Data collection and study procedures

A retrospective, observational study design was used to extract deidentified EMR data from the Arkansas Clinical Data Repository (AR-CDR) [48] on all CHF patients who received care at the University of Arkansas for Medical Sciences (UAMS) between January 1, 2014 and December 31, 2021. The EMR has proven to be a rich source of data for understanding care delivery teams and their impact on the care outcomes of patients with cardiovascular and other chronic diseases [18, 48]. UAMS is Arkansas’ only academic medical center and only Level One Trauma Center. The AR-CDR contains data on patient care encounters at UAMS’ main campus in Little Rock, AR, a 535-bed hospital (i.e., 431 adult beds, 64 newborn bassinets, 40 psychiatry beds) [49]. The AR-CDR also contains data on over 30 of UAMS’ regional clinics in rural and metropolitan areas of Northeast, Northwest, Central, Southeast, and Southwest Arkansas. Fig 2 reflects annual patient care utilization at UAMS [49]. Eligibility criteria limited subjects to patients who were aged 18–105 with CHF. CHF diagnosis was identified using the International Classification of Diseases Rv.10 (ICD-10) codes. The study procedures (Protocol number: 262593) were approved by the Institutional Review Board at UAMS.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. UAMS patient care utilization.

https://doi.org/10.1371/journal.pone.0286363.g002

Study variables

The AR-CDR data contained all patient, CDT, and outcome variables. Patient variables included deidentified patient ID, age, sex, race, ethnicity, clinical diagnosis, comorbidities, medical history, medications, and heart failure phenotype data. Care outcome variables included the number of hospitalizations (i.e., patient admitted to the hospital without an admission within the preceding 30 days), the number of readmissions (i.e., patient admitted to the hospital within 30 days of a hospital discharge), the number of days between readmissions, and the high charge care encounter (i.e., the amount of dollars billed to the patient by the hospital for medical care rendered). All patient encounter dates including primary care, hospitalizations, and follow-up periods for readmission were based on the date of first entry in the EMR (i.e., the date of care). Time lapses between initial hospitalization and readmission varied between patients due to factors including patient’s need for care. CDT variables, the AR-CDR data contained an edge list that associated patient encounters with the specific clinicians (i.e., clinical roles) who provided care during each encounter. The AR-CDR data indicated a total of up to 60 clinical roles on the CDTs of CHF patients. However, only clinical roles that have been associated with hospitalizations, readmissions, days between readmissions, and high charge encounters were included: physician, resident, advanced nurse practitioner, registered nurse, licensed practical nurse, patient care technician, care manager, social worker, physical therapist, and occupational therapist.

Statistical analysis

Binomial logistic regression was used to ascertain the effects of age, sex, race, ethnicity, and CDT composition on the likelihood that participants had a hospitalization, a high number of readmissions, a high number of days between readmissions, and a high charge care encounter. Within the regressions, odd ratios represented associations between variables. All continuous care outcome variables and age were dichotomized at the median to distinguish a high/low threshold, supporting the dichotomy assumption of the binomial logistic regression. For example; a “high charge encounter” was defined as a charge above the median charge for a care encounter (i.e., $22,403) and was transformed to “1”, representing “yes; high charge”, and a charge below the median was transformed to “0”, representing “No; low charge”. These transformations were replicated for hospitalizations, readmissions, and days between readmissions, and patient age. Furthermore, a Mann Whitney-U test was used to identify differences between AA and CA care outcomes. All patient, CDT, and care outcome variables were treated as continuous and mutually exclusive within the Mann Whitney-U analysis.

Results stratified by heart failure severity

Heart failure patients with preserved ejection fraction.

There were 707 patients with ejection fraction of greater than or equal to 50 percent (Table 7), classified as heart failure with preserved ejection fraction (i.e., HFpEF) which indicated patients with the least severe cases of heart failure [51, 52, 54].

