Days Spent at Home and Mortality After Critical Q14 Illness A Cluster Analysis Using Nationwide Data

BACKGROUND: Beyond the question of short-term survival, days spent at home could be considered a patient-centered outcome in critical care trials. RESEARCHQUESTION: What are the days spent at home and health care trajectories during the year after surviving critical illness? STUDY DESIGN AND METHODS: Data were extracted on adult survivors spending at least 2 nights in a French ICU during 2018 who were treated with invasive mechanical ventilation or vasopressors or inotropes. Trauma, burn, organ transplant, stroke, and neurosurgical patients were excluded. Stays at home, death, and hospitalizations were reported before and after ICU stay, using state sequence analysis. An unsupervised clustering method was performed to identify cohorts based on post-ICU trajectories. RESULTS: Of 77,132 ICU survivors, 89% returned home. In the year after discharge, these patients spent a median of 330 (interquartile range [IQR], 283-349) days at home. At 1 year, 77% of patients were still at home and 17% had died. Fifty-one percent had been re-hospitalized, and 10% required a further ICU admission. Forty-eight percent used rehabilitation facilities, and 5.7%, hospital at home. Three clusters of patients with distinct post-ICU trajectories were identi ﬁ ed. Patients in cluster 1 (68% of total) survived and spent most of the year at home (338 [323-354] days). Patients in cluster 2 (18%) had more complex trajectories, but most could return home (91%), spending 242 (174-277) days at home. Patients in cluster 3 (14%) died, with only 37% returning home for 45 (15-90) days. INTERPRETATION: Many patients had complex health care trajectories after surviving critical illness. Wide variations in the ability to return home after ICU discharge were observed between clusters, which represents an important patient-centered outcome

2][3][4][5] Beyond the legitimate question of shortterm survival, outcomes research in critical care is increasingly focusing on longer-term survival and quality of life.In this regard, the post-ICU syndrome (PICS) has been increasingly recognized as a major clinical entity. 1,6,7Defined as impairments in cognition, mental health, and physical function after critical care, it affects 33% to 99% of survivors. 8[14] The main objective of this study was to report outcomes and health care trajectories during the year after ICU discharge in adult patients admitted to French ICUs in 2018, using state sequence analysis.We focused on their ability to return home, days spent at home, and health care trajectories.

Data Source
France has a mandatory public health insurance system that covers the entire population, ie, 67 million inhabitants.The National Health Data System (Système National des Données de Santé, SNDS) comprehensively collects anonymized individual health care consumption data, reimbursed to beneficiaries of the various French public health insurance schemes. 15The SNDS includes outpatient data (pharmacy reimbursement claims, health care professional visits, and laboratory or imaging claims) and is linked to data collected on public and private hospital admissions via the Programme de Médicalisation des Systèmes d'Information, the national hospital discharge database.The Programme de Médicalisation des Systèmes d'Information comprises information regarding admissions to acute care hospitals (ACH), psychiatric wards (PW) use of rehabilitation facilities (RF) and hospital at home (HAH).The SNDS is also linked to a specific database for skilled nursing homes (SNH).The SNDS collects demographic data, date of death, and long-term chronic diseases eligible for 100% reimbursement of health care expenditure.Hospital stays in ACH are classified by the Groupes Homogènes de Malades system, a French adaptation of diagnosis-related groups.Long-term chronic diseases and hospital diagnoses are coded according to the International Statistical Classification of Diseases, 10 th Revision.Procedures are coded according to the Classification Commune des Actes Médicaux, a French classification of medical procedures.
The use of SNDS data by the Caisse Nationale de l'Assurance Maladie (CNAM), the French National Health Insurance Fund, has been approved by decree and by the French data protection authority (Commission Nationale de l'Informatique et des Libertés).CNAM has permanent access to SNDS data in application of the provisions of article R. 1461-12 of the French public health code.

