Healthcare utilization and guideline-directed medical therapy in heart failure patients with reduced ejection fraction

This study used data from the Optum^® Integrated File, which contains data from both claims and electronic health records for patients across the USA [7]. To be included in this integrated file, patients must be enrolled in United Healthcare insurance and have at least one hospital encounter in the electronic health records data in the window of data availability (2007–Q3 2019). The combination of claims and clinical data provides a comprehensive view of a patient’s clinical interactions with the healthcare system. Optum data provides a continuum of treatment and cost information, such as medications by therapeutic area, provider notes with treatment rationale and cost by procedure and condition. For this analysis, we utilized the electronic health record data to assess ejection fraction since these measurements would be recorded in the hospital setting. We utilized the claims data to assess comorbidities, drug utilization and outcomes since this information is consistently contained in the longitudinal United Healthcare payer database versus the inconsistent information from the electronic health records portion of the integrated data.

Patients in the Optum Integrated File from 1 January 2007 through 30 June 2020 with a record of HFrEF were identified by a series of patient selection queries. HFrEF was measured by either: a record of an ejection fraction of ≤40% from the Humedica electronic health records portion of the Optum integrated database; or diagnosis of systolic HF via International Classification of Diseases (ICD) coding followed by at least one prescription fill for an angiotensin receptor neprilysin inhibitor (ARNI), which is almost exclusively used in patients with HFrEF, from the United Healthcare payer database. Patients were required to be 18 years of age or older and were excluded if they had Stage V chronic kidney disease (CKD) or end stage renal disease. Patients were required to have 6 months of continuous enrollment in the payer database both before and after their HFrEF index date to ensure information consistency in the payer database to capture comorbid conditions, HF drug utilization and other study variables. The initial 6 months after HFrEF diagnosis was considered a landmark period, during which drug regimens could be added and titrated according to recommendations [8]. Data used to perform this analysis was de-identified and accessed in compliance with the Health Insurance Portability and Accountability Act. As a retrospective analysis of a de-identified database, the research was exempt from Institutional Review Board (IRB) review under 45 Code of Federal Regulations (CFR) 46.101(b)(4).

Patients with a prescription record of both a BB and either ACE-I, angiotensin receptor blocker (ARB) or ARNI during the 6-month landmark period following their HFrEF diagnosis were considered to be on at least minimal GDMT and assigned to the GDMT cohort. Patients not meeting criteria were assigned to the not on GDMT cohort even if the they had a record of being on one of the two classes of medications. The GDMT cohort was further classified by patients having a record of filling prescriptions in both classes (BB and ACEI/ARB/ARNI) concurrently within the 6-month landmark period, which was labeled GDMT STRICT in the analysis. For GDMT STRICT, a patient had to have at least 1 day of overlapping prescriptions with up to a 7 day gap in both classes.

Outcomes of interest for this analysis included the following healthcare utilization end points, on an annualized basis: all cause hospitalizations, cardiovascular (CV) related hospitalizations, hospital days, emergency department (ED) visits, outpatient visits and office visits. All outcomes were measured up to 2 years from the end of the landmark period (6 months after a patient’s diagnosis of HFrEF in which drug regimens were assessed).

The covariates for this analysis included patient demographics and comorbidities. Demographics included age, sex, race and region. Comorbidities of interest were identified using diagnosis codes that appeared in the 6-month period before a patient’s index diagnosis of HFrEF. Since patients with concomitant diseases may be contraindicated for certain GDMT therapies, patients with a record of the following comorbid conditions prior to their HFrEF diagnosis were considered: atrial fibrillation, hypertension, coronary artery disease and chronic kidney disease. Finally, although diuretics are not part of the strict definition of GDMT, patients with a record of loop diuretic usage during the landmark period were also followed.

All cohorts (GDMT, GDMT STRICT and not on GDMT) were described by patient characteristics and specific comorbid conditions. Descriptive analytics were presented as mean and standard deviation for continuous variables or count and percentage for categorical variables. All healthcare utilization outcomes were estimated for the GDMT versus the not on GDMT cohort using general linear models with multiple data distribution considerations. All models were adjusted for patient demographics with the following chronic conditions: atrial fibrillation, hypertension, coronary artery disease, chronic kidney disease and record of diuretic use. All utilization models were repeated for the GDMT STRICT versus not on GDMT cohort. Model results are reported as least square mean estimates and p-values.

The major strength of this study is that it characterizes HF by ejection fraction, which many large retrospective studies cannot due to the limited types of data available in an administrative claims database. In addition, there is limited data on the association between GDMT and healthcare utilization costs. Much of the current research to date has cohorts stratified by ejection fraction, type of medication, gender or a given comorbidity. This study uses a cohort of nonoptimally treated patients to more meaningfully examine the direct relationship between GDMT and utilization outcomes such as days spent in the hospital. In this analysis, claims data are combined with electronic health records and adjustments were made for patient characteristics known to affect prognosis and; therefore, intensity of care. The broad selection criteria employed in the use of a large dataset also helped preserve generalizability.

As with all studies that rely on automated sources of data, it is possible that parameters such as billing codes could be biased and proxies could fail to capture certain factors difficult to ascertain from the available clinical data. Our use of a proxy (prescription of ARNI) for reduced EF as an alternative to echocardiogram may skew our cohorts toward a higher percentage of patients on GDMT, since ARNI is one qualifying component of the drug regimen. However, 94.2% of our HFrEF population of interest had a record of an echocardiogram result of an ejection fraction of ≤40% and only a small percentage (7%) relied on ICD coding and a record of taking an ARNI to meet inclusion criteria. We were unable to adjust for vital signs, laboratory data, region and other potential confounding measures due to the lack of clinical parameters in this real world data source. Consequently, we were unable to differentiate patients that are not eligible for GDMT (due to heart rate, blood pressure, renal function, etc.), also recognizing that inability to tolerate GDMT portends worse outcomes. However, we adjusted for baseline differences in many comorbid conditions that may deem patients ineligible for GDMT, for example, chronic kidney disease.

Finally, a continuous enrollment both 6 months prior to and after HFrEF diagnosis was needed for inclusion. This stringent criterion was necessary to ensure data collection on baseline comorbidities and medication regimens, but may induce bias toward healthier patients (i.e., those with early death during the first 6 months would not be included in the analysis). We expect that the relationship between GDMT and healthcare utilization should be preserved, as the same continuous enrollment periods were applied to each cohort. Patient adherence may also be inadequately captured by the use of prescription fills to determine medication status, as not everyone who fills their prescription is taking the drugs as intended. We treated cohort status as static during the outcome period; although, it is possible that patients could have changed cohorts by altering their drug regimen more than 6 months after their HFrEF index date to either be on GDMT or not on GDMT. This cohort crossover could have biased our results. Given the patients represented in our analysis, results are generalizable only to the commercially insured patient population.