Causal association of social determinants of health with mortality and dynamic impact in patients with chronic heart failure: a longitudinal study using marginal structure models

Abstract

Aims

It is unclear whether there is a causal relationship between social determinants of health (SDoH) and mortality in patients with chronic heart failure (CHF). In this study, we assessed causality using inverse probability weighting (IPW) of marginal structural models (MSMs) during the course of CHF.

Methods and results

A total of 1377 patients with CHF were enrolled in this multi-centre, prospective cohort study from September 2017 onwards. The social domain and two dimensions of the chronic heart failure patient-reported outcomes measure (CHF-PROM) were used to assess SDoH, social support, and support utilization in these patients. CHF-PROM and mortality information were obtained at 1, 3, and 6 months following discharge and every 6 months thereafter at regular follow-up visits. The impact of SDoH, social support, and support utilization on mortality was analysed by logistic regression and IPW of MSMs. Logistic regression showed that SDoH, social support, and support utilization at baseline were not associated with mortality. After adjustment for confounders, MSMs showed that SDoH and social support were not associated with mortality at baseline. In contrast, low support utilization at baseline and unfavourable SDoH, low social support, and low support utilization during follow-up increased the risk of death.

Conclusion

Using follow-up data and analysis of MSMs, we found that long-term out-of-hospital effects of SDoH, but not one-time effects, were risk factors for mortality in patients with CHF. SDoH should be assessed during the entire course of CHF to prolong patient survival.

Registration

Chinese Clinical Trial Registry, ChiCTR2100043337 (https://www.chictr.org.cn/showproj.html?proj=64980).

Graphical Abstract

Open in new tabDownload slide

Introduction

Chronic heart failure (CHF) is a complex clinical syndrome caused by structural and/or functional abnormalities of the heart and has a high prevalence and a poor prognosis.¹ Data from the Global Burden of Disease Study show that heart failure affects more than 64 million people worldwide, with the age-standardized prevalence increasing by 0.6% over 3 years.² Various pharmacological and mechanical therapies have been developed to address traditional risk factors. However, patients still have a substantial residual risk.³ Evaluation of the impact of non-traditional risk factors on mortality is important for developing effective management measures to reduce the residual risk and mortality in patients with heart failure.⁴

A new report on performance metrics, published collaboratively by the American College of Cardiology/American Heart Association, points to the assessment of social determinants of health (SDoH) as a new quality measure.⁵ SDoH are important non-traditional risk factors associated with healthcare inequities and have a significant impact on disease outcomes.^6,7 For example, the Government of Canada defines SDoH as including factors such as income and social status, social support, and coping skills. Guidelines for managing heart failure recommend use of patient-reported outcomes questionnaires to assess health status in patients with heart failure.^8,9

There is a growing body of evidence showing that SDoH are a risk factor for cardiovascular disease, but the research findings have not always been consistent. Some research has shown that SDoH increase the mortality risk in patients with cardiovascular disease.^10,11 However, another study found that a larger burden of unfavourable SDoH was not associated with a significantly increased mortality risk for patients.¹² These research findings present paradoxes because they focus solely on cross-sectional studies at baseline, neglecting the influence of confounders or inadequate control of confounders. One-time data cannot reflect the real status of SDoH. Moreover, logistic and Cox regression methods cannot address the problem of time-dependent confounding in longitudinal follow-up data and do not fully utilize the information on changes in confounders over time. Clarifying the relationship between SDoH and mortality in patients with CHF may guide the development of effective disease management strategies. We have performed a cohort study of patients with CHF in three hospitals in China. We collected information on the dynamic changes in SDoH during the course of the disease to determine the causal effect of SDoH on mortality in these patients. Inverse probability weighting (IPW) of marginal structural models (MSMs) was also used to assess the relationship between SDoH at baseline, follow-up, and death. The aim of the study was to determine if there is a causal relationship between SDoH and mortality in patients with CHF using longitudinal data in the hope of identifying new intervention targets for clinical treatment and management of CHF.

Methods

Study design and participants

This study is a multi-centre prospective cohort study and included patients from three medical centres in the Shanxi Province of the People’s Republic of China who were enrolled between 1 September 2017 and 30 August 2022. The study is registered in the China Clinical Trial Registry as ChiCTR2100043337.

