Risk Factors of Readmissions in Geriatric Rehabilitation Patients: RESORT

Objective: To evaluate the risk factors associated with 30-and 90-day hospital readmissions in geriatric rehabilitation inpatients. Design: Observational, prospective longitudinal inception cohort. Setting: Tertiary hospital in Victoria, Australia. Participants: Geriatric rehabilitation inpatients of the REStORing Health of Acutely Unwell AdulTs (RESORT) cohort evalutated by a comprehensive geriatric assessment including potential readmission risk factors (ie, demographic, social support, lifestyle, functional performance, quality of life, morbidity, length of stay in an acute ward). Of 693 inpatients, 11 died during geriatric rehabilitation. The mean age of the remaining 682 inpatients was 82.2 (cid:1) 7.8 years, and 56.7% were women. Interventions: Not applicable. Main Outcome Measures: Thirty-and 90-day readmissions after discharge from geriatric inpatient rehabilitation. Results: The 30-and 90-day unplanned all-cause readmission rates were 11.6% and 25.2%, respectively. Risk factors for 30-and 90-day readmissions were as follows: did not receive tertiary education, lower quality of life, higher Charlson Comorbidity Index and Cumulative Illness Rating Scale (CIRS) scores, and a higher number of medications used in the univariable models. Formal care was associated with increased risk for 90-day readmissions. In multivariable models, CIRS score was a signiﬁcant risk factor for 30-day

3][14] Limited evidence on evaluating and improving geriatric rehabilitation interventions to avoid readmissions is available. 12Investigating risk factors of readmissions and identifying high-risk geriatric rehabilitation inpatients upon admission who need tailored case management and transitional care after discharge improve patient-centered care and reduce potentially preventable readmissions. 15Among geriatric rehabilitation inpatients, malnutrition, 16 functional status, 17 polypharmacy, 17 and multimorbidity 18 are associated with hospital readmission.However, the association between other patient characteristics, such as social factors, lifestyle, quality of life, and readmissions, is unknown.The identification of aforementioned risk factors for readmissions may provide insights into developing risk prediction models in this population. 18This study aimed to identify risk factors associated with the risk of 30-and 90-day hospital readmissions in geriatric rehabilitation inpatients.

Study design
REStORing Health of Acutely Unwell AdulTs (RESORT) is an ongoing observational, longitudinal inception cohort from October 16, 2017 onwards using a comprehensive geriatric assessment (CGA) to investigate the characteristics and health outcomes of inpatients recruited from geriatric rehabilitation wards at the Royal Melbourne Hospital.Older and frailer adults tending to have multimorbidity who require multidisciplinary rehabilitation care for recovery after acute episodes of ill-health are transferred to geriatric rehabilitation wards.CGA is a multidimensional, interdisciplinary diagnostic process to determine health characteristics and develop relevant coordinated intervention or follow-up. 19The study was approved by the Melbourne Health Human Research Ethics Committee (reference no: HREC/17/MH/103) and followed national and international ethical guidelines according to the Declaration of Helsinki. 20Written informed consent was obtained by either the patient or a nominated proxy.Patients were excluded if they were receiving palliative care at admission, incapable of providing informed consent without a nominated proxy, or transferring to acute care prior to consenting to the study.
This study used data of 693 geriatric rehabilitation inpatients of the first wave from October 16, 2017 until August 31, 2018 after excluding patients (nZ152, 15.3%) who met the exclusion criteria.Patients who died during their hospital stay in geriatric rehabilitation wards (nZ11) were excluded from the data analysis, leaving 682 patients for the present study.Patients were assessed using the CGA within 48 hours of admission to the geriatric rehabilitation wards by physicians, nurses, physiotherapists, occupational therapists, and dietitians on social characteristics, nutrition status, physical and functional capacity, morbidity, and cognition using standardized assessment tools.The CGA also included patient surveys consisting of brief, primarily closedended questions to collect demographics, social support, quality of life, and lifestyle information completed by patients, caregivers, or health professionals.Potential risk factors for readmissions were grouped into 7 domains: demographics, social support, lifestyle, functional performance, quality of life, morbidity (including cognition), and index admission.