Table 7 reflects HFpEF patients, a heart failure phenotype representing patients with the least severe cases (i.e., near normal pumping) of heart failure. Model 1 of Table 7 reflects HFpEF patient hospitalization risk (i.e., being admitted to the hospital without an admission within the preceding 30 days) over the 7-year study period. Hospitalizations for HFpEF patients (n = 707) had a mean of 53.61, a median of 51, and a mode of 96. The model explained 50.7% of the variance in hospitalizations and correctly classified 84.9% of cases. Additionally, HFpEF patients with an advanced practice registered nurse or a registered nurse on their CDT were 55% and 85% less likely to have a hospitalization, respectively. However, HFpEF patients with specific clinical roles on their CDTs were more likely to have a hospitalization: a physician (i.e., 2.07 times more likely), care manager (i.e., 7.81 times more likely), and a physical therapist (i.e., 27.45 times more likely). Additionally, HFpEF patients with renal failure were also 29% more likely to have a hospitalization.

Model 2 of Table 7 reflects HFpEF patient readmission risk (i.e., being admitted to the hospital within 30 days of a discharge) over the 7-year study period. Readmissions for HFpEF patients had a mean of 7.81, a median of 2, and a mode of 0. The model explained 36.5% of the variance in readmissions and correctly classified 72.2% of cases. HFpEF patients with specific clinical roles on their CDT were less likely to have a high number of readmissions: a registered nurse (i.e., 52%), patient care technician (i.e., 25%), care manager (i.e., 18%), social worker (i.e., 24%), or an occupational therapist (i.e., 30%). HFpEF patients who were on a thiazide diuretic were 49% less likely to have a high number of readmissions. However, HFpEF patients were more likely to have a high number of readmissions if they had anemia deficiency (i.e., 3.11 times more likely), coagulopathy (i.e., 2.61 times more likely), metastatic cancer (i.e., 66%), obesity (i.e., 3.26 times more likely), peripheral vascular disease (i.e., 99%), renal failure (i.e., 63%), or a tumor without metastasis (i.e., 67%). In addition, HFpEF patients were more likely to have a high number of readmissions if they were on an angiotensin receptor blocker (i.e., 65%), a calcium channel blocker (i.e., 21%), or an antihypertensive (i.e., 43%).

Model 3 of Table 7 reflects HFpEF patient risk of having a high number of days between readmissions (i.e., more days at home) over the 7-year study period. Days between readmissions for HFpEF patients had a mean of 30.01, a median of 29.35, and a mode of 48.13. The model explained 17% of the variance in the number of days between readmissions and correctly classified 65.6% of cases. HFpEF patients were more likely to have a high number of days between readmissions (i.e., more days at home) if they were on a diuretic (i.e., 20%), or had a resident (i.e., 34%) or registered nurse (i.e., 47%) on their CDT. HFpEF patients were less likely to have a high number of days between readmissions (i.e., less days at home) if they had anemia deficiency (i.e., 39%), coagulopathy (i.e., 40%), metastatic cancer (i.e., 57%), peripheral vascular disease (i.e., 40%), renal failure (i.e., 41%), or if they were on an angiotensin receptor blocker (i.e., 20%).

Model 4 of Table 7 reflects HFpEF patient risk for a high charge encounter over the 7-year study period. Care encounter charges for HFpEF patients had a mean of $52,835.96, a median of $52,892.97, and a mode of $91,208. The model explained 41.6% of the variance in charges per care encounter and correctly classified 76.1% of cases. HFpEF patients were less likely to have a high charge care encounter if they had metastatic cancer (i.e., 19%). However, HFpEF patients with specific comorbidities and medical histories were more likely to have a high charge care encounter: coagulopathy (i.e., 65%), metastatic cancer (i.e., 53%), or renal failure (i.e., 23%). HFpEF patients with specific clinical roles on their CDT were more likely to have a high charge care encounter: a physician (i.e., 93%), resident (i.e., 22%), advanced nurse practitioner (i.e., 57%), registered nurse (i.e., 3.33 times more likely), patient care technician (i.e., 2.63 times more likely), care manager (i.e., 2.09 times more likely), social worker (i.e., 89%), physical therapist (i.e., 4.99 times more likely), or an occupational therapist (i.e., 43%).

