Advertisement

Factors Associated With Discharge Destination in Older Patients: Finnish Community Hospital Cohort Study

Open AccessPublished:August 09, 2022DOI:https://doi.org/10.1016/j.jamda.2022.07.004

      Abstract

      Objectives

      Primary care physician-led community hospitals provide basic hospital care for older people in Finland. Yet little is known of the outcomes of the care. We investigated factors associated with discharge destination after hospitalization in a community hospital and the role of active rehabilitation during the stay.

      Design

      Prospective observational study.

      Setting and Participants

      Short-term community hospital stays of older adults (≥65 years) living in the Kuopio University Hospital district in central and eastern Finland.

      Methods

      Data on short-term (1-31 days) hospital stays from 51 community hospitals were collected with an electronic survey between January and June 2016. Physicians, secretaries, and rehabilitation staff from each community hospital completed the data collection form. Discharge destination was defined as home, residential care or death, and active rehabilitation as frequency of rehabilitation at least once a day. Analyses were conducted using the Bayesian approach and the BayesiaLab 9.1 tool.

      Results

      Data of 11,628 community hospital stays were analyzed. The patients' mean age was 81.6 years (SD 7.9), and 57.5% were women. A younger age (65-74 years), a high number of rehabilitation staff (>2 per 10 patients), and receiving rehabilitation at least once a day were associated with discharging patients to their own homes. Daily rehabilitation was associated with returning to home in all patient groups.

      Conclusions and Implications

      Older patients admitted to a community hospital for any reason may benefit from active rehabilitation. The role of community hospitals in the acute care and rehabilitation of older patients is important in aging societies.

      Keywords

      The current population of Finland is 5.55 million. There are 5 university hospital districts, 22 specialized care hospital districts, and about 200 general practitioner–run local community hospitals. Local hospitals are part of the primary health care system and use shared premises with health centers. In 2016, half of inpatient days in Finnish somatic hospitals were provided by community hospitals.
      Statistical information on welfare and health in Finland.
      The Finnish hospital system is publicly financed.
      • Keskimäki I.
      • Tynkkynen L.K.
      • Reissel E.
      • et al.
      Finland: Health system review.
      Community hospitals are designated to care for older people who usually have multiple illnesses and needs. Acute diseases are the main reason for care, but there are also other tasks such as treatment of chronic conditions, treatment after hospitalization in specialized health care, rehabilitation, and end-of-life care.
      • Keskimäki I.
      • Tynkkynen L.K.
      • Reissel E.
      • et al.
      Finland: Health system review.
      ,
      • Saari H.
      • Lönnroos E.
      • Mäntyselkä P.
      • et al.
      What are the characteristics of short-term inpatient care in Finnish primary care hospital? Article in Finnish.
      Local hospitals serve as consultation units for home care and play an important role in supporting older people living at home.
      Regardless of the reason for admission, hospitalization represents a risk for functional decline in older people.
      • Covinsky K.E.
      • Pierluissi E.
      • Johnston C.B.
      Hospitalization-associated disability: “She was probably able to ambulate, but I’m not sure.”.
      • Chodos A.H.
      • Kushel M.B.
      • Greysen S.R.
      • et al.
      Hospitalization-associated disability in adults admitted to a safety-net hospital.
      • Hirsch C.H.
      • Sommers L.
      • Olsen A.
      • et al.
      The natural history of functional morbidity in hospitalized older patients.
      Early rehabilitation for hospitalized older people may improve functioning without additional costs.
      • Martinez-Velilla N.
      • Cadore E.L.
      • Casas-Herrero A.
      • et al.
      Physical activity and early rehabilitation in hospitalized elderly medical patients: systematic review of randomized clinical trials.
      ,
      • Barnes D.E.
      • Palmer R.M.
      • Kresevic D.M.
      • et al.
      Acute care for elders units produced shorter hospital stays at lower cost while maintaining patients' functional status.
      In Finland, there are national guidelines and recommendations for the management of specific health conditions such as stroke or hip fracture, but rehabilitation recommendations for several other conditions of older people are incomplete or missing.
      Finnish Medical Society Duodecim
      Current care guidelines.
      Community hospitals seek to provide basic hospital care for older people and enhance their functional recovery and return to their own homes. However, as far as we know, there is little research on factors associated with discharge destination or outcome of rehabilitation in local community hospitals or comparable primary care physician–led wards.
      • Young J.
      • Green J.
      • Forster A.
      • et al.
      Postacute care for older people in community hospitals: a multicenter randomized, controlled trial.
      ,
      • Green J.
      • Young J.
      • Forster A.
      • et al.
      Effects of locality based community hospital care on independence in older people needing rehabilitation: randomised controlled trial.
      The aim of this study was to analyze which factors were associated with discharge destinations after short-term (up to 1 month) inpatient care in local community hospitals. In addition, we estimated the specific impact of rehabilitation on the discharge of the patients. The study was conducted in real-life settings in central and eastern Finland and using the Bayesian approach.

      Methods

      Setting

      This study was conducted in the catchment area of Kuopio University Hospital (KUH) in central and eastern Finland. The population of the KUH area was 813,000 and represented 15% of the total population of Finland. There were 55 community hospitals in the KUH area in 2016, and all of them were invited to participate. The principal investigator contacted the chief physicians of the community hospitals, and written information about the study was sent to the community hospitals. A total of 51 community hospitals run by 65 municipalities participated in the study. The data collection was carried out between January and June 2016, the collection time was 2-4 months (mean 3.2 SD 1.0) per hospital unit. The study was approved by the Ethics Committee of the Hospital District of the Northern Savo (approval number 340/2015).

      Patient Selection

      Community hospital stays of patients aged ≥65 years were included in the study. Patient selection was restricted to short-term care. Patients with community hospital stays lasting 1-31 days were included.

