The pre-discharge variables analysed were classified into the following six categories: 1) patients’ demographic, social and economic characteristics; 2) patients’ clinical characteristics; 3) patients’ clinical history; 4) patients’ attitude and perception; 5) environmental, social and hospital characteristics; and 6) admission and discharge characteristics. The sections below report the results for each of these groups of variables.
Patients’ demographic, social and economic characteristics
Among patients’ demographic, social and economic characteristics the main results are synthetized in Table
3.
Table 3
Synthesis of the main significant results regarding patients’ demographic, social and economic characteristics
Age | 15/44 | Mixed direction (10) | Older age protective (8)b |
Gender | 13/46 | Mixed direction (10) | Mixed direction (8) |
Marital status | 9/28 | Being married protective factor (5) | Being married protective factor (5) |
Living situation/number of cohabitant/residential stability | 5/20 | Mixed direction (4) Homelessness risk factor (1) | Mixed direction (4) |
Education level | 4/14 | Mixed direction (3) | For involuntary hospitalization: education protective factor (1) |
Employment status | 5/15 | Unemployment risk factor (5) | No significant results |
Ethnical group/immigration status | 6/29 | Being black risk factor (2) | Mixed direction (6) |
Financial status | 1/6 | Higher financial means protective factor (1) | Higher financial means protective factor (1) |
Receiving benefits (pension or for a service-connected disability or other welfare benefits) | 5/6 | Receiving benefits risk factor (3)c | Service-connected disability risk factor (1) |
Forensic and violence issues | 1/3 | Violence history protective factor (1) | No significant results |
Military situation | 1/2 | No significant results | Non-service connected disability and highest income or a non-veteran protective factor (1) |
In eight cases [
15,
16,
21,
33,
48‐
51], risk of readmission was associated with younger
age at multivariate level, but only in four cases a significant decrease in risk was found with age in all the analyses performed. However, some occurrences of non-monotonic behaviour (two at multivariate level) emerged [
24,
26,
52,
53] and a higher risk for older age was found but, when also multivariate analysis was performed, in no case age remained significant [
23,
34,
47].
As for
gender, in multivariate analysis, a consistently higher risk for men resulted in four papers [
31,
40,
52,
54], while, in four cases [
9,
15,
55,
56] a higher risk for female patients was found.
Concerning
marital status, being married (including also cohabitee/partner in a few studies) proved somehow protective in nine papers [
21,
23‐
25,
33,
48,
51,
57,
58] (in four cases only in bivariate analysis). In Wong and Chung [
48], the result actually just pointed out an increase in the risk for singles (but only in bivariate analysis), while in Bernardo et al. [
58] and Grinshpoon et al. [
51] (in the case of affective but not in that of schizophrenic patients) for divorced people.
As for
living situation, in terms of place (mainly, whether owning a home, living in an institution or being homeless), and of household composition (i.e., with whom the patient is living, especially whether alone or not), most of the papers analysing such variables did not meet statistical significance. Living in care (vs alone or with family) was found as a protective factor in Dixon et al. [
30], and Russo et al. [
59] found homelessness as a risk factor at bivariate level, while living alone was found as protective in Priebe [
14] and in Adams [
60]. In Ono et al. [
18], the variable “number of cohabitants” was considered, a larger number turning out to be a protective factor for readmission.
At bivariate level two articles on all patients with psychiatric disorders found a protective role for higher
education (i.e., a higher risk of readmission for patients with primary education or illiteracy, and lower for those with university degree [
57]; a low level of education turned out as a risk-increasing factor [
36]), while one paper found a lower educational level as a protective factor for readmission [
58]. The only significant association found in multivariate analysis (in one paper for subjects who were hospitalized involuntarily) highlighted that the number of years of education was associated with a decrease in readmission risk [
61].
A protective behaviour of
employment was found in five papers, but only in bivariate analysis. Being a skilled worker turned out to be a protective factor while being unemployed a risk factor [
57]; full-time employment turned out as protective as well vs part-time employment, receiving social assistance or being unemployed [
58]. Patients who were either employed or students showed a lower readmission risk [
46]; also an increased risk was found for patients not in employment vs those who were employed (including subsistence and in the Army forces) [
60] and unemployment was found as a risk factor for early readmission [
29].
