HIM-Interchange

Introduction

Subacute and non-acute services are specialised multidisciplinary hospital-based wards or programs that primarily focus on the optimisation of patient functioning and quality of life or the provision of support for patients with impairment, activity limitation or restrictions due to a health condition. The role of these services is therefore inherently different from that of inpatient acute care, as a ‘person’s functioning may relate to their whole body or a body part, the whole person, or the whole person in a social context, and to impairment of a body function or structure, activity limitation and/or participation restriction’ (Australian Institute of Health and Welfare 2013). Subacute and non-acute care consists of five separate care types; rehabilitation, palliative care, geriatric evaluation and management (GEM), psychogeriatric care, and non-acute care. Due to an ageing population and an increased prevalence of chronic disease and disability, hospital-based services for subacute and non-acute patients will continue to significantly contribute to the demand for healthcare services in Australia.

The total number of subacute and non-acute separations from public and private hospitals increased from 462,000 to 610,000 between 2013–14 and 2017–18, equating to annual growth of 7.2% (AIHW, 2019). In 2017–18, subacute and non-acute care accounted for 4% of separations in the public and private hospital system, with rehabilitation care accounting for 46% of all subacute and non-acute care separations in public hospitals and 80% in private hospitals. Additionally, due to Australia’s ageing population, increased pressure is placed on GEM services in particular, as a result of increased patient demand (Schofield and Earnest 2006). Since 2013–14 the number of patients utilising GEM services at public hospitals has increased by an average of 1.7% annually. The type of care required for subacute and non-acute patients in public hospitals results in the average length of stay being 10.5 days longer than patients admitted to acute public hospitals (AIHW 2019). To ensure system sustainability, healthcare administrators need to effectively manage the increased demand for services with accurate funding allocations. To ensure adequate pricing of the provision of care, a new version of the subacute and non-acute classification has been developed by the Independent Hospital Pricing Authority (IHPA). This new classification version better captures the relationship between patient characteristics and costs.

Subacute and non-acute care is currently classified using the Australian National Subacute and Non-Acute Patient (AN-SNAP) Classification Version 4 (V4) which has been used by IHPA to price admitted subacute and non-acute care since 1 July 2016 for the purpose of activity based funding (ABF). IHPA undertakes regular review of all ABF classifications to ensure they reflect changes to patient complexity and cost over time. In 2018–19, IHPA commenced a review into the subacute and non-acute classification, resulting in the development of a refined classification. Classification refinement addressed changes in clinical practice and provision of care, in addition to utilising improved volume and coverage of national subacute data collected since 2015 (IHPA 2021). The refinement of the AN-SNAP classification aimed to enhance the existing AN-SNAP V4 to better capture patient complexity and resource consumption for ongoing ABF purposes.

The refinement of the AN-SNAP classification re-assessed the statistical performance of V4 and explored the inclusion of new variables in order to optimise future performance. Investigation into patient comorbidities was prioritised, due to being identified in AN-SNAP V4 development as a potential cost driver for both rehabilitation and GEM care, particularly in older patients (Green et al. 2015). This resulted in patient frailty, or risk of frailty, being identified as a focus for investigation (for the purpose of this report frailty has been considered a patient comorbidity).

AN-SNAP Classification

The AN-SNAP classification sorts episodes of care into groups by using splitting variables. These variables can either be categorical, meaning they describe a ‘quality’ or ‘characteristic’ of something, or numeric meaning they describe a measurable quantity such as a number. The variables are applied to the groups in a particular order to progressively break the groups down into meaningful classes, with each step in this process referred to as a ‘split’. In addition to this, if a numeric variable is used to split a group, it does this using set numbers as upper and/or lower thresholds for the group (IHPA 2021).

In AN-SNAP, the episodes are first grouped using episode type (admitted or non-admitted) as a categorical variable. Following this, the next step is to split those two groups into smaller groups using another categorical variable, which is care type. Subsequently, there are several further splits using a mix of categorical and numeric variables. The variables used in the AN-SNAP V4 classification are listed in Table 1.

Table 1: AN-SNAP V4 variables

Figure 1 shows the high-level structure of the admitted branch of AN-SNAP V4 which represents how the classification sorts episodes of care into various final groupings, also referred to as end classes. The admitted branch of AN-SNAP V4 was included in the development for V5, however the non-admitted branch was excluded as the classification is not used to price non-admitted subacute care in Australia.