      Data Collection

      We developed a structured and standardized electronic survey for the data collection. Most of the items in the survey were similar to the records routinely collected for the Finnish Care Register (HILMO).,
      • Kajantie M.
      • Manderbacka K.
      • McCallum A.
      • et al.
      How to Carry Out Register-Based Health Services Research in Finland? Compiling Complex Study Data in the REDD Project.
      Furthermore, the primary reason for hospital stay, content of the care, and whether a patient received rehabilitation were inquired. Physicians responsible for treatment, ward secretaries, and rehabilitation professionals were responsible for data collection. Most respondents were visited, and all received written instructions for how to complete the survey. The survey was completed at the end of each patient’s hospital stay and included no personal identifiers.
      The survey consisted of 25 questions on the following items: demographics (age, sex, home municipality), community hospital identifier, where the patient came from to the community hospital (secondary or tertiary hospital, residential care, home with or without home care), the primary reason for the hospital stay (acute diseases, chronic diseases, assessment of symptoms, diagnostic investigations), underlying diagnoses, content or goal of the care (management of multimorbidity, continuing care, care of medical complication, starting a new treatment, diabetes management, treatment of pain, wound care, psychiatric care, substance abuse treatment, treatment of poisoning, terminal care, scheduled interval care) and discharge. Home care refers to the status of being a long-term recipient of services at home as defined by legislation. Finnish home care services include support and assistance in activities of daily living, home nursing, physician and hospital-at-home services, rehabilitation, and end-of-life care.
      • Auvinen K.J.
      • Räisänen J.
      • Voutilainen A.
      • et al.
      Interprofessional medication assessment has effects on the quality of medication among home care patients: randomized controlled intervention study.

      Outcome and Intervention Variables

      Discharge destination was used as an outcome variable and it was defined as follows: discharged home, discharged to a residential care, or died during the hospital stay. Residential care included public and private nursing and care homes providing accommodation and round-the-clock care services for older adults.
      Rehabilitation served as an intervention variable in Bayesian analyses. We asked whether patients received rehabilitation according to a specific plan during their community hospital stay and how frequent the rehabilitation sessions were. Rehabilitation performed at least once a day was defined as active rehabilitation.