As for
ethnical group, being black was found to be significantly associated with a higher risk of readmission in two papers in multivariate analyses: when examining the 5-year readmission risk (vs white patients) [
26] and the 60-day readmission risk (vs native American and Asian patients, only for some subgroups analysed) [
49]. In Phibbs et al. [
24], on the contrary, being black (vs white) turned out as a protective factor. The other ethnical group meeting significant results in the literature was the Hispanic one, associated with a lower risk of readmission at 8–30 days (compared with white patients) in Mark et al. [
52], but with a higher risk (compared with white and other non-black patients) in Stahler et al. [
32] and (compared with white and black patients) in Becker and Shafer [
33].
Among
socioeconomic factors, income, socioeconomic status and financial status were not significantly associated to readmission in five papers [
19,
25,
28,
48,
55], while higher financial means were found as a protective factor in Owen et al. [
62]. The variable “presence of a disability support pension” resulted as a risk factor (only in bivariate association) in Callaly et al. [
29], as well as being in receipt of welfare benefits in Priebe et al. [
14]. In Phibbs et al. [
24] service-connected disability turned out as a risk factor at multivariate level, while contrasting results emerged at bivariate level [
21,
26].
Finally, variables related to forensic and violence issues were analysed in three papers, but only in Wong and Chung [
48] violence history was associated with a decreased risk of readmission (only at bivariate level). Other variables related to
military service were analysed (years of active duty service, branch of service, military rank), but only a composite indicator - being either a “means-test C” (i.e., non-service connected disability and highest income) or a non-veteran - was found as a protective factor [
24].
Patients’ clinical characteristics
Diagnosis, defined as primary psychiatric diagnosis, was the main clinical characteristic of the patients analysed, but different grouping methods were adopted trough the papers. Results turned out to be not significant in 18 cases. Due to the large amount of information, only the main significant results reported in multivariate analysis are presented in the text. Having a psychotic disorder resulted in an increased risk to being readmitted in two papers [
52,
56], having a mood disorder or a substance abuse diagnosis in one [
52], and personality disorder in one paper [
54]. In Swartz et al. [
61], having psychosis compared with affective disorders resulted in a decreased risk of readmission only for one of the two sub-groups of patients being discharged to an outpatient commitment group. In Sanchez et al. [
55], having a secondary psychiatric diagnosis (the primary being a medical condition) was a protective factor compared with having bipolar disorder as the primary diagnosis. Among severe mental disorders, in Thompson et al. [
63] schizo-affective disorders increased the risk compared with other schizophrenic disorders.
When explicitly examined, the presence of a secondary diagnosis of substance abuse or dependence (or substance abuse complications) resulted in an increase of the risk of readmission in some multivariate analysis [
52,
55,
59], while decreasing risk in one study [
50]. Substance abuse patients with mental and behavioural disorders due to psychoactive substance use were more likely to be readmitted [
15,
21‐
24]. Moreover, in Phibbs et al. [
24], differences among type of substance of abuse emerged and in Kim et al. [
26], a major depressive disorder diagnosis (versus “other depression diagnosis”) and a tobacco use disorder were negatively associated with hospital readmission.
Finally, psychiatric comorbidity with other psychiatric diagnoses was also explicitly examined with non-homogeneous results. Number of psychiatric diagnoses was significant in one paper [
15]. Presence of a personality disorder when resulted significant increased the risk of readmission at multivariate level in [
9,
33]. A study by Stahler et al. [
32] found that having a chief complaint of depression decreased the risk of readmission among patients with dual diagnosis.
Physical comorbidity has been studied as possible predictor as well: results have been reported in another review of the CEPHOS-LINK project [
11].
