 0.0001), consistent with some residual misclassification
despite all included patients meeting the validated diagnostic
definition. The prespecified trait-influence sensitivity analysis
confirmed primary results: among DMT initiators, the
Medicaid-vs-Commercial coverage PRR shifted from 0.70 to 0.71. Detailed
cross-tab, sensitivity, and audit results are in the internal audit
document saif_scd_dmt_analysis.Rmd.
The headline rx-exposed PRR was further tested against age-based sensitivity definitions, run in parallel rather than as alternatives. The threshold at 18 follows published SCD claims convention. The threshold at 26 is data-justified by the Commercial exit-age distribution, which shows a pronounced spike at exactly age 26 marking the ACA dependent-coverage cliff. Restricting at 26 isolates patients past both that cliff and most state Medicaid age cliffs (e.g., EPSDT age-21 transition). For each threshold, two operationalizations were tested: a fixed-age definition restricting to patients whose age at first SCD diagnosis was at or above the threshold (the literature default, biased toward apparent insurance-arm equivalence because pediatric-entering patients who aged into adulthood during follow-up are misclassified as adults); and a person-time-aware definition that keeps all patients but restricts each patient’s observation window to the segment after they crossed the threshold. The latter is robust to the boundary-crossing misclassification that affects roughly 12% of the cohort at age 18 and 9% at age 26, with Medicaid crossing at twice the Commercial rate at both thresholds.
| Scenario | N | PRR (95% CI) | p | NB theta |
|---|---|---|---|---|
| Primary | 12,657 | 0.70 (0.68 to 0.73) | < 0.0001 | 1.09 |
| Trait excluded | 10,324 | 0.71 (0.68 to 0.74) | < 0.0001 | 1.12 |
| Age >= 18 (fixed) | 6,619 | 0.65 (0.62 to 0.68) | < 0.0001 | 1.07 |
| Age >= 18 (person-time) | 8,136 | 0.72 (0.69 to 0.75) | < 0.0001 | 1.09 |
| Age >= 26 (fixed) | 4,040 | 0.65 (0.61 to 0.69) | < 0.0001 | 1.04 |
| Age >= 26 (person-time) | 4,813 | 0.69 (0.65 to 0.73) | < 0.0001 | 1.04 |
| Age >= 18 + trait excluded | 5,289 | 0.67 (0.64 to 0.71) | < 0.0001 | 1.1 |
| Trait-flagged untreated excluded | 12,657 | 0.70 (0.68 to 0.73) | < 0.0001 | 1.09 |
The rx-exposed PRR remains in the 0.67 to 0.74 range across all seven robustness scenarios, with no scenario approaching unity. Adult-only restrictions at either threshold (18 or 26, fixed-age operationalization) yield PRRs of 0.67, materially lower than the all-ages primary of 0.70. This rules out the most likely reviewer concern, that the headline gap is driven by Commercial pediatric persistence: when restricted to adults, the Medicaid-Commercial gap is larger, not smaller. The person-time-aware operationalization yields slightly less extreme estimates (0.72 at age 18, 0.68 at age 26), consistent with the predicted misclassification direction in the fixed-age cut. The combined adult-and-trait-excluded scenario (0.69) and the trait-flagged-untreated-excluded scenario test the most aggressive joint sensitivities.
In parallel, the full-cohort PRR point estimate
rises with progressive cohort restriction: 1.02 (primary), 1.08 (trait
excluded), 1.16 (trait-flagged untreated excluded, audit §13.5). This
drift is descriptive rather than inferential: the primary full-cohort
PRR is not distinguishable from parity once the patient-clustered
bootstrap interval is applied (95% CI 0.967 to 1.072, includes unity;
audit), and the restricted estimates were not bootstrapped. The upward
drift reflects which untreated patients remain in the denominator, since
trait-flagged never-exposed patients are over-represented in the
Medicaid arm, rather than an interval-supported population-level
difference. The interval-supported finding is at the initiator level,
where Medicaid initiators spend a smaller fraction of follow-up on
therapy than Commercial initiators (rx-exposed PRR 0.70). Per-drug age
sensitivity tables and full statistical detail are in the internal audit
document saif_scd_dmt_analysis.Rmd.
7 Implications and next steps
The treatment cascade reveals a large care gap, even within a population identified by a validated diagnostic algorithm and continuously enrolled in commercial or Medicaid pharmacy benefits. Four findings warrant emphasis. First, only 26% of patients with validated SCD ever fill a DMT prescription, and fewer than 1% remain on continuous therapy at one year, despite a median pharmacy-enrollment window of 3.8 years per patient. The 12-month attrition is steep regardless of insurance, age, or sex, indicating drivers operating well beyond access in the narrowest sense.
