Study design, population, and outcomes
This is a retrospective cohort study that included patients aged 80 years or older who were admitted to each of the three participating hospitals—Soroka University Medical Center (SUMC) in Israel, Sir Charles Gairdner Hospital (SCGH) in Australia, and Beth Israel Deaconess Medical Center (BIDMC) in Boston, United States of America (USA). The three hospitals are all major tertiary level, academic, referral hospitals in their cities. Admissions in Israel and Australia were analyzed between the years 2006 and 2015 and in Boston between the years 2009–2015. These patients were sub-grouped into those who were admitted to an ICU and those who were admitted to a general ward. Only the first admission of every patient within the defined period of the study was analyzed. Within the index admission, only the first department and the first ICU admission were included in the analysis. Readmissions were considered as admissions occurring 6 months or less after discharge.
Exclusion criteria included: hospitalization for less than 24 h; admission for psychiatric conditions; admission for acute coronary syndrome as a primary diagnosis; advanced directives of CMO (comfort measures only) on day one of ICU admission; cardiac surgery patients (as elective patients) and absence of laboratory data on the admission.
The primary outcomes were in-hospital mortality and all-cause mortality at 6, 12, 18, and 24 months. Secondary outcomes were incidence of all ICU admissions among this age group in each medical center; ICU and ward mortality rates; ICU and hospital length of stay; and home discharge rates.
Clinical definitions and data sources
We used the Charlson’s comorbidity index to quantify the severity of comorbidities and identified them by ICD-9 and ICD-10 codes during the hospitalizations and by data obtained from community clinics [14]. The Laboratory‐Based Acute Physiology Score (LAPS) is a validated score that predicts in-hospital mortality in various primary diagnoses based on laboratory data only [15]. The score ranges between a minimum of 0 and a theoretical maximum of 256. For the LAPS calculation, we included the first lab results of every test in the first admission.
The Australian data were linked using data from the hospital admission dataset, the state mortality dataset (with follow-up of 2 years after discharge), and the pathology provider.
Due to technical difficulties, we could not obtain long-term mortality data of the BIDMC cohort. Thus, for comparison on long-term mortality rates, we used the data of the Israeli and Australian cohorts and compared it to the ‘Medical Information Mart for Intensive Care’ (MIMIC-III) database between 2009 and 2012 [16]. This comparison is valid since MIMIC-III data are derived only from BIDMC, taken from years included in our study, and from the same ICUs, and therefore including an unbiased part of the same cohort.
ICU capacities and admission policies
In all countries, ICU beds were calculated only if they are a part of a designated ICU unit with a team that is specialized in intensive care.
At SUMC, Beer Sheva, Israel, there are 827 adult beds and 24 ICU beds, respectively (ICU beds are 2.9% of all hospital beds, a ratio of 34:1). There are 16 general ICU beds and 8 medical ICU beds. This hospital is characterized by a highly restrictive triage policy. The admissions are at the discretion of the admitting intensive care specialist. Patients who are believed to have an irreversible clinical condition are mechanically ventilated in the internal medicine departments in most cases. ICU patients typically present with an acute need for mechanical ventilation or hemodynamic instability. SUMC is the only tertiary medical center in the city and serves as the tertiary center for a population of approximately 1 million people. It is the only hospital in the region of Southern Israel that provides cardiothoracic surgery, vascular surgery, and neurosurgery services.
At SCGH, Perth, Western Australia (WA), there are 600 adult and 23 adult ICU beds, respectively (ICU beds are 3.8% of all hospital beds, a ratio of 25:1). All the ICU beds are mixed with no differentiation between types of ICU. This ratio leads to a more restrictive ICU admission policy with ICU admission triaged on the basis of acuity, comorbidities, and prognosis. The admissions are at the discretion of the admitting intensive care specialist. No mechanical ventilation occurs outside of the single designated ICU areas in the hospital. There are 3 major referral ICUs in Perth, a city of 2 million people. SCGH is the only hospital in the state of WA providing neuro-interventional radiology services and liver transplantation. It also includes cardiothoracic surgery, vascular surgery, and neurosurgery services.
At BIDMC, Boston, USA, there are 621 adult and 77 ICU beds, respectively (ICU beds are 13% of all hospital beds, a ratio of 8:1). The breakup of ICU beds at BIDMC: 28 medical, 8 neurology, 18 trauma and surgical, 15 cardiovascular surgical, and 8 cardiac. This hospital is characterized by a liberal ICU admission policy wherein no broad-based exclusion criteria are implemented. When one of the ICUs is full (surgical-ICU, trauma-ICU, three medical-ICUs, cardiovascular-ICU, and cardiac-ICU), the patients are admitted in another suitable ICU or to the post-anesthesia care unit (PACU) as a “boarder”. This facilitates a policy of no refusal of ICU care for patients that are evaluated as requiring it. BIDMC is a tertiary level academic medical center serving Eastern Massachusetts and serves a catchment area of ~1 million people. Within the city of Boston (population of 4.5 million people) there are 5 other tertiary level academic medical centers and quite a few satellite hospitals within close proximity. Tertiary level centers in Boston provide similar services with few exceptions.
At SUMC and SCGH, ICU admissions are at the discretion of the admitting intensive care specialist and are “closed units” for the purposes of management, although teams are welcome to visit and provide their expertise and share in the care of the patient. In all centers, the medical insurance of the patient does not affect the ICU admission policy. Lawsuits are uncommon in intensive care in all countries.
Statistical analysis
Demographics and baseline clinical characteristics are presented as mean (SD) or median (IQR) as appropriate. When appropriate, we made univariate comparisons using χ2-test for categorical variables and using one-way ANOVA or Kruskal–Wallis test for quantitative variables.
We used multivariable logistic regression to characterize the factors determining ICU admission in the American cohort. Variables with a significance of < 0.1 in the univariate analysis were inserted to the multivariable model. This model has a c-statistic of 0.8 demonstrating good discrimination. With the results of the logistic regression, we calculated a score for all patients, based on sex, age, Charlson score, LAPS, and primary admission diagnosis. The purpose of the score was to predict the probability of Israeli and Australian patients to be admitted to the ICU, based on the American model. All patients were divided into quintiles of probability to be admitted to the ICU. In every quintile, we compared the actual admission allocation and the in-hospital mortality rates between the different countries, both in the wards and in the ICUs.
Patients in the 5th quintile in the Israeli and Australian cohort, i.e., patients with the highest probability to be admitted to the ICU, according to the American model, were compared according to the primary and secondary outcomes. Mortality data were linked with the state death database in Australia, obtained from the MIMIC database for the USA, and from the state database in Israel. We calculated survival according to the Kaplan–Meier method with log-rank test.
A p value of 0.05 or less (two-sided) was considered statistically significant. Statistical analyses were performed with Stata, version 13.0 (StataCorp, Texas, USA) and SPSS, version 25 (IBM Corp).