In terms of
suicide, in Lyons [
7] suicide potential as a reason of admission decreased the risk of readmission at 1 year, but not at 30 days or at 6 months. In Kim et al. [
26] a history of suicide attempt increased the risk of readmission in one paper at bivariate analysis, but resulted not significant in other two papers [
48,
58]. In Monnelly [
16], when at least a sign of instability during hospitalisation was reported, the risk of readmission increased although suicide alone was found not significant. Finally, in Wong and Chung [
48] family history of suicide seemed to make this group of patients more vulnerable, indicating a relatively higher risk of readmission in bivariate analysis because of further mental deterioration provoked by this social stress.
Lower Global Assessment of Functioning (
GAF) [
64] scores resulted in an increase of the risk of readmission as measured at admission ([
25,
43] - at bivariate level; [
50] – at multivariate level) and in the previous 4 months before admission [
61]; and in one paper [
16] (at the bivariate level) when GAF was measured at discharge. When previous GAF was evaluated significance was found for the lowest value in the prior year (only in bivariate analysis) [
25]. A greater severity corresponded to a lower risk of readmission, but only when comparing readmission vs nursing home disposition, while no significant differences emerged between readmission to hospital and continuous stay in the community [
19]. Patient clinical status was also analysed through
other scales of functioning or psychopathology, together with
measures of cognitive status,
quality of life,
psychosocial problems or history of behavioural problems (e.g., aggression). At least one significant association with readmission was found in 12 papers (in four papers only at bivariate level [
9,
19,
58,
62]). Few studies used different versions of the Brief Psychiatric Rating Scale (BPRS) [
65]. When BPRS resulted significant, readmitted patients had a higher score on 24-item BPRS at discharge [
66], but direction of the significant association resulted reversed using a 23-item version of BPRS at admission at bivariate level [
59]. At multivariate level, higher scores in the anxiety index of The Symptom Checklist 90 Revised [
67] and in the Behavior and Symptom Identification Scale [
68] measured at hospital admission increased the risk of readmission [
46,
69].
In Lyons et al. [
7] using “the Severity of Psychiatric Illness scale” and “The Acuity of Psychiatric Illness scale”, the 30-day readmission risk increased for higher level of self-care impairment, 6-month readmission risk for higher clinical status scores at admission and higher level of severity of symptoms and 1-year readmission risk for self-care impairment, severity of symptoms and premorbid dysfunction level.
More psychosocial problems evaluated at discharge using DSM Axis IV [
64] were found associated to readmission, but only at bivariate level [
19], while one of their items (economic problems) turned out as a risk factor in multivariate analyses [
49]. Other different measures of functioning resulted significant in some papers at bivariate and multivariate level. In this latter case, activity of daily living dysfunction was found as a risk factor ([
50] and, for women with dementia, both at admission and at discharge [
18]).
One paper [
59] analysed the quality of life, finding a lower risk of psychiatric readmission for patients: with more social contacts and frequency of contacts with family (by telephone) and visits with family and with friends, with higher global life satisfaction reported both at admission (also at multivariate level) and at discharge, and with more satisfaction for each of the following subscales: living arrangements, family relations, social relations, leisure activities, personal safety, and finances.
Cognitive impairment resulted associated to readmission in patients who were hospitalised in a ward for dementia but only at bivariate analysis and in late readmission vs control or early readmission with differences between genders [
18], with late readmission more likely for women and less likely for men with higher cognitive function.
In a few papers different
proxies of severity as a subjective evaluation by staff members were analysed, resulting not significant in two papers [
45,
49]. In other studies, a poor versus fair or good prognosis increased the risk of readmission [
63] in multivariate analysis, as well as, at bivariate level, requiring extensive assistance [
40] and (considering early vs late readmission) having any active symptomatology and affective symptoms (across all diagnoses) or presence of psychotic symptoms at discharge (only among patients with schizophrenic/schizoaffective disorders) [
17].
Table
4 synthetizes the main results for this group of variables.