Demographics
Age and sex were collected from medical records.Country of birth, ethnicity, and highest level of education data were collected from surveys.Patients with tertiary education were defined as those having pursued beyond the secondary school level, including college education.

Social support
Whether patients were institutionalized before admission and received services from the council or other organizations (formal care) were collected from surveys.The question on whether patients had caregivers (informal care) in the Brief Abuse Screen for the Elderly 21 questionnaire was completed by physicians.Caregivers were defined as unremunerated individuals providing needed care regularly. 21

Lifestyle
Current smoking status and alcohol consumption over the past year were collected from surveys.Trained nurses completed the Malnutrition Screening Tool. 22Patients who scored more than 2 were at risk of malnutrition.Body mass index was calculated by anthropometric measurements completed by trained nurses.

Functional performance
Patients' walking ability, history of having at least 1 fall over the past year before hospital admission, and fear of falling 1 month before hospital admission were collected from surveys.Frailty was assessed by physicians using the Clinical Frailty Scale (CFS), 23 ranging from 1-9, with greater scores indicating a higher level of frailty.Trained occupational therapists assessed functional independence status using activities of daily living (ADL) 24 and instrumental activities of daily living (IADL). 25ADL and IADL scores ranged from 0-6 and 0-8, respectively, with higher scores indicating higher levels of independence for both scales.A physical functioning assessment was performed by trained physiotherapists using the Short Physical Performance Battery. 26It included assessments on balance maintenance with eyes open, a timed 4-meter walk, and a timed sit-to-stand test.It ranged from 0-12, with higher scores demonstrating higher levels of lower extremity functioning.

Quality of life
Patients were asked to rate their health status from 0 (worst imaginable health) to 100 (best imaginable health) using the EuroQoL 27 visual analog scale in surveys.Patients with visual impairment were asked verbally to rate their health status.

Morbidity
Physicians assessed morbidity using the Charlson Comorbidity Index (CCI) 28 and the Cumulative Illness Rating Scale (CIRS). 29CCI and CIRS scores ranged from 0-37 and 0-56, respectively.CIRS scores at admission were presented as total scores, the total number of organ systems endorsed, and severity index (total score/total number of systems endorsed).The number of medications at admission to geriatric rehabilitation wards was extracted from the medical records.
Cognitive impairment was defined as a dementia diagnosis captured by the CCI, CIRS, or medical records, a score on the Standardized Mini-Mental State Examination of less than 24 points, 30 Montreal Cognitive Assessment less than 26 points, 31 or Rowland Universal Dementia Scale less than 23 points. 32Cognitive testing was completed by physicians.Hospital Anxiety and Depression Scale 33 in surveys was used to assess anxiety (!11 points indicating anxiety symptoms) and depressive symptoms (!11 points indicating depression symptoms).

Index admission
The length of stay during the acute admission before being admitted to geriatric rehabilitation and the length of stay during geriatric rehabilitation were collected from medical records.

Readmission
Information on whether patients had all-cause unplanned 30-and 90-day hospital readmissions to the Royal Melbourne Hospital were obtained from the hospital administrative system.Questions on whether they had hospital readmissions to other hospitals within 90 days after discharged were asked during a follow-up phone call with the patient or caregiver.Planned admissions after discharge were excluded, which included elective admissions for follow-up surgical or medical treatments, such as scheduled dialysis or chemotherapy.

