Correspondence to Dr Samrachana Adhikari; [email protected]
STRENGTHS AND LIMITATIONS OF THIS STUDY
By linking pharmacy data from Surescripts with electronic health record (EHR), we were able to overcome important prior limitations of insurance claims data to assess medication adherence, which are unable to capture lapses in prescription or primary non-adherence when a medication is prescribed but never filled.
Compared with relying on insurance claims data alone, our linked EHR-pharmacy database with Surescripts data will have more complete information about medication fills irrespective of patients’ insurance plan, because relying on information available through insurance claims limits adherence data to beneficiaries of specific insurance plans.
We further augment patient-level EHR data with linked neighbourhood-level social determinants of health data, which allowed us to investigate factors beyond individual characteristics traditionally available in EHR and has the potential to strengthen non-adherence predictions as well as identify social factors that might be targeted for adherence interventions for patients with heart failure.
The cohort is derived from a single health system based in greater New York City and thus may not reflect the underlying population, due to selection of patients into the EHR; additionally, because patients in our cohort reside in mostly urban metropolitan areas, our findings may not be generalisable to patients who live in non-urban areas.
Considering the heterogeneity in disease presentation and treatment for the broader cardiovascular diseases, our findings from this cohort of patients with heart failure may have limited generalisability to patients with other cardiovascular diseases.
Introduction
Over six million people in the USA suffer from heart failure,1 a highly morbid and expensive condition, with an estimated 5 year mortality of up to 50%,2 and an associated US$30 billion in healthcare expenditures nationwide.1 Heart failure is also a leading cause of hospitalisation and accounts for nearly one million hospital admissions annually. Over one-fifth of such hospitalisations result in a repeat hospital admission within 30 days.3 4 Evidence-based medications can improve outcomes for these patients. Clinical practice guidelines for heart failure with reduced ejection fraction recommend multiple classes of medications. These include a beta-blocker; an ACE inhibitor, angiotensin receptor blocker (ARB) or angiotensin receptor neprilysin inhibitor (ARNI), collectively ACEi/ARB/ARNI; a mineralocorticoid receptor antagonist (MRA) and a sodium-glucose cotransporter 2 inhibitor (SGLT2i).2 These four classes of medications have been proven to reduce mortality and hospitalisation for patients with heart failure,5–9 and together, are estimated to result in over 50% reduction in mortality.10 Nonetheless, adherence to these medications is necessary for them to be effective in practice.2
Medication non-adherence in heart failure patients is common,11–14 particularly among vulnerable patient populations. A number of multilevel factors influence medication adherence.15 There are significant variations in rates of adherence by individual patient characteristics, healthcare team and provider characteristics, and neighbourhood social determinants.11 In particular, black and Hispanic patients,16 17 patients with low socioeconomic status,18 those who lack social support, and have complex treatment regimen19 20 often face higher barriers to maintaining optimal adherence. These barriers are further exacerbated by therapy-related factors, such as copayment21 and neighbourhood-level factors, such as pharmacy access.22
The WHO defines adherence as the extent to which a person’s medication-taking behaviour corresponds with recommendations from a healthcare professional.15 Direct patient observation is the most accurate method of determining adherence; however, it is generally not feasible on a population level. Pharmacy refill data, either from insurance claims, or directly from pharmacies, are popular indicators to measure medication adherence retroactively in a large scale. However, their utility in clinical care has been limited for a number of reasons. First, they may not be readily available at the point of care; because they lack information needed to capture temporary discontinuations in prescription or when medications are prescribed but never filled.23 Second, the databases may be limited to patients with specific insurance providers or pharmacy of choice.24 As such, medication information available through the insurance claims are limited to patients who are the beneficiaries of the insurance plan. And, such data alone do not include important clinical information, such as laboratory values, vital signs or echocardiogram results. Alternatively, connecting electronic health record (EHR) to the retail filling data from the pharmacies could provide a more comprehensive and inclusive data source to measure adherence.25 Linkage of EHR with pharmacy databases will also have more complete information about medications regardless of the patient’s insurance plan.25 Additionally, embedding pharmacy fill data within an EHR could allow for capturing adherence in a real-time manner that can be accessible during clinical care.
Prior studies have demonstrated clinic-based or community-based interventions, such as behavioural counselling26 and coaching on medication adherence27 28 can improve adherence. However, opportunities for such interventions are frequently missed, as providers are often unable to recognise risk patterns for medication non-adherence.29 Machine learning-based predictive models, which have been shown to support provider decision making,30 31 can support providers in identifying patients with high likelihood of non-adherence. Prior models to predict adherence, however, have been limited by data availability. For instance, models that incorporate both pharmacy dispense records and insurance claims data can achieve moderate predictive accuracy for identifying risk for future pharmacy fill non-adherence, but they lack many important data elements available in the EHR32 33 and other relevant variables known to influence adherence, including neighbourhood social determinants of health.34 35
We have established an EHR-based cohort of heart failure patients with comprehensive data elements from multiple sources to improve on existing medication adherence prediction models. In our cohort, data from the EHR are linked with pharmacy fill data for real time incorporation of prescription fills and have been supplemented with linked social determinants data to incorporate neighbourhood-level factors. These data can be used to retrospectively develop models for making accurate prediction of medication non-adherence for heart failure patients, and will have the capability to identify barriers to adherence, which can assist in developing and implementing interventions at multiple levels to address this important issue.36
Cohort description
Design and study setting
We developed a retrospective cohort study of patients seen at New York University (NYU) Langone Health, a large health system that includes four acute care hospitals and over 350 ambulatory clinics in New York City (NYC), the surrounding metropolitan area and Florida. The health system uses a single EHR, Epic (Epic Systems, Verona, Wisconsin, USA).