Table 4
Synthesis of the main results regarding patients’ clinical characteristics
Psychiatric Diagnosis | 28/46 | Mixed results and different diagnostic groups compared (20) | Mixed results and different diagnostic groups compared (17) |
Suicide attempt or gesture (history or risk during hospitalization) | 3/6 | A history of suicide attempt (1) and a family history of suicide (1) risk factors | Suicide potential protective factor (1) |
GAFb | 6/11 | Measured in different moments (4) | Measured in different moments (3) |
Subjective prognosis and risk score | 3/5 | Symptomatology at discharge (1) and patients required heavy care risk factor (1) | Poor prognosis risk factor (1) |
Finally, antipsychotic and substance use prescription fill in 6 months before the index hospitalization resulted associated with readmission [
52] as well as the number of medications filled during the year before but with a non-monotonic association [
26].
Patients’ clinical history
Admission history turned out to be significantly associated with readmission in 32 out of 37 studies, resulting in 31 cases as a risk factor. In 20 of these studies such relationship was found in all the multivariate analyses performed, while in one other case only in some of the different multivariate regressions performed; only in one case association was found at bivariate but not at multivariate level [
66]. In just one study and only in bivariate analyses [
14], a negative relationship was found between having been previously hospitalized and readmission risk.
Duration of illness was considered in four papers [
25,
37,
55,
57]. Two papers [
25,
57], found a significant association (with length of illness being a risk factor for readmission, only in bivariate analyses). In Wong and Chung [
48], a decrease in the risk of readmission was found for older
age at onset. A recent French study [
38] compared three groups: late and early onset geriatric patients and young adults. In this case, late onset turned out to be a risk factor (while the lowest risk was found for young adults). In Ng et al. [
66] an
index admission corresponding to the first onset of illness was found as a protective factor towards readmission within 6 months from discharge, but only in bivariate analysis (the authors suggesting that being at the first onset was associated with a lower risk of readmission due to compliance of medication), while in another study [
19] no significant association between first onset and readmission was found for older adults hospitalized for depression.
Number of hospital days in a given period before index admission was found associated to higher risk ([
25] and, only in bivariate analysis, [
26]) while, in Moos et al. [
21], it turned out as non-significant. The average length of hospital stay in previous admissions was also considered in one study, turning out to be non-significantly related to readmission [
48].
Several measures of
non-
hospital pre-
admission contacts with health services were analysed. Being known to the mental health service before index admission [
9], previous use of outpatient mental health services [
23,
26,
50,
52], and preadmission relationship with a mental health practitioner [
31] were found as predictors of readmission in multivariate analyses. Three papers considered outpatient medical visits before index admission [
21‐
23]; Moos et al. [
21,
23] found them to be a significant risk factor at multivariate level. Moos et al. [
22,
23] also analysed at multivariate level the effect of prior inpatient treatment for a medical condition: it was associated with an increased risk of readmission in both studies.
Table
5 synthetizes the main results for this group of variables.
Table 5
Synthesis of the main results regarding patients’ clinical history
Previous admissions | 32/37 | Previous admissions risk factor (23)b | Previous admissions risk factor (21) |
Duration of illness | 2/4 | Higher length of illness risk factor (2) | No significant results |
Age at onset | 2/6 | Mixed direction (2) | No significant results |
Whether index admission corresponded to first onset/episode | 1/2 | First onset protective factor (1) | No significant results |
Number of previous hospital days/average previous LoS | 2/4 | Number of previous hospital days risk factor (2) | Number of previous hospital days risk factor (1) |
Previous use of health services | 8/10 | Increasing risk with service use (3) | Increasing risk with service use (8) |
Contextual factors: environmental, social and hospital
Environmental factors such as hospital location and variables related to neighbourhood environment characteristics, health system factors and social context factors (family and caregivers relationships) were considered in this category.
A comparison between
urban (or metropolitan) and
rural (or non-urban) areas was performed in five papers. An urban setting was found as a risk-increasing factor in one study [
52], while a higher risk for rural areas was found in another study, where however only bivariate analysis was performed [
57]. Some papers analysed differences in readmission risk related to
hospital or discharge location, but are referred to specific national situations; in particular, Kim et al. [
26] and Adams [
60] compared US regions and Lin et al. [
34] Taiwanese regions.
