Incidences and variations of hospital acquired venous thromboembolism in Australian hospitals: a population-based study

We used records from the NSW Admitted Patient Data Collection (APDC) database. It includes all admitted patients in NSW public and private, acute, sub-acute and non-acute facilities, and is used for health services planning, funding and research. Reporting to APDC is mandatory at all hospitals. The APDC includes information on patient demographics, medical conditions and procedures, hospital characteristics, and separations (discharges, transfers and deaths) from all public and private healthcare facilities in NSW. Each episode of care in the APDC is assigned with up to 55 codes for morbidities (principal diagnosis and comorbidities) based on the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD-10-AM) 4th edition [25]. The codes are assigned by trained and qualified clinical coders based on clinical notes and according to standards [26]. The quality of data was found reasonable, but varied across data elements and hospitals [27‐29].

In NSW public hospitals accommodated 62 % of all patients’ admissions, with a greater proportion of overnight stays (70 %), but a similar same-day stays (51 %) compared to private hospitals. Over 80 % of funding for public and private hospitals were provided by government and health insurers respectively [30]. All admissions, between 1st January 2002 to 31st December 2009, from 104 of 497 NSW acute public and private hospitals (11,408,808, admissions; 71 %) were included. We excluded community, non-acute, psychiatric and rehabilitation facilities, nursing home and hospices, and the two children’s hospitals.

Included were patients who had a hospital length of stay (LOS) of at least two days, aged between 18 and 90 years (inclusive), and were not transferred to another acute care facility (4,089,144 episodes (35.8 %)).

Based on ICD-10-AM, we considered the codes I26 (.0) and I26.9 for PE and I80.1, I80.2, I80.3, I80.8, I80.9, I82.8, and I82.9 for DVT cases - a total of nine VTE diagnosis codes. VTE codes were chosen according to well-established VTE related measures within a quality and safety context [12, 31, 32] and existing published coding procedures [19, 22]. Cases with no VTE code as the principal diagnosis were identified as patients at risk and those who had at least one VTE code in secondary diagnoses were identified as HA-VTE cases. It resulted in exclusion of patients with VTE as principal diagnoses from both numerator and denominator. Obstetric patients, identified by major diagnosis category using Australian Refined Diagnosis Related Groups (AR-DRG) [25], and those surgical patients who only had a procedure for interruption of vena cava (ICD codes: 348000, 3533000 and 3533001) were excluded from the study population as suggested within a patient safety context in Australia [32] and elsewhere [31, 33]. The discharge status (alive or deceased) was used to derive the case fatality associated with a HA-VTE diagnosis. HA-VTE cases and related deaths were respectively presented as incidence rate (per 1000 admissions) and case fatility.

For all admissions, two sets of patient- and hospital-related covariates were considered. Patient demographic variables included age, gender, country of birth, marital status, patient socio-economic status, principal diagnostic disease groups (the ten most common groups based on Elixhauser comorbidities [34]), and length of stay within the study population. We utilised a postcode-level advantage and disadvantage index of Socio-Economic Indices for Areas (SEIFA) with the lower values indicating more disadvantaged areas [35]. SEIFA scores were categorised into four classes (1st quartile = most disadvantaged areas and 4th quartile = most advantaged areas). The disease groups were identified using principal diagnostic codes (ICD-10-AM) at admissions through the methodology developed by Quan et al. [34]. Admissions with any surgical codes recored as the principal procedure were categorised as surgical admissions and those with no procedural data were categorised as medical admissions. Using relevant procedure codes from ICD-10-AM (Additional file 1), we defined six major surgical procedures including coronary-artery bypass graft (CABG), abdominal aortic aneurysm (AAA) repair, total hip replacement, total knee replacement, cholecystectomy, and other surgical procedures.

Hospital covariates included the local health district (metropolitan, rural and regional NSW), the hospital type (public and private), and associated peer groups. Public hospital peer groups included “A1”: principal referral group, usually teaching hospitals; “A3”: ungrouped acute; “B”: major metropolitan and non-metropolitan; “C1”: district group 1; and, “C2”: district group 2. Private hospital peer groups included “21”: major; and “22”: district group hospitals. Hospital peer groups contained similar type and sized hospitals, ranging from those treating more than 25,000 acute case-mix weighted separations per annum in principal referral groups through to treating 2000+ (but less than 5000) acute case-mix weighted separations per annum in district group [36].

We employed Poisson mixed models to evaluate adjusted incidence and case fatality rates and rate ratios for both study outcomes after including all patients and hospital-related characteristics. A random intercept term was utilised to incorporate any clustering effect at hospital-level. Stratified models were constructed to derive estimates for specific admission types (medical and surgical) and hospital peer groups (public and private) separately. To investigate the temporal pattern of the outcomes, calendar years were entered into the model as indicator variables, with 2002 as the reference year. A model with the year as a continuous variable was also examined for linear trends. We derived specific trends for hospital types, admission type, peer group and surgery type using separate models. Adjusted rates for specific years were derived by multiplying yearly-adjusted rate ratios to the crude rates observed in the reference year.