Data sources
The primary data source was the EHR at NYU Langone Health. The EHR contains demographic information including patient address; clinical variables such as comorbidities and vital signs; laboratory findings; structured data from relevant tests such as ejection fraction (EF) from echocardiograms; medication orders; primary pharmacy including address; healthcare utilisation including within-hospital hospitalisations, emergency visits and outpatient clinic visits; and provider characteristics such as specialty. Additionally, the EHR has pharmacy fill information through linkage to pharmacy and pharmacy benefits manager data available through Surescripts (Surescripts, Arlington, Virginia, USA), which supports electronic prescriptions. Additionally, we used a number of publicly available datasets to assess neighbourhood-level social determinants of health across domains of social and economic factors, built environment, pharmacy and health landscape, and environmental factors. These data sources included the decennial Census, American Community Survey, National Neighborhood Data Archive and the Centers for Disease Control and Prevention.37 For a large subset of patients who resided in urban areas, including NYC, additional social determinants of health data elements were collected from NYC government agencies.
Inclusion and exclusion criteria
Our population included all patients aged 18 years and older with heart failure who were seen in at least one clinical encounter (outpatient visit, telehealth visit, emergency department visit or hospitalisation) at NYU Langone Health between 1 April 2021 and 31 October 2022. Heart failure was defined as having either: (1) a heart failure diagnosis, as defined by International Classification of Diseases, 10th Revision (ICD-10) codes endorsed by the Center for Medicare and Medicaid Services38 and similar to our prior work3 39–41 or (2) a reduced EF, defined as less than or equal to 40%. We included patients with an active prescription from at least one of the following four medication classes: beta-blocker, ACEi/ARB/ARNI, MRA or SGLT-2 classes. Patients were excluded if they did not have a geocodable address or lived outside of the USA. For interpretability and reliability of neighbourhood social determinants of health data, we limit our cohort to patients with geocodable addresses. However, we will plan on sensitivity analyses including patients with missing address information as well.
Study variables
Predictors
We derived predictors primarily from the EHR with additional neighbourhood social determinants of health data from linkages to publicly available data sources. The predictors were informed by the conceptual model of factors that affect medication adherence (figure 1, adapted from the WHO42), which divides predictors of adherence into five categories: social determinants of health, healthcare team and system-related factors, condition-related factors, therapy-related factors and patient-related factors.43 We mapped the factors from the WHO conceptual model onto a socioecological model to conceptualise the ways in which these five categories of predictors influence adherence both directly and via their nested interrelationships. Socioecological models are frequently used as conceptual frameworks in the context of both cardiovascular disease and medication adherence.44–46
Enlarge this image.Figure 1. Conceptual diagram linking various determinants (adapted from the WHO definitions) to medication adherence. The five factors from the WHO conceptual model were mapped onto a socioecological model to conceptualise the ways in which these five categories influence adherence both directly and via their nested interrelationships.