Patients were selected according to strict eligibility criteria. The inclusion criteria were as follows: age 18 years or older; diagnosis of CHF in accordance with the 2021 European Society of Cardiology heart failure guidelines¹; New York Heart Association functional class II–IV; and treatment for heart failure within the past month. Patients who had experienced acute cardiovascular events within the past 2 months and those with a life expectancy of less than 1 year because of another disease were excluded.

Instruments used for measurements

The CHF-patient-reported outcomes measure (CHF-PROM) is divided into four domains, 12 dimensions, and 57 entries¹³ and has been confirmed to be reliable and valid.¹³ Specific scale information is detailed in Supplementary material online, Table S1 in the Supplementary Materials. In this study, the social domain of CHF-PROM was used to assess SDoH status. The social domain was divided into the social support dimension and the support utilization dimension. The specific questionnaire items are shown in Table 1.

Each item was measured on a Likert scale ranging from 1 to 5 to reflect the frequency of occurrence of each issue during the past 2 weeks (1, never; 2, occasionally; 3, approximately half of the time; 4, often; and 5, almost every day). The final SDoH, social support, and social utilization scores were calculated by adding the scores of the corresponding items. A lower score represented worse SDoH status.

We defined exposure as unfavourable SDoH, low social support, and low support utilization in patients with CHF. The overall scores for the SDoH, social support dimension, and support utilization dimension were 40, 25, and 15, respectively; patients were grouped according to the median of the three scores, which were 27, 17, and 11, respectively. Patients were divided into an unfavourable SDoH group (<27) and a favourable SDoH group (≥27), a low social support group (<17) and a high social support group (≥17), and a low support utilization group (<11) and a high support utilization group (≥11). The quality of the CHF-PROM data were assessed using Cronbach’s alpha, which yielded coefficients of 0.868, 0.726, and 0.809, respectively, for SDoH, social support, and support utilization.

Data collection procedures

Baseline was defined as the first hospitalization for patients diagnosed with CHF during the study period. General information and CHF-PROM scores were assessed at baseline during hospitalization. CHF-PROM scores and outcomes were collected by face-to-face counselling or by telephone at months 1, 3, and 6 after discharge and every 6 months thereafter (Figure 1). Professionally trained individuals entered all data into the system to ensure their quality.

General information

General information included patient demographics, clinical information, comorbidities, and treatments. Demographics included age, sex, body mass index (BMI), marital status, education, occupation, household income, and type of health insurance. Clinical information included heart rate, blood pressure, and medical history. Treatments included medication and revascularization. These variables are shown in detail in Table 2. Age, BMI, heart rate, and blood pressure were treated as continuous variables, and all other variables were transformed into categorical variables.

Outcome data

The outcome was defined as all-cause death during follow-up and included both cardiac (attributable to cardiovascular disease such as heart failure, acute myocardial infarction, or arrhythmia) and non-cardiac death (unrelated to organic or functional cardiac disease). Information on deaths was obtained from regular telephone follow-ups and inquiries made to the death information system for reports on the cause of death registrations in Shanxi Province based on the patient’s identification number.

Ethical considerations

The study protocol was approved by the Ethics Committee of Shanxi Medical University (approval number 2021GLL103) and performed in accordance with the principles of the Declaration of Helsinki. All patients provided written informed consent before enrolment.

Statistical analyses

Processing of missing values

MSMs require sufficient repeated-measures datum points to estimate model parameters for reliable results, and deletion of patients with fewer than three follow-up visits avoids model overfitting, which produces unstable results.¹⁴ Patients with fewer than three follow-up visits, those missing more than 30% of PROM data, and those who refused follow-up were excluded. Missing data included both non-time-varying and time-varying variables. For missing values in the time-varying PROM variables, the average value of the two points in time before and after deletion was used to fill in the values.¹⁵ Missing values for non-time-varying variables were implemented using the MissForest package in R 4.3.2.¹⁶

Descriptive analysis

Data that were not normally distributed are shown as the median (inter-quartile range). Qualitative data are presented as the number (percentage). Differences between variables in the death and survival groups were examined using t-tests, χ² tests, or Mann–Whitney U tests depending on the type of data distribution. All statistical analyses were performed using SPSS version 26 software (IBM Corp., Armonk, NY, USA). All statistical tests were two-tailed. A P-value of <0.05 was considered statistically significant.