      Statistical Analysis

      We utilized the Bayesian approach to analyze the data. The advantages of Bayesian analysis (BA), especially in complicated data, have been widely discussed.
      • Kekolahti P.
      Bayesian Network Analysis of Mobile Service and Device Usage. Aalto University Publication Series Doctoral Dissertations, 201/2019, Department of Communications and Networking.
      • Darwiche A.
      Bayesian networks.
      • Rautiainen E.
      • Ryynänen O.P.
      • Laatikainen T.
      • et al.
      Factors associated with 5-year costs of care among a cohort of alcohol use disorder patients: a bayesian network model.
      An important feature of BA is the possibility to make assumptions in causal inference from observational data.
      • Pearl J.
      • Mackenzie D.
      The Book of Why: The New Science of Cause and Effect.
      ,
      • Cox Jr., L.A.
      Modernizing the Bradford Hill criteria for assessing causal relationships in observational data.
      The visual form of a Bayesian network (BN) uses a Directed Acyclic Graph (DAG), which consists of nodes representing random variables, and arcs between the nodes representing the existence of a statistical dependency between 2 nodes. A conditional probability table (CPT) is attached for each node to describe the size of this statistical dependency (a local conditional dependency) between 2 nodes. Arcs between nodes are directed from a starting node (called a parent) and an ending node (called a child). Arc direction can represent both noncausal (predictive or explanative) or causal modeling. A DAG allows testing and identifying causal and noncausal relationships among the variables involved in the study.
      • Pearl J.
      • Mackenzie D.
      The Book of Why: The New Science of Cause and Effect.
      The analyst can control for a certain variable or a combination of variables when testing the causal effect of a variable of interest on the target variable.
      • Conrady S.
      • Jouffe L.
      Bayesian Networks and BayesiaLab – A Practical Introduction for Researchers.
      A DAG is constructed either manually or with machine learning based on observational data. The analyst can set restrictions for the structural learning algorithm to avoid illogical connections or to add theoretically justified connections. In this study, we used the expert-assisted machine learning by adding an arc we considered to be hypothetically relevant.
      • Darwiche A.
      Bayesian networks.
      ,
      • Ryynänen O.P.
      • Leppänen T.
      • Kekolahti P.
      • et al.
      Bayesian network model to evaluate the effectiveness of continuous positive airway pressure treatment of sleep apnea.
      We analyzed the data using the BayesiaLab 9.1 tool (Bayesialab). The analysis was done with a data set of 11,628 individuals. In total, 2436 (0.4% of the data set) numerical values of the data set were missing, being the type missing at random. We performed missing value imputations by predicting missing values using a structural equation model (EM) algorithm.
      • Myers J.W.
      • Laskey K.B.
      • DeJong K.A.
      Learning bayesian networks from in complete data using evolutionary algorithms.
      The variable Age indicating the patient’s age, was discreted into 3 groups with 10-year intervals. The variable Rehabilitation staff/10 pts was discreted as follows: (1) number of staff per 10 patients ≤ 0.7, (2) number of staff per 10 patients 0.8-2.0, (3) number of staff per 10 patients over 2.0, (4) number of staff per 10 patients not defined (indicating that rehabilitation staff was only partially available in hospital units). All other variables in the final data set were discrete.
      The outcome variable Outcome discharge was set as a target node (a dependent variable). We used the variable Active rehabilitation as an intervention variable. The original data set consisted of 45 independent variables.
      We used the TabooEQ (Equivalence Classes) unsupervised learning algorithm to construct a noncausal Bayesian network to find the relevant dependences between all the 48 variables (46 independent variables, the intervention variable, and the outcome). The TabooEQ algorithm did not recognize an arc Active rehabilitation → Outcome discharge. Because, according to our hypothesis, we wanted to measure the causal effect of the intervention (Active rehabilitation) on the outcome (Outcome discharge), we manually added an arc between those variables.
      • Ryynänen O.P.
      • Leppänen T.
      • Kekolahti P.
      • et al.
      Bayesian network model to evaluate the effectiveness of continuous positive airway pressure treatment of sleep apnea.
      We dropped out all variables not connected with the outcome or the intervention variable, as well as other variables with only a marginal connection with the outcome or the intervention variable.
      The result was a BN consisting of 10 variables, including the intervention variable and the outcome. Arcs between variables were noncausal, except the manually added arc Active rehabilitation →Outcome discharge. We analyzed the robustness of the arcs by using the Jackknife resampling method in BayesiaLab. We only used arcs that were present in all networks obtained from the Jackknife resampling.