We preprocessed EHR data elements available in each visit in order to obtain the predictors listed in table 1. Demographic variables were collected on each visit. As a measure of heart failure severity, blood pressure, heart rate, laboratory values and EF from echocardiogram were included whenever available. Individual comorbidities (guided by Charlson and Elixhauser Comorbidity Indices) were derived from problem list, medical history and encounter diagnoses using ICD-10 codes. These ICD-10 codes are presented in online supplemental file 1. We additionally derived a number of therapy-related variables from EHR, including number of heart failure medications prescribed, length of prescription, frequency and dosing of medications, and initiation of new medications. Prior adherence based on proportion of days covered (PDC) measure47 for the prior 6 months to heart failure medications was included when available. We also measured healthcare and system-related factors. For example, outpatient provider continuity was measured for each patient using the usual provider of care index, which we defined as the ratio of the number of visits with the most frequently seen cardiologist, internal medicine or medicine subspecialty provider to the total number of visits to these providers over a 1-year period.48–50
Table 1Predictors derived from EPIC EHR by categories in the conceptual model
| Category | Variable | Definition |
| Patient-related factors | Age | Age at the time of visit |
| Sex | Male/female | |
| Self-reported race | Black, White, Asian, other, missing | |
| Ethnicity | Hispanic/Latino, not Hispanic/Latino | |
| Language | First language of patient | |
| Missed appointments | Number of missed appointments in the past year prior to the visit | |
| Encounter type | In-person or a telemedicine visit | |
| Condition-related factors | Hospitalisations | Number of hospital admissions in previous year prior to the visit |
| Heart failure (HF) hospitalisations | Number of HF-specific hospital admissions in previous year | |
| Clinic visits | Number of clinic visits in the last year | |
| HF clinic visits | Number of clinic visits related to HF in the last year | |
| ED visits | Number of ED visits in the last year | |
| HF ED visits | Number of ED visits related to HF in the last year | |
| Potassium | Most recent value available for the visit | |
| Sodium | Most recent value available for the visit | |
| Creatine | Most recent value available for the visit | |
| eGFR | Most recent value available for the visit | |
| BPN | Most recent value available for the visit | |
| NT-ProBNP | Most recent value available for the visit | |
| Haemoglobin | Most recent value available for the visit | |
| Ejection fraction | Most recent value available for the visit | |
| Heart rate | Most recent value available for the visit | |
| Blood pressure | Most recent value available for the visit | |
| Comorbidities | Diagnosed conditions (Elixhauser ComorbidityIndex aside from HF that are documented in problem list and/or medical history; | |
| Therapy-related factors | Number of medications prescribed | All medications prescribed, including HF medications |
| Number of days on medications | Length of prescription for HF medication | |
| Frequency of medication changes | How often provider changes the prescribed HF medications | |
| Presence of medication allergies | Number of medication allergies | |
| Medication initiation or renewal | Initiation of a new HF medication or renewal of an HF medication | |
| Dosing frequency | How many times a patient has to take their HF medication per day (once daily, twice daily, etc) | |
| Insurance type | Patient’s insurance coverage/status | |
| Healthcare team and system | Medicare drug coverage formularies | Medicare coverage for the prescribed HF medication (yes or no) |
| Copayment amount | Patient medication copayments | |
| Prior authorisations | For each HF medication whether prior authorisation is needed | |
| Provider specialty | Main practising specialty of the provider | |
| Sub specialty training | Other specialty trainings beyond main practice | |
| Insurance mix | Insurance plans accepted by the provider | |
| Provider Continuity Index | Ratio of the number of visits with the most frequently-seen cardiologist, internal medicine or medicine subspecialty provider to the total number of visits to these providers over a 1-year period | |
| Frequency of follow-ups | Whether the patient was seen at least once in that last 6 months prior to the visit |
BPN, Brain natriuretic peptide; ED, emergency department; eGFR, estimated glomerular filtration rate; EHR, electronic health record; HF, heart failure; NT-ProBNP, N-terminal pro-brain natriuretic peptide.
In order to link individuals to neighbourhood-level social determinants of health, we used their home address to map each individual to publicly available community-level social determinants data. We aimed to capture both compositional and contextual neighbourhood-level determinants of health across domains of social and economic factors, built environment, pharmacy and health landscape, and environmental factors (figure 2). Compositional place effects, such as median household income, are the result of aggregating individual-level characteristics of a neighbourhood’s population while contextual place effects, such as levels of air pollution are only measured at the neighbourhood level.51
Enlarge this image.Figure 2. Selected census tract-level variables from four primary domains influencing heart failure and medication adherence along with related publicly available data sources derived for our cohort.
We geocoded patient addresses using geocodio, a secured programme within NYU Langone firewall that returns the longitude and latitude of any address from the USA. The longitude and latitude were then linked with the corresponding census tract based on the 2010 census boundaries, to align with the boundaries for neighbourhood variables measured prior to 2020 census. These census tracts based on the patient’s home address were then linked to publicly available datasets to obtain neighbourhood-level social determinants. Using various publicly available data sources, we computed neighbourhood-level variables related to five domains of neighbourhood-level social determinants of health.52 These domains include economic stability (eg, poverty rate, rent burden, median household income), healthcare access and quality (eg, availability of ambulatory healthcare services, availability of pharmacy), social and community context (eg, racial/ethnic diversity, internet access), neighbourhood and built environment (eg, means of transportation to work, access to public transit, access to greenspace) and education access and quality (eg, neighbourhood-level educational attainment). Further, census tracts were assigned to urban, suburban and rural community types, using the definitions from the Rural Urban Commuting Area codes.53 We also geocoded patient’s pharmacy location information in the EHR and calculated its geographical distance from the geocoded address of the patient. Figure 2 displays various interconnected neighbourhood-level social determinants along with related publicly available data sources currently considered in our study. These data elements will continue to be expanded for our cohort as we identify new sources.
