We conducted a retrospective cohort study with routinely collected administrative hospital and mortality data.
settings and data
New South Wales (NSW) is Australia’s most populous state with 7.2 million inhabitants in 201226. We used the patient (hospital) data collection recruited in NSW, linked to mortality data for the period 1 January 2008 to 31 March 2013 all public and private hospitalizations ending in discharge, transfer, type change or death. Hospital admissions were coded using the International Statistical Classification of Diseases and Related Problems, Tenth Revision, Australian Revision (ICD-10-AM) and Australian Sophisticated Diagnosis Related Group (AR-DRG) codes. The Center for Health Record Linkage combined the two data sets using probabilistic methods with false positive and false negative rates of 0.5%28.
study team formation
Our study replicated the inclusion criteria of the original HFRS publication 9 with a cohort of NSW residents aged 75 years and older who had at least one unplanned acute hospital admission between 1 January 2010 and 31 December 2012. For admissions that end with a type of change (e.g. from acute to sub-acute care) or transfer, related stay times, the admission data and the admission status from the first care segment as well as the separation dates and separation dates from the last care segment were structured based on the type. We selected a random hospitalization as the “index” admission for each patient.
predictions and results
We classified the two main analysis variables of interest, multimorbidity and vulnerable risk, using the ICD-10-AM to index admissions and all hospitalization codes over the past two years.
Long-term conditions were identified from a list of 29 chronic conditions from the Charlson and Elixhauser indices supplementing the major morbidity from the most recent systematic reviews (Supplementary Table S4). Multimorbidity was defined as having at least two conditions from this list.
We calculated a continuous HFRS using the 109 ICD-10 codes from Gilbert et al.9 adapted to the Australian modification (ICD-10-AM) (Supplementary Table S5). HFRS captures vulnerabilities as well as functional deficits and comorbidities associated with symptoms. The presence of each of the 109 ICD-10 codes was detected from patients’ hospital records, a weight was assigned, and the weights were summed across all codes to obtain HFRS 9. We constructed dichotomous vulnerable groups with low vulnerability (HFRS < 5) and advanced vulnerability risk (HFRS 5, combining moderate and high vulnerabilities) using the Gilbert et al.
We constructed a composite variable of multimorbidity and prone risk with four categories: neither polymorbidity nor high risk of malignancy, only high risk of malignancy, multiple morbidity, and those with both multimorbidity and high risk of malignancy.
Other covariates are age at index entry (in five-year groups), sex, level of socioeconomic status based on the Index of Socioeconomic Gains and Disadvantages (IRSAD) compared to the Australian Statistics (SEIFA) socioeconomic index areas, and number of hospital admissions (none, one, two or more) in the last two years.
Interesting results include: mortality rates within 30 days of index inclusion; prolonged hospital stay (> 10 days in hospital); Unscheduled readmissions (for patients discharged alive) within 30 days of discharge, consistent with the original HFRS development study. Results AR-DRG procedure, based on admission type, grouped into medical (without OR intervention), surgical (with significant OR intervention), and other (with non-OR intervention).
We used descriptive statistics to compare demographic characteristics and raw outcome ratios between polymorbidity and at-risk groups.
We constructed a randomized Poisson intercept model to measure the association of outcomes with multimorbidity, weakly accounting for within-hospital clustering, and adjusted for age, gender, socioeconomic status, and number of prior admissions. Due to the high frequency of the results, the effects are given as a relative risk (RR).
We calculated and presented the interaction analyzes recommended by Knoll and Vanderweil16. Interactions were estimated on an additive scale using relative excess risk due to interactions (RERIRR), with adjustment for cluster data. RERIRR = 0 means no interaction (exactly additive), RERIRR > 0 indicates more interaction than additive, and RERIRR < 0 means less interaction than additive. Interactions were assessed on a multifactorial scale by including the interaction term in the fitted Poisson model, including both the main effect (multimorbidity and fritility) and the interaction term (multimorbidity * fritility). The significance of an interaction is indicated on the multiplier scale when the relative risk of the interaction duration differs from 1, and on the additive scale when the RERI differs from 0.
We used SAS version 9.4 (SAS Institute Inc., Cary, NC) for data management, analysis, and charting.
We have received ethical approval from the NSW Population and Health Services Research ethics committees (ref 2009/03/141) and the Aboriginal Health and Medical Research Council (ref 684/09) with exceptions for written informed consent. The study was conducted in accordance with the Australian National Health and Medical Research Council’s National Statement on Ethical Conduct in Human Research.
The records used in this article are available from the NSW Department of Health and the Birth, Death and Marriage Register, NSW, Australia. The dataset was created with permission and specific ethical approval of each source data restorer. Due to their strictly confidential character, the authors are not permitted to pass on data sets from individual entities. The data are available to researchers upon request and are subject to the approval processes of data administrators and ethics committees, as detailed on the NSW Center for Health Records Linkage website (