Marginal structural model

We performed a correlation analysis to screen for general information on variables associated with both exposure and outcome. We defined sets of confounders screened for associations of unfavourable SDoH, low social support, and low support utilization with mortality as sets 1, 2, and 3, respectively. Set (a) represented baseline confounders and set (b) represented confounders during long-term follow-up. The screened confounders were used to construct the MSM.

This model has been widely used for causal inference in real-world research, especially for analysis of repeated measurements in longitudinal studies.¹⁷ The core step of MSM is to control for confounders by assigning each individual a certain weight using IPW to obtain a virtual population. Analysis of this virtual population provides an unbiased estimation of treatment effects. This analysis ensures that the relationship between treatment and outcomes remains consistent with the original population rather than affected by confounders.^18,19

The probability of treatment levels for individuals must be estimated before using IPW. This study used classical logistic regression analysis to estimate treatment probability. Predicted probabilities of patients receiving treatment were calculated using each confounding factor as an independent variable and the exposure factor as the outcome.²⁰

Based on the probability estimation, the IPWs of patients were further calculated to balance confounding factors and construct a comparable virtual population. The following formula was used:

where A(k) indicates whether an individual receives exposure at time point k (unfavourable SDoH = 1, low social support = 1, low support utilization = 1), X represents baseline confounders, and L represents time-dependent confounders. Hence, the numerator and denominator represent the probability that an individual receives exposure at time point k under the influence of baseline and time-dependent confounders, respectively. The IPW can be adjusted for treatment-level groups (e.g. unfavourable and high SDoH groups) to have the same distribution in population sub-groups characterized by various confounders. Percentile truncation was applied to prevent extreme weights from occurring during the calculations. The IPW obtained above was truncated (replacing the extreme values on both sides of the 2.5 and 97.5 percentiles with the values of those percentile points) to obtain stabilized weights for each individual.²¹

Estimates of effect

We further applied a generalized estimating equation model to estimate the effect values and 95% confidence intervals (CIs) between conditions such as SDoH and outcome events for patients. Unfavourable SDoH, low social support, and low support utilization were used as independent variables and patient death as the outcome, with truncated stabilized weights included in the model. The effect values obtained from the MSMs represent the true causal effects after controlling for time-dependent confounders.

Results

Descriptive analysis

There were 1497 study participants at baseline. During this period, 120 patients (8.02%) were excluded. Forty-eight patients were excluded for refusal to attend for follow-up visits, 24 because of death before making three follow-up visits, 32 for making fewer than three follow-up visits for other reasons, 10 for deciding the trial was too demanding and long, and 6 for not being contactable by telephone. Finally, data for 1377 patients (91.98%) were analysed.

The patient characteristics at baseline are shown in Table 2. A total of 183 patients (13.29%) died during the follow-up period. Deceased individuals were older, had a lower BMI, were more likely to have a lower educational level, and had a higher likelihood of being unmarried or widowed. They also had a higher prevalence of comorbidities, such as atrial fibrillation, valvular heart disease, chronic renal failure, and cancer. Patients who died were less likely to receive statins, β-receptor blockers, angiotensin-converting enzyme inhibitors/angiotensin receptor blockers (ACEIs/ARBs), or revascularization and were more likely to be treated with diuretics.

Dynamic changes in SDoH scores

The dynamic changes in SDoH reflected by the social domain scores of the CHF-PROM and the proportions of patients are shown in Figure 2. The scores for SDoH, social support, and support utilization initially increased, continued to increase gradually, peaked at month 42 post-discharge, and then declined sharply (Figure 2A). This trend was synchronized with the change in SDoH scores (Figure 2B).

Screening for confounding factors

The results of screening for confounding factors are shown in Supplementary material online, Figure S1 in the Supplementary Materials. Univariate analysis at baseline showed that the confounders in set 1(a) were statins and ACEIs/ARBs. Set 2(a) was empty, and the confounders in set 3(a) were age, atrial fibrillation, statins, and ACEIs/ARBs.

