This investigation sought to quantify the risk factors for short-term readmission in patients with heart failure (HF). Electronic databases were systematically searched for studies reporting relative risk, odds ratio, and hazard ratio for the combined primary outcome of all-cause hospital readmission or all-cause mortality ≤90 days from discharge of patients with HF. Clinical characteristics, study design, type and incidence of outcome, univariable effect sizes for each risk factor, and their associated 95% confidence intervals were extracted. Each univariable effect size was pooled and computed in a separate meta-analysis using random-effects models weighted by inverse variance. The frequency of significance of each risk factor in multivariable models was also assessed to confirm their independence. Sixty-nine studies (2,038,524 patients) were included and 144 factors were reported, including 32 reported more than twice. The significant associations of the combined primary outcome were chronic lung disease, chronic kidney disease, atherosclerotic vascular disease (peripheral, coronary, and cerebrovascular), diabetes, anemia, lower systolic blood pressure, previous admission, multidisciplinary treatment, and use of beta-blockade and angiotensin-converting enzyme inhibition or angiotensin receptor blockade. In multivariable analyses, most of these variables remained independently associated with the combined primary outcome. However, age, male gender, black race, hypertension, dyslipidemia, smoking, atrial fibrillation, cancer, and uses of diuretics, aldosterone antagonists, and digoxin were not significant. In conclusion, noncardiovascular co-morbidities, poor physical condition, history of admission, and failure to use evidence-based medication are more strongly associated with 90-day readmission or death than standard risks in patients with HF.
Heart failure (HF) is a leading cause of hospital readmission in patients aged >65 years. Despite improvement in outcomes with medication, readmission rates after HF admission are still increasing. This poses significant problems including impaired quality of life and increased costs and resource utilization. The recent inclusion of 30-day all-cause readmission or death as a major focus of quality improvement and payment reform attests to the seriousness of readmission in patients with HF as a health economic problem. Many of these readmissions are predictable and, therefore, possibly preventable. On these grounds, it is important for clinicians to identify which patients may be at highest risk of being readmitted. Although various strategies have been used to limit readmissions, the processes of care in these programs have varied substantially and not all have been associated with lower readmission rates. A better understanding of the risks for readmission is essential for more effective targeting of disease management strategies. Readmission is prone to occur relatively early, especially in the transition phase from hospital to home, and the 90-day risk of readmission has been widely published. In this study, we sought to quantify risk factors for the prediction of all-cause hospital readmission or all-cause mortality ≤90 days from discharge.
Methods
We followed the Meta-analysis of Observational Studies in Epidemiology criteria for performing and reporting the present meta-analysis ( Appendix 1 ). The electronic databases PubMed, Scopus, PsychINFO, and the Evidence-Based Medicine Reviews on Ovid were searched using the medical subject heading (MeSH) terms patient readmission, risk, and HF. First, we performed a search using the MeSH term “patient readmission” and the key words “readmi$” and “rehosp$” (using ‘‘$’’ for truncation). Second, we searched using the MeSH term “risk” and the key words, “model$,” “predict$,” “use$,” “util$,” and “risk$.” Third, we searched using the MeSH term “heart failure, congestive.” Combination of the term results from the patient readmission, risk, and HF searches included 4,536 articles. This search result was narrowed down using the follow-up descriptors, “30-day,” “60-day,” and “90-day.” The final search included a total of 882 articles (30-day readmission: 518 articles, 60-day readmission: 107 articles, 90-day readmission: 257 articles). Additional relevant articles were obtained from scrutiny of the reference lists of articles identified in the search, including a past systematic review. From these lists, studies were included if they met each of the following criteria:
- 1)
studies of a full-length original article in a peer-reviewed English language journal;
- 2)
studies done in human adults >18 years;
- 3)
patients with an index hospitalization caused or complicated by HF;
- 4)
studies reporting quantitative effect sizes in relative risk (RR), odds ratio (OR), or hazard ratio (HR) with 95% confidence intervals relating to primary outcome or studies reporting actual numbers of the patients with and without risk and the event rates in each group for the calculation of RR or univariable OR.
Studies reporting only adjusted effect size for individual risks without covariate information in the multivariable models were excluded because other independent predictors in those models were unknown, and the assessment of the heterogeneity of the adjusted analyses was impossible. The univariable effect sizes of low-frequency (reported in <3 studies) risk factors were also excluded.