      We calculated the mutual information (MI) between each variable and the outcome variable, as well as maximum and minimum Bayes factors (BF) in order to clarify the strength of the dependency and its confidence level for each variable state. The MI between 2 variables (X and Y) shows how much the knowing of variable Y reduces the uncertainty about variable X. MI is a measure that is not dependent on the variables’ linearity.
      • Korb K.B.
      • Nicholson A.E.
      Bayesian Artificial Intelligence.
      ,
      • Yumoto F.
      • Wood R.L.
      • Tractenberg R.E.
      Machine learning to evaluate the quality of patient reported epidemiological data. Survey Research Methods Section.
      BF is a Bayesian approach for hypothesis testing. A high BF value indicates a stronger association, and a lower BF indicates uncertainty. Kass and Raftery (1995) give the following interpretation for BF: log10K = 0 to 0.5 not worth more than a bare mention, 0.5 to 1 substantial, 1 to 2 strong, and >2 decisive.
      • Kass R.E.
      • Raftery A.E.
      Bayes factors.
      BayesiaLab enables controlling for (ie, to fix the distribution) any variable and their combinations, for example, fixing the variable Terminal care of the state to “yes” = 100% and state “no” = 0%. This means that the model changes; accordingly, the state distribution of the outcome variable in the previous hypothetical case that all patients were in the state of Terminal care is true.
      To identify a causal connection between the intervention variable and the outcome, technically 2 alternatives exist. First, the intervention variable can be isolated from unwanted associational backdoor paths between the intervention variable and the outcome. This method requires full knowledge about causal associations of the BN.
      • Pearl J.
      • Mackenzie D.
      The Book of Why: The New Science of Cause and Effect.
      ,
      • Elwert F.
      • Winship C.
      Endogenous selection bias: The problem of conditioning on a collider variable.
      In this study, the true causal network between the variables and the outcome variable remains unknown. Therefore, we applied VanderWeele’s and Shiptser’s modified disjunctive confounder criteria (DCC) for estimating a direct causal effect of an intervention variable on the outcome variable from an otherwise noncausal Bayesian network.
      • VanderWeele T.J.
      • Shpitser I.
      A new criterion for confounder selection.
      VanderWeele and Shiptser defined the original DCC as controlling for each variable that is a cause of the treatment, or of the target, or both.
      • VanderWeele T.J.
      • Shpitser I.
      A new criterion for confounder selection.
      The variables to be controlled must have been present before intervention. No unmeasured confounder is allowed to have a direct effect on the intervention or on the outcome. VanderWeele added 2 additional qualifications to DCC for practical use for confounder controlling and renamed it as “modified DCC.”
      • VanderWeele T.J.
      Principles of confounder selection.
      Additional definitions are (1) excluding from this set any variable known to be an instrumental variable and (2) including variables that do not satisfy the criterion but are good proxies for unidentified common reasons for treatment.
      • Pearl J.
      • Mackenzie D.
      The Book of Why: The New Science of Cause and Effect.
      ,
      • Elwert F.
      • Winship C.
      Endogenous selection bias: The problem of conditioning on a collider variable.
      • VanderWeele T.J.
      • Shpitser I.
      A new criterion for confounder selection.
      • VanderWeele T.J.
      Principles of confounder selection.

      Results

      A total of 11,628 community hospital stays were analyzed. The patient characteristics and variables used in Bayesian analyses are shown in Table 1. The mean age of the patients was 81.6 (SD 7.9) years and 57.5% were women. The majority of patients came to a community hospital from home (n = 6894; 59.5%); nearly a third came from a secondary or tertiary care hospital (n = 3348, 28.9%), and 9.5% (n = 1100) from a residential care facility. The median of length of community hospital stay was 5 days (interquartile range 2-8). Regarding the first underlying diagnoses, the 4 most common categories were cardiovascular diseases, 18.4% (n = 2143); respiratory system diseases, 15.7% (n = 1830); symptoms, 8.9% (n = 1031); and injuries, 8.5% (n = 993).
      Table 1Explanations and Distributions of the Variables of Short-Term Community Hospital Stays Analyzed in BayesLab
      VariableVariable Distribution, n (%)
      Age, y (BN)65-104 (min-max)

      81.6 ± 7.9 (mean ± SD)

      65-74 = 2360 (20.4%)

      75-84 = 4578 (39.5%)

      ≥85 = 4657 (40.2%)
      SexFemale = 6578 (57.5%)

      Male = 4869 (42.5%)
      Where coming from (BN)Referral hospital (secondary or tertiary care) = 3348 (28.9%)

      Another health center = 243 (2.1%)

      Home, with home care = 3684 (31.8%)

      Home, no home care = 3210 (27.7%)

      Residential facility = 1100 (9.5%)
      Length of stay, d≤5 = 6613 (56.9%)

      6-10 = 2819 (24.2%)

      11-15 = 1168 (10.0%)

      16-31 = 1017 (8.7%)
      Community with referral hospital (BN)

      Secondary or tertiary care hospital which is in same locality as the community hospital
      No = 8475 (72.9%)