Study Population Selection
Inclusion criteria were patients aged 18 years and older, admitted to a French adult ICU between January 1, 2018 and December 31, 2018, for at least 2 consecutive nights, requiring invasive mechanical ventilation or vasopressors or inotropes.Patients admitted to the ICU for trauma, burn injuries, organ transplant, stroke, or intracranial surgery were excluded from analysis because these causes of admission intrinsically affect the patient's ability to return home and reflect highly specific populations with distinct trajectories.Patients with no health care reimbursement in 2017 or with data linkage problems were also excluded.For each patient, if several ACH admissions met the selection criteria, the first was considered the index stay.Because the study focused on post-ICU trajectories, only patients discharged alive after the index ICU admission were included.

Data Collection and Statistical Analysis
Data were extracted on age, sex, and selected preexisting comorbidities identified by algorithms applied to the patients' 2017 data.These algorithms, developed by CNAM, combine inpatient diagnoses, longterm disease information, and pharmacy reimbursement claims, and are applied annually to each beneficiary, providing information on 58 health conditions. 16r the index stay, the cause of hospitalization (based on the Groupes Homogènes de Malades classification, summarized into 10 categories), ICU procedures, length of ICU stay, and the Simplified Acute Physiology Score II (SAPS II) on ICU admission were identified.
For each patient, a daily state sequence was created to analyze care pathways in the 365 days preceding and the 365 days after ICU discharge (baseline date, e- Fig 1).A sequence refers to the daily succession of the different events (states) defining the patient's trajectory.Subsequent states were collected: death, hospital stays in ACH (at least 1 night), RF, PW (only full-time hospitalization), HAH, and SNH stays.Among ACH stays, ICU admissions were specifically identified.When neither hospitalized nor deceased, patients were considered to be at home.Patients with SNH stays after the index stay were considered at home if they were already in a SNH before the index stay.When multiple states overlapped on the same day, we used the following priority rule to define the chosen state: Death > ICU > ACH > RF > PW > HAH > SNH > Home.To facilitate visualization and clustering operations (see later

Take-home Points
Study question: How many days did patients spend at home during the year after surviving critical illness?Results: Among patients surviving an ICU stay, 89% returned home (for a median 330 [interquartile range (IQR), 283-349] days) but with wide variability between clusters, and only 37% of patients who ultimately died after ICU discharge could return home for 45 (15-90) days.Interpretation: Many patients had complex health care trajectories after surviving critical illness, with large variability in their ability to return home.

Q6
), the daily state sequence was aggregated in a weekly state sequence, selecting the most frequent state presented by each patient during each week of the year before and after ICU discharge (52 weeks per year).If an equal number of events occurred during the same week for concurrent states, the previously described priority rule was used to define the weekly state.
Descriptive statistics were reported based on the individual daily state sequences (for each state: number of patients with at least one admission, number of admissions, cumulative length of admissions).Health care trajectories before and after ICU discharge were represented using distribution plots (transversal distribution of the different states each week) and sequence index plots (superposition of all the individual weekly state sequences).
Based on the post-ICU weekly state sequences, an unsupervised clustering method was used to identify groups of patients with similar trajectories after ICU discharge.The (dis)similarity between sequences was first measured: pairwise distances were computed between individual sequences by optimal matching, using the Longest Common Subsequence method. 17The partition around medoids clustering algorithm was then applied, using previously computed distances. 18This algorithm is intended to find a prespecified number of k representative sequences, called medoids, and attributes other sequences to the closest medoid.It aims to reduce the sum of dissimilarities between the medoid (center of the cluster) and the attributed sequences.The number of clusters was determined according to both statistical criteria and clinical appraisal of the clustering results.Sensitivity analyses using other sequence dissimilarity methods (optimal matching with different costs, Hamming distance) 17 and clustering algorithms (hierarchical ascendant classification, using Ward's method) were conducted.Overall, results were similar across analyses, and the approach combining longest common subsequence and partition around medoids was chosen for its robustness.
A multinomial logistic regression model was then used to assess baseline factors associated with the subsequently created clusters.Multivariable analysis was adjusted for baseline factors considered clinically relevant.
Data from the year before ICU discharge that was used as an exploratory analysis aimed to describe whether it differed from post-ICU clusters.
Results are presented as percentage or median and interquartile range (IQR) or ORs and 95% CI.Analyses were performed using SAS software (version 9.4, SAS Institute Inc) and R version 3.5.2(packages TraMineR version 2.0-11 and WeightedCluster version 1.4 for sequence analysis and clustering). 17,18

