Hospital readmission after critical care survival

Survivors of critical illness frequently require increased healthcare resources after hospital discharge. We undertook a systematic review and meta-analysis to assess hospital re-admission rates following critical care admission and to explore potential re-admission risk factors. We searched the MEDLINE, Embase and CINAHL databases on 05 March 2020. Our search strategy incorporated controlled vocabulary and text words for hospital re-admission and critical illness, limited to English language. Two reviewers independently applied pre-de ﬁ ned eligibility criteria and assessed quality using the Newcastle Ottawa Score checklist and extracted data. Primary outcome was acute hospital re-admission in the year after critical care discharge. Of the 8851 studies screened, 87 met inclusion criteria and 41 were used within the meta-analysis. The analysis incorporated data from 3,897,597 individual patients and 741,664 re-admission episodes. Pooled estimates for hospital re-admission after critical illness were 16.9% (95%CI: 13.3 – 21.2%) at 30 days; 31.0% (95%CI: 24.3 – 38.6%) at 90 days; 29.6% (95%CI: 24.5 – 35.2%) at six months; and 53.3% (95% CI: 44.4 – 62.0%) at 12 months. Signi ﬁ cant heterogeneity was observed across included studies. Three risk factor contributed to excess acute care rehospitalisation one year after discharge: the presence of comorbidities; events during initial hospitalisation (e.g. the presence of delirium and duration of mechanical ventilation); and subsequent infection during the post-hospital discharge period. Hospital re-admission is common in survivors of critical illness. Careful attention to the management of pre-existing comorbidities during transitions of care may help reduce healthcare utilisation after critical care discharge. Future research should determine if targeted interventions for at-risk critical care survivors can reduce the risk of subsequent rehospitalisation.


Introduction
Survivorship after critical illness brings challenges to patients and their primary caregivers in the months after hospital discharge [1,2]. These include physical, social, emotional and cognitive problems [3][4][5][6]. Critical care survivors frequently require access to outpatient and acute inpatient hospital resources in the post-discharge period [7,8]. Hospital re-admission may cause distress for individual patients and their caregivers; and increase strain on the healthcare system [9,10]. For patients who survive critical care, it is not currently clear what proportion of hospital re-admissions are potentially preventable nor the proportion that indicate terminal decline, as observed in other sub-groups of the population (e.g. older adults) [11].
A greater understanding of the use of healthcare resources across the clinical recovery continuum, as well as delineation of potential modifiable risk factors, may help support the individual patient as well as the healthcare system. There is therefore a need to synthesise the current evidence base, to inform future interventional work in the field.
We conducted a systematic review and meta-analysis to understand the frequency of hospital re-admission after critical care survival. A secondary objective was to evaluate potential risk factors for re-admission. We hypothesised there would be a high hospital re-admission rate in the months following discharge and that prior health status would play an important contributory role to the use of healthcare resources.

Methods
No ethical approvals were sought for this secondary analysis of previously published data. This systematic review was prospectively registered and conducted and reported according to the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines [12]. The search strategy was formulated according to the CoCoPop (condition, context and population) mnemonic which is recommended for systematic reviews designed to address prevalence and incidence data (Table 1) [13].
Eligible studies had a randomised controlled trial, cohort or case-control design. Only studies in which > 50% of the study population had been admitted to a critical care environment were included. Narrative reviews, editorials, case reports, duplicate publications, qualitative studies and conference abstracts were excluded. We also excluded studies that were limited to children or neonates and those that reported re-admission to a critical care environment during the same hospital encounter. In addition, we excluded specialist ICU populations (e.g. cardiothoracic and neurosurgical) from inclusion in the meta-analysis as the focus was the general critical care population only. Data on the type of critical care population, including re-admission rates and risk factors for hospital re-admission, are detailed in the online Supporting Information (Table S1) We included studies that met the following criteria: adults (aged >18 yrs); inclusion of hospital re-admission data; and studies where more than 50% of the population being studied had been admitted to a critical care environment. Each study was independently reviewed for eligibility by two clinicians, first by title and abstract review followed by full-text review. Eligibility disagreements were resolved by a third reviewer. We used the Covidence software package (v2619) to undertake the study selection phase and data extraction. When two or more studies reported data from the same patient cohort, the most relevant article was chosen. Of note, a small number of publications included patients from the same cohort but the studies reported hospital re-admissions at different timepoints. If a study cohort reported on the same cohort of patients but included different longitudinal re-admission data, both studies were analysed.
Re-admission rate, within the context of this review, was defined as the number of patients re-admitted to hospital after initial discharge at least once during the study follow- Cohort study quality was assessed using the Newcastle Ottawa Score checklist [12]. This consists of three main domains to assess the quality and risk of bias. These are as follows: patient selection (cohort data source, representativeness and ascertainment of exposure to the outcome of interest); comparability of cohort; and outcome assessment (including adequate follow-up time, acquisition of outcome and adequacy of follow-up). We assessed for the risk of bias in the randomised controlled trials in this analysis using the Cochrane risk of bias methodology [14].
Data on risk of bias and overall quality assessment are presented in the online Supporting Information (Table S2).
Reviewer agreement was assessed with the j statistic and was interpreted according to Landis and Koch guidelines [15]. Data from eligible studies were pooled for the primary outcomes (hospital re-admission). Pooling was undertaken at the four most frequently reported time frames in the literature: 30 days; 90 days; 6 months; and 12 months. Other data were not included in the metaanalysis due to limited data available at these time-points.
We also included a sub-group analysis of studies that examined hospital re-admission in patients who had prolonged exposure to critical care, defined as patients ventilated for, or with a critical care stay, of >7 days. One study also included the definition: 'ventilation for 4 days with a tracheostomy in place, or ventilation for 21 days without a tracheostomy. 'After reviewer discussion, this was included in the prolonged exposure cohort. We limited inclusion to this component of meta-analysis to readmission rates at 12 months after hospital discharge.