      Yes = 3153 (27.1%)
      Type of communityRural = 3457 (29.7%)

      Urban = 8171 (70.3%)
      Rehabilitation resources/10 patients (BN)≤0.7 = 7048 (60.6%)

      >0.7-2 = 2183 (18.9%)

      >2 = 309 (2.7%)

      Not defined 2087 (18.0%)
      Did the patient receive rehabilitation?No = 9452 (81.3%)

      Yes = 2160 (18.6%)
      Primary reason for care
       New diagnosis (BN)

      Confirmed during the hospital stay
      No = 6721 (57.8%)Yes = 4907 (42.2%)
       New diagnosis not confirmed

      New symptoms/nonspecified condition
      No = 10,892 (93.9%)
      Yes = 703 (6.1%)
       Chronic disease (BN)No = 9196 (79.1%)

      Yes = 2432 (20.9%)
       Chronic disease and need for rehabilitationNo = 11,419 (98.5%)

      Yes = 176 (1.5%)
       Assessment and investigations of long-lasting symptomsNo = 10,664 (91.7%)

      Yes = 964 (8.3%)
       Due to investigationNo = 11,353 (97.9%)

      Yes = 242 (2.1%)
       Continuing care after stay in secondary or tertiary care hospital; treatment in progressNo = 9496 (20.0%)

      Yes = 4063 (18.0%)
       Continuing care after stay in secondary or tertiary care hospital; need for rehabilitationNo = 10,783 (93.0%)

      Yes = 812 (7.0%)
      Diagnostic VariablesICD-10 Diagnoses
      Other infection (BN)A00-B99 Certain infectious and parasitic diseasesNo = 10,928 (94.0%)

      Yes = 700 (6.0%)
      Neoplasm (BN)C00-D48 NeoplasmsNo 0 = 11,037 (94.9%)

      Yes 1 = 591 (5.1%)
      Dementia (BN)F00-F09 Organic, including symptomatic, mental disorders

      G30-G32 Other degenerative diseases of the nervous system
      No = 10,956 (94.2%)

      Yes = 672 (5.8%)
      Alcohol abuse (BN)F 10 Mental and behavioral disorders due to use of alcoholNo = 11,490 (98.8%)

      Yes = 138 (1.2%)
      Heart failure (BN)I 50 Heart failureNo = 10,958 (94.2%)

      Yes = 670 (5.8%)
      Other circulatory system disease (BN)I00-I99 Diseases of the circulatory system, excluding I 50 Heart failureNo = 10,155 (87.3%)

      Yes = 1473 (12.7%)
      Pneumonia (BN)J10.0; J11.0; J12-J18; J20-J22 Pneumonia and other acute lower respiratory infectionsNo = 10,308 (88.6%)

      Yes = 1320 (11.4%)
      Digestive system disease (BN)K00-K93 Diseases of the digestive systemNo = 11,094 (95.4%)

      Yes = 534 (4.6%)
      Musculoskeletal system disease (BN)M00-M99 Diseases of musculoskeletal system or connective tissueNo = 10,909 (93.8%)

      Yes = 719 (6.2%)
      Urinary tract infection (BN)N10-N12; N30; N34; N39 Infectious renal tubulo-interstitial disease or other infectious urinary system diseaseNo = 10,981 (94.4%)

      Yes = 647 (5.6%)
      Symptomatic (BN)R00-R99 Symptoms, signs or abnormal clinical or laboratory findings, not elsewhere classifiedNo = 10,597 (91.1%)

      Yes = 1031 (8.9%)
      Injury (BN)S00-T98 Injury, poisoning and certain other consequences of external causesNo = 10,635 (91.5%)

      Yes = 993 (8.5%)
      Other (BN)A disease not classified elsewhereNo = 9488 (81.6%)

      Yes = 2140 (18.4%)
      Primary Content of the Care
      Treatment of acute infection or traumaNo = 7017 (70.5%)

      Yes = 4578 (39.5%)
      Treatment after stay in secondary or tertiary care hospitalNo = 10,352 (89.0%)

      Yes = 1276 (11.0%)
      Terminal care (BN)No = 11,310 (97.3%)

      Yes = 318 (2.7%)
      Scheduled interval careNo = 11,530 (96.8%)

      Yes = 374 (3.2%)
      Treatment of painNo = 11,254 (96.2%)

      Yes = 446 (3.8%)
      Treatment of mental confusionNo = 11,336 (97.8%)

      Yes = 259 (2.2%)
      Treatment of psychiatric problemsNo = 11,535 (99.2%)

      Yes = 93 (0.2%)
      Preplanned treatmentNo = 11,449 (98.5%)

      Yes = 179 (1.5%)
      Management of multimorbidityNo = 11,025 (94.8%)

      Yes = 603 (5.2%)
      Treatment of poisoningNo = 11,614 (99.9%)

      Yes = 14 (0.1%)
      Detoxification and substance abuse treatmentNo = 11,535 (99.2%)

      Yes = 93 (0.8%)
      Management and care of diabetesNo = 11,595 (99.7%)

      Yes = 33 (3.8%)
      Starting a new treatmentNo = 11,449 (98.5%)

      Yes = 179 (1.5%)
      Treatment or care complicationsNo = 11,576 (99.6%)

      Yes 0 52 (0.4%)
      RehabilitationNo = 11,122 (95.6%)

      Yes = 506 (4.4%)
      Wound careNo = 11,542 (99.3%)

