Background:

The morbidity, mortality and costs associated with hypertension and heart failure in the US are enormous. In the US population, the positive role of having a usual source of care (USOC) on the receipt of preventative services is known. However, associations between USOC and hypertension control and whether a differential association across age groups exists is unknown in the US population.

Heart failure affects patients in myriad ways: economically, physically, socially and emotionally. Heart failure negatively impacts health-related quality of life (HRQOL), but age-related differences in HRQOL (and specifically emotional health) are unknown. Finally, studies are mixed on whether increased monitoring and nurse coaching can impact emotional health post-discharge in recently hospitalized heart failure patients.

Methods:

To assess the relationship between USOC and hypertension control, I use data from the National Health and Nutrition Examination Survey (NHANES) from 2007-2012. I utilize multivariable logistic regression to evaluate the association between having a USOC and hypertension control. The differential effect of USOC on hypertension control by age is assessed using predicted marginal effects for various age groups within this model and then analyzing pairwise comparisons of the marginal effects.

To examine age-related differences in emotional health in recently hospitalized heart failure patients, I analyze longitudinal data from the BEAT-HF study—a multicenter trial comparing the impact of wireless remote monitoring and nurse coaching versus usual care on emotional health for patients hospitalized with heart failure. Multivariable linear regression and mixed effects models are utilized to evaluate whether there are baseline and longitudinal differences in emotional health across age groups. To analyze whether the association between age and emotional health outcomes is mediated by physical health and/or social health I use a multi-step regression model allowing for cross-equation error correlation (“seemingly unrelated regression”) and structural equation modeling. To assess the intervention effect on emotional health in the study, I utilize mixed effects linear regression controlling for treatment arm and hospital level random effects.

Results:

In adjusted analyses, those with a USOC had a higher odds of hypertension control [OR=3.89, 95%CI (2.15-6.98)]. The marginal effect of having a USOC is associated with a 30 percentage-point higher probability of controlled blood pressure compared to those without a USOC [marginal difference in probability=0.30, 95%CI (0.19-0.41)]. In tests of pairwise comparisons of marginal effects, there was a 7-8 percentage point difference in marginal effect of USOC on hypertension control in the youngest group (compared to all middle age groups) which was statistically significantly lower. In terms of the US population this difference amounts to 70,000-80,000 fewer young individuals with controlled hypertension per million individuals with hypertension. There was also a 3-4 percentage point difference in marginal effect in the oldest age group (compared to all middle age groups) which was statistically significantly lower. This difference amounts to 30,000-40,000 fewer older individuals with controlled hypertension per million in the US hypertension population.

In the BEAT-HF trial, older individuals had better emotional health in multivariable linear regression models controlling for demographic and clinical characteristics [lower scores indicating better emotional health; β=-1.9, 95% CI (-3, -0.8)]. The effect of age on emotional health was partially mediated by physical health in all models (Barron and Kenney multi-step regression, seemingly unrelated regression with simultaneous regression equations and correlated error terms, and structural equation modeling). The mixed effects analysis for the intervention’s effect on emotional health showed a small but statistically significant effect at 180 days [=-1.3, 95%CI(-2.2, -0.02)]. By Cohen’s rules of thumb the standardized difference in groups approaches a “small” effect size (adjusted effect size(ES)=0.17 vs. “small” ES=0.2), but is below it. In mixed effects models using tests of interaction, there was no differential effect of treatment by age or social isolation.

Conclusion:

Having a usual source of care is significantly associated with improved hypertension control in the US population. The variation in the association across age groups has important implications in targeting age-specific anti-hypertensive strategies to reduce the burden of hypertension in the US population.

Older patients with heart failure in this study have better emotional health than younger patients. This may be related to increased coping or acceptance of limitations, since older patients overall had more comorbidities and a higher proportion of NYHA class III heart failure. Both treatment and control groups had improved emotional health scores in the post-discharge period, but the telemonitoring and nurse coaching intervention had small positive effects on emotional health at 180 days. Treatment non-adherence may have minimized the effect on emotional health, but this large-scale randomized controlled trial likely gives an accurate assessment of the real-world effect of telemonitoring and nurse coaching on a broad heart failure population.