Furthermore, Fig 4 illustrates median differences between AAs (n = 39,076) and CAs (n = 39,187) in total care encounters, each group accounting for approximately 48% of total care encounters. As demonstrated by Fig 4A, AAs had a significantly higher number of hospitalizations (p = .001) than CAs (i.e., 6 vs. 4, respectively). As a group, AAs had a significantly higher number of readmissions (p = .001) than CAs (i.e., 56 vs. 40, respectively; Fig 4B). In addition, AAs had a significantly higher number of days between readmissions [days at home] (p = .001) than CAs (i.e., 28.34 vs. 23.70, respectively; Fig 4C). However, AAs had significantly lower care encounter charges (p = .001) than CAs (i.e., $14,249.22 v. $33,213.78, respectively; Fig 4D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Mann Whitney-U median differences between CAs and AAs.

https://doi.org/10.1371/journal.pone.0286363.g004

CDT composition influence on care outcomes

The EMR is the ecosystem for care delivery, providing wide-ranging and longitudinal data and opportunities for stratifying patient risk of poor outcomes [38]. Recent studies have called for EMR-driven examinations of the optimal CDT composition (i.e., size, roles, structures, and interaction) to support high-quality personalized care [22, 23, 38]. The most widely accepted model of CDT composition is the Veterans Health Administration’s Patient Aligned Care Teams model [39]. While fundamental to the domain, variation exists in how primary care clinics have operationalized the model─ lacking specificity in expressing which specific CDT clinical role are highly associated with patient risk of hospitalization, readmission, and high charge encounters [22, 39]. Our findings suggested that CHF patients may have benefited the most from the care of registered nurses more than any other CDT role. Notably, the registered nurse was the most heavily engaged clinical role found among care delivery encounters of CHF patients. Among all patients, registered nurses were the only CDT role consistently associated with a 30% decreased risk of hospitalization and a 31% decrease in risk of having a high number of readmissions during the 7-year study period. In the most severe heart failure cases, HFrEF patients with a registered nurse on their CDT were 88% less likely to have hospitalization and were 50% less likely to have a high number of readmissions. Similar decreases in risk of hospitalization and readmission were also found in the less severe cases of heart failure (i.e., HFmrEF and HFpEF). However, the possible reasons for the association between registered nurses and improved care outcomes have been thoroughly evaluated in other studies. The key role of the registered nurse in the management of CHF is largely focused on patient monitoring (i.e., early intervention, medication management, outpatient follow-up, self-care education, and care coordination) [55]. Registered nurses are also responsible for initial assessment and triage, discharge planning, and end-of-life care [55–57].

However, there is variation in the scope of practice among registered nurses, influenced by various factors (e.g., educational level of the nurse, state of practice, and clinical environment and resources). However, registered nurses generally educate patients on the causes of CHF and facilitate discussions on lifestyle modifications (e.g., eliminating/reducing sodium intake and increasing physical activity). With responsibilities that require such high levels of engagement with patients, registered nurses also act as an early alert system, creating more complex and compounded care when required for high-need CHF patients. For example, a registered nurse’s triage results (e.g., abnormal electrocardiogram read) could indicate the need to immediately engage more CDT members (e.g., the code team or cardiologist). The findings affirm that registered nurses are the most impactful (i.e., highly associated with positive patient outcomes) frontline providers. The findings also emphasize the value of the responsibilities undertaken by registered nurses (e.g., managing CHF risk factors, symptom management, and education and counseling). These responsibilities help explain why registered nurses were the most frequent clinical role found among care delivery encounters and solidify their role in the development and implementation of future interventions that target decreasing hospitalizations in CHF patients.