Development Process

The development process for AN-SNAP V5 was informed by a number of overarching principles with three key principles being: clinical coherence, resource homogeneity and classification soundness. This ensured that patient episodes were grouped based on similar clinical characteristics and level of resource utilisation, in addition to the classification being statistically stable and robust. IHPA undertook the development of the classification in two stages. The first stage focused on reviewing the performance of the existing AN-SNAP V4 variables using the most recent data available, and the second stage investigated the possible inclusion of new variables. Across both development stages significant stakeholder consultation occurred with the Commonwealth and state and territory health departments, clinicians and other organisation across the subacute sector.

Stage one – Analysis of the existing AN-SNAP V4 classification

AN-SNAP V4 performance was assessed using activity and cost data from the financial year that the classification was implemented (2015–16) to the most current data available in 2017–18. Analysis was undertaken to understand the extent any update could improve the grouping of episodes and how well the classification explained the cost of an episode of care. To ensure statistical decision making was clinically and technically supported, IHPA undertook extensive stakeholder engagement with its Subacute Care Working Group and Clinical Advisory Committee and established additional specialist rehabilitation and GEM clinical subgroups which provided further clinical expertise to support the decision-making process.

Analysis was conducted on rehabilitation, palliative care, GEM, psychogeriatric, and non-acute care types. The primarily statistical measure used throughout the classification refinement was the Reduction in Deviance (RID) statistic. RID was used in order to assess the statistical performance of the classification as a whole, or for a relevant subsection of a classification. RID was utilised as it measures how much variability in cost is explained by the system; the higher the RID value, the higher the amount of cost variation that is explained by the classification. The analysis demonstrated that the classification’s statistical performance could only be marginally improved to 53.8% RID by adjusting the existing variables (IHPA, 2021), which was an overall increase of 0.1% RID. This supported the decision to explore potential new variables for inclusion in the classification.

Stage two – Development and consideration of new variables

To identify potential new and clinically relevant variables that could improve statistical performance, IHPA consulted with clinical specialists and other key stakeholders involved in the provision of subacute and non-acute care. Patient comorbidities, and in particular frailty, was identified as a priority area for a potential new variable. During development, clinicians noted the increasing proportion of frail patients across the subacute sector, but despite the change in the patient casemix it was acknowledged that conceptualising and consistently measuring frailty would be challenging (Theo et al., 2013). This was supported in the findings from the AN-SNAP V4 Final Report (Green et al. 2015), which identified patient comorbidities as a potential cost driver for patient complexity that required further investigation. The AN-SNAP V4 classification did not incorporate a comorbidity related measure into the classification. Therefore, IHPA was required to focus on approaches that could use data already collected as a proxy for comorbidities such as frailty.

To identify existing measures of frailty, a literature review was conducted. IHPA identified an international frailty study that produced a proxy method for capturing patient frailty: Development and validation of a Hospital Frailty Risk Score focusing on older people in acute care settings using electronic hospital records: an observational study (Gilbert et al., 2018). In this study, Gilbert et al. used cluster analysis to identify a distinct group of patients in hospital with characteristics of frailty that could be identified on the basis of International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes, and created a Hospital Frailty Risk Score using these ICD-10 codes. The study’s findings showed that older patients, i.e. ≥ 75 years old, with frailty characteristics were at a higher risk of adverse health outcomes and were therefore more complex (Gilbert et al. 2018).

To investigate the Hospital Frailty Risk Score, IHPA needed to ensure it was fit-for-purpose for ABF in Australia. First, the ICD-10 codes identified by Gilbert et al. (2018) in the Hospital Frailty Risk Score were mapped to a total of 109 International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD-10-AM) codes. These codes were reviewed by IHPA Clinical Coders to determine whether any should be excluded from being assigned a higher complexity marker (IHPA 2021). The intent was to identify codes that may be ill-defined or effectively duplicated other coded information and remove them. Excluding these codes avoided double-counting and unjustified increases in the risk score. The ICD-10-AM codes remaining then formed the basis of IHPA’s proposed recognition of frailty, referred to as the Frailty Related Index of Comorbidities (FRIC).

IHPA also investigated alternative measures for capturing patient comorbidities, such as the Charleston Comorbidity Index (CCI). The CCI is an internationally validated approach to measuring disease burden by quantifying the effect of comorbid illnesses on patient outcomes (Sundararajan et al. 2004). Following the identification of these potential new variables, various methods were utilised to assess inclusion in the classification including ‘shortlisting’ potential new variables using the Classification and Regression Tree (CART) algorithm. The CART algorithm was applied to determine the appropriate place in the classification structure based off explanatory power with cost.

Following the investigation of potential new variables, a decision was made to progress with a new frailty measure, however, alternative measures were not progressed as they were outperformed by existing AN-SNAP V4 variables. Once determined, IHPA simulated the inclusion of the FRIC into the classification and subsequently applied numeric thresholds for the purpose of determining final end class splits for classification grouping.