Random-effect meta-analysis with Clopper Pearson 95%
CIs and 95% prediction intervals (PI) was used to obtain an estimate of the effect size for the primary outcome measure (hospital re-admission). Data were pooled across the entire population and reported from each study. Patients who died in hospital after critical care admission were not included within re-admission rate calculations. Random-effects metaregression log odds were used to estimate pooled proportions of hospital re-admission, including time to readmission (30 days, 90 days, 6 and 12 months); location of study (Europe, Asia, South America, Canada and USA); type of critical care admission (surgical, medical or mixed); and study type (multicentre or single-centre). The I 2 statistic was used to assess study heterogeneity. The I 2 represents the percentage of total variance across studies that was attributable to heterogeneity rather than change.
Heterogeneity was defined as I 2 >50%. Analysis was performed using R (V4.10) and data visualisation was undertaken using the R Package ggplot2. All data produced for this analysis are provided in the online Supporting Information (Table S1). The full R code is included in the online Supporting Information (Appendix S2).

Results
Our search strategy identified 9524 records. After duplicates were removed, 8851 were screened for inclusion. Of these, 8540 were excluded based on the title or abstract. Therefore, 87 studies met the eligibility criteria and were included in this analysis ( Fig. 1) . The j value for agreement on full text was excellent (0.90, p <0.01). We excluded specialist ICU populations (e.g. cardiothoracic and neurosurgical) from inclusion in the meta-analysis as the focus was the general critical care population only. Therefore, 41 studies were included in the meta-analysis. were multicentre in nature ( Table 2). The full characteristics and outcomes of studies included are presented in the online Supporting Information (Table S1). A summary of the main features of the included studies is presented in Table 2.

Risk of bias
The quality assessment for the included studies is shown in the online Supporting Information (Table S2). The  Figure   S1). These plots suggested that there was heterogeneity of the reported pooled proportions from studies included in the meta-analysis.

Meta-analysis: hospital re-admission following critical illness
For the meta-analysis, only hospital re-admissions up to 12 months post-discharge were included, as these were the most frequently reported outcomes. We did not include studies that reported ICU re-admission in isolation or ICU re-admission within the same hospital encounter.
Therefore, 41 studies were included in the meta-  Figure S2). We undertook a further sensitivity analysis for those studies deemed to be at very high risk of bias (Newcastle Ottawa Score ≤ 3 or those deemed to be at high risk of bias using the Cochrane Risk of bias methodology). Again, this yielded no difference in the synthesised results (online Supporting Information, Figure S3).

Risk factors for hospital re-admission
Utilising study data included in the pooled meta-analysis, 28 studies reported risk factors for re-admission. Adverse events during the initial hospitalisation were also cited as risk factor for re-admission in 12 (42.9%) of these studies. Risk factors  (Fig. 3). There was evidence of heterogeneity across the studies (I 2 = 79%, p < 0.01, s 2 = 0.3). Risk factors for re-admission in the prolonged stay cohort were explored in five studies [42,49,75,92,100]. One study reported that prolonged ventilation was a risk factor for re-admission at 6 and 12 months post-discharge [42], while another reported that those patients with shorter critical care stays were at a higher risk of re-admission at 30 days post-discharge [75]. Three studies reported that either infection or sepsis was the most common reason for readmission in this sub-group [49,92,100].

Discussion
This review has shown that acute rehospitalisation following     Figure 3 Rate and timing of rehospitalisation in long-term stay patients. Random-effect meta-analysis of proportions by rehospitalisation interval reported.

12-month rehospitalisation
More work is required to understand how best to support these patients in the post-hospital discharge phase.
We identified that multimorbidity before critical illness and baseline frailty were risk factors for hospital readmission. This is consistent with previous qualitative research highlighting the relationship between complex health and psychosocial needs and hospital re-admission, especially in the context of multimorbidity and polypharmacy [9]. There are a number of potential clinical

Supporting Information
Additional supporting information may be found online via the journal website.     Appendix S1. Review search strategy.