Decreasing hospitalizations in AA patients is particularly impactful to CHF patients because heart failure-related hospitalizations for AA patients [both men and women] are nearly 2.5 times higher than CA patients, nationally [58–61]. Yet, the CDTs of CHF patients generally include a diverse set of care types and clinical roles [41–43]. Research suggests patients benefit from the composite of different types of clinical training offered by the variation and unique contributions of each clinical role to the patient’s care plan [41–43]. The only known study in the domain evaluated associations between primary care team communication, interaction, and costs for patients with cardiovascular disease in 155 CDT members (e.g., physicians, advanced nurse practitioners, registered nurses) at 6 primary care clinics [18]. The study found that CDTs that were more interconnected, less centralized, and had a shared team vision were more likely to deliver high quality cardiovascular care at a lower cost than CDTs who were not [18]. However, the study did not investigate the extent to which specific clinical roles were associated with improving hospitalizations or care costs. Studies have also been conducted on other chronic diseases, posing similar questions and yielding similar results [62, 63]. Even with the additional contribution of this study, the contributions of all clinical roles can not be minimized or devalued because the literature related to CDT composition is in its infancy. However, this finding stimulates the need for future research investigating the role of specific clinical roles in designing, risk stratifying, contributing to, and operationalizing the care plans of AA patients with CHF. Additional evidence will advance the framework for determining if patients have “access to high-quality care” by requiring patients to have access to the clinical roles that are associated with improved care outcomes in specific diseases. Correspondingly; as a recommendation for change in clinical practice, increasing the number of registered nurses on CDTs of AA patients with CHF may decrease hospitalizations and increase days between readmissions. In addition, hospitals and clinics that provide care for large populations of African Americans should consider optimizing the role of the registered nurse by ensuring that registered nurses are more specialized in CHF care delivery and prepared to trigger early alert systems within their organization’s current risk-based care guidelines.

Implications for population health management

While the study design and results could not determine that CDT composition caused better or worse outcomes for CHF patients, the results emphasized how heavily associated clinical roles are with a CHF patient’s risk of poor or improved care outcomes. For example, among all patients, CHF patients with a physician on their CDT were 3.02 times more likely to have a hospitalization during the 7-year period. When stratified by heart failure severity, the findings on hospitalization risk were relatively consistent among HFrEF (i.e., 2.97 times more likely), HFmrEF (i.e., 3.23 times more likely), and HFpEF patients (i.e., 2.07 times more likely). CHF patients with a resident on their CDTs were 11% more likely to have a high number of days between readmissions (i.e., more days at home) and 30% more likely to have a high charge encounter, but 13% less likely to have a high number of readmissions. These results do not indicate that physicians and residents caused poor or better outcomes; rather, that they were associated with the outcomes. More specifically, these results indicate that these physicians and residents followed a model of care delivery that required them to heavily engage with high-need patients (i.e., physicians and residents are generally more heavily associated with the sickest patients). However, as evidence of this, we would have expected to observe significant differences between HFrEF and HFpEF patients in hospitalizations and readmissions. Yet, the results did not reveal any statistically significant indication of this. One of the most prevalent and recent models of care delivery is the Comprehensive Care Physician Model, where physicians focus on providing care to patients who are at an increased risk of hospitalization, simultaneously supporting inpatient and outpatient care [57]. This care model was an evolution of the Traditional Model of care delivery (i.e., 1990’s) in which risk of hospitalization was not used to evaluate the extent to which physicians [or residents] should have engaged in the care of high-risk patients [64].