Stahler et al. [
32], considered many variables related to
neighbourhood environment characteristics and services distances, finding a higher risk of readmission for patients who lived in close proximity to a Narcotics Anonymous meeting location and a lower one for patients living in areas with higher educational attainment.
Unavailability of resources, measured in terms of either absence of services and resources required by the patient in the geographic area to which the patient had access, or a waiting list making them non-usable, was also measured but resulted as not significant [
40].
Physician gender and experience (using age as a proxy) were examined with bivariate analysis, gender turning out to be non-significant and experience being protective [
34]. The same study analysed also
other hospital-
level variables and found that being discharged from medical centres or not-for-profit hospitals was a protective factor, while patients discharged from regional and public hospitals had the highest readmission rates. In Mark et al. [
52], lower median length of stay and higher annual mean number of stays for Medicaid patients with mental or substance use disorder (M/SUD) or some psychiatric/psychological procedures (interviews, consultations and evaluations; somatotherapy, individual psychotherapy) turned out as risk factors and other psychiatric/psychological procedures (other psychotherapy and counselling, alcohol and drug rehabilitation and detoxification) as protective factors, with the annual mean number of stays of patients with M/SUD diagnosis and the median LoS being significant also in the multivariate analyses.
We also considered two economic issues partially related to the health system characteristics, but analysed at individual level. As for papers related to
payment/
reimbursement mechanisms and
insurance, Medicaid was found as a protective factor (vs commercial insurance) in Kolbasovsky [
45] while mixed results emerged in bivariate analysis [
52,
56].
Among variables related to
social support, at multivariate level, insufficient emotional and practical support of caregivers increased the risk of being readmitted [
40], as well as did maladaptive family system functioning [
20] and social support unreliability [
50]. Also, for women with dementia, having caregivers who felt burdened by care responsibilities increased risk of late readmission versus no readmission [
18].
At bivariate level, criticism of family member and caregiver’s over-estimation of their own ability to provide assistance and emotional support, more family involvement, attendance of a carer at the discharge planning, perceived treatment support reported significant results [
7,
25,
35,
37,
40,
57]. On the contrary, presence and extent of social support network, pre-discharge contacts with family or non-government psychosocial support organisations, change in the support system preceding hospitalisation and family conflict resulted non-significant.
Table
7 synthetizes the main results.
Table 7
Synthesis of the main results regarding contextual factors: environmental, social and hospital
Urban/metropolitan vs ruralb | 2/5 | Mixed direction (2) | Urban residence risk factor (1) |
Environmental variables, services distance and availability of resources | 1/2 | No significant association | Living in close proximity to a Narcotics Anonymous meeting location risk factor while living in areas with higher educational attainment protective factor (1) |
Physician characteristics and other hospital-level variables | 2/2 | Different variables analysed and found significant (2) | Number of Medicaid patients with mental or substance use disorder (1) and shorter median LoS (1) risk factor |
Fee-for-service or capitated Medicaid plan or (type of) insurance coverage | 3/4 | Mixed direction (2) | Medicaid (vs commercial insurance) protective factor (1) |
Social support | 9/14 | Social support protective factor (6)c (different variables analysed) | Social support protective factor (4) (different variables analysed) |
Admission and discharge characteristics
Length of stay was examined in many studies. In Ono et al. [
18], higher values of LoS turned out to be a risk factor for early readmission (in the first 3 months), but a protective factor towards late readmission (from the 4th to the 24th month), such results being confirmed also in multivariate analyses. In four studies [
26,
28,
46,
53], only at bivariate level, a longer LoS resulted as a risk factor of being readmitted. In four studies a longer LoS turned out as a protective factor in multivariate analysis: toward readmission at 28–30 days for patients with different psychiatric diagnoses [
34,
54], and at 4–5 years for patients respectively with substance use disorders or schizophrenia [
21,
33]. Moreover, a longer LoS turned out as a protective factor also in three papers only performing bivariate analysis [
36,
44,
57].