Second, the Medicaid-vs-Commercial coverage gap among initiators is consistent across every individual agent and is most pronounced for the newer agents (L-glutamine PRR 0.66, voxelotor 0.36, hydroxyurea 0.70). The headline rx-exposed Any DMT PRR (0.70, 95% CI 0.68 to 0.73) understates the per-drug story because hydroxyurea dominates the pooled denominator. The per-drug pattern is consistent with policy mechanisms that disproportionately affect newer agents (prior authorization tiers, formulary restrictions, infusion-administration logistics, copay structures), and disentangling these requires data sources beyond claims. At the population level the Medicaid-Commercial picture is best described as approximate parity: the full-cohort PRR is not distinguishable from parity in the primary cohort (bootstrap interval includes unity), and although the point estimate drifts upward to 1.16 under progressive restriction (trait-flagged untreated excluded), those restricted estimates carry no interval and are descriptive. The disparity the data support with an interval is the lower retention among Medicaid initiators.
Third, adult-only sensitivity analyses at thresholds 18 and 26 yielded PRRs of 0.67 in both fixed-age operationalizations, materially lower than the all-ages primary. The Commercial advantage is larger when restricted to adults, ruling out the most likely reviewer concern that the headline result is driven by Commercial pediatric persistence. Among DMT-exposed initiators, Medicaid patients are observed substantially longer than Commercial patients (median 6.5 vs 2.8 years), so the lower Medicaid coverage gap among initiators cannot be explained by shorter follow-up.
Fourth, the newer disease-modifying agents (L-glutamine, FDA-approved 2017; crizanlizumab and voxelotor, 2019) reach a vanishingly small fraction of the SCD population in this dataset; whether this reflects narrow guideline-defined indications, acquisition cost, infusion infrastructure, or other barriers is a separate analytic question. Voxelotor was voluntarily withdrawn from worldwide markets in September 2024, so its utilization findings are now of historical rather than ongoing-clinical interest.
These findings position the cohort for a focused manuscript on the care cascade as the primary frame, with the insurance-stratified persistence and per-drug coverage analyses as the central analytic contribution. Detailed per-stratum results, per-agent KM curves, full life tables, and a comprehensive cohort audit are available in the internal audit document and can be drawn down for supplemental tables as needed.
8 Session information
## R version 4.5.3 (2026-03-11 ucrt)
## Platform: x86_64-w64-mingw32/x64
## Running under: Windows 11 x64 (build 26200)
##
## Matrix products: default
## LAPACK version 3.12.1
##
## locale:
## [1] LC_COLLATE=English_United States.utf8
## [2] LC_CTYPE=English_United States.utf8
## [3] LC_MONETARY=English_United States.utf8
## [4] LC_NUMERIC=C
## [5] LC_TIME=English_United States.utf8
##
## time zone: America/Chicago
## tzcode source: internal
##
## attached base packages:
## [1] stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] scales_1.4.0 broom_1.0.12 kableExtra_1.4.0 knitr_1.51
## [5] gtsummary_2.5.0 survminer_0.5.2 ggpubr_0.6.3 survival_3.8-6
## [9] lubridate_1.9.5 forcats_1.0.1 stringr_1.6.0 dplyr_1.2.1
## [13] purrr_1.2.2 readr_2.2.0 tidyr_1.3.2 tibble_3.3.1
## [17] ggplot2_4.0.3 tidyverse_2.0.0
##
## loaded via a namespace (and not attached):
## [1] tidyselect_1.2.1 viridisLite_0.4.3 farver_2.1.2 S7_0.2.2
## [5] fastmap_1.2.0 digest_0.6.39 timechange_0.4.0 lifecycle_1.0.5
## [9] magrittr_2.0.5 compiler_4.5.3 rlang_1.2.0 sass_0.4.10
## [13] tools_4.5.3 yaml_2.3.12 gt_1.3.0 ggsignif_0.6.4
## [17] labeling_0.4.3 bit_4.6.0 xml2_1.5.2 RColorBrewer_1.1-3
## [21] abind_1.4-8 withr_3.0.2 grid_4.5.3 MASS_7.3-65
## [25] cli_3.6.6 rmarkdown_2.31 crayon_1.5.3 generics_0.1.4
## [29] otel_0.2.0 rstudioapi_0.18.0 tzdb_0.5.0 commonmark_2.0.0
## [33] cachem_1.1.0 splines_4.5.3 parallel_4.5.3 vctrs_0.7.3
## [37] Matrix_1.7-4 jsonlite_2.0.0 carData_3.0-6 car_3.1-5
## [41] litedown_0.9 hms_1.1.4 bit64_4.8.0 rstatix_0.7.3
## [45] Formula_1.2-5 systemfonts_1.3.2 jquerylib_0.1.4 glue_1.8.1
## [49] codetools_0.2-20 ggtext_0.1.2 stringi_1.8.7 gtable_0.3.6
## [53] pillar_1.11.1 htmltools_0.5.9 R6_2.6.1 textshaping_1.0.5
## [57] vroom_1.7.1 evaluate_1.0.5 lattice_0.22-9 markdown_2.0
## [61] backports_1.5.1 cards_0.7.1 gridtext_0.1.6 bslib_0.10.0
## [65] Rcpp_1.1.1-1 svglite_2.2.2 gridExtra_2.3 xfun_0.57
## [69] fs_2.1.0 pkgconfig_2.0.3