Healthcare quality and mortality among patients hospitalized for heart failure by hospital level in Beijing, China

Study design and data collection

We invited all the affiliated hospitals and teaching hospitals of the Capital Medical University, including 10 tertiary hospitals and 13 secondary hospitals, of which five typical tertiary hospitals and four typical secondary hospitals agreed to participate in this study. All these participating hospitals provided inpatient care to HF patients within existing cardiovascular departments. In China, hospitals are graded into community health centres, secondary hospitals and tertiary hospitals, according to their functionality, size and specialization. Tertiary hospitals generally have a large number of beds and provide comprehensive medical services. Secondary hospitals usually provide general medical service to catchment areas of 300 000–500 000 local inhabitants. Characteristics of participating tertiary hospitals and secondary hospitals are shown in Table A1. All patients aged 18 years or older and hospitalized with a primary discharge diagnosis of HF during 1 January 2014 to 31 December 2015 were included in this study. For patients hospitalized twice or more in the participating hospitals, only the most recent inpatient hospitalization was included in the analysis.

Data were collected through chart review by trained staff. Information on patients' demographic characteristics, medical history, clinical presentation, laboratory examination, echocardiography, in-hospital treatments, discharge diagnoses, and medications at discharge were collected. We used insurance types as a proxy for patients' socio-economic status. The database was linked to the Death Surveillance System in Beijing to obtain all patient's survival information after admission. Ethical approvals were acquired from each participating hospital. Informed consent was waived in this study.

Metrics used in the measurement of healthcare quality in patients with heart failure

We reported three inpatient performance measures: (i) proportion of patients who had an LVEF assessment, (ii) proportion of β-blocker prescription at discharge in patients with LVSD, defined as LVEF < 40%, and (iii) proportion of ACEI or ARB prescription at discharge in patients with LVSD. All these indices were recommended by the American Heart Association/American College of Cardiology/American Medical Association HF performance measurement guideline.¹³ We also evaluated two additional quality measures: (i) proportion of patients with B-type natriuretic peptide (BNP) /N-terminal pro BNP tested and (ii) proportion of mineralocorticoid receptor antagonists (MRAs) prescription at discharge in patients with LVSD^{12, 13, 15} (Table 1). For measures of drug prescriptions at discharge, patients with contraindication or patients who died during their hospitalization were excluded.

Clinical outcomes

Patients' mortality data were acquired from the Beijing Death Surveillance System through data linkage, which records survival status of all Beijing citizens. Patients with the same name initials, sex, and birth dates in the health record and in the surveillance system were considered as the same person. Patients not recorded in the death surveillance system were considered as alive. A 30 day mortality rate was defined as the proportion of patients who died from any reason within 30 days after being admitted to a hospital with HF as the primary discharge diagnosis. The 90 day and 1 year mortality rates were defined as death from any cause within 90 days or within 1 year after hospital admission.

Statistical analysis

Patients' baseline characteristics, clinical outcomes and hospital performance measures were analysed to assess differences between secondary and tertiary hospitals. We summarized skewed continuous variables as median with interquartile ranges, normally distributed data as mean with standard deviations and categorical variables as frequencies and percentages. Student's t test/Wilcoxon rank-sum test, and χ² test were used to compare continuous or categorical data between two hospital groups.

Propensity score-weighted hierarchical generalized linear mixed models with logit link were performed to compare 30 day, 90 day, and 1 year mortality rates between secondary and tertiary hospitals. Propensity scores, which represent the probability of being admitted to tertiary hospital compared with a secondary hospital, were acquired using a logistic model that adjusted for 11 patient-level risk factors that associated with patients' outcomes,¹⁶ including age, sex, LVEF, New York Heart Association class, serum creatinine, diabetes, systolic blood pressure, body mass index, HF duration, current smoking, and chronic obstructive pulmonary disease. Weights of models were derived from the inverse of propensity score or the inverse of 1 minus propensity score.¹⁷ Additional adjustment of patient-level risk factors that used in the propensity score calculation was also performed in modelling process. Although these covariates were already in the propensity score model, this double-adjustment approach reduced selection bias as thoroughly as possible following the principle of ‘double robustness’.¹⁸ Four models were set up sequentially. The 11 patient-level risk factors mentioned earlier were adjusted in Model 1. The variables including heart rate >100 beats per minute, anaemia (haemoglobin <130 g/L in male patients, haemoglobin <120 g/L in female patients), medical history of coronary artery disease (defined as a history of myocardial infarction, percutaneous coronary intervention, or coronary artery bypass graft surgery), cardiomyopathy, valvular heart disease, and stroke were further adjusted in Model 2. Health insurance and marriage were added to previously adjusted risk factors for Model 3. Discharge prescription of β-blocker, ACEI/ARB, and MRAs were added to previous risk factors for Model 4. Clustering of patients and their care within hospitals was treated as random effects in all hierarchical generalized linear mixed models. Odds ratios (ORs) and their 95% confidence intervals (CIs) were presented to show the associations of predictors and clinical outcomes.