Adherence from pharmacy fill data
The primary outcome was medication adherence at 6 months following a clinical encounter (outpatient visit, telehealth visit or hospitalisation) for four evidence-based therapies for heart failure: beta-blockers, ACEi/ARB/ARNI, MRA and SGLT2i. A common measure of adherence from refill data is the proportion of days covered or PDC.54 PDC was defined as the ratio of the number of days covered by medication fills of a given class within a 6-month time period and the number of days in that time period,47 and was calculated through the use of EHR data, to determine whether a patient had an active prescription for a medication, and Surescripts data, to identify medication fills. Consistent with convention,47 55 a patient was defined as being adherent to therapy when the average PDC across classes was above 80%.55 A PDC >80% is widely used as a threshold for good adherence because it is associated with reductions in clinically meaningful outcomes,56 57 and is the level used by health plans for quality improvement purposes.58 The current cohort was followed until 30 April 2023 to ensure everyone had at least 6 months follow-up of prescription fill data.
Patient and public involvement
None.
Findings to date
A flow chart describing inclusion and exclusion is shown in figure 3. Of the 45 621 eligible patients who were >18 years old with diagnosis of heart failure or reduced EF (≤40) since 2017, 43 935 patients had at least one encounter between 1 April 2021 and 31 October 2022. Among them, 39 986 patients had active medication prescriptions for any of the 4 guideline-directed medication therapies (GDMTs) during the study period. Fifteen patients who did not have a geocodable address (eg, missing street information, PO box) and eight patients who resided in territories outside of the continental USA were excluded. Thus, our final cohort included 39 963 patients.
Enlarge this image.Figure 3. Flow diagram with inclusion exclusion criteria. ARB/ARNI; angiotensin receptor blocker/angiotensin receptor neprilysin inhibitor; GDMTs, guideline directed medication therapies; MRA, mineralocorticoid receptor antagonist; SGLT2i, sodium-glucose cotransporter 2 inhibitor.
Individual patient characteristics are summarised in table 2. While the predictors from the EHR were based on the time of the clinical encounter, for patients with multiple encounters, to create patient-level summary we considered demographic variables from the first encounter and clinical variables from the last encounter, when these variables were available in the study time period. The average age of the cohort was 73±14 years old, 44% were female and 48% of the cohort were current or former smokers. Seventy-four per cent of patients had Medicare insurance, followed by 18% on commercial insurances and 7% on Medicaid. The self-reported race/ethnicty was 65% non-Hispanic White, 13% non-Hispanic Black, 10% Hispanic and 4% non-Hispanic Asian. About 7% of patients had either missing or unknown self-reported race and ethnicity information. Missingness on smoking status and insurance was low (<1%). The most common comorbid conditions observed were uncomplicated hypertension (77%), cardiac arrhythmias (56%), obesity (33%) and valvular disease (33%). During the interval of 2 years, 33 606 (84%) patients had an active prescription of beta blocker, 32 626 (82%) had ACEi/ARB/ARNI, 11 611 (29%) had MRA and 7472 (19%) had an active prescription of SGLT2i. Average number of heart failure GDMTs per patient was 2.13±0.89 during the study period. Ninety-nine per cent of the patients in the cohort resided in urban metropolitan areas and 94% resided in New York state.
Table 2Summary of selected characteristics of the 39 963 patients in the cohort with at least one encounter between 1 April 2021 and 31 October 2022
| Age, mean (SD) years | 73.00 (13.56) |
| Race and ethnicity, n (%) | |
| Non-Hispanic White | 25 894 (64.8%) |
| Non-Hispanic Black | 5115 (12.8%) |
| Non-Hispanic Asian | 1693 (4.2%) |
| Non-Hispanic Pacific Islander/Native Hawaiian/American Indian | 69 (0.2%) |
| Non-Hispanic others | 12 (0.0%) |
| Non-Hispanic multiracial | 143 (0.4%) |
| Hispanic | 4089 (10.2%) |
| Refused/unknown | 2620 (6.6%) |
| Missing | 328 (0.8%) |
| Language, n (%) | |
| English | 33 680 (84.3%) |
| Spanish | 1826 (4.6%) |
| Russian | 2169 (5.4%) |
| Chinese | 365 (0.9%) |
| Bengali | 124 (0.3%) |
| Smoking status, n (%) | |
| Current | 2281 (5.7%) |
| Former | 17 108 (42.8%) |
| Never | 20 435 (51.1%) |
| Unknown | 88 (0.2%) |
| Missing | 51 (0.1%) |
| Insurance, n (%) | |
| Medicare | 29 707 (74.3%) |
| Medicaid | 2940 (7.4%) |
| Commercial | 7152 (17.9%) |
| Other | 54 (0.1%) |
| Missing | 110 (0.3%) |
| Sex, n (%) | |
| Male | 22 228 (55.6%) |
| Female | 17 733 (44.4%) |
| Unknown | 2 (0.0%) |
| 10 common comorbidities, n (%) | |
| Hypertension, uncomplicated | 30 621 (76.8%) |
| Cardiac arrhythmias | 22 496 (56.4%) |
| Obesity | 13 105 (32.9%) |
| Valvular disease | 13 303 (33.4%) |
| Peripheral vascular disorders | 12 868 (32.3%) |
| Diabetes, uncomplicated | 12 041 (30.2%) |
| Renal failure | 10 389 (26.0%) |
| Fluid and electrolyte disorders | 10 162 (25.5%) |
| Chronic pulmonary disease | 10 030 (25.1%) |
| Diabetes, complicated | 8968 (22.5%) |
| Heart failure medication prescription, n (%) | |
| Betablocker | 33 606 (84.3%) |
| ACEi/ARB/ARNI | 32 626 (81.8%) |
| MRA | 11 611 (29.1%) |
| SGLT2i | 7472 (18.7%) |
| Number of heart failure medication prescription by person, mean (SD) | 2.13 (0.89) |
| Community type, n (%) | |
| Urban metropolitan area | 39 567 (99.0%) |
| Suburban area | 315 (0.8%) |
| Rural area | 80 (0.2%) |
ARB, angiotensin receptor blocker; ARNI, angiotensin receptor neprilysin inhibitor; MRA, mineralocorticoid receptor antagonist; SGLT2i, sodium-glucose cotransporter 2 inhibitor.