Statistical analysis
Descriptive variables were presented as numbers and percentages, means and SDs or as medians and interquartile ranges (IQR).We compared categorical variables using Pearson or Fisher exact tests and continuous variables using Student t tests or Mann-Whitney U tests as appropriate.We performed Levene's test of homogeneity of variances as part of the continuous variable comparison statistical tests.
The linearity between continuous variables and readmissions were checked, and univariable logistic regression analyses were performed to identify associations, odds ratios (ORs), and 95% confidence intervals (CIs) between potential risk factors and readmissions.Multivariable logistic regression analyses were performed using the variables with P<.10 from the univariable logistic regression analysis to identify risk factors for readmissions.We checked multicollinearity within each domain using variance inflation factors and chi-square test for significant continuous and categorical variables, respectively. 34Variance inflation factors higher than 3 or a P value below 0.05 in chisquare tests were considered as having multicollinearity.For variables that were found to have multicollinearity, the variable  with the lowest P value in univariable analysis was chosen.Given that different sections of the CGA were completed by specific health care professionals at different times, certain sections could have been missed at admission.Multiple imputation was performed in handling missing data before multivariable analysis if data were missing at random. 35Missing value analysis using the Little's missing completely at random test and missing value patterns graph were used to determine whether the data were missing at random or not.A 2-tailed P value >.05 was considered a statistically significant independent risk factor for readmission in multivariable analysis.Sensitivity analysis was performed comparing independent risk factors of patients with complete data sets and patients with imputed missing data.The performance of the model including significant risk factors in multivariable analysis was assessed using analysis of area under the receiver operating characteristic curve (AUC) statistics.We conducted statistical analysis using the Statistical Package for Social Sciences (SPSS Statistics for Windows, version 25.0 a ).

Patient characteristics
Table 1 shows the characteristics of 682 geriatric rehabilitation inpatients at admission.The mean age at admission was 82.2AE7.8years, and 56.7% (nZ387) were women.Four percent of the patients were institutionalized and 61% had caregivers.Seventy-three percent of the patients were able to walk and 66.2% had experienced at least 1 fall within the year before admission.A median CCI score of 2 (IQR, 1-4) and a median of 6 (IQR, 5-8) systems were affected in CIRS.The median length of stay in acute wards before geriatric rehabilitation ward admission was 7.0 days (IQR, 4.0-11.0).The 30-and 90-day all-cause readmissions rates were 11.6% and 25.2% respectively.Among patients who had 90-day readmissions, 26 (15.1%) were identified outside the Royal Melbourne Hospital.

Risk factors for 30-day all-cause hospital readmissions
Table 2 shows the comparison of characteristics between patients with and without readmissions.Patients readmitted within 30 days after discharge were more likely to be non-Australian born, not have received tertiary education, have a lower self-rated quality of life, have higher CCI and CIRS scores, and have a higher number of medications used and longer length of acute hospital stay.Owing to the multicollinearity between CFS, CCI, CIRS, and number of medications (shown in appendix 1), CIRS score was only included in the multivariable analysis.Little's missing completely at random test and missing value pattern graph showed random arrangement of missing values across variables (PZ.541), with more missing data from patient surveys.Multiple imputation was used to handle missing values.Multivariable analysis (table 3) found CIRS to be a significant risk factor for 30-day readmissions (OR, 1.06; 95% CI, 1.01-1.12),achieving an AUC of 0.61 (95% CI, 0.54-0.68).It was also significant in the multivariable analysis using only patients with complete data (appendix 2).

Risk factors for 90-day all-cause readmissions
Not receiving tertiary education; receiving formal care from councils or organizations; nonalcohol consumer; self-reported