Health Care Trajectories After ICU Discharge
Three clusters were identified based on the patients' post-ICU trajectories (Tables 2 and 3). Figure 1 shows the state distribution plots and sequence index plots of care pathways before and after ICU discharge for all patients and for each of the three clusters.
Regarding other stays, 48% of patients were admitted at least once to RF for 29 (21-54) days, 5.7% had HAH stays, 5.2% were admitted to a PW, and 2% to a SNH facility.At 1 year post-ICU discharge, 77% of patients were at home and 17% had died.
Over the year preceding the index ICU stay, 99% were at home for a median cumulative duration of 351 (333-358) days, 56% were hospitalized in an ACH for a median 7 (3-18) days, and 4.9% had been admitted to an ICU for a median 4 (2-9) days (e-Table 1 and Figure 2).

Health Care Trajectories Among Clusters
We identified three clusters with very distinct characteristics and outcomes (Figs 1 and 2).Cluster 1 (n ¼ 52,254, 68%) was characterized by an early return to home and, mostly, a hospital-free trajectory for the year after ICU discharge.Patients could be discharged home in 99.8% of cases, for a median 338 (323-354) days.The median time before home discharge was 13 (6-28) days.At 1 year, 98% were still alive and 95% were at home.Nonetheless, 47% required rehospitalization in an ACH a median 2 (1-3) times for a median of 8 (3-17) days; 6.5% were readmitted to an ICU for a median 4 (2-8) days, and 42% were admitted to RFs for 22 (19-29) days.Cluster 1 included the highest rate of patients admitted to a PW (6.3%, for 33 (14-79) days).HAH or SNH   1).

Risk Factors to Belong to a Cluster
Patients in clusters 2 and 3 were older and had more comorbidities than patients in cluster 1 (Table 3).They were less frequently hospitalized for cardiac surgery, had longer index ICU stays, had higher SAPS II scores, and more frequently required renal replacement therapy, blood transfusion, mechanical circulatory support, gastrostomy, and tracheotomy than patients in cluster 1.
Compared with cluster 1, patients in cluster 2 were more often women, whereas patients in cluster 3 were more often men.These results were confirmed for variables included in the multivariable analysis (Table 3