The primary outcome was the combined end point of all-cause hospital readmission or all-cause mortality ≤90 days from discharge, and a secondary analysis was undertaken of outcomes at the combined end point of all-cause hospital readmission or all-cause mortality ≤30 days from discharge and only all-cause hospital readmission ≤90 days from discharge. We combined studies with outcomes including death and those with outcomes excluding death (readmission only) because the short-term mortality after discharge was relatively low (<5%). The data concerning the individual study populations were extracted independently by 2 reviewers (MS and KN). All discrepancies were reviewed and resolved by consensus. Data for continuous variables were extracted as a weighted mean of the articles reporting the continuous variables data. The detail definitions of risks defined on a categorical basis were also extracted from each study. Covariates included in the multivariable models were extracted to assess the way for the adjustment and the independence of risk variables used for adjusted analysis. When the variable was significant, risk tendency (more risk or less risk) was also confirmed in reference to the adjusted effect size. Model performance was also extracted, if available.
The reported RR, univariable OR, or univariable HR for each risk variable was pooled and analyzed individually using random-effect models weighted by inverse variance. When RR and univariable OR were not reported, RR and univariable OR and their 95% confidence intervals were calculated from the actual number of patients in the groups with and without risk and the event rates in each group and then combined with reported RR and univariable OR. The calculated risk estimation of individual risk variables was then pooled into a combined analysis for each RR, univariable OR, and univariable HR to demonstrate overall risk profile and to compare individual risk impact. Heterogeneity was described by I 2 . Publication bias was assessed visually by funnel plots of effect estimates and sample size. From the studies reporting the multivariable models, the number of model candidates and the number of independent predictors were counted. Then, the frequency of risk factors being identified as an independent predictor was expressed as the percent significance (i.e., number of significant risks × 100/number of model candidates). Study quality was assessed by the Newcastle-Ottawa scale (0 to 9 points) for cohort and case-control studies or the Jadad scale (0 to 5 points) for randomized controlled trials. Statistical analysis was performed using RevMan5.2 (Cochrane Information Management System, Oxford, UK) with 2-tailed p values <0.05 considered significant.
Results
The process of article selection based on Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines is presented in Figure 1 ; 69 articles were used for the systematic review and 57 articles for the meta-analysis. There were a total of 2,038,524 subjects, aged 52 to 81 years (weighted mean 79 years), 34% to 99% were men (weighted mean 43%), and most patients were Caucasian (weighted mean 72%). Most studies were based in the United States (56 studies: 81%), and 46 studies (67%) were observational. The weighted mean follow-up time was 36 days. The event rates for the 90- and 30-day combined outcomes of all-cause readmission or all-cause death were 14% to 49% (weighted mean 33%) and 3% to 32% (weighted mean 26%). The event rate for the only 90-day all-cause readmission was 21% to 47% (weighted mean 30%).
The baseline characteristics of included studies ( Appendices 2 and 3 ) show that co-morbidities were of variable prevalence, although there were missing data for demographic variables. Of the 52 articles (75%) that reported hypertension data or diabetes data, 60% (weighted mean) had hypertension and 28% (weighted mean) diabetes mellitus. In 39 articles (57%) reporting ejection fraction data, the weighted mean of ejection fraction was 38%. Of the 33 articles (48%) reporting angiotensin-converting enzyme inhibitors (ACEi) or angiotensin receptor blockers (ARB) data, 49% (weighted mean) were taking these medications, and of the 26 articles (38%) reporting β-blocker data, 60% (weighted mean) were treated with these agents.
One hundred forty-four variables including 30 socioeconomic variables and 35 co-morbidities were associated with the primary outcome ( Appendix 4 ). Of these, there were 32 clinical variables in which a univariable effect size was reported in ≥3 studies. Physical data and serum markers that were expressed as a categorical variable were unsuitable for meta-analysis because of a variety of cutoff values.
Table 1 summarizes pooled RR, univariable OR, and univariable HR for the risk variables (details are listed in Appendix 5 ). For risk variables reported as RR, chronic lung disease was significantly associated with the primary outcome, followed by chronic kidney disease, cognitive impairment, ischemic heart disease, post-coronary artery bypass graft surgery, myocardial infarction, cerebrovascular disease, and diabetes. Also, a multidisciplinary approach and use of β blockers were negatively associated with the primary outcome. For risk variables reported in univariable OR, in addition to the earlier mentioned variables, previous admission, anemia, post-percutaneous coronary intervention, peripheral vascular disease, depression, ejection fraction <40%, and use of ACEi/ARB were significantly associated with the primary outcome. Finally, risk variables reported in univariable HR showed that chronic lung disease, diabetes, and systolic blood pressure had significant associations with the primary outcome. However, age, male gender, black race, hypertension, atrial fibrillation, dyslipidemia, smoking, hyponatremia, cancer, living alone, and use of diuretics, aldosterone antagonists, digoxin, and antiplatelets were not significantly associated with the primary outcome.