Strengths and limitations
There are many notable strengths of the cohort we have established. Our cohort is among the first large cohorts that uses EHR and linked pharmacy fill data to assess medication adherence, which enables us to use both medication orders and medication fills in consideration of adherence at a population level. As a result, we are able to overcome important limitations of prior studies that use insurance claims data without EHR linkages to study medication adherence. They have limited utility for real-time clinical interventions for several reasons. First, such data are unable to capture lapses in prescription or primary non-adherence, when a medication is prescribed but never filled. Second, insurance claims data are only available for patients covered under a given insurance plan, and therefore, medication information available through the claims are limited to patients who are the beneficiaries of the plan. Surescripts, on the other hand, supports electronic prescriptions from EHRs to pharmacies and also links EHRs with medication fill information from pharmacies and pharmacy benefit managers. Thus, as we have shown in our previous work,25 EHR-pharmacy database will have more complete information about medication fills regardless of the insurance plan. Finally, there is typically a lag in obtaining insurance claims data in most clinical settings, making them less useful in understanding current adherence measurement and for real-time predictions. Surescripts data are updated in the EHR in real time, which will make the findings relevant to real-world point-of-care interventions. We further augment patient-level EHR data with linked neighbourhood-level social determinants of health data, including pharmacy characteristics and census tract-level information. This allows us to investigate factors beyond individual characteristics, traditionally available in EHR and examine multilevel risk factors, which has the potential to strengthen our predictions as well as identify social factors that might be target for adherence interventions.
Our cohort also has some important limitations. It is derived from a single health system based in greater NYC and thus may not reflect the underlying population, due to selection of patients into the EHR.59 We are missing some potentially important variables, such as individual-level social determinant of health data, which are generally not captured in EHR. The pharmacy fill data available through Surescripts may not capture all medication fills, although we found that Surescripts data contained over 90% of fills found in insurance claims.25 Further, because our cohort is limited to a single health system, we will not capture PDC information for patients who may be receiving additional care from providers in a different health system. As there is significant heterogeneity between types of heart failure and underlying aetiology of heart failure, which may impact adherence, combination of all types of heart failure within one cohort may also mask some of the important mechanisms impacting non-adherence. On the other hand, with a particular focus on heart failure, our findings will not be generalisable to the broader cardiovascular disease class. While understanding medication non-adherence in broader cardiovascular disease is also important, we decided to keep the focus of our cohort on heart failure because of considerable heterogeneity in disease presentation, treatment and medication use for the broader cardiovascular diseases.
Future plans
The established cohort with multilevel predictors will be used to support a primary study to train and validate a machine learning model to predict medication adherence, as well as to support ancillary studies related to medication adherence in heart failure. We further plan on including data from an additional hospital system in NYC for external validation which will improve the generalisability of our findings.
Collaboration
Additional questions or access to data for ethically approved research studies can be requested by directly contacting the corresponding author at [email protected].
Data availability statement
Data are available on reasonable request. Data may be made available from the corresponding author with reasonable request, pending institutional and IRB approval and after completing a data sharing agreement.
Ethics statements
Patient consent for publication
Not applicable.
Ethics approval
This study was approved by the New York University Institutional Review Board with a waiver of informed consent (IBR approval number: s19-00131).
Contributors SA and SBB developed the study design and methods. SA drafted the manuscript. SK, XL, TN and CF supported data acquisition and analysis for this work. SA, SBB, AM, JD, RC and IK contributed to interpretation of data. SK, XL, TN, CF, AM, JD, RC, IK and SBB provided critical review for important intellectual content. All authors provided final approval of the version to be published and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. SA and SBB are the guarantors of the study.
Funding This work is supported by funding from NHLBI (R01HL155149) to SBB and SA. AM also declares funding support from American Heart Association (AHA 23CDA1042602), outside the scope of the present work. RC is supported by NSF (1845487).
Disclaimer The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.