Modes were used to impute missing values of categorical variables, and medians were imputed for missing values of continuous variables. Haemoglobin level had the highest proportion of missing values at 5%, which were replaced with the median value of 132 g/L. A two-sided P value less than 0.05 was considered statistically significant. All statistical analyses were performed using SAS software version 9.4 (SAS Institution Inc., Cary, NC, USA).

Baseline characteristics

We enrolled 2663 patients hospitalized due to HF (1413 from four secondary hospitals and 1250 from five tertiary hospitals). Table 2 outlines the baseline characteristics of the patients. Patients hospitalized in secondary hospitals were older (median age: 74 years vs. 72 years, P < 0.001) and with larger proportion of female patients (52.7% vs. 43.3%, P < 0.001) than those hospitalized in tertiary hospitals. A significantly higher proportion of patients in secondary hospitals (55.5% vs. 4.4%, P < 0.001) was covered by the New Rural Cooperative Medical Scheme (NRCMS, which covered rural residents in China). Prevalence of coronary artery disease, cardiomyopathy, atrial fibrillation, hypertension, diabetes, anaemia, and kidney dysfunction (estimated glomerular filtration rate <60 mL/min/1.73m²) were higher in patients hospitalized in tertiary hospitals.

Performance measures

Table 3 demonstrates rates of individual performance measures by hospital level. Although proportions of patients having BNP/N-terminal pro BNP tested were similar in secondary and tertiary hospitals (85.9% vs. 87.7%, P = 0.180), a lower proportion of patients in secondary hospitals had LVEF evaluated (73.2% vs. 90.1%, P < 0.001). A lower proportion of patients with LVSD wad prescribed β-blockers or combined β-blockers and ACEI/ARB at discharge in secondary hospitals (44.3% vs. 72.4%, and 30.1% vs. 49.3%, respectively, both P < 0.001). Although not statistically significant, proportion of patients receiving ACEI/ARB was also lower in secondary hospitals (57.1% vs. 64.4%, P = 0.070). In contrast, a slightly higher proportion of eligible patients in secondary hospitals were prescribed MRAs at discharge (85.9% vs. 80.3%, P = 0.072).

Clinical outcomes

Before adjustment, mortality rates at 30 days, 90 days, and 1 year were significantly higher in secondary hospitals than their counterparts in tertiary hospitals (5.5% vs. 3.0%, P = 0.001; 10.8% vs. 5.0%, P < 0.001; and 21.0% vs. 12.1%, P < 0.001). In Model 1, patients in secondary hospitals had a higher risk of 90 day mortality (OR: 2.06, 95% CI: 1.10–3.84, P = 0.024) and 1 year mortality (OR: 1.64, 95% CI: 1.02–2.62, P = 0.039), but not statistically significant lower risk of 30 day mortality (OR: 1.49, 95% CI: 0.63–3.52, P = 0.368). The associations remained consistent in Models 2, 3, and 4 (Figure 1).

Implications for health disparities

Differences in health insurance and extent of coverage may partially account for the heterogeneity of hospital-level healthcare quality.¹¹ China's government instituted universal health insurance coverage to improve access to and affordability of public healthcare, and social health insurance programmes have successfully covered more than 97% Chinese residents.²⁵ However, different social health insurance programmes have different reimbursement policies, which influences patients' care-seeking behaviour from different levels of hospitals.²⁶ For example, in Beijing, urban employers are covered by Urban Employ Based Medical Insurance, under which reimbursement for inpatient care in secondary hospitals and tertiary hospitals are similar [92% (in secondary hospital) vs. 90% (in tertiary hospital) between RMB 30 000 and 40 000 (US $4471–$5961); 97% vs. 95% above RMB 40 000 (US $5961)²⁷]. On the other hand, rural residents are covered by the NRCMS, under which the reimbursement rates for inpatient care in secondary hospitals and tertiary hospitals are quite different [70% (in secondary hospital) vs. 60% (in tertiary hospital) between RMB 20 000 and 50 000 (US $2980–$7452); 80% vs. 67% above RMB 50 000 (US $7452)²⁷]. The current study shows that 55.5% secondary hospital inpatients were covered by NRCMS, while corresponding number was 4.4% in tertiary hospitals. Rural HF residents, especially female patients with low socio-economic status, had limited accessibility to high-quality medical treatment, which may exacerbate health inequities. Future research should be conducted to reduce variation of health service in different hospitals.