We did not include either the Elixhauser or the Charlson Index comorbidities in the models due to an approximately unexpected 50 % drop (see Additional file 2) observed in the indices among our study population in recent years [21], and also reported geographical variations and biases in the coding that may lead to misleading results [37, 38]. For example, it was reported that there is an increased under-recording of diabetes conditions in APDC datasets between 2008 and 2012 [39].

To study the variation of outcomes across hospitals within each hospital group, hospital-specific random intercept components were extracted from stratified Poisson mixed models constructed for each hospital group, then ranked and categorised into five classes at 20 % incremental quintiles. To obtain adjusted differences between those with the highest and those with the lowest HA-VTE and HA-VTE case fatality rates, the adjusted classes were entered into a Poisson model including patient characteristics covariates. We then used Pearson correlation to assess the association of hospital performances between HA-VTE and post HA-VTE deaths, based on the hospital-specific random intercepts. All analyses were performed in R package version 3.0.0 [40] and Stata™ 11.0 [41].

Of the 3,331,677 admissions during 2002–2009 with a median LOS of five days (1^st and 3^rd quartiles: 3, 9 days), 38,161 patients developed a HA-VTE, resulting in an incidence rate of 11.4 per 1000 patients (median LOS = 12, quartiles: 7–22 days; Table 1). Among them, 3716 (9.7 %) died in hospitals (median LOS = 13, quartiles: 7–25 days). Patients who underwent surgery (elective or non-elective, 83 % of all included patients) accounted for 93.3 % of HA-VTE cases and were 2.3 times more likely to develop HA-VTE compared to medical patients, but had a similar mortality rate. Compared to females, male patients were less likely to develop a HA-VTE (IRR = 0.93) but more likely to die (IRR = 1.25). HA-VTE incidence and associated case fatality were higher among elderly. Patients who stayed longer in hospitals had higher HA-VTE incidences but lower fatalities.

Patients who were born in Europe (except the UK), Asia and North Africa experienced a lower risk of post-operative HA-VTE but a similar risk of death. Higher socio-economic status (quartiles of SEIFA) was associated with a higher risk of post HA-VTE death, in particular for surgical patients. Surgical and medical patients admitted with pulmonary circulation or coagulopathy disorders had the highest HA-VTE incidences, followed by cancer patients who had the highest fatality rate. Patients who underwent total knee replacement, AAA repair and total hip replacement surgeries had a higher risk of HA-VTE. However, post HA-VTE mortality was lower amongst orthopaedic surgical patients, compared to other surgical patients (see Additional file 3).

For medical patients, HA-VTE incidence was significantly higher in private hospitals compared to public hospitals (IRR = 1.76), whereas for surgical patients it was similar (see Additional files 3 and 4). The risk of death in private hospitals was almost half of the risk in public hospitals, in particular for surgical patients.

Among public hospitals, patients from principal referral hospitals were more likely to acquire VTE during hospitalisation in comparison to patients from smaller hospitals. However, patients admitted to smaller hospitals, were at a higher risk of death. This pattern was almost consistent across medical and surgical patients (see Additional files 3 and 4). No differences in outcomes were observed between private hospital peer groups.

HA-VTE incidence rate increased by 14 % between 2002 and 2004 reaching an adjusted incidence rate of 12.4 per 1000 patients, and then declined during the rest of the study period (Fig. 1). The increase in the HA-VTE incidence rate between 2002 and 2004 occurred in surgical but not medical patients in all hospitals (Fig. 2). However, in private hospitals, the incidence rate of HA-VTE peaked later, with greater increments of 44 and 40 % in 2006 and 2007 for surgical and medical patients, respectively (Fig. 2 and Additional file 4). Higher variations were observed in stratified trends - by public versus private, types of patients, hospital peer groups and surgical types (see Additional file 5).

The adjusted HA-VTE case fatality rate significantly declined by 22 % from 11.7 to 9.1 % over the study period (Fig. 1). Stratified analysis illustrated in Fig. 3 showed similar declines between public (2009 vs 2002: IRR = 0.74) and private hospitals (2009 vs 2002: IRR = 0.88), surgical (IRR = 0.80) and medical patients (IRR = 0.59). Large principal referral and metropolitan hospitals (coded as A1 and B) exhibited a significant decrease in the HA-VTE associated fatality rate over the study period (see Additional file 5).

Table 2 shows that private hospitals exhibited a larger variation between those hospitals with the lowest, and those with the highest HA-VTE rate, and associated fatality rates within each peer groups compared to public hospitals peer groups. The major private hospital peer group showed the highest variability of HA-VTA between the top and bottom 20 % percentile performers (IRR = 5.33), followed by district group 2 public hospitals (IRR = 2.94) and district private hospital groups (IRR = 2.44). For HA-VTE case fatality, the major private hospital group had the largest variability (IRR = 6.23), followed by the district private hospital group (IRR = 5.72) and then the district group 1 public hospitals (IRR = 3.77). These within group variations were almost replicated for surgical patients, whereas for medical patients, the district group 1 public hospital showed the highest variability for HA-VTE rates (IRR = 4.32) and principal referral public hospital group had the highest variability for HA-VTE case fatality (IRR = 15.65). The significant negative correlations between individual hospital HA-VTE rates and HA-VTEs case fatality rates for private hospital peer groups implied that hospitals with the highest HA-VTE rate tended to have a lower rate of subsequent death. There were no such associations within other public hospitals peer groups.