1 Bui AL, Horwich TB, Fonarow GC. Epidemiology and risk profile of heart failure. Nat Rev Cardiol 2011; 8: 30–41. doi:10.1038/nrcardio.2010.165
2 Yancy CW, Jessup M, Bozkurt B, et al. 2013 ACCF/AHA guideline for the management of heart failure: a report of the American college of cardiology foundation/American heart Association task force on practice guidelines. J Am Coll Cardiol 2013; 62: e147–239. doi:10.1016/j.jacc.2013.05.019
3 Blecker S, Herrin J, Li L, et al. Trends in hospital readmission of Medicare-covered patients with heart failure. J Am Coll Cardiol 2019; 73: 1004–12. doi:10.1016/j.jacc.2018.12.040
4 Suter LG, Li S-X, Grady JN, et al. National patterns of risk-standardized mortality and readmission after hospitalization for acute myocardial infarction, heart failure, and pneumonia: update on publicly reported outcomes measures based on the 2013 release. J Gen Intern Med 2014; 29: 1333–40. doi:10.1007/s11606-014-2862-5
5 CONSENSUS Trial Study Group. Effects of enalapril on mortality in severe congestive heart failure. results of the cooperative North Scandinavian enalapril survival study (CONSENSUS). N Engl J Med 1987; 316: 1429–35. doi:10.1056/NEJM198706043162301
6 Garg R, Yusuf S. Overview of randomized trials of angiotensin-converting enzyme inhibitors on mortality and morbidity in patients with heart failure. JAMA 1995; 273: 1450–6.
7 Granger CB, McMurray JJV, Yusuf S, et al. Effects of Candesartan in patients with chronic heart failure and reduced left-ventricular systolic function intolerant to angiotensin-converting-enzyme inhibitors: the CHARM-alternative trial. Lancet 2003; 362: 772–6. doi:10.1016/S0140-6736(03)14284-5
8 Poole-Wilson PA, Swedberg K, Cleland JGF, et al. Comparison of Carvedilol and metoprolol on clinical outcomes in patients with chronic heart failure in the Carvedilol or metoprolol European trial (COMET): randomised controlled trial. Lancet 2003; 362: 7–13. doi:10.1016/S0140-6736(03)13800-7
9 McMurray JJV, Packer M, Desai AS, et al. Angiotensin-Neprilysin inhibition versus enalapril in heart failure. N Engl J Med 2014; 371: 993–1004. doi:10.1056/NEJMoa1409077
10 Vaduganathan M, Claggett BL, Jhund PS, et al. Estimating lifetime benefits of comprehensive disease-modifying pharmacological therapies in patients with heart failure with reduced ejection fraction: a comparative analysis of three randomised controlled trials. The Lancet 2020; 396: 121–8. doi:10.1016/S0140-6736(20)30748-0
11 Zhang Y, Wu S-H, Fendrick AM, et al. Variation in medication adherence in heart failure. JAMA Intern Med 2013; 173: 468. doi:10.1001/jamainternmed.2013.2509
12 Sueta CA, Rodgers JE, Chang PP, et al. Medication adherence based on part D claims for patients with heart failure after hospitalization (from the Atherosclerosis risk in communities study). Am J Cardiol 2015; 116: 413–9. doi:10.1016/j.amjcard.2015.04.058
13 Chang LL, Xu H, DeVore AD, et al. Timing of Postdischarge follow-up and medication adherence among patients with heart failure. JAHA 2018; 7. doi:10.1161/JAHA.117.007998
14 Curtis LH, Mi X, Qualls LG, et al. Transitional adherence and persistence in the use of aldosterone antagonist therapy in patients with heart failure. American Heart Journal 2013; 165: 979–986. doi:10.1016/j.ahj.2013.03.007
15 Choudhry NK, Kronish IM, Vongpatanasin W, et al. Medication adherence and blood pressure control: a scientific statement from the American heart Association. Hypertension 2022; 79: e1–14. doi:10.1161/HYP.0000000000000203
16 Zhang Y, Baik SH. Race/Ethnicity, disability, and medication adherence among Medicare beneficiaries with heart failure. J Gen Intern Med 2014; 29: 602–7. doi:10.1007/s11606-013-2692-x
17 Dickson VV, Knafl GJ, Riegel B. Predictors of medication Nonadherence differ among black and white patients with heart failure. Res Nurs Health 2015; 38: 289–300. doi:10.1002/nur.21663 Available: http://doi.wiley.com/10.1002/nur.v38.4
18 Osborn CY, Kripalani S, Goggins KM, et al. Financial strain is associated with medication Nonadherence and worse self-rated health among cardiovascular patients. J Health Care Poor Underserved 2017; 28: 499–513. doi:10.1353/hpu.2017.0036
19 George J, Shalansky SJ. Predictors of refill Non‐Adherence in patients with heart failure. Br J Clin Pharmacol 2007; 63: 488–93. doi:10.1111/j.1365-2125.2006.02800.x
20 Cheen MHH, Tan YZ, Oh LF, et al. Prevalence of and factors associated with primary medication Non‐Adherence in chronic disease: a systematic review and Meta‐Analysis. Int J Clin Pract 2019; 73: e13350. doi:10.1111/ijcp.13350