Discussion
In this large retrospective population study of critically ill adults surviving an admission to a French ICU in 2018, 89% returned home for a median duration of 330 (283-349) days, and 17% died over the year after ICU discharge.Rehospitalizations in acute care units and ICUs were needed for 51% and 10% of patients, respectively.There was wide heterogeneity in their ability to return home.We identified three clusters reflecting three distinct post-ICU trajectories.Many patients had complex trajectories with alternating periods at home and hospital.Most patients who died during the year after discharge could not return home, and those who did managed to stay home for only a short period.
We previously reported that ICU survivors had a high risk of dying over subsequent years. 19In this study, we confirmed an ICU mortality of approximately 20% and an additional mortality rate of 17% in the year after ICU discharge.Albeit important, mortality may not be the worst outcome considered by patients or their relatives. 14,20,21Several observational studies reported poor quality of life of altered functional status after surviving a critical illness. 4,5,22,23Only a few randomized controlled trials have explored functional outcomes as a crucial end point.The conventional ventilation or extracorporeal membrane oxygenation for severe adult No. of stays, median (IQR) a 1 ( Patients with SNH stays after the index stay were considered at home if they were already in SNH before the index stay.c Among deceased patients.respiratory failure trial, for instance, explored the impact of extracorporeal membrane oxygenation in patients with severe acute respiratory distress syndrome on death or severe disability at 6 months. 24Recently, ability to return home and hospital-free days have been proposed as significant patient-centered outcomes in ICU survivors. 25,26Implementing strategies to accelerate and improve recovery and the ability to return home are advocated by both critical illness survivors and clinicians. 27ality of life in ICU survivors should be viewed as a main goal of ICU management.In the year preceding the ICU discharge, 99% of our cohort were at home for a median of 351 (333-358) days, although 56% were hospitalized in an acute care unit for 7 (3-18) days during 1 (1-3) stays, and 5% had an ICU stay.During the year after ICU discharge, 89% of the patients returned home, 51% were re-hospitalized in an ACH for 11 (4-25) days, and 10% spent 5 (2-11) days in an ICU.Of note, half of the patients were admitted at least once to an RF for a median of 29 (21-54) days, which represents a significant increase in health care resource utilization.The days spent at home during the year preceding admission was not a major discriminant of post-ICU trajectories.
Large variations in postdischarge trajectories were identified in the three different clusters.Cluster 1 gathered survivors who returned home after ICU discharge and survived, although many required several acute hospitalizations, and 42% were admitted to an RF.Cluster 2 included patients who had more complex health care trajectories, with 71% requiring a new acute hospitalization and 17% an ICU readmission.Most were admitted to long-term care facilities.Cluster 3 mainly comprised patients who died during the year after ICU discharge, with only 37% able to return home (43% including receipt of HAH) for a short period.Patients in clusters 2 and 3 were more likely to have prolonged ICU stays (>10 days), receive renal replacement therapy, or have a tracheotomy or gastrostomy performed. 28Most had complex trajectories with large utilization of health care resources.In the general population in France in 2018, approximately 15/ 1,000 inhabitants were admitted for hospitalizations in RF, and 107/1,000 inhabitants for overnight hospitalizations in ACH.Admissions to RF vary across hospitals and regions, depending on ease of access or the population profile, and these are decided on a case-bycase basis by physicians, with no specific economic or clinical criteria.Our study reports at a national scale the trajectories of patients after ICU discharge and adds to the literature of post-ICU outcomes.Among 1,083 Medicare survivors of sepsis, of whom only 38% required ICU admission, 63% were readmitted in the first year after discharge, spending a median 16 days (IQR, 2-45) in an inpatient health care facility. 25Among patients with septic shock, only a third of survivors had not returned to independent living by 6 months after discharge.In our study, the identification of clusters of patients provides important insights into the population more likely to return home after an ICU stay and those more likely to have complex trajectories with a requirement for complex care.Although these data are not intended for decision-making at the individual level, they nonetheless provide valuable information on the health care intensity of different populations after an ICU admission.Most of the patients who died during the year after ICU discharge never went home, except for short periods, and they spent most of their time in acute care units and RFs.Of note, 98% of them were at home in the year before ICU admission, excluding such criteria as a predictor of post-ICU outcomes.These results reinforce the need for accurate predictive and prognostic tools in patients discharged from the ICU. 292][33][34][35][36][37] This has been reported as an umbrella syndrome-PICS-corresponding to a global health impairment that includes physical, psychological, and cognitive symptoms after critical illness. 8  prospective, multicenter cohort, a history of anxiety or depression, prolonged duration of mechanical ventilation, and inability of a home discharge were associated with long-term disability. 12e methodology used in our study has several strengths.First, we used a nationwide administrative database with an excellent capture of health care utilization.Approximately 7% of patients were excluded because of absence of reimbursed health care or linkage issues making it impossible to follow health care trajectories.We excluded patients with specific causes of index hospital stay admission that could have had a major impact on post-ICU trajectories.Sequence analysis allowed us to analyze health care trajectories, considering the different states and their chronological progression, and could thus complement the focus on specific outcomes.Different sequence dissimilarity measures were tested, as well as a hierarchical ascendant clustering method.Although the three-cluster typology remained broadly similar, clusters 1 and 2 could vary in size.
Limitations of our study include the observational design, which prevents any causal association.This study focused on hospitalization data to define health care trajectories; ambulatory care requirements were not analyzed.Moreover, clinical information, functional status, or markers of quality of life are not directly available in the SNDS to assess whether patients who returned home were independent for daily life activities.Hospital bed availability, regional resource differences, and health care provider preferences could have impacted health care trajectories.
Overall, our study highlights the use of a massive claim database to explore long-term outcomes in critically ill patients, including the probability of returning home, which is a major patient-centered outcome.Future articles may further detail predictors of such long-term outcomes.

