Innovative approaches to analysing aged care falls incident data: international classification for patient safety and correspondence analysis

Abstract

Falls in residential aged care facilities (RACFs) are common and can have significant health consequences. Understanding how and why falls occur in RACFs is an essential step to design targeted fall prevention and intervention programmes; however, little is known about falls’ mechanisms in RACFs. This study aims to use international incident classification systems and novel analysis techniques to describe factors that contribute to falls requiring hospitalization in RACFs. Retrospective data of falls assessed by nurses as requiring hospitalization from 429 residents in 22 Australian RACFs in 2019 were used. Data were reviewed using a modified version of the International Classification for Patient Safety (ICPS), which categorizes patient safety into incident types and contributing factors using a three-tiered structure. The ICPS codes were summarized using the descriptive statistics. The association between assigned ICPS codes were analysed using correspondence analysis. Six hundred and three falls assessed as requiring hospitalization were classified into 659 incident types, with the most common incident type being ‘patient incidents’ (injury sustained/adverse effect in the health care system) (603, 91.5%) at Level 1, ‘falls’ (601, 91.2%) at Level 2, and ‘falls involving bedrooms’ (214, 32.5%) at Level 3. The 603 falls had 1082 contributing factors, with the most common contributing factor being ‘patient factors’ (events affected by factors associated with the patient) (982, 90.8%) at Level 1, ‘patient not elsewhere classified’ (characteristics of the patient contributed to the incident not classified elsewhere) (571, 52.8%) at Level 2, and ‘loss of balance’ (361, 33.4%) at Level 3. In a correspondence analysis, three dimensions were responsible for 81.2% of the variation in falls incidents and environmental and organizational factors were important factors contributing to falls. The application of the ICPS demonstrated that personal factors (e.g. pre-existing physical and psychological health or impairment) were the most common contributing factors to falls assessed as requiring hospitalization, while the correspondence analysis highlighted the role of environmental and organizational factors. The results signal the need for more research into multifactorial falls prevention interventions in RACFs.

Introduction

Falls and falls-induced injuries are frequent among older adults residing in residential aged care facilities (RACFs) (i.e. nursing homes, care homes, or long-term care facilities) [1]. Falls can lead to severe clinical consequences, including functional impairment, disability, premature death [2, 3], and poor quality of life [4], and might cause heavy financial burden on RACF residents and their families [5]. In Australia, more than half of residents in RACFs fall at least once per year [6] and approximately half of all resident hospital admissions were attributable to falls [7].

Reporting of incidents, including falls, has been traditionally poor and less transparent in Australian RACFs despite the considerable magnitude of costs and outcomes associated with falls. Recently, the Australian government has mandated the reporting of falls in RACFs through a quality indicator programme [8], following eight other countries with established falls reporting systems [9]. While this programme may improve data collection and reporting, this progress in isolation is unlikely to shift the dial on the number of falls occurring in RACFs. To inform improved falls prevention and intervention programmes, a more nuanced and evidence-based understanding of how and why falls occur in RACFs is needed [10].

One major challenge in understanding of patient safety incidents, such as falls, is the inconsistent and unstandardized use of language in describing incidents, which reduces the ability to aggregate, organize, and compare information across aged care providers. To address this difficulty, the Conceptual Framework for International Classification for Patient Safety (ICPS) was introduced. ICPS is a standardized set of concepts with agreed definitions and preferred terms to enable internationally acceptable classification of the major events associated with patient safety ( Appendix 1) [11]. This conceptual framework facilitates standardized description, comparison, measurement, interpretation, and dissemination of information to improve patient care. The framework uses a three-tiered decision tree structure to attribute incident type/s and contributing factor/s to each patient safety issue. Incident characteristics refers to information regarding the circumstances surrounding the incident such as where and when the incident occurred and who was involved. Contributing factor are the circumstance, action or influence which is thought to have played a role in the origin or development of an incident or to increase the risk of an incident. As such, each incident type and contributing factor have three levels of description. For example, a resident fall in their room due to low blood pressure could be described as: incident type, Level 1: patient accident, Level 2: falls, and Level 3: fall involving bedroom; contributing factor, Level 1: patient factors, Level 2: physical and psychological health or impairment, and Level 3: physical disease or impairment. The ICPS has been operationalized in other healthcare settings to classify adverse events in adult liver and kidney transplant surgeries [12], factors attributable to medical or surgical complication deaths [13], and medication error classification standards in nursing care [14]. However, to the best of our knowledge, there is no other study that has applied the ICPS to classify falls incident data in RACFs.