21 Mukhopadhyay A, Adhikari S, Li X, et al. Association between Copayment amount and filling of medications for angiotensin receptor Neprilysin inhibitors in patients with heart failure. J Am Heart Assoc 2022; 11: e027662. doi:10.1161/JAHA.122.027662
22 Pednekar P, Peterson AM, Heller D. Comparing medication adherence rates between Pharmacy desert and non-desert areas among elderly in Pennsylvania. Value in Health 2016; 19: A205–6. doi:10.1016/j.jval.2016.03.1253
23 Go AS, Fan D, Sung SH, et al. Contemporary rates and correlates of Statin use and adherence in nondiabetic adults with cardiovascular risk factors: the KP CHAMP study. Am Heart J 2017; 194: 25–38. doi:10.1016/j.ahj.2017.08.013
24 Tyree PT, Lind BK, Lafferty WE. Challenges of using medical insurance claims data for utilization analysis. Am J Med Qual 2006; 21: 269–75. doi:10.1177/1062860606288774
25 Blecker S, Adhikari S, Zhang H, et al. Validation of EHR medication fill data obtained through electronic linkage with pharmacies. JMCP 2021; 27: 1482–7. doi:10.18553/jmcp.2021.27.10.1482
26 Whitlock E. Evaluating primary care behavioral counseling interventions: an evidence-based approach. Am J Prev Med 2002; 22: 267–84. doi:10.1016/S0749-3797(02)00415-4
27 Wu J-R, Cummings DM, Li Q, et al. The effect of a practice-based Multicomponent intervention that includes health coaching on medication adherence and blood pressure control in rural primary care. J Clin Hypertens (Greenwich) 2018; 20: 757–64. doi:10.1111/jch.13265
28 Tülüce D, Kutlutürkan S. The effect of health coaching on treatment adherence, self-efficacy, and quality of life in patients with chronic obstructive pulmonary disease. Int J of Nursing Practice 2018; 24. doi:10.1111/ijn.12661 Available: https://onlinelibrary.wiley.com/toc/1440172x/24/4
29 Meddings J, Kerr EA, Heisler M, et al. Physician assessments of medication adherence and decisions to intensify medications for patients with uncontrolled blood pressure: still no better than a coin toss. BMC Health Serv Res 2012; 12: 270. doi:10.1186/1472-6963-12-270
30 Wissel BD, Greiner HM, Glauser TA, et al. Prospective validation of a machine learning model that uses provider notes to identify candidates for Resective epilepsy surgery. Epilepsia 2020; 61: 39–48. doi:10.1111/epi.16398
31 Shashikumar SP, Wardi G, Paul P, et al. Development and prospective validation of a deep learning algorithm for predicting need for mechanical ventilation. Chest 2021; 159: 2264–73. doi:10.1016/j.chest.2020.12.009
32 Franklin JM, Krumme AA, Shrank WH, et al. Predicting adherence trajectory using initial patterns of medication filling. Am J Manag Care 2015; 21: e537–44.
33 Lauffenburger JC, Franklin JM, Krumme AA, et al. Predicting adherence to chronic disease medications in patients with long-term initial medication fills using indicators of clinical events and health behaviors. J Manag Care Spec Pharm 2018; 24: 469–77. doi:10.18553/jmcp.2018.24.5.469
34 Karanasiou GS, Tripoliti EE, Papadopoulos TG, et al. Predicting adherence of patients with HF through machine learning techniques. Healthc Technol Lett 2016; 3: 165–70. doi:10.1049/htl.2016.0041
35 Son Y-J, Kim H-G, Kim E-H, et al. Application of support vector machine for prediction of medication adherence in heart failure patients. Healthc Inform Res 2010; 16: 253–9. doi:10.4258/hir.2010.16.4.253
36 Kini V, Ho PM. Interventions to improve medication adherence: A review. JAMA 2018; 320: 2461–73. doi:10.1001/jama.2018.19271
37 Cantor MN, Chandras R, Pulgarin C. FACETS: using open data to measure community social determinants of health. J Am Med Inform Assoc 2018; 25: 419–22. doi:10.1093/jamia/ocx117
38 Centers for Medicare & Medicaid Services, Measure Methodology, Available: https://www.cms.gov/Medicare/Quality-Initiatives-Patient-Assessment-Instruments/HospitalQualityInits/Measure-Methodology
39 Blecker S, Anderson GF, Herbert R, et al. Hospice care and resource utilization in Medicare beneficiaries with heart failure. Med Care 2011; 49: 985–91. doi:10.1097/MLR.0b013e318235c221
40 Blecker S, Katz SD, Horwitz LI, et al. Comparison of approaches for heart failure case identification from electronic health record data. JAMA Cardiol 2016; 1: 1014–20. doi:10.1001/jamacardio.2016.3236
41 Blecker S, Paul M, Taksler G, et al. Heart failure-associated hospitalizations in the United States. J Am Coll Cardiol 2013; 61: 1259–67. doi:10.1016/j.jacc.2012.12.038
