We analysed the Victorian Admitted Episodes Dataset (VAED) between 01 July 2005 and 30 June 2008, inclusive (fiscal years 2005/06, 2006/07 and 2007/08), to estimate the incidence of fall-related hip fracture in community-dwelling people aged 65+ years in Victoria and determine patients’ comorbidity profiles. The VAED is an administrative and clinical data collection of admitted patient episodes in acute hospitals in Victoria, Australia’s second most populous state [16
]. This data collection is managed by the Victorian Department of Health (DOH) and used to support casemix funding, epidemiological research, health services planning and policy development [16
]. The collection is subject to regular audits which indicate good-to-excellent diagnosis and procedure coding quality [16
]. The most recent published audits included, among other diagnoses and procedures, Charlson comorbidities, external cause of falls, hip fracture diagnosis and hip replacement [17
Each patient within a hospital is identified by a unique, hospital generated patient identifier and each episode has a unique hospital derived episode number; however, the VAED lacks a system-wide UPI and does not capture date of injury information [16
]. Episodes containing the principal mechanism of injury indicating a fall (W00–W19 in the International Classification of Diseases, Tenth Revision, Australian Modification (ICD-10-AM)) [19
], the age at admission of 65+ years and the principle diagnosis indicating an injury (S00 to T75 or T79 in ICD-10-AM) [19
] were extracted from the VAED to form an unlinked dataset. The S00 to T75 or T79 range was specified in order to exclude injuries due to medical care procedures [21
summarises the data extraction process. The unlinked dataset was internally linked by the DOH using stepwise deterministic linkage and person-identifying variables (such as sex, date of birth, country of birth, postcode, and Medicare number and suffix) to produce a linked dataset for the present study [22
]. The linkage process and linkage quality have been described in detail elsewhere [22
]. Briefly, this process consists of nine steps, including standardisation of linkage variables, determination of the quality of coding and quality assessment of linked data [22
]. A DOH study on the quality of VAED internal linkage for the period 1995–2000 found that the quality of coding was high and the false positive rate, defined as the rate of incorrectly matched records, was low (between 1% to 2%) [22
]. However, the report indicated that the false negative rate, measured as the percentage of unmatched inter-hospital transfer records from the same patients, was high (15%). A more recent assessment of the quality of VAED internal linkage is not publicly available.
A flow chart of data extraction process. ICD–10–AM – International Classification of Diseases, Tenth Revision, Australian Modification. VAED: Victorian Admitted Episode Dataset.
Within the unlinked dataset we used a standard approach of identifying incident fall-related hip fractures [14
] (hereafter referred to as the base case) (Table
)—records were selected if the principal diagnosis was hip fracture (S72.0 to S72.2 in ICD–10–AM [19
]), the admission source was coded as “private residence/accommodation” and the discharge status was other than in-hospital death. The category “private residence/accommodation” includes people living in their own home or private accommodation and excludes residents in nursing homes [16
]. We excluded records indicating readmission within 30
days of discharge; however, given the lack of a UPI in the VAED and the lack of a hospital site identifier for private hospitals this was only possible for patients admitted to the same public hospitals during the study period (64.7% of patients according to the linked dataset). Patients admitted to the same public hospitals during the study period were significantly younger (median age 76
years; interquartile range ((IQR) 60–84) than those admitted to different public hospitals during the study period (median age 81
years; IQR 72–86) (nonparametric equality-of-medians test p <0.001).
Selection criteria for linked and unlinked Victorian Admitted Episodes Dataset
For the linked dataset, we used the same principal diagnosis range and admission source category as those for unlinked data, but disregarded discharge status. We further refined our identification by including only records showing emergency hospital admission for acute care with no hip revision procedure code(s) (Table
]. Due to the lack of date of injury in the VAED [18
], the lack of ICD–10–AM codes on laterality of fractures (Saad P. Disease classification developer, National Centre for Classification in Health (Australia). Personal communication. 30 April 2010) and the inaccuracy of fracture type classification [23
], we developed additional criteria to distinguish between the first and subsequent fall-related hip fracture in the same patients. We assumed, based on a literature review, that the minimum time gap (clearance period) between incident fall-related hip fractures in the same patient would be 120
] and that all principal external cause codes (mechanism of injury, place of occurrence and activity being undertaken when injured) would differ between different fall-related hip fractures in the same patient. We also performed the analysis without the criterion for external cause codes; however, we found only seven cases that would be identified as incident cases if we omitted this criterion.
We defined the method of identifying incident fall-related hip fracture from linked data as the reference standard and calculated the sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) for the base case using standard definitions [26
]. We conducted a sensitivity analysis of the base case under four scenarios (Table
)— S1 included records showing the discharge status as in-hospital death; S2 excluded records indicating readmission within 120
days of discharge; S3 excluded records indicating readmissions within 120
days of discharge, records with only hip revision procedure code(s), and those with the care type coded as non-acute and the category of admission coded as non-emergency; S4: was the same as S3 but also included records showing the discharge status as in-hospital death (i.e. S4 employed the same selection criteria as the reference standard).
We calculated age-specific hospital admission rates for fall-related hip fracture in community-dwelling people aged 65+ years using Victorian population estimates for relevant years [27
]. The denominator for each age group was obtained by subtracting the number of residents in nursing homes from the population estimate for this group. We directly standardised rates of hospitalisation to the 2001 Australian standard populations [31
Patients’ comorbidities were classified using the Deyo adaptation of the Charlson Comorbidity Index (CCI) because this index was constructed using administrative data similar to those collected for the VAED, and validated using the VAED [32
]. We also estimated the prevalence of other risk factors for falling and fall-related fractures, including osteoporosis, Parkinson’s disease, visual impairment, deafness and delirium, using ICD–10–AM codes also tested on the VAED [33
]. We distinguished comorbidities from adverse events that arose during hospitalisation by utilising a condition-onset flag available in the VAED [16
]. The extent to which comorbidity prevalence estimates by unlinked data differed from those by linked data was assessed in absolute terms by performing pairwise comparisons. For patients in the linked dataset with more than one hospitalisation for a fall-related hip fracture, we optimised comorbidity ascertainment by defining the first multiday record as the index hospitalisation and searching this record as well as looking back at previous record(s) for the presence of comorbidities (hereafter referred to as lookback) [34
]. Comorbidity was deemed to be present if it was coded in one or more of these records. The median period of lookback was 565
days (IQR 274–836).
The Monash University Human Research Ethics Committee granted approval for this study. We conducted all analyses in Stata version 10 [35
]. We evaluated equality of proportions using two-sample chi square tests of proportions or Fisher's exact test, as appropriate. For skewed continuous variables, we compared medians using nonparametric K-sample tests on the equality of medians [35
]. A multivariable Poisson regression model controlling for age and sex was used to assess the existence of a trend in fall-related hip fracture hospitalisation rates over time. All tests were two tailed. The level of significance was 5%.