This retrospective cohort study used the Optum Research Database (ORD) of administrative medical and pharmacy insurance claims for commercial and Medicare Advantage enrollees in the USA. The ORD, which is fully deidentified and used in a Health Insurance Portability and Accountability Act of 1996 (HIPAA)-compliant manner, comprises medical and pharmacy claims data (including linked enrollment) from 1993 to present on more than 90 million lives. This secondary data source covered approximately 8% of the US commercially insured population and 18% of the Medicare Advantage population (with medical and pharmacy claims). Claims include associated health-plan and patient-paid amounts. Additional data, which include socioeconomic (SES) and social determinants of health (SDoH) characteristics that are not available in typical claims data, were linked from additional Optum data sources using unique patient identification numbers. Patients with missing SES or SDoH data remained in the study population. As no identifiable protected health information was accessed in the conduct of this study, institutional review board approval or waiver of approval was not required.

Patients eligible for the study had evidence of incident HR-MDS between 1 January 2017 and 30 April 2022. Evidence of HR-MDS was defined as having at least one inpatient medical claim with an International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) diagnostic code for HR-MDS (D46.20–D46.22) or at least two outpatient claims with an ICD-10-CM diagnostic code for MDS (D46.0–46.9, D46.A–D46.Z) or HR-MDS on separate dates but within 1 year and with at least one of these claims coded as HR-MDS [14]. The earliest date of HR-MDS diagnosis was the index date. Patients were required to be at least 18 years old and have continuous health plan enrollment for at least 12 months prior to (baseline period) and at least 3 months on or after the index date (or less if due to death) (follow-up period). They were followed until earliest AML diagnosis, death, disenrollment, or study end date, whichever occurred first. Patients were excluded if they had any claims with a diagnosis of HR-MDS in the 12-month baseline period to approximate a “newly diagnosed” HR-MDS study population. In addition, patients were excluded if they had evidence of a stem cell transplant, other primary cancers (except nonmelanoma skin cancer), metastatic disease, acute myeloid leukemia (AML), myelofibrosis, or pregnancy in the baseline period. Patients who had evidence of MDS but no diagnoses of HR-MDS during baseline were retained in the study.

Patient demographic, SES, and clinical characteristics were measured during the baseline period. Patients were also evaluated for a prior MDS diagnosis defined as either having a claim for MDS prior to the set of claims used to establish the index date or receiving active treatment for MDS during the baseline period.

The study measured social determinants of health (SDoH) using indices for five SDoH domains: financial stress, food insecurity, housing insecurity, social isolation, and transportation difficulty. Each SDoH index is based on predictive models that use a combination of individual-level consumer data and publicly available data by geography to predict the likelihood of an individual having unmet needs in each of the SDoH domains. Optum uses proprietary steps to review and validate a model for each SDoH domain. Briefly, Lasso regression is used to reduce dimensionality, an XGBoost model is used to choose the top important features, a correlation analysis is used to identify multicollinearity, and top features are selected for the final model [15]. Each model is validated on unseen data, and the area under the curve is reviewed for sensitivity and specificity levels. The SDoH indices are scaled from 0 to 9, with 0 indicating the highest level of unmet need—or patients having more insecurity—and 9 indicating the lowest risk of insecurity for that domain. In this study, patients were further categorized into tertiles of a high level of unmet need (0–3), moderate level of unmet need (4–6), and no unmet need (≥ 7). The transportation access measure was developed and is presented for patients aged ≥ 65 years only.

MDS-related signs, symptoms, and diagnostic procedures were selected on the basis of clinical relevance and identified using codes. MDS-related signs/symptoms identified using diagnosis codes included anemia, neutropenia, bleeding, thrombocytopenia, bruising, dyspnea, infections, Sweet syndrome, and fatigue. The time between the earliest claim for each sign/symptom and the index date was calculated for each patient. Patients were then classified by the earliest sign/symptom present during the baseline period, and the time from earliest sign/symptom to the index date was presented stratified across demographic and clinical characteristics. Similarly, the earliest of claims with procedure codes for a bone marrow biopsy and aspiration, blood cell counts, cytogenic analysis, flow cytometry, immunohistochemistry, gene sequencing, and human leukocyte antigen (HLA) typing were identified for each patient. Time between each procedure and the index date was calculated, and patients were classified by the earliest diagnostic procedure present during the 12-month baseline period. Time from HR-MDS diagnosis to progression of AML was assessed.

Patients were classified on the basis of treatments received during the variable follow-up period. If patients received multiple treatments, they were classified into categories on the basis of the following order of more definitive treatment: (1) HSCT, (2) active treatment (azacitidine, decitabine, decitabine and cedazuridine, lenalidomide, cyclosporine, thalidomide, anti-thymocyte globulin, clofarabine, daunorubicin, idarubicin, cytarabine, fludarabine, venetoclax, and topotecan), (3) MDS supportive care treatments (hydroxyurea and hematopoietic growth factors), and (4) none of the treatments in categories 1–3.

Baseline characteristics and treatment patterns were analyzed descriptively. Length of time between the initial presence of a sign or symptom during the 12-month baseline period and the HR-MDS diagnosis was classified into a longer (> 6 months) versus shorter time to diagnosis. A multivariable logistic regression model was used to identify baseline demographic, SES, SDoH, and clinical characteristics that were associated with a longer versus shorter time to diagnosis. A multivariable proportional hazards regression model was used to describe the association between baseline demographic, SES, SDoH, and clinical characteristics and progression to AML. A backward stepwise selection process was used to identify characteristics with the strongest relationships to remain in each final model. Covariables remained in the model if they had a p-value < 0.1. We examined variance inflation factors (VIFs) in the final models to confirm that there were no issues with multicollinearity in the final model.

A p-value of < 0.05 was considered statistically significant, and all analyses were conducted using SAS statistical software version 9.4 (SAS Institute, Cary, NC, USA).