      Yes = 86 (0.7%)
      Variables in Bayes Network Model are marked with the abbreviation BN.
      In the final model, there were 8 independent variables with an association relevant to the outcome Discharge destination or intervention variable Active rehabilitation (Figure 1). The overall prediction performance of the BN was 80.1%, ROC index = 76.9%, and R2 = 0.34.
      Figure thumbnail gr1
      Fig. 1Bayesian Network Model. Variables associated with the outcome variable Outcome discharge or intervention variable Active rehabilitation. Variables are presented with a conditional probability table (CPT).
      The MI between the states of the variables and the outcome are presented in Table 2. The strongest associations were seen between variables Diagnosis → Where coming from (overall contribution 19.1%) and Where coming from → Outcome discharge (overall contribution 9.1%). Outcome discharge = Home was most common among patients who came to a community hospital from their homes and did not receive regular home care services (MI = 0.196). Having no terminal care (MI = 0.03), being younger (65-74 years) (MI = 0.002), and receiving active rehabilitation (MI = 0.001) showed a strong association with discharge to home. The BF was >1 in all associations indicating strong association.
      Table 2Local Analyses of Mutual Information Between Target Variable and Independent Variables
      NodeMutual InformationMax Bayes FactorMin Bayes Factor
      Outcome discharge = Home (69.9%)
       Where coming from0.1959home, no home care1.2459residential facility0.0283
       Terminal care0.0306no1.0236yes0.1619
       Age group0.001865-74 y1.036485 y or older0.9599
       Active rehabilitation0.0010yes1.0713no0.9916
       Chronic disease0.0003yes1.0278no0.9926
       New diagnosis0.0003no1.0111yes0.9848
       Diagnosis0.0002dementia1.0171pneumonia0.9885
       Rehabilitation staff/10 patients0.0000>2 (3/4)1.0238not defined (4/4)0.9970
       Community with referral hospital0.0000yes1.0019no0.9993
      Outcome discharge = Residential facility (25.2%)
       Where coming from0.1673residential facility3.5919home, no home care0.4351
       Terminal care0.0037no1.0186yes0.3375
       Age group0.001985 y or older1.107365-74 y0.9008
       Chronic disease0.0015no1.0394yes0.8511
       New diagnosis0.0014yes1.0880no0.9357
       Diagnosis0.0005urinary tract infection1.0658dementia0.9158
       Active rehabilitation0.0000no1.0026yes0.9776
       Rehabilitation staff/10 patients0.0000≤2 (2/4)1.0019>2 (3/4)0.9645
       Community with referral hospital0.0000yes1.0073yes0.9973
      Outcome discharge = Died (4.9 %)
       Terminal care0.0840yes16.3284no0.5690
       Where coming from0.0106another health center1.6355home, no home care0.3926
       Active rehabilitation0.0055no1.1063yes0.0995
       New diagnosis0.0015no1.1709yes0.7659
       Chronic disease0.0012yes1.3672no0.9028
       Diagnosis0.0007dementia1.1872urinary tract infection0.8262
       Rehabilitation staff/10 patients0.0001not defined1.0416>2 (3/4)0.8434
       Community with referral hospital0.0001no1.0242yes0.9350
       Age group0.000085 y or older1.021075-84 y0.9835
      The results are presented separately for all 3 states of the target. The column Max Bayes Factor presents the states with the highest Bayes Factor (BF) values. Correspondingly, the column Min Bayes Factor presents states with the lowest BF values. Only factors having Bayes factor >1 are presented.
      As expected, coming from a residential facility was most strongly associated with Outcome discharge = Residential facility (MI = 0.167, BF = 3.59). In addition, discharging to a residential facility was associated with no terminal care (MI = 0.004), age ≥85 years (MI = 0.002), urinary tract infection (MI = 0.0005), and not receiving active rehabilitation (MI = 0.000). The BF was >1 in all these variables.
      Outcome discharge = Died was self-evidently associated with terminal care (MI = 0.84, BF = 16.3). Other variable values associated with Outcome discharge = Died were the following: transfer from another primary care hospital (MI = 0.01), not receiving active rehabilitation (MI = 0.001), and presence of chronic disease (MI = 0.001).
      The number of rehabilitation staff was associated with discharge destination (Table 2). A high number of rehabilitation staff (>2 per 10 patients) was associated with Outcome discharge = Home. Conversely, a lower or unspecified number of rehabilitation staff were associated with Outcome discharge = Residential facility or Death. The number of rehabilitation staff showed MI less than 0.001 with all the outcome variable values. However, the BF >1 indicated a strong association.
      Receiving active rehabilitation increased the rate of patients being discharged to their own homes and reduced the rate of deaths in all patient groups (Table 3). Patients whose hospital stay was caused by urinary tract infection, heart failure, or dementia benefited the most from active rehabilitation.
      Table 3States of the Outcome Variable (Outcome Discharge) When the Variable Active Rehabilitation Is Unfixed and Independent Variable States Are Fixed to the Alternative Yes = 100%
      No Fixation of Active rehabilitationFixation of Intervention Variable Active Rehabilitation yes = 100%
      HomeResidential CareDiedHomeResidential CareDied
      No fixation of independent variables69.925.24.975.024.50.5
      Fixations of independent variables
       Terminal care = yes12.79.677.7nonenonenone
       Where coming from: referral hospital66.125.98.073.925.50.6
       Where coming from: other health center58.633.38.166.733.30.0
       Where coming from: residential facility11.682.06.414.785.30.0
       Where coming from: home, with home care79.616.83.681.517.90.6
       Where coming from: home, no home care87.111.01.987.912.10.0
       Dg heart failure70.822.66.677.221.41.4
       Dg pneumonia60.334.25.563.536.10.4
       Dg alcohol abuse76.519.54.0nonenonenone
       Dg dementia70.225.74.178.920.60.5
       Dg neoplasm71.422.66.074.924.40.7
       Dg digestive system disease69.424.75.976.023.50.5
       Dg urinary tract infection63.931.24.973.925.60.5
       Dg symptomatic73.322.44.376.922.70.4
       Dg other infection70.524.74.872.527.00.5
       Dg injury70.225.24.674.325.20.5
       Dg musculoskeletal system disease74.621.73.775.224.30.5
       Dg other circulatory system disease70.024.45.674.724.70.6
       Dg other73.022.34.777.522.00.5
       Rehabilitation staff/10 patients ≤0.769.725.25.176.324.20.5
       Rehabilitation staff/10 patients ≤2.070.325.24.574.525.00.5
       Rehabilitation staff/10 patients >2.072.124.33.675.623.90.5
       Rehabilitation staff not defined69.925.24.974.924.60.5
       Community with referral hospital: yes70.225.24.674.425.20.4
       Community with referral hospital: no69.825.15.175.525.00.5
       Age group: 65-74 y72.522.74.876.023.50.5
       Age group: 75-84 y71.523.74.875.723.80.5
       Age group: 85 y or older67.127.95.073.925.60.5
       Chronic disease: yes73.922.23.977.721.90.4
      Dg, Diagnosis.
      Three columns in the right show the same independent variables’ fixations when the Active rehabilitation is fixed to value yes = 100%.
      To test the role of active rehabilitation in patient groups assumed to be high risk, certain characteristics of the patients were combined (Table 4). The favorable effect (Discharge destination = Home) of active rehabilitation was seen in all these patient groups.
      Table 4States of the Outcome Variable (Outcome Discharge) in Selective Combinations
      Fixations of Selected Patient Group CombinationsNo Fixation of Active RehabilitationFixation of Intervention Variable Active Rehabilitation yes = 100%
      HomeResidential CareDiedHomeResidential CareDied
      Dg∗ pneumonia + Age group 85 y or older54.739.65.759.040.60.4
      Dg urinary tract infection + Age group 85 y or older60.234.85.073.026.50.5
      Dg dementia + Age group 85 y or older67.528.34.278.321.20.5
      Dg pneumonia + Age group 75-84 y63.431.35.365.937.70.4
      Where coming from = referral hospital + Dg injury64.425.99.773.925.50.6
      Age group 65-74 y + Dg injury70.523.26.375.424.10.5
      Dg, Diagnosis.
      Finally, to estimate the effect of active rehabilitation, we fixed all variables to their original distributions according to vanderWeele.
      • VanderWeele T.J.
      Principles of confounder selection.
      Among the patients who did not receive active rehabilitation, discharge destinations were broken down as follows: home 69.4%, residential facility 25.3%, and died 5.4%. The corresponding figures for active rehabilitation were 73.9%, 25.3%, and 0.8%, respectively.