Univariate analysis during follow-up showed that the confounders in set 1(b) were BMI, education, valvular heart disease, ACEIs/ARBs, statins, diuretics, and revascularization. Set 2(b) confounders were age, BMI, valvular heart disease, cancer, ACEIs/ARBs, statins, and revascularization. Set 3(b) confounders were BMI, marital status, education, chronic renal failure, ACEIs/ARBs, statins, diuretics, and revascularization.

Estimation of causality

At baseline, logistic regression modelling showed no statistically significant association of mortality with SDoH, social support, or support utilization. IPW of MSMs also found that SDoH and social support were not associated with mortality. IPW of MSMs only showed a statistically significant association of low support utilization at baseline with mortality (hazard ratio 1.44, P = 0.04). The results for SDoH, social support, and support utilization at baseline are shown in Supplementary material online, Table S2 in the Supplementary Materials.

The results for SDoH, social support, and support utilization during follow-up are presented in Figure 3. Logistic regression modelling showed that patients with favourable SDoH had an increased mortality risk. Revascularization had the greatest effect on patient mortality. MSM also showed that patients with unfavourable SDoH had an increased risk of death. Logistic regression modelling results showed that patients with higher social support had a higher risk of death. Revascularization had the greatest impact on patient mortality. MSM showed that less social support was associated with a higher risk of death. Logistic regression modelling did not identify a significant association of support utilization with death outcomes. Patients with a lower BMI had a higher risk of death, and those who underwent revascularization had a lower risk of death. MSM showed that patients with low support utilization had a higher risk of death.

Discussion

Patients with CHF face additional challenges related to social factors. In this study, only low support utilization at baseline was identified to have a causal relationship with mortality. However, during the course of the disease, unfavourable SDoH, low social support, and low support utilization all contributed to the risk of death. This study is one of the few to investigate the causal relationship between SDoH throughout the course of disease and death in patients with CHF. Its findings provide new targets and ideas for residual risk to improve the prognosis of CHF by intervening on non-traditional risk factors for SDoH.

Randomized controlled trials face significant challenges in terms of controlling for social factors, while conventional observational cohort studies often lack an adequate strategy for handling time-dependent confounders. IPW of MSMs can effectively reflect the true causal relationship between SDoH and mortality outcomes in patients with CHF. In this study, we performed causal analyses at both baseline and during follow-up and corrected the effects of time-dependent confounders. At baseline, the results showed that SDoH was not associated with death in patients with CHF, similar to the findings of a cohort study in the USA.¹² Other studies showed that unfavourable SDoH was a risk factor for death in patients with cardiovascular disease.^10,11 The results of the research to date are inconsistent because they have focused solely on cross-sectional studies at baseline, neglecting the influence of confounders or not controlling adequately for confounders. This study also confirmed the causality between low support utilization at baseline and death. A secondary analysis of sub-groups in the COACH study noted that patients with CHF and low support utilization had poor self-care behaviours, which further affected their prognosis.²²

MSMs were used to analyse the data in this cohort study, and the results show causality between SDoH, social support, support utilization, and death. This causal relationship may be attributable to the fact that unfavourable SDoH have a negative impact on patients with CHF during the course of their disease.^23,24 The MSM results indicate that patients with more unfavourable SDoH have a higher risk of death. Patients with unfavourable SDoH are in unfriendly social environments in which they receive poor social support and medical resources.²⁵ Furthermore, during the course of the disease, unfavourable SDoH cause or worsen negative emotions in patients, which further affect self-management of CHF and increase the risk of death.²⁵ An observational study of 119 patients with CHF that included 6 years of follow-up found an association between social isolation and death,²⁶ which is consistent with our present findings. Our study had a larger sample size and focused on social risk factors, confirming that low support utilization by patients with CHF increased their risk of death. Higher utilization of social support plays a positive role in the self-management of patients with CHF by enhancing self-efficacy.²⁷ Caregivers and clinicians providing social support should prioritize patients’ utilization of this support. Emphasis on support utilization could reduce unfavourable SDoH and improve the prognosis in patients with CHF. Overall, the impact of unfavourable SDoH on patient outcomes is crucial for a considerable period after discharge. Therefore, outpatient management and long-term monitoring of SDoH are important in patients with CHF. Patients in unfavourable social environments after discharge should be monitored carefully, considering that they may be more affected by a variety of adverse factors. Furthermore, there is a need to develop appropriate intervention strategies for risk factors associated with unfavourable SDoH, low social support, and low support utilization to improve patient outcomes.

