Predictors of psychiatric rehospitalization among elderly patients

The population of Hong Kong and the proportion of elderly people have been increasing rapidly. The aim of this retrospective cohort study is to determine predictive factors for psychiatric rehospitalization within 2 years among elderly patients who were discharged from psychiatric wards, in attempt to reduce their rehospitalization rate and to reintegrate them into the community. Patients aged 65 and over, who were discharged from psychiatric wards of Pamela Youde Nethersole Eastern Hospital from 1 March 2010 to 29 February 2012, were identified. Rehospitalization within 2 years after discharge was the primary outcome measure, and the time to rehospitalization was measured as the secondary outcome. Patients were subgrouped into readmitted and non-readmitted groups. Logistic regression and Cox regression analyses were applied to the potential predictive factors with odds ratios and hazard ratios obtained, respectively, for the significant findings. Kaplan-Meier survival curves were plotted for graphical representation of the study results in survival analysis. 368 individuals satisfying the study criteria were identified. The same four factors were shown to be significantly associated with rehospitalization in both multiple logistic regression and Cox regression survival analysis. Referral to other psychiatric disciplines upon discharge (p< 0.001, OR=0.325, HR=0.405) was associated with a lower rehospitalization risk and correlated to a longer time to rehospitalization. History of suicidal behaviors (p< 0.001, OR=4.906, HR=3.161), history of violent behaviors (p< 0.001, OR=5.443, HR=3.935) and greater number of previous psychiatric admissions (p< 0.001, OR=1.250, HR=1.121) were associated with a higher rehospitalization risk and predicted earlier rehospitalization. The rehospitalization rate of elderly patients was 5.2% at 1 month, 9.5% at 3 months, 15.0% at 6 months, 17.1% at 1 year, 18.8% at 1.5 year and 20.9% at 2 years.


Background
The population in Hong Kong and the proportion of elderly people are increasing rapidly. Pamela Youde Nethersole Eastern Hospital is a regional hospital under the Hospital Authority serving the eastern district of Hong Kong Island. The characteristics of elderly patients were different from that of general adults, for example, more of them suffered from cognitive disorders, they were more often hospitalized due to medical comorbidities, and they might require placement such as old age homes after discharge. Despite many psychiatric readmission studies were already available in the literature, surprisingly very few of them were targeted to elderly patients only.
Rehospitalization had been regarded as a useful indicator to measure quality of care provided by hospitals worldwide 1 , and no such data was available for elderly psychiatric patients in Hong Kong. Therefore this study was conducted to identify risk factors in attempt to reduce rehospitalization, so mental health services could be utilized more effectively in view of of the increasing needs from the aging population.

Methods
The aim of this retrospective cohort study is to determine predictors for psychiatric rehospitalization over 2 years among elderly patients after they were discharged from psychiatric wards.

Inclusion criteria
Patients aged 65 and over, who were discharged from the psychiatric wards of Pamela Youde Nethersole Eastern Hospital from 1 March 2010 to 29 February 2012, formed the study population. For patients having repeated discharge episodes during the study period, only the first discharge episode was included as the index episode for that patient to avoid duplication of data from the same individuals.

Exclusion criteria
Patients who died during inpatient stay at the index episode, or had not received mental health care in any psychiatric clinics under the Hospital Authority after discharge, including those who were followed up by private psychiatrists or overseas psychiatric care systems, were excluded due to a lack of information to analyze their outcome.

Dependent variable
Rehospitalization was defined as the primary outcome.

Independent variables
Socio-demographic factors included age (upon discharge), gender, ethnicity, marital status, education level and type of residence (upon discharge). Clinical factors included priority follow-up (PFU) status (see Appendix for more details), length of inpatient stay, primary psychiatric diagnosis (as determined from the medical coding with ICD-9-CM in CMS by respective case doctors), presence of psychiatric comorbidities (as reflected by more than one psychiatric diagnosis recorded in CMS in the index episode), number of chronic physical illnesses (which required regular outpatient follow ups by other specialties), MMSE scores, referral to other psychiatric disciplines upon discharge (including community psychiatric nurses, clinical psychologists, social workers or day hospital), history of suicidal behaviors (including suicide attempts and self-harm behaviors in lifetime), history of violent behaviors (in lifetime) and number of previous psychiatric admissions.

Null hypothesis
The null hypothesis was that none of the identified factors are associated with rehospitalization.

Univariate tests and multiple logistic regression analysis
After descriptive statistical studies, univariate tests were conducted on all independent variables to identify possible significant factors for subsequent analysis. For categorical factors, a Chi-squared test was performed, and Fisher's exact test was conducted for factors with an expected cell count less than 5. All the continuous factors in this study were not normally distributed as determined by the Shapiro-Wilk test and Mann-Whitney U test was conducted. Significant and marginally significant factors with a p value of <0.100 identified in the univariate analysis were included in the subsequent multiple logistic regression analysis to identify significant factors (with a p value of <0.050) predicting rehospitalization in 2 years after discharge.