      Discussion

      Our study identified the factors associated with the discharge of older patients after short-term hospital stays in community hospitals. Moreover, associations between rehabilitation and discharge were analyzed using Bayesian network models. This method enabled us to investigate the impact of rehabilitation in unselected patient samples and a real-life context.
      In the present study, a vast majority of the patients were discharged to their own homes, a quarter to residential care facilities and 5% of patients died during their hospitalization. There were few other studies with comparable settings and reporting on discharge destinations.
      • Brusco N.K.
      • Taylor N.F.
      • Hornung I.
      • et al.
      Factors that predict discharge destination for patients in transitional care: A prospective observational cohort study
      • Chen C.
      • Koh G.C.
      • Naidoo N.
      • et al.
      Trends in length of stay, functional outcomes, and discharge destination stratified by disease type for inpatient rehabilitation in Singapore community hospitals from 1996 to 2005.
      • Chen C.
      • Naidoo N.
      • Er B.
      • et al.
      Factors associated with nursing home placement of all patients admitted for inpatient rehabilitation in Singapore community hospitals from 1996 to 2005: a disease stratified analysis.
      Furthermore, previous community hospital research has mainly focused on nonacute care,
      • Heaney D.
      • Black C.
      • O'donnell C.A.
      • et al.
      Community hospitals—the place of local service provision in a modernising NHS: an integrative thematic literature review.
      • Winpenny E.M.
      • Corbett J.
      • Miani C.
      • et al.
      Community hospitals in selected high income countries: a scoping review of approaches and models.
      • Pitchforth E.
      • Nolte E.
      • Corbett J.
      • et al.
      Community Hospitals and Their Services in the NHS: Identifying Transferable Learning From International Developments – Scoping Review, Systematic Review, Country Reports and Case Studies.
      whereas our study assessed older patients needing short-term care and for acute reasons in most cases. Thus, discharge to home was more common in the present study than in most previous studies.
      A younger age, a higher number of rehabilitation staff per patient, and daily rehabilitation were associated with discharging patients to their own homes. Consequently, a lower number of rehabilitation staff and less frequent rehabilitation were associated with discharging to a nursing home or death during the hospital stay. The favorable effect of active rehabilitation was identified in all patient groups, including patients with urinary tract infections, heart failure, and dementia, both in unadjusted and adjusted analyses.
      Patients coming to a community hospital from a secondary or tertiary hospital showed a poorer outcome than patients coming directly from their own homes or from other care or residential facilities. This can reflect differences in the disease severity. The variable “Where coming from” can be seen as a proxy of disease severity. Patients coming from a secondary or tertiary care hospital were presumably more seriously ill than patients coming from health centers or from their homes. Some of our findings were rather self-evident, such as high mortality among terminal care patients. Patients who came to a hospital from a nursing home or other residential settings are presumably discharged back to the same settings. However, this was a real-life study on short-term community hospital care, and based on this, it was important to include all the patients in the analyses. Despite the substantial heterogeneity of the patients, favorable effects of active rehabilitation were notable.
      Unlike the majority of research that investigated rehabilitation from the perspective of certain diseases or rehabilitation intervention,
      • de Morton N.A.
      • Keating J.L.
      • Jeffs K.
      Exercise for acutely hospitalised older medical patients.
      • Kosse N.M.
      • Dutmer A.L.
      • Dasenbrock L.
      • et al.
      Effectiveness and feasibility of early physical rehabilitation programs for geriatric hospitalized patients: a systematic review.
      • Kanach F.A.
      • Pastva A.M.
      • Hall
      • et al.
      Effects of structured exercise interventions for older adults hospitalized with acute medical illness: a systematic review.
      • McKelvie S.
      • Hall A.M.
      • Richmond H.R.
      • et al.
      Improving the rehabilitation of older people after emergency hospital admission.
      • de Morton N.A.
      • Keating J.L.
      • Jeffs K.
      The effect of exercise on outcomes for older acute medical inpatients compared with control or alternative treatments: a systematic review of randomized controlled trials.
      • Scheerman K.
      • Raaijmakers K.
      • Otten R.H.J.
      • et al.
      Effect of physical interventions on physical performance and physical activity in older patients during hospitalization: a systematic review.
      • Valenzuela P.L.
      • Morales J.S.
      • Castillo-García A.
      • et al.
      Effects of exercise interventions on the functional status of acutely hospitalised older adults: a systematic review and meta-analysis.
      • Ellis G.
      • Gardner M.
      • Tsiachristas A.
      • et al.
      Comprehensive geriatric assessment for older adults admitted to hospital.
      our study was carried out in a usual care context, in real-life clinical settings and without patient selection. The content of rehabilitation varied and was determined individually according to the patients’ needs and local rehabilitation resources. Consequently, our findings are not fully comparable with the previous research.
      To our knowledge, this was the first study using BA to investigate factors associated with the discharge destination and outcome of rehabilitation in community hospital settings. We considered BA to be useful for this complicated data set. A beneficial feature of BA was the possibility to estimate causal effects.
      A strength of our study was the high participation rate of community hospitals and good compliance in the data collection. Thus, our study provided comprehensive data on the reasons for treatment and contents of short-term primary care hospital stays. There were few interruptions in data collection and, similarly, hospitals were operating normally at the time of the data collection.
      We identified several limitations in our study as well. Some unmeasured variables may have had an effect on the discharge, or on whether the patient received active rehabilitation. Use of the disjunctive confounder criterion requires, in addition to the use of pretreatment variables only, that no unmeasured confounder have a direct effect on treatment selection, outcome, or both. We believe that the patient’s disease severity, physical and cognitive functioning, frailty status, and assumed prognosis were potential and partially overlapping confounders. Diagnoses and coming from a secondary or tertiary hospital are proxies for disease severity. In addition, the patient’s assumed rehabilitation capacity probably influenced the choice and activity of rehabilitation.