Hospital readmissions following acute exacerbations of chronic obstructive pulmonary disease (COPD) are frequent occurrences, resulting following one in five discharges. Readmissions are costly and also put patients at risk for adverse outcomes stemming from ongoing hospitalizations. Under the Hospital Readmissions Reduction Program, the Centers for Medicare and Medicaid Services have imposed penalties on excess readmissions for COPD, among other conditions. In this dissertation, we explore the effect of comorbidity, operationalized by the Charlson and Elixhauser indices, on readmission odds using multilevel logistic regression modelling and employing information criteria in order to determine which provides better fit in estimating readmissions. The Elixhauser index having prevailed, we estimated the readmission odds for returns related to an ongoing COPD diagnosis, versus for all other causes. We found that only 45% of returns were for ongoing COPD, and that those who returned for other reasons had substantially higher burdens of comorbidity. Finally, we evaluated the effect of the implementation of the Hospital Readmissions Reduction Policy on readmission rates for COPD, controlling for secular trends. We found that the addition of COPD to the list of penalized conditions did not significantly affect the already down-trending rates of readmissions when compared to non-penalized conditions or compared to the other penalized conditions of congestive heart failure, myocardial infarction, or pneumonia. Readmissions for COPD remain an ongoing problem for healthcare burden and cost, but overall readmissions do seem to be lessening. Understanding the effect of comorbidity on readmissions will help health systems better plan their efforts to improve care, and for policies like the Hospital Readmissions Reduction Program to adjust their expected readmission rates more fairly when applying penalties.

This dissertation evaluates the national burden of heart failure (HF) hospitalizations. HF is the leading cause of admission among all cardiovascular conditions. Yet, information is lacking on the factors associated with hospital expenditures, differences in utilization by gender and ethnicity, and the complexity of hospitalized patients. Due to an aging demographic, the prevalence of HF is projected to continue to increase in the future. This dissertation is divided into three projects that describe different aspects of hospital utilization for heart failure.

The National Inpatient Sample (NIS) provided through the Healthcare Cost and Utilization Project (HCUP) were used for all three projects. The NIS datasets were obtained for the years between 2002 and 2013. Each year of the NIS contains a sample of 7 to 8 million hospital discharges representing 20% of all hospitalizations nationally. The NIS is the largest all-payer database that uses administrative data to estimate hospital utilization nationally.

The first project describes the factors associated with hospitalization costs in the highest quantile compared to the lowest quantile. Hospital cost data was estimated based on charges using established conversion methods. A multinomial logistic model adjusted for patient and hospital factors as well as sampling strategy was used to identify factors associated with higher cost hospitalizations. Select patient demographics and comorbidities were associated with the highest hospitalization costs. When controlling for all patient related factors, hospital characteristics and region were still associated with higher hospitalization costs.

The second project reports on the trends in age-standardized hospital utilization by gender and ethnicity. Shifting age demographics make reporting trends in crude hospitalization rates inaccurate. The purpose of this research project was to standardize hospital utilization rates and contrast them by gender and ethnic subgroups. The difference in hospital utilization rates for men has increased relative to women. The difference in hospital utilization for blacks is nearly two and a half times that of whites and the relative difference has not changed over the recent decade. Hispanics have lowered their hospital utilization rates for HF. Asians have consistently had the lowest HF hospitalization burden.

The third project describes the shifts in comorbid conditions among hospitalized HF patients. While research attention has focused on reducing repeat HF admissions, shifts in the complexity of care for HF patients have not been described. Over the last decade, diabetes, obesity, and renal disease have increased in prevalence among hospitalized HF patients. Average Elixhauser comorbidity scores have increased for all gender and ethnic groups. The severity of comorbid illnesses among HF patients continues to increase, prompting the need for more effective management of these complex patients.

Background

Methods

Results

Conclusion

Objective

Data sources/study setting

Study design

Principal findings

Conclusions