Interpretation
Most patients surviving a critical illness could return home.Many patients had complex health care trajectories compared with the year before their index ICU admission, but most patients who died after ICU discharge never return home or remain there for short periods, highlighting the need to better identify this subgroup of patients.Days at home should be considered an important patient-centered outcome in future critical care trials.

]
acute care hospital; IQR ¼ interquartile range; SAPS II ¼ Simplified Acute Physiology Score II. a Including contiguous ACH stays.

Figure 1 -FLA 5 . 6 . 0 DTD
Figure 1 -State distribution plot (A) and sequence index plots (B, C, D) of health care trajectories during the 52 weeks before and after ICU discharge, for all patients and by cluster.A, State distribution plots (transversal distribution of the different states each week); B, C, D, Sequence index plots (superposition of the longitudinal individual sequences of patients): unsorted (B), sorted by states from start of the post-ICU trajectory (C) and sorted by states from end (D).*Patients already admitted to SNH before ICU admission are represented as "Home" after ICU discharge.In all figures, the x axis represents week numbering, before and after ICU discharge.The baseline date (ICU discharge) is set at the beginning of week 1.In A, the y axis represents the proportion of patients.In B, C, and D, one line represents one patient sequence.SNH ¼ skilled nursing home.

Figure 3 -Figure 2 -FLA 5 . 6 . 0 DTD
Figure3-Factors associated with being in clusters 2 or 3 in multinomial logistic regression models, with cluster 1 taken as a reference.

TABLE 1 ]
Q10Characteristics of Patients and Index Stays

TABLE 1 ]
(Continued)The progression of "ACHto-Home" (23%) or "ACH-to-RF-to-Home" (17%) were the two main distinct state sequences in this cluster.Over the year preceding ICU discharge, 99% of patients in this cluster were at home for a median of 354 (343-359) days.Fifty-four percent had been hospitalized in an ACH for 6 (2-14) days, and 5.8% stayed in RFs (e-Table1).

TABLE 2 ]
Description of Hospital and SNH Stays, Home Stays, and Death in the Year (365 Days) After ICU Discharge, for All Patients and by Cluster ACH ¼ acute care hospital; HAH ¼ hospital at home; IQR ¼ interquartile range; LOS ¼ length of stay; PW ¼ psychiatric ward; RF ¼ rehabilitation facilities; SNH ¼ skilled nursing home.
a Among patients with at least one stay.b

TABLE 3 ]
Characteristics of Patients and Index Stays by Cluster, and Factors Associated With Clusters in Multinomial Logistic Regression Models

TABLE 3
38,39may explain the high utilization of health care resources after ICU discharge, especially RF and psychiatric hospitalization.This last form of hospitalization was needed by 6% of cluster 1 patients.In the French and European Outcome reGistry in Intensive Care Units (FROG-ICU) cohort, 22% and 19%, respectively, showed symptoms of anxiety or depression.38,39In a