      Conclusion and Implications

      Our study showed the real-life outcome of rehabilitation among unselected older patients in community hospital settings. Active rehabilitation in community hospitals enhances the patients’ likelihood of returning home. Community hospitals know the local service system and have the possibility to tailor care and rehabilitation for older patients according to their individual needs. The role of community hospitals in the acute care and rehabilitation of older adults as well as support for health services given at home is valuable and increasingly important in rapidly aging societies.
      Older patients hospitalized for any reason, for example, acute infection or dementia, may benefit from active rehabilitation. Our findings indicate that the availability and intensity of rehabilitation should be increased. Local community hospitals can serve this purpose.

      References

      1. Statistical information on welfare and health in Finland.
        • Keskimäki I.
        • Tynkkynen L.K.
        • Reissel E.
        • et al.
        Finland: Health system review.
        2019 (World Health Organization. Regional Office for Europe, European Observatory on Health Systems and Policies)
        • Saari H.
        • Lönnroos E.
        • Mäntyselkä P.
        • et al.
        What are the characteristics of short-term inpatient care in Finnish primary care hospital? Article in Finnish.
        Finn Med J. 2019; 44: 2506-2518
        • Covinsky K.E.
        • Pierluissi E.
        • Johnston C.B.
        Hospitalization-associated disability: “She was probably able to ambulate, but I’m not sure.”.
        JAMA. 2011; 306: 1782-1793
        • Chodos A.H.
        • Kushel M.B.
        • Greysen S.R.
        • et al.
        Hospitalization-associated disability in adults admitted to a safety-net hospital.
        J Gen Intern Med. 2015; 30: 1765-1772
        • Hirsch C.H.
        • Sommers L.
        • Olsen A.
        • et al.
        The natural history of functional morbidity in hospitalized older patients.
        J Am Geriatr Soc. 1990; 38: 1296-1303
        • Martinez-Velilla N.
        • Cadore E.L.
        • Casas-Herrero A.
        • et al.
        Physical activity and early rehabilitation in hospitalized elderly medical patients: systematic review of randomized clinical trials.
        J Nutr Health Aging. 2016; 20: 738-751
        • Barnes D.E.
        • Palmer R.M.
        • Kresevic D.M.
        • et al.
        Acute care for elders units produced shorter hospital stays at lower cost while maintaining patients' functional status.
        Health Aff (Millwood). 2012; 31: 1227-1236
        • Finnish Medical Society Duodecim
        Current care guidelines.
        www.kaypahoito.fi
        Date accessed: June 2, 2022
        • Young J.
        • Green J.
        • Forster A.
        • et al.
        Postacute care for older people in community hospitals: a multicenter randomized, controlled trial.
        J Am Geriatr Soc. 2007; 55: 1995-2002
        • Green J.
        • Young J.
        • Forster A.
        • et al.
        Effects of locality based community hospital care on independence in older people needing rehabilitation: randomised controlled trial.
        BMJ. 2005; 331: 317-322
      2. Care register for health care.
        • Kajantie M.
        • Manderbacka K.
        • McCallum A.
        • et al.
        How to Carry Out Register-Based Health Services Research in Finland? Compiling Complex Study Data in the REDD Project.
        1st ed. STAKES, 2006
        • Auvinen K.J.
        • Räisänen J.
        • Voutilainen A.
        • et al.
        Interprofessional medication assessment has effects on the quality of medication among home care patients: randomized controlled intervention study.
        J Am Med Dir Assoc. 2021; 22: 74-81
        • Kekolahti P.
        Bayesian Network Analysis of Mobile Service and Device Usage. Aalto University Publication Series Doctoral Dissertations, 201/2019, Department of Communications and Networking.
        2019
        • Darwiche A.
        Bayesian networks.
        Commun ACM. 2010; 53: 80
        • Rautiainen E.
        • Ryynänen O.P.
        • Laatikainen T.
        • et al.
        Factors associated with 5-year costs of care among a cohort of alcohol use disorder patients: a bayesian network model.
        Healthc Inform Res. 2020; 26: 129-145
        • Pearl J.
        • Mackenzie D.
        The Book of Why: The New Science of Cause and Effect.
        1st ed. Basic Books, 2018
        • Cox Jr., L.A.
        Modernizing the Bradford Hill criteria for assessing causal relationships in observational data.
        Crit Rev Toxicol. 2018; 48: 682-712
        • Conrady S.
        • Jouffe L.
        Bayesian Networks and BayesiaLab – A Practical Introduction for Researchers.
        1st ed. Bayesia USA, 2015
        • Ryynänen O.P.
        • Leppänen T.
        • Kekolahti P.
        • et al.
        Bayesian network model to evaluate the effectiveness of continuous positive airway pressure treatment of sleep apnea.
        Healthc Inform Res. 2018; 24: 346-358
        • Myers J.W.
        • Laskey K.B.
        • DeJong K.A.
        Learning bayesian networks from in complete data using evolutionary algorithms.
        Proc Genet Evol Comput Conf. 1999; : 458-465
        • Korb K.B.
        • Nicholson A.E.
        Bayesian Artificial Intelligence.
        2nd ed. Chapman & Hall/CRC, 2004
        • Yumoto F.
        • Wood R.L.
        • Tractenberg R.E.
        Machine learning to evaluate the quality of patient reported epidemiological data. Survey Research Methods Section.
        JSM Proc. 2018; : 2583-2595
        • Kass R.E.
        • Raftery A.E.
        Bayes factors.
        J Am Stat Assoc. 1995; 90: 773-795
        • Elwert F.
        • Winship C.
        Endogenous selection bias: The problem of conditioning on a collider variable.
        Annu Rev Sociol. 2014; 40: 31-53
        • VanderWeele T.J.
        • Shpitser I.
        A new criterion for confounder selection.
        Biometrics. 2011; 67: 1406-1413
        • VanderWeele T.J.
        Principles of confounder selection.
        Eur J Epidemiol. 2019; 34: 211-219
        • Brusco N.K.
        • Taylor N.F.
        • Hornung I.
        • et al.
        • Factors that predict discharge destination for patients in transitional care: A prospective observational cohort study
        Aust Health Rev. 2012; 36: 430-436
        • Chen C.
        • Koh G.C.
        • Naidoo N.
        • et al.
        Trends in length of stay, functional outcomes, and discharge destination stratified by disease type for inpatient rehabilitation in Singapore community hospitals from 1996 to 2005.
        Arch Phys Med Rehabil. 2013; 94: 1342-1351
        • Chen C.
        • Naidoo N.
        • Er B.
        • et al.
        Factors associated with nursing home placement of all patients admitted for inpatient rehabilitation in Singapore community hospitals from 1996 to 2005: a disease stratified analysis.
        PLoS One. 2013; 8: e82697
        • Heaney D.
        • Black C.
        • O'donnell C.A.
        • et al.
        Community hospitals—the place of local service provision in a modernising NHS: an integrative thematic literature review.
        BMC Public Health. 2006; 6: 309
        • Winpenny E.M.
        • Corbett J.
        • Miani C.
        • et al.
        Community hospitals in selected high income countries: a scoping review of approaches and models.
        Int J Integr Care. 2016; 16: 1-13
        • Pitchforth E.
        • Nolte E.
        • Corbett J.
        • et al.
        Community Hospitals and Their Services in the NHS: Identifying Transferable Learning From International Developments – Scoping Review, Systematic Review, Country Reports and Case Studies.
        NIHR Journals Library. 2017; 5: 1-219
        • de Morton N.A.
        • Keating J.L.
        • Jeffs K.
        Exercise for acutely hospitalised older medical patients.
        Cochrane Database Syst Rev. 2007; 2007: CD005955
        • Kosse N.M.
        • Dutmer A.L.
        • Dasenbrock L.
        • et al.
        Effectiveness and feasibility of early physical rehabilitation programs for geriatric hospitalized patients: a systematic review.
        BMC Geriatr. 2013; 13: 107
        • Kanach F.A.
        • Pastva A.M.
        • Hall
        • et al.
        Effects of structured exercise interventions for older adults hospitalized with acute medical illness: a systematic review.
        J Aging Phys Act. 2018; 26: 284-303
        • McKelvie S.
        • Hall A.M.
        • Richmond H.R.
        • et al.
        Improving the rehabilitation of older people after emergency hospital admission.
        Maturitas. 2018; 111: 20-23
        • de Morton N.A.
        • Keating J.L.
        • Jeffs K.
        The effect of exercise on outcomes for older acute medical inpatients compared with control or alternative treatments: a systematic review of randomized controlled trials.
        Clin Rehabil. 2007; 21: 3-16
        • Scheerman K.
        • Raaijmakers K.
        • Otten R.H.J.
        • et al.
        Effect of physical interventions on physical performance and physical activity in older patients during hospitalization: a systematic review.
        BMC Geriatr. 2018; 18: 288
        • Valenzuela P.L.
        • Morales J.S.
        • Castillo-García A.
        • et al.
        Effects of exercise interventions on the functional status of acutely hospitalised older adults: a systematic review and meta-analysis.
        Ageing Res Rev. 2020; 61: 101076
        • Ellis G.
        • Gardner M.
        • Tsiachristas A.
        • et al.
        Comprehensive geriatric assessment for older adults admitted to hospital.
        Cochrane Database Syst Rev. 2017; 9: CD006211