This study aimed to provide the first application of the ICPS to classify individual incident characteristics and contributing factors of falls assessed as requiring hospitalizations in RACFs and to use novel techniques to analyse associations between coded incident types and contributing factors. Results from this analysis could be used to consistently inform facilities of fall characteristics to inform tailored and optimized fall prevention and intervention programmes in RACFs.

Methods

A retrospective cohort study of reported falls requiring hospitalization in 2019 from 22 RACFs administered by one aged care provider in New South Wales, Australia, was conducted.

Inclusion/exclusion criteria

Data relating to all residents ≥65 years old from participating facilities were considered in this analysis. Resident data were excluded if they did not experience at least one fall assessed by the nurses as requiring a hospital visit (emergency department visit or admission) between 31 January and 31 December 2019 (Fig. 1).

Data sources and linkage

Data were sourced from the aged care provider’s clinical information system. The data were deidentified by the provider and included two datasets: (i) residents’ profiles and (ii) all falls incidents. The resident’s profile contains information about the demographic and clinical characteristics of the resident, including length of stay, age, and health comorbidities. The falls incidents datasets were recorded on incident registry forms. These forms document the details of the falls, including categorical variables (e.g. facility, type of incident, location, hospitalization, and severity assessment code score), continuous variables (e.g. time and date), and open text fields (e.g. incident details and outcomes of incident review). We selected data of falls assessed as requiring hospitalization because they are generally the most serious and costly type of falls. Each falls incident was linked to the resident’s profile using a unique identifier developed by the aged care provider. The incident registry is completed at the time of an event by the staff member involved. This could be a care worker, enrolled nurse, or registered nurse. Once a staff member completes an incident form, it is reviewed and signed off by their manager.

Data collection

This study applied the World Health Organization’s (WHO) ICPS Data Dictionary Version 2—Australian Modification 2018 [15] to identify the incident characteristics, incident types, and contributing factors responsible for falls. The modified version of the WHO ICPS consists of seven sections: unique identifiers, patient characteristics, incident characteristics, incident type, contributing factors, organization preventive actions, and coronial recommendations [15]. Among these components, incident characteristics, incident types, and contributing factors were relevant in our study, which comprise a three-tiered (i.e. Levels 1, 2, and 3) hierarchical classification framework (Fig. 2) ( Appendix 1).

Four researchers (K.S., A.N., P.P., and I.M.) reviewed the data dictionary of ICPS to gain an in-depth understanding of the definitions of each concept and familiarized themselves with the incident registry before applying the classification (Fig. 3). Each researcher first applied the modified ICPS to 20 random cases, which were analysed independently by three researchers (K.S., P.P., and A.N.). This was followed by a debriefing session that enabled detailed discussions on classification including making consensus decisions on what modifications need to be made to the tool to increase its usability in the aged care setting. This process resulted in some additions to the Level 3 categories incident types, such as ‘bedroom’ and ‘common areas’, and contributing factors, such as ‘patient had a recent medical change’ and ‘resident not using physical aid’. All changes are listed in  Appendix 1. The remaining incidents were coded (by P.P.) and confirmed independently (by K.S. and I.M.), with discussions held at every 50th incident to discuss uncertainties and discrepancies (by K.S., A.N., P.P., and I.M.).

Data analysis

Descriptive statistics were used to describe the categorized incident types and contributing factors using Statistical Analysis System 9.4 software. Correspondence analysis was conducted in R software [16] using R package ‘FactoMineR’ [17] to investigate the association between incident types and contributing factors.

Correspondence analysis is a useful exploratory statistical technique to visualize the relationships between categorical variables [18]. This method provides the ability to plot the incident types and the contributing factors in a geometrical space, by considering the associations between the two variables. Furthermore, correspondence analysis can effectively reduce the number of attributes by proposing synthetic dimensions that consolidate variables with a high-dimensional Euclidean space in a data matrix. This allows for a more simplified and interpretable representation of the relationships between the variables. The first dimension accounts for the greatest proportion of variance in the dataset, and each subsequent dimension explains the next greatest proportion of the remaining variance. We used the chi-squared distance to calculate the distance between points on the plot and the singular values generated by the singular value decomposition to the variance in each dimension [19].

In this study, we included incident types with more than five occurrences in the correspondence analysis. Therefore, only two incident types, fall incidents and clinical process/procedure, were included in the correspondence analysis. Fall incidents were analysed using Level 3 ICPS codes, while clinical process/procedure was analysed at Level 1 due to low frequency. Where falls had multiple contributing factors and incident types, multiple records were created. A contingency table was created for the incident records considering incident types as rows and contributing factors as columns. The correspondence analysis was performed on the contingency table, and the first three dimensions were explored and analysed using symmetric and asymmetric biplots. Asymmetric biplots were plotted using standard coordinates for the incidents and principal coordinates for the contributing factors, while symmetric biplots were plotted using principal coordinates of both incident types and contributing factors [20]. Therefore, the symmetric biplot could be used to interpret the relationship between contributing factors or incident types. This is achieved by analysing the proximity between the plotted points and their positions on the dimensions. The relationship between any contributing factors with an incident type could be identified using the asymmetric biplot, where the strength of the relationship can be inferred based on the acute angle formed between the contributing factor and the incident type.