42 World Health Organization. Adherence to long-term therapies: evidence for action,. 2003 Available: https://www.who.int/chp/knowledge/publications/adherence_full_report.pdf
43 Golden SD, Earp JAL. Social ecological approaches to individuals and their contexts. Health Educ Behav 2012; 39: 364–72. doi:10.1177/1090198111418634
44 Mueller M, Purnell TS, Mensah GA, et al. Reducing racial and ethnic disparities in hypertension prevention and control: what will it take to translate research into practice and policy Am J Hypertens 2015; 28: 699–716. doi:10.1093/ajh/hpu233
45 Berben L, Dobbels F, Engberg S, et al. An ecological perspective on medication adherence. West J Nurs Res 2012; 34: 635–53. doi:10.1177/0193945911434518
46 Poon IO, Ogboe CW. Applying the social-ecological approach to evaluate diabetes medication management in older people. Sr Care Pharm 2021; 36: 548–55. doi:10.4140/TCP.n.2021.548
47 Martin BC, Wiley-Exley EK, Richards S, et al. Contrasting measures of adherence with simple drug use, medication switching, and therapeutic duplication. Ann Pharmacother 2009; 43: 36–44. doi:10.1345/aph.1K671
48 Pollack CE, Hussey PS, Rudin RS, et al. Measuring care continuity: A comparison of claims-based methods. Med Care 2016; 54: e30–4. doi:10.1097/MLR.0000000000000018
49 Warren JR, Falster MO, Tran B, et al. Association of continuity of primary care and Statin adherence. PLoS One 2015; 10: e0140008. doi:10.1371/journal.pone.0140008
50 Barker I, Steventon A, Deeny SR. Association between continuity of care in general practice and hospital admissions for ambulatory care sensitive conditions: cross sectional study of routinely collected, person level data. BMJ 2017; 356: j84. doi:10.1136/bmj.j84
51 Macintyre S, Ellaway A, Cummins S. Place effects on health: how can we Conceptualise, Operationalise and measure them Social Science & Medicine 2002; 55: 125–39. doi:10.1016/S0277-9536(01)00214-3
52 Healthy People 2030, U.S. Department of Health and Human Services, Office of Disease Prevention and Health Promotion, Available: https://health.gov/healthypeople/priority-areas/social-determinants-health
53 Butler MA. Rural-urban continuum codes for Metro and Nonmetro counties. US Department of Agriculture, economic research service, Agriculture and rural economy division. 1990.
54 Malo S, Aguilar-Palacio I, Feja C, et al. Different approaches to the assessment of adherence and persistence with cardiovascular-disease preventive medications. Curr Med Res Opin 2017; 33: 1329–36. doi:10.1080/03007995.2017.1321534
55 Choudhry NK, Shrank WH, Levin RL, et al. Measuring concurrent adherence to multiple related medications. Am J Manag Care 2009; 15: 457–64.
56 Carnicelli AP, Li Z, Greiner MA, et al. Sacubitril/valsartan adherence and Postdischarge outcomes among patients hospitalized for heart failure with reduced ejection fraction. JACC: Heart Failure 2021; 9: 876–86. doi:10.1016/j.jchf.2021.06.018
57 Karve S, Cleves MA, Helm M, et al. Good and poor adherence: optimal cut-point for adherence measures using administrative claims data. Curr Med Res Opin 2009; 25: 2303–10. doi:10.1185/03007990903126833
58 Mathews R, Wang W, Kaltenbach LA, et al. Hospital variation in adherence rates to secondary prevention medications and the implications on quality. Circulation 2018; 137: 2128–38. doi:10.1161/CIRCULATIONAHA.117.029160
59 Bower JK, Patel S, Rudy JE, et al. Addressing bias in electronic health record-based surveillance of cardiovascular disease risk: finding the signal through the noise. Curr Epidemiol Rep 2017; 4: 346–52. doi:10.1007/s40471-017-0130-z
© 2023 Author(s) (or their employer(s)) 2023. Re-use permitted under CC BY-NC. No commercial re-use. See rights and permissions. Published by BMJ. http://creativecommons.org/licenses/by-nc/4.0/ This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/ . Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Details
; Mukhyopadhyay, Amrita 1 ; Kolzoff, Samuel 2 ; Li, Xiyue 1 ; Nadel, Talia 1
; Cassidy Fitchett 1 ; Chunara, Rumi 3
; Dodson, John 1 ; Kronish, Ian 4
; Blecker, Saul B 1 1 New York University Grossman School of Medicine, New York City, New York, USA
2 University of Utah Hospital, Salt Lake City, Utah, USA
3 New York University, New York City, New York, USA
4 Center Behavioral Cardiovascular Health, Columbia University Medical Center, New York City, New York, USA