| Factors | No of articles | RR [95%CI] | I 2 | No of articles | Univariate OR [95%CI] | I 2 | No of articles | Univariate HR [95%CI] | I 2 |
|---|---|---|---|---|---|---|---|---|---|
| Age (per 1 year) | 0 | NA | 1 | NA | 6 | 1.00 [0.99, 1.02] | 62% | ||
| Male | 20 | 0.98 [0.95, 1.01] | 0% | 24 | 0.98 [0.95, 1.01] | 0% | 6 | 1.22 [0.95, 1.57] | 85% |
| Black race ∗ | 9 | 1.23 [0.93, 1.61] | 81% | 11 | 1.24 [0.99, 1.56] | 76% | 2 | NA | |
| Single or living alone | 2 | NA | 3 | 1.18 [0.85, 1.64] | 58% | 0 | NA | ||
| Systolic blood pressure (per 10 mm Hg) | 0 | NA | 0 | NA | 3 | 0.91 [0.88, 0.95] | 0% | ||
| Prior all-cause admission in past year | 2 | NA | 3 | 1.73 [1.40, 2.15] | 70% | 0 | NA | ||
| Diabetes mellitus | 14 | 1.18 [1.10, 1.27] | 50% | 17 | 1.26 [1.16, 1.36] | 40% | 6 | 1.30 [1.06, 1.60] | 24% |
| Hypertension | 10 | 0.93 [0.74, 1.18] | 83% | 12 | 0.88 [0.69, 1.13] | 79% | 4 | 1.05 [0.74, 1.49] | 37% |
| Dyslipidaemia | 1 | NA | 2 | NA | 3 | 0.58 [0.29, 1.17] | 44% | ||
| Smoker ∗ | 6 | 1.08 [0.89, 1.32] | 68% | 7 | 1.10 [0.87, 1.38] | 67% | 1 | NA | |
| Ischaemic heart disease | 10 | 1.25 [1.08, 1.44] | 62% | 13 | 1.28 [1.12, 1.47] | 43% | 4 | 1.21 [0.81, 1.82] | 57% |
| Myocardial infarction | 9 | 1.19 [1.06, 1.33] | 57% | 10 | 1.21 [1.09, 1.34] | 38% | 0 | NA | |
| Atrial fibrillation | 7 | 0.99 [0.79, 1.25] | 57% | 11 | 1.18 [0.96, 1.44] | 45% | 4 | 1.09 [0.76, 1.56] | 58% |
| Chronic kidney disease ∗ | 8 | 1.33 [1.18, 1.50] | 55% | 10 | 1.45 [1.26, 1.67] | 41% | 1 | NA | |
| Chronic lung disease | 7 | 1.47 [1.22, 1.76] | 84% | 9 | 1.57 [1.27, 1.93] | 77% | 6 | 1.38 [1.13, 1.70] | 75% |
| Cerebrovascular disease | 5 | 1.16 [1.10, 1.22] | 0% | 7 | 1.35 [1.03, 1.76] | 43% | 1 | NA | |
| Anaemia ∗ | 2 | NA | 3 | 1.73 [1.28, 2.35] | 0% | 2 | NA | ||
| Cognitive impairment | 5 | 1.28 [1.14, 1.43] | 18% | 5 | 1.40 [1.15, 1.70] | 25% | 1 | NA | |
| Depression | 5 | 1.25 [0.95, 1.63] | 87% | 6 | 1.39 [1.05, 1.83] | 83% | 2 | NA | |
| Peripheral vascular disease | 2 | NA | 3 | 1.35 [1.21, 1.50] | 19% | 1 | NA | ||
| Cancer | 4 | 1.12 [0.73, 1.71] | 87% | 5 | 1.17 [0.70, 1.96] | 85% | 2 | NA | |
| Hyponatraemia ∗ | 3 | 1.14 [1.10, 1.17] | 0% | 4 | 1.47 [0.89, 2.42] | 67% | 0 | NA | |
| Post PCI | 2 | NA | 4 | 1.64 [1.17, 2.31] | 39% | 0 | NA | ||
| Post coronary bypass | 3 | 1.21 [1.05, 1.40] | 0% | 4 | 1.23 [1.06, 1.43] | 0% | 0 | NA | |
| Beta-blocker | 9 | 0.75 [0.66, 0.85] | 2% | 10 | 0.74 [0.66, 0.83] | 0% | 4 | 0.82 [0.59, 1.14] | 55% |
| ACEi/ARB | 9 | 0.87 [0.70, 1.08] | 64% | 12 | 0.76 [0.59, 0.98] | 65% | 5 | 0.86 [0.48, 1.55] | 82% |
| Diuretics | 6 | 1.21 [0.77, 1.89] | 60% | 7 | 1.32 [0.82, 2.12] | 55% | 2 | NA | |
| Aldosterone antagonist | 3 | 0.88 [0.56, 1.40] | 0% | 4 | 0.87 [0.55, 1.36] | 0% | 2 | NA | |
| Digoxin | 5 | 0.92 [0.76, 1.12] | 62% | 6 | 0.89 [0.72, 1.09] | 53% | 1 | NA | |
| Antiplatelets | 1 | NA | 3 | 0.74 [0.37, 1.49] | 69% | 1 | NA | ||
| Multidisciplinary intervention ∗ | 3 | 0.66 [0.49, 0.88] | 9% | 3 | 0.52 [0.35, 0.78] | 0% | 0 | NA | |
| Ejection fraction < 40% | 5 | 1.11 [0.96, 1.29] | 38% | 6 | 1.31 [1.00, 1.70] | 63% | 1 | NA |
∗ The definition of race, smoking, chronic kidney disease, anemia, hyponatremia, and multidisciplinary intervention were followed according to each study. Details were described in Appendix 5 .