Results

Demographics

Across the 22 included RACFs, 3221 residents were admitted in 2019. A total of 603 falls assessed as requiring hospitalization were recorded from 429 residents. Among these 429 residents, nearly two-thirds were female (64.8%). On average, residents had 10.6 falls since admission into aged care with a mean length of stay of 2.7 years (range 0–25 years). Almost three quarters of residents had a diagnosis of dementia and/or cognitive impairment (72.7%) (Table 1).

Falls

Residents on average experienced 1.4 falls assessed as requiring hospitalization during 2019, with a range from 1 to 10. Falls assessed as requiring hospitalization most frequently occurred in residents’ rooms (n = 371, 61.5%) or in communal areas (e.g. corridor, outdoor area, and dining room) (n = 133, 22.1%). Bathrooms (n =92, 15.3%) and external locations (e.g. family residence and transport) (n = 5, 0.8%) were less common fall locations ( Appendix 2, Table A1).

Incident type

Using the modified ICPS, the 603 falls assessed as requiring hospitalization were coded to 659 incident types. As shown in Table 2, at Level 1, the most common incident type classified was patient incidents (n = 603, 91.5%), followed by clinical process/procedure (n = 48, 7.3%), medication (n = 4, 0.6%), infrastructure/buildings (n = 3, 0.5%), and clinical administration (n = 1, 0.2%). Patient incidents were further classified as either a fall (n = 601, 91.2%) or fall through blunt force (n = 2, 0.3%) at Level 2 (Table 2) and were further classified as falls involving the bedroom (n = 214, 32.5%), bed (n = 107, 16.2%), bathroom (n = 69, 10.5%), toilet (n = 22, 3.3%), steps (n = 1, 0.15%), transportation (n = 3, 0.5%), and other individuals (n = 1, 0.2%) at Level 3 ( Appendix 2, Table A2). For the 48 clinical process/procedure incidents, they were classified as screening/prevention/routine check-up (n = 43, 6.5%) and day-to-day general patient healthcare and observations (n = 5, 0.8%) at Level 2 and were most frequently coded into delayed screening (n = 41, 6.2%) (Level 3) but were coded less frequently as incomplete (n = 4, 0.6%) or delayed (n = 1, 0.2%) general healthcare and observations at Level 3. Examples of delayed screening include instances where a resident had a myocardial infarction, glycaemic episode, pain, acute cognitive decline, agitation, stroke, incontinence, constipation, hypotensive episode, aspirated, gastric reflux, drowsiness, and recurrent falls over 24 h or was generally described as unwell at the time of the fall.

Contributing factors

In total, 1082 contributing factors were associated with the 603 falls. As shown in Table 3, Level 1 contributing factors frequently included patient factors (n = 982, 90.8%), followed by work/environment (n = 64, 5.9%), organization service (n = 14, 1.3%), unknown (n = 13, 1.2%), and staff factors (n = 9, 0.8%). Within patient factors, patient factors not elsewhere classified (n = 571, 52.8%) and pre-existing impairments (n = 331, 30.6%) were common contributing factors at Level 2, while communication issues (n = 80, 7.4%) were less common. At Level 3, within patient factors, loss of balance (n = 361, 33.4%), recent medical change (n = 125, 11.6%), physical disease (n = 264, 24.4%), and psychological health (n = 66, 6.1%) were common contributing factors, while intentionally not disclosing information (n = 55, 5.08%) was less common ( Appendix 2, Table A3). Examples of unintentionally disclosing information included resident’s lack of insight about their abilities and forgetting to ring the call button, while intentional reasons for not calling staff included resident choosing to not comply with care, impulsiveness, lack of trust and fear, and attempts to self-care or move without assistance.

Correspondence analysis

A correspondence analysis was conducted on 1114 records (a record is an incident type and/or contributing factor for a fall) ( Appendix 3, Table A1). A significant relationship was found between the incident types (rows) and contributing factors (columns) (χ² = 181.747, df = 90, P-value ≤ .001). Three dimensions were identified as critical in explaining the variances of falls incidents analysed. These three dimensions were ‘Dimension 1’ primarily driven by incidents of falls involving communal spaces and outdoor areas and contributing factors of work environment, resident communication issues, and resident not using physical aides; ‘Dimension 2’ primarily driven by incident types of clinical processes, falls involving chair or wheelchair, and unspecified falls along with significant and contributing factors of resident with a recent medical change, loss of balance, and communication issues; and ‘Dimension 3’ primarily driven by incident types involving bed, bedroom, physical aides, and bathrooms along with contributing factors such as resident attending physical environment and resident-related physical and psychological health ( Appendix 3  Tables A3 and A4). These three dimensions from the correspondence analysis in total explained 81.16% of the inertia (i.e. total variance) ( Appendix 3, Table A2).