Risk stratification (RS), the core of population health management, is nationally recognized as the clinician-driven process of categorizing patients into tiers based on health and social factors to reflect the potential for unplanned healthcare utilization and expensive encounters [13]. In primary care settings [the gateway for all other specialties], processes for RS are relatively new [16, 18]. Among consistently applied RS approaches, the process of tiering patients (i.e., highly-complex, high-risk, rising-risk, low-risk) involves three sequential steps; characterizing the patient population, systematically assigning risk levels, and mitigating risk [13, 16, 26–32]. Systematically assigning risk levels fuels the planning, development, and implementation of personalized care plans, and is the most critical component of population health management [31, 32]. The "one-size-fits-all" approach to care planning and delivery (i.e., same level of resources provided to all patients) is clinically ineffective and wasteful [32]. Assigning and prioritizing individual patients’ needs based on risks promotes the efficient allocation of clinical resources (i.e., clinical team), reserving high-intensity resources for patients who are highly-complex and high-need [32]. Of the current RS approaches, computational tools are employed; yet, clinical intuition (i.e, experience-based, decision-making) is the most widely-used practice [29–31]. Consequently, the subjectivity in how CDTs employ their clinical intuition may promote disparities in the outcomes of cardiovascular disease patients (i.e., women, AAs, elderly) [29–31]. Thus, clinical intuition, alone, is not a consistent, sufficient, and comprehensive RS approach. RS can be approached more comprehensively by working in concert with computational tools, constituting the fundamental theorem of biomedical informatics [33]. Current computational algorithms for CHF include the: Adjusted Clinical Groups, Charlson Comorbidity, Elixhauser Comorbidity Indices, Hierarchical Condition Categories, and Framingham Risk Score [34–36]. Broadly, current tools rely on aggregated patient demographic data (e.g., age, sex, race, marital status), comorbidities (e.g., number and type of chronic conditions), patient panels (e.g., glucose, lipid, systolic and diastolic levels), and behavioral (e.g., diet, physical activity, smoking) to compute a single risk score that assigns patients into risk tiers [16, 17, 34–36]. However, in solely relying on such indicators [ignoring care delivery influences], current tools have not optimized their utility in comprehensively approaching RS [13, 28, 37].

The statistically significant CDT findings suggest that current tools could be expanded to support clinical staffing needs. Hospitals, clinics, and community health centers could improve clinical efficiency with computational tools that can accurately predict how allocating specific clinical roles to patients with the greatest needs would impact and potentially improve care outcomes. More broadly, now that we know that CDT composition is associated with care outcomes, CDT variables can be built into EMRs and data warehouses as a real-time, hospital-level surveillance tool for stratifying patient risk of poor outcomes.

Limitations

The study data was generated from Arkansas’ only academic health center. While the data reflects patient encounters throughout Arkansas (i.e., 30 of UAMS’ regional clinics in Northeast, Northwest, Central, Southeast, and Southwest Arkansas), findings may not be generalizable to healthcare systems throughout the United States. The lack of heart failure phenotype data on 4,298 of the study’s patients may explain the large mean differences in hospitalizations between entire study population (Table 4) and the subgroups when the analysis was stratified (Tables 5–7). The lack of data may also have impacted the results of the odds ratios in Tables 5–7. However, confidence intervals are provided to approximate any potential variation. Furthermore, the use of heart failure phenotype, alone, did not provide a comprehensive indication of heart failure severity. Consistent with the wide-recognized use of the New York Heart Association Classification, both objective and subjective assessments of heart failure severity would have provided a stronger indication of severity. Clinician-driven, subjective assessments (e.g., observations of shortness of breath, tiredness, and lack of appetite) and other pseudo indicators (e.g., functional status, frailty, and oxygen support needs) were not available to support this study. While this study found CDT roles were statistically significant predictors of care outcomes, larger-scale studies are needed to assess the effectiveness of expanding the current set of risk factors to include CDT composition as a predictor of patient hospitalizations, readmissions, and charges. This study also limited its inquiry to specific clinical roles and did not comprehensively assess associations between outcome variables and all possible roles which contributed to the care of CHF patients. Furthermore, we recognize that cardiologists and other specialty physicians have a vital role in caring for CHF patients. However, the data infrastructure of the AR-CDR aggregated all types of physicians into the role of “physician”. Therefore, our results did not distinguish between generalist and specialty physicians (i.e., cardiologist). Finally, a potential constraint of future research will be the lack of access to more national, EMR data that can match CHF patients with their CDTs. Therefore, it is further recommended that cardiovascular clinical data registries expand by adding CDT-specific data elements (i.e., de-identified CDT numbers, size, structure, and interactions among other CDT) that will more globally support clinical and clinical research studies in the domain. Provisionally, additional smaller-scale, healthcare system-level studies of CDT will help expand the domain.