Of the articles for systematic review, 36 provided multivariable models for predicting the primary outcome ( Appendix 6 ). Table 2 lists the frequency of independent predictors for the primary outcome. Risk tendency was consistent for most factors except for anemia and depression. Previous admission, higher blood urea nitrogen, hyponatremia, use of ACEi/ARB, lower systolic blood pressure, and cerebrovascular disease were the most frequent independent risks for the primary outcome. Chronic lung disease, chronic kidney disease, higher creatinine, ischemic heart disease, diabetes, cognitive impairment, and anemia, which had relatively strong association with the primary outcome in the univariable meta-analysis, showed a moderate frequency of being recognized as independent predictors. A past diagnosis of HF and long length of hospital stay were 2 administrative variables that were relatively frequently identified as independent risks. Age, cancer, and hypertension with no significant association with the primary outcome in the univariable meta-analysis were independent with modest frequency. In contrast, black race, higher brain natriuretic peptide, male gender, atrial fibrillation, depression, use of diuretics, and left ventricular systolic dysfunction were less frequently identified as independent.
| Factors | Number of model candidates | Number of significance | Number of not-significance | Percent significance ((Number of significance/Number of model candidates) × 100) | |||
|---|---|---|---|---|---|---|---|
| More risk | Less risk | Consistency | |||||
| 1 | Age | 19 | 7 | 0 | Yes | 12 | 37 |
| 2 | Ischaemic heart disease ∗ | 14 | 7 | 0 | Yes | 7 | 50 |
| 3 | Male gender | 13 | 2 | 0 | Yes | 11 | 15 |
| 4 | Chronic lung disease ∗ | 11 | 6 | 0 | Yes | 5 | 55 |
| 5 | Lower systolic blood pressure | 11 | 9 | 0 | Yes | 2 | 82 |
| 6 | Diabetes mellitus | 10 | 4 | 0 | Yes | 6 | 40 |
| 7 | Higher BUN | 10 | 9 | 0 | Yes | 1 | 90 |
| 8 | Anaemia ∗ | 8 | 5 | 1 | No | 2 | 50 † |
| 9 | Higher BNP ∗ | 8 | 1 | 0 | Yes | 7 | 13 |
| 10 | Prior admission ∗ | 8 | 8 | 0 | Yes | 0 | 100 |
| 11 | Chronic kidney disease | 7 | 3 | 0 | Yes | 4 | 43 |
| 12 | LV systolic dysfunction ∗ | 7 | 2 | 0 | Yes | 5 | 29 |
| 13 | Hyponatraemia ∗ | 7 | 6 | 0 | Yes | 1 | 86 |
| 14 | ACEi/ARB | 6 | 0 | 5 | Yes | 1 | 83 |
| 15 | Previous diagnosis of heart failure | 6 | 4 | 0 | Yes | 2 | 67 |
| 16 | Atrial fibrillation | 6 | 1 | 0 | Yes | 5 | 17 |
| 17 | Cancer | 5 | 2 | 0 | Yes | 3 | 40 |
| 18 | Higher creatinine | 5 | 2 | 0 | Yes | 3 | 40 |
| 19 | Depression | 5 | 2 | 1 | No | 2 | 20 † |
| 20 | Black race | 5 | 0 | 0 | Yes | 5 | 0 |
| 21 | Rales | 5 | 3 | 0 | Yes | 2 | 60 |
| 22 | Cerebrovascular disease | 4 | 3 | 0 | Yes | 1 | 75 |
| 23 | Hypertension | 4 | 0 | 2 | Yes | 2 | 50 |
| 24 | Diuretics | 4 | 1 | 0 | Yes | 3 | 25 |
| 25 | Longer LOS | 3 | 0 | 2 | Yes | 1 | 67 |
| 26 | Cognitive impairment | 3 | 1 | 0 | Yes | 2 | 33 |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