The
legal status of the index admission was considered among the potential predictors in nine papers, with a higher risk for voluntarily admitted patients found in Hendryx et al. [
49] (vs court-order admitted patients) and (but only in bivariate analysis) Russo et al. [
59]. In this last study, readmission rates for chronic patients assigned to a locked unit decreased.
As for
type of discharge, escape from hospital or discharge against medical advice increased risk of readmission in two papers [
30,
57] and in one study the 90-day readmission risk increased for discharge referral to other centres due to remission versus discharged on medical advice, but not for discharged against medical advice [
55]. Adequacy of discharge planning (as evaluated by a social worker) turned out instead as a protective factor [
40] in multivariate analyses, as well as having a discharge plan sent to the GP on discharge from the index admission [
9,
29].
Discharge destination - planned during admission - in terms of accommodation (e.g., community centres, home) resulted non-associated to readmission, apart from being followed by the social welfare services which increased risk of readmission compared with referral to relatives [
48] as well as (in bivariate analysis) having an assigned service in community [
33]. Moreover, one paper reported a decreased risk for patients assigned to an outpatient (vs control) commitment group, both alone and in interaction with psychotic diagnosis [
61].
Complications during hospitalisation for patients suffering from dementia resulted as not significant in multivariate analysis but increased risk for early readmission when looking at bivariate associations [
18]. In Monnelly [
16], when at least a sign of instability during hospitalisation (i.e., use of restraints, use of seclusion, requiring orders for close observation in the 3 days before discharge, active psychotic behaviour, suicide attempts or gesture, assault within 5 days of discharge, receiving p.r.n. medications - not including hypnotics - or not) was reported, the risk of readmission increased (as well as for each sign separately, apart from those occurring most rarely, i.e., suicide and assault), while in [
54] admissions involving reportable aggressive incidents were found as more likely to lead to readmission (in both papers, also in multivariate analyses).
Six papers [
17,
28,
35,
39,
48,
66] analysed different aspects of
pharmacological treatment (such as dosage or medication prescription), but it resulted significant in only the three of them discussed below and always in bivariate associations. Being on depot injectable antipsychotic medication turned out as a risk factor, while using atypical antipsychotic medication was protective towards readmission [
66]. Prescription of atypical antipsychotic was again found as a protective factor (while no significance was found for depot) [
35]. Receiving mood stabilizers was found as a protective factor for patients with bipolar disorder, while receiving antipsychotic medications for those with depressive psychosis [
17].
Intensive case management (ICM) was found as protective versus control group in multivariate analysis [
45]. Other three papers analysed
other interventions during hospitalisation finding significant results only at bivariate level: in one receiving ECT during the hospital stay reduced the risk of early readmission [
19] - this variable resulted not significant in another paper [
66] -, an intervention of advanced directives (a statement of a person’s preferences for treatment during admission) versus control group did not result in statistically different readmission risks [
42].
Table
8 synthetizes the main results.
Table 8
Synthesis of the main results regarding admission and discharge characteristics
Length of stay | 13/33 | Mixed direction (11) | Protective factor (4), mixed results (1) |
Involuntary admission | 2/9 | Involuntary admission protective factor (2) | Involuntary admission protective factor (1) |
Type of discharge | 6/10 | Discharge plan sent to GP (2), located (1), rated as adequate (1) and discharge on medical advice (2) protective factors | Discharge plan sent to GP (2), rated as adequate (1), discharge on medical advice (2) protective factors |
Referral made at discharge/discharge destination | 3/6 | Being followed by social welfare services (1), having an assigned service in community (1) risk factors | Being followed by social welfare services (1) risk factor; patients assigned to an outpatient (vs control) commitment group protective factor (1) |
Complications during hospitalization | 3/3 | Complications during hospitalization risk factor (3) | Complications during hospitalization risk factor (2) |
Treatment and clinical practice | 5/9 | Atypical antipsychotic (2), receiving mood stabilizers at discharge (1), antipsychotic medications (1), ECT in the hospital stay (1) protective factors; on depot injectable antipsychotic (1) risk factor | Intensive case management services protective factor (1) |