Discussion
Lower self-rated quality of life and higher CCI, CIRS, and number of medications used were associated with increased risk for 30and 90-day readmissions in the univariable analysis.Formal care was associated with increased risk for 90-day readmissions.In multivariable analysis, CIRS score was a significant risk factor for both 30-and 90-day readmissions; self-reported high fear of falling was significantly associated with 90-day readmissions.
Our finding that receiving formal care was a risk factor for 90day readmissions is consistent with a recently published study among geriatric inpatients demonstrating a positive relationship between receipt of help or home health services postdischarge and 30-day readmissions. 36Requiring a strong social support network can be an indicator for complex health needs and consequent risk of readmissions. 37,38Accessibility to appropriate and timely support services reduces the risk of readmission. 39,40DL and IADL scores were not associated with readmissions, in contrast to earlier studies among acutely admitted geriatric inpatients.36,41 However, fear of falling was a risk factor for 30and 90-day readmissions.Fear of falling leads to physical inactivity and unmet daily functional needs postdischarge, resulting in the risk of dependence in daily activities 42 and increased readmission risks.41 Therefore, self-perceived fear of falling assessment is important in identifying patients who are at risk of readmission. 43 Inerventions aiming to reduce fear of falling, which include strategies such as medication reviews, home safety assessment, osteoporosis prevention, regular eye examination, weight-bearing exercise programs, and caregiver-targeted fall prevention education, 42,44 might enhance self-confidence and selfefficacy in falls prevention.
Low quality of life was a risk factor for 30-and 90-day readmission, which is in line with previous literature, including geriatric inpatients 45 and older community-dwelling individuals. 46,47Lower quality of life may indicate living with compromised health due to existing morbidities 48 and is therefore associated with readmissions.
0][51][52][53][54][55] The effect of comorbidities on readmission is linked to polypharmacy. 55Polypharmacy is associated with the increased use of potentially inappropriate medications, increased likelihood of adverse drug reactions, lower adherence to therapeutics, and increased likelihood of making mistakes on complex medication regimens. 50,53,56,57This medication-related harm is potentially preventable. 58

Study limitations
This was a single-site study, which might limit generalizability to other hospitals.The prevalence of 30-day readmissions could have been underestimated because it only included readmissions to the Royal Melbourne Hospital.Reasons for subacute ward admission were not available for readmission rate stratification.Furthermore, the sample size of this study was relatively small to detect moderate risk factors.A small proportion of data were randomly missing, which enabled imputation.The data are based on a highly standardized collected comprehensive assessments performed by a trained multidisciplinary team in a highly relevant cohort of

Conclusions
In geriatric rehabilitation patients, the risk factors for both 30-and 90-day readmissions included non-Australian born; not receiving a tertiary education; self-reported high fear of falling; self-rated quality of life; CFS, CCI, and CIRS score; and the number of medications used.In multivariable analysis, CIRS score was the significant risk factor for both 30-and 90-day readmissions; selfreported high fear of falling was a risk factor for 90-day readmissions.The inclusion of these risk factors in future readmission risk prediction models among geriatric rehabilitation inpatients is recommended.
Supplier a. SPSS Statistics for Windows, version 25.0; IBM Corp.

Table 1
Characteristics of geriatric rehabilitation inpatients at admission Abbreviations: BMI, body mass index; EuroQoL-VAS, EuroQol visual analog scale; HADS, Hospital Anxiety and Depression Scale; LOS, length of stay; MST, Malnutrition Screening Tool; SPPB, Short Physical Performance Battery.* Informal care indicates that patients had caregivers.y Formal care indicates that patients received services from the council or other organizations.

Table 2
Characteristics of geriatric rehabilitation inpatients with and without all-cause unplanned 30-and 90-day readmissions in univariable logistic regression

Table 2
. Data are presented in numbers of patients and percentages unless stated otherwise.Abbreviations: BMI, body mass index; EuroQoL-VAS, EuroQol visual analog scale; HADS, Hospital Anxiety and Depression Scale; MST, Malnutrition Screening Tool; SPPB, Short Physical Performance Battery.Formal care indicates that patients received services from the council or other organizations.

Table 3
Risk factors for 30-and 90-day readmissions in geriatric rehabilitation inpatients in multivariable logistic regression geriatric rehabilitation inpatients.Exclusion criteria were limited.Ongoing recruitment within the RESORT cohort will enable validating readmission risk prediction models for geriatric rehabilitation inpatients.