Screening for confounders associated with exposure and outcomes is a critical step in MSMs. The confounders identified in this study may have mediated the causal relationship between SDoH and mortality in patients with CHF. We found that confounders associated with all three exposures (unfavourable SDoH, low social support, and low support utilization) and death were low BMI and no ACEIs/ARBs, statin, or revascularization therapy. Confounders associated with outcomes and exposure included low BMI, low education, single status, and no ACEIs/ARBs, statin, or revascularization therapy. Previous research has confirmed that high BMI is associated with a better prognosis in patients with CHF.²⁸ The Chinese Longitudinal Healthy Longevity Survey showed that patients with low BMI had more unfavourable SDoH,²⁹ which is consistent with our findings. Clinicians in China should monitor patients with CHF and a low BMI very carefully. Our study also found that using ACEIs/ARBs and statins improved outcomes in patients with CHF, as suggested by the 2022 American Heart Association/American College of Cardiology/Heart Failure Society of America guideline for the management of heart failure. Moreover, the Chinese patient population receiving guideline-recommended medications such as ACEIs/ARBs at discharge is low compared with contemporary observational data from the USA and Europe. In our study, the percentage of patients who were taking ACEIs/ARBs was 40%, which is similar to that in other studies performed in China, including the China PEACE Retrospective Heart Failure Study (51.5%)³⁰ and the Heart Failure Registry of Patient Outcomes study (43.3%).³¹ This implies that there is a large number of patients in China who are not receiving effective drug therapy as recommended by the current clinical guidelines. These patients may have unfavourable SDoH status. Our next step is to promote guideline-directed medical therapy for suitable patients with heart failure. In contrast, use of statins in patients with CHF is limited to those who also have coronary artery disease.⁸ Our present findings suggest that patients who are not taking ACEIs/ARBs or statins have lower socioeconomic status. Such patients have poor access to health care, less motivation, and are less likely to take medications as prescribed.³² Our study also found that revascularization is an effective measure for reducing patient mortality. A cohort study of patients with CHF at the Tyumen Cardiology Research Center confirmed that most patients who underwent percutaneous coronary intervention had higher levels of social support,³³ which is consistent with our findings, suggesting a need to focus on patients who do not undergo revascularization.

Despite our careful data collection and analysis, this study had some limitations. First, the study data were primarily from the Shanxi Province of China, which limits their generalizability. Therefore, validation is required in other populations. Second, 120 patients were lost to follow-up. This loss may have affected the internal validity of our results. Third, use of MSMs assumed that all confounders were identified. However, unobserved confounders may have been present in the model.

In conclusion, this study using MSM provides evidence of a causal relationship between SDoH and death in patients with CHF. We found that the long-term out-of-hospital causal effects, but not one-time effects, of SDoH were risk factors for CHF-related mortality. This increase in risk highlights the importance of clinicians and caregivers addressing unfavourable SDoH, increasing social support, and enhancing utilization of support throughout the course of CHF to improve the prognosis of the disease.

Supplementary material

Supplementary material is available at European Journal of Cardiovascular Nursing online.

Acknowledgements

We thank Liwen Bianji (Edanz) (www.liwenbianji.cn) for editing the English text of a draft of this manuscript.

Authors’ contributions

Jing Tian and Yanbo Zhang contributed to the conception and design of the study. Yujing Wang, Yongfeng Lv, Jingjing Yan, Yajing Wang, and Jing Tian performed material preparation, data collection and analysis. Yujing Wang wrote the first draft of the manuscript and all authors commented on previous versions of the manuscript. Guisheng Song was involved in revising this article. All authors read and approved the final manuscript. All authors declare no conflict of interest.

Funding

National Nature Science Foundation of China (grant numbers 82103958 and 82173631); the Shanxi Science and Technology Innovation Talent Team Project (grant number 202204051001026) and the Shanxi Province Science and Technology Strategic Research Special Project (grant number 202104031402136).

Data availability

The data underlying this article cannot be shared publicly because they include information that can identify patients. The data will be shared upon reasonable request to the corresponding author.

References

Author notes