AN-SNAP V5

The updated classification developed for subacute and non-acute care, AN-SNAP Version 5 (V5), has improved the statistical performance of the classification by introducing a number of changes. The most significant update is the inclusion of the FRIC. In addition to this, a new impairment type has been added for knee, hip and shoulder joint replacements. Finally, existing variables in the classification have been reordered or expanded, including the use of the Health of the Nation Outcome Scales (HoNOS) 65+.

Overall, the AN-SNAP V5 has a RID of 54.3%, which is an improvement of 0.4% when compared to AN-SNAP V4. The largest performance increases are evident in the GEM and non-acute care types with a RID increase of 2.6% and 2.0% respectively. Table 2 provides a summary of all the changes from AN-SNAP V4 to V5.

Table 2: Variable changes from AN-SNAP V4 to V5

Key Classification Changes

Frailty Related Index of Comorbidities

In IHPA’s two-stage development of AN-SNAP V5, a proxy for patient frailty was identified as being valuable to improve how patient complexity is captured in the classification in addition to enhancing statistical performance. As a result, IHPA included the FRIC for AN-SNAP V5 to measure frailty for the GEM and non-acute care types, which had the largest statistical improvement when incorporating the measure. The FRIC was also considered for inclusion within the rehabilitation care type. However, there was a lack of clinical support for inclusion at this stage.

The FRIC consists of ICD-10-AM codes that were identified as appropriate proxies for frailty by IHPA, with each diagnosis code assigned an individual score. The relevant diagnosis code scores assigned for each episode of subacute patient care are summed to calculate the total FRIC score. The FRIC demonstrated a considerable improvement in classification performance, specifically for the GEM and non-acute care types. Performance for these care types increased by a RID of 2.6% and 2.0% respectively.

Knee, hip and shoulder replacement

During the development of AN-SNAP V4 a decision was made to combine impairment types ‘replacements’ and ‘other orthopaedic conditions’ into one impairment type group. This was supported by statistical analysis at the time which showed little difference in costs. Subsequently, due to support from clinicians and improved volume of cost data that showed variation in costs, IHPA has reinstated a specific impairment type group for ‘replacement’ in the admitted adult rehabilitation classes for AN-SNAP V5. This provides three classes to classify rehabilitation activity following knee, hip and shoulder replacements.

Modification to existing variables

AN-SNAP V5 switches the order that the Function Independence Measure™ (FIM) sub-scores (Weighted FIM™ Motor and FIM™ Cognition) are applied for the brain dysfunction impairment type under the rehabilitation care. In AN-SNAP V5, brain dysfunction episodes are first grouped by FIM™ Cognition scores and then by WFIM™ Motor scores. Changing the order was a clinician led proposal first suggested by the AN-SNAP V5 development rehabilitation clinical subgroup. Analysis using the CART model showed this order in the brain dysfunction impairment type was better at predicting cost with analysis demonstrating a RID of 15.1%, which was a considerable improvement from the AN-SNAP V4 baseline of 13.1%. Therefore, a decision was made under the brain dysfunction impairment type to switch the variable order to FIM™ Cognition and then Weighted FIM™ Motor.

The other significant change to an existing variable was to use the HoNOS 65+ total score for psychogeriatric care. HoNOS is a clinician rated instrument comprising 12 simple scales measuring behaviour, impairment, symptoms and social functioning, with total and individual item scores. Analysis showed that splitting the psychogeriatric short term care episodes into two end-classes, based on HoNOS 65+ total score, outperformed the five classes used in AN-SNAP V4 (which used specific item scores for overactive behaviour and activities of daily living) by 0.5% RID. IHPA therefore decided that, on balance, it was appropriate to adopt the better performing variable. This also more closely aligned the approach used in the psychogeriatric care type to that used for the Australian Mental Health Care Classification.

Conclusion

AN-SNAP V5 is considered a modest refinement from V4. It has been developed over a two-year period through detailed statistical analysis and assessment, as well as consultation with clinical experts and other key stakeholders. The overall structure of AN-SNAP V5 remains the same compared to V4. However, a number of changes have been introduced to better capture patient complexity, and therefore resource consumption, in the public hospital system.

IHPA will continue to refine and improve AN-SNAP in order to capture patient comorbidities such as frailty, which are integral to understanding the changing costs of delivering subacute and non-acute care. It is apparent that the number of subacute and non-acute patient episodes is increasing in Australia each year, with the implementation of AN-SNAP V5 aiming to ensure that sustainable funding is provided for these services in the future through ABF.