Survival analysis
Survival analysis was further conducted to identify factors predicting earlier rehospitalization. The rehospitalization rates at different time points within the 2 years were calculated. Univariate analysis with simple Cox regression was performed to all variables. Those with a p value of <0.100 were included in the subsequent multiple Cox regression analysis, to determine variables that significantly (with a p value of <0.050) correlated with time to rehospitalization. Kaplan-Meier survival curves were also plotted. The coding for each variable is explained in the top row. Blank rows correspond to the discharge episodes that were excluded according to the exclusion criteria. Concerning the socio-demographic characteristics of the study population, the median age was 76 and male to female ratio was 1 to 1.61. 98.1% of individuals were Chinese. 47% of the study population was married, 39.7% was widowed, 7.9% was divorced or separated and 5.4% was single. For education level, 41.6% was less than primary, 31% was primary, 20.6% was secondary and only 6.8% was tertiary or above. 52.7% of the study population lived at home with family or friends, 33.7% was living in placement, 9% lived alone at home and 4.6% lived with maid at home. Regarding the clinical characteristics of the study population, the median length of inpatient stay for the study population was 27 days.   Table 1 and Table 2 respectively. Results of subsequent multiple logistic regression are shown in Table 3. Four significant factors were identified with their odds ratios calculated. Referral to other psychiatric disciplines upon discharge (p< 0.001, OR=0.325) was associated with a lower risk of rehospitalization, while the other three factors including history of suicidal behaviors (p< 0.001, OR=4.906), history of violence (p< 0.001, OR=5.443) and number of previous admissions (p< 0.001, OR=1.250) were associated with higher risk of rehospitalization. The overall accuracy of this predictor model was 83.4%. Survival analysis with Cox regression was then conducted to examine factors that predicted earlier rehospitalization. For individuals who did not readmit but were deceased before completion of the 2 year follow up period, the date of death was taken as the date of censoring. Taking different time points in the Kaplan-Meier survival curve as plotted in Figure 2, the cumulative readmission rate of the study population was 5.2% at 1 month, 9.5% at 3 months, 15.0% at 6 months, 17.1% at 1 year, 18.8% at 1.5 year and lastly 20.9% at 2 years upon completion of the follow up period. Most of the rehospitalization occurred within the first 6 months after discharge where the slope of the curve was steepest.   Univariate analysis by simple Cox regression was performed and the results are shown in Table 4. The significant results of subsequent multiple Cox regression survival analyses are presented in Table 5. The same four significant factors were identified in the multiple Cox regression analysis as that in the previous logistic regression analysis. Hazard ratios were calculated and Kaplan-Meier survival curves were plotted. Referral to other psychiatric disciplines upon discharge (p< 0.001, HR=0.405) correlated to a longer time to rehospitalization (Figure 3

Discussion
This study involved Chinese and non-Chinese elderly psychiatric patients in Hong Kong, and by comparison between the two groups, ethnicity was not identified as a significant factor. However, it should be noted that too few cases were non-Chinese and so the statistical comparison between them might not be meaningful. In comparison to other studies involving psychogeriatric patients only, most studies did not find type of residence to be a significant factor in psychiatric readmission, except 1 study 22 which reported higher  readmission risk in patients with little or no supervision in living arrangements following discharge. However, this study did not find any significant differences in outcome regarding rehospitalization between patients who lived alone, with others, or in placement.
The same four clinical factors were shown to be significantly associated with rehospitalization with a p value of <0.001 in both multiple logistic regression and Cox regression survival analysis, including referral to other psychiatric disciplines upon discharge, history of suicidal behaviors, history of violence and number of previous psychiatric admissions. With adequate aftercare services 6,7,29,30 by referral to other disciplines for support upon discharge, patients were more likely to comply with treatment plans and have their problems managed in outpatient settings, which reduced the need for psychiatric rehospitalization. Moreover, a study 12 had reported that patients with previous suicidal attempts were more often readmitted, and similar findings had been shown in studies that patients having violence behaviors previously were at higher risk of readmission 11,25,28,31 . Psychiatric rehospitalization was often required when patients displayed risks of harming self or others. Furthermore, studies had extensively reported that patients with greater number of previous admissions were predisposed to rehospitalization 7,8,15,20,25,32,33 . These patients might be more often rehospitalized due to suboptimal control on symptoms relating to their mental health problems and inadequate support in the community.
Concerning factors that were more specific to the elderly, 1 reviewed article 27 reported that elderly patients without medical comorbidity were more likely rehospitalized, however, number of chronic illnesses was not identified as a significant risk factor. Elderly patients were more likely to be hospitalized due to poor physical conditions. Sometimes psychiatric symptoms had already been treated in general  wards, but readmission to general wards were not accounted in this study. Besides, having a cognitive disorder or a lower MMSE score would not affect the risk of rehospitalization. However, it was noted that over one-fourth (26.6%) of the study population did not have their MMSE scores documented, and therefore the results might differ if the MMSE scores from all study individuals were available.

Limitations
Firstly, this was a retrospective study and the correctness of data depended heavily on the information in CMS. Some variables could have been underestimated or changed during the follow up period. Secondly, the study population involved the psychiatric unit in a regional hospital in Hong Kong only. The findings might not be generalized to other parts of the world.

Strengths
Data on patients' characteristics were retrieved through medical records rather than self-reporting from patients and this could minimize recall bias. Besides, this study involved elderly patients only and the findings were more specific towards psychogeriatric patients.

Implications
These findings were important to the daily practice of psychiatrists and especially the psychogeriatricians, since most of the previous rehospitalization studies in the literature were not specifically targeted to elderly patients. There was very little information available regarding risk factors that predicted psychiatric rehospitalization of elderly patients in Hong Kong prior to this study. In view of the aging population and increasing need for psychiatric services in Hong Kong, this study could help in identifying elderly patients with a high rehospitalization risk for more intensive treatments and better discharge planning based on their risk factors, including history of suicidal behaviors, history of violent behaviors and greater number of previous psychiatric admissions. Factors including cognitive disorders, lower MMSE scores, comorbid chronic physical illnesses, marital status, type of residence, age, gender, ethnicity, PFU status, education level, length of inpatient stay, and presence of psychiatric comorbidities did not affect the risk of rehospitalization in the elderly psychiatric patients. On the other hand, referral to other psychiatric disciplines upon discharge was highly encouraged to prevent rehospitalization.