Symmetric plots of Dimensions 1, 2, and 3 ( Appendix 3, Fig. A1) demonstrate relationships between contributing factor or incident types. Based on the analysis of Dimension 1, it can be observed that certain contributing factors exhibit a tendency to be related to one another. For instance, factors related to the work environment and instances where residents do not use physical aides demonstrate a degree of relationship. Similarly, organizational factors exhibit a tendency to be related to instances where residents attend to the physical environment and communication issues ( Appendix 3, Fig. A1A).

The five most weighted incident types and contributing factors were plotted in asymmetric plots ( Appendix 3, Fig. A2). The angle between incident types and contributing factors represents the strength of the relationships with acute angles indicating a stronger relationship. Acute angles in Fig. A2A indicate that there are strong relationships between clinical processes and recent medical changes, between outdoor and communal area falls and work environmental factors (e.g. light and noise) and resident not using physical aide, and between falls involving wheelchair and toilets and resident communication issues. In  Appendix 3, Fig. A2b also shows strong relationships between falls involving wheelchairs and loss of balance as well as communication issues, resident physical and psychological health, and falls in the bedroom and resident attending to the physical environment.

Discussion

Statement of principle findings

While falls are common in RACFs, little research has been conducted in this setting to understand how and why falls occur [21]. In this study, we applied the modified ICPS for the first time to classify fall incidents using data from 22 Australian RACFs. The results showed that personal factors such as resident pre-existing psychological or physical health were the most common contributing factors to falls assessed as requiring hospitalization. In addition, the correspondence analysis highlighted the role of environmental factors (e.g. lighting in communal spaces) and organizational factors (e.g. workforce and supervision) in common incident types. These results provide new information about the mechanism of falls in RACFs and may be valuable in informing optimized fall prevention and intervention programmes for improved quality-of-life outcomes of RACF populations.

Strengths and limitations

The application of the ICPS in RACFs is an innovative way to understand falls in neat categories at a high level. This analysis is significant as it is the first, to our knowledge, to apply these methods in the RACF setting. This study is limited by its sample size, restraining further analysis of incident types of contributing factors at Level 3 in ICPS. The study is also limited by the quality of incident reports extracted from electronic health records and possible biases of staff completing incident reports. These factors may have resulted in some incident types and contributing factors not being captured, resulting in an under-reporting of our findings. However, additional quality checks and blinded reviews minimized bias of authors applying the ICPS.

Interpretation within the context of the wider literature

Direct comparison of results to other literature is limited. However, the results broadly echo previous studies demonstrating that individual, environmental, and organizational factors are all critical components influencing falls [22, 23].

Falls can result from a combination of intrinsic and extrinsic risk factors. Extrinsic risk factors encompass environmental and organizational elements, while intrinsic factors involve individual characteristics. Our study found that there is a strong relationship between falls involving wheelchairs, toilets, and resident communication issues. A systematic review exploring the relationship of falls and communication disability in adults hospital patients demonstrated that there is some evidence that communication disability is associated with an increased risk of falls and suggested that communication functions should also be included on falls risk assessment checklists and falls prevention tools although further research is needed [24]. A recent study of community-dwelling adults aged ≥55 years found that less than one-third knew of the potential effects on some medications in increasing the risk of falls [25]. It is important that residents have an understanding of the fall risks associated with the medication they take.

Implications for policy, practice, and research

The results suggest that multifactorial falls prevention interventions may be the most appropriate to address the range of underlying factors contributing to falls. For example, measures at the individual level (e.g. staff assistance in transfer and close supervision, routine check on medication regimens, and falls education to increase compliance) might be more effective in preventing falls if they are appropriately integrated with efforts to reduce fall risk factors at the environmental and organizational level (e.g. minimizing trip hazards in the bedroom, designing purposively designed communal areas with appropriate lighting, flooring, and sound, and maximizing face-to-face care from nursing and care staff). Evidence on multifactorial, environmental, or organizational interventions to prevent falls in RACFs is limited. Instead, evidence currently focuses on individual factors such as exercise [26]. The lack of evidence on actionable next steps limits the immediate practicality of incident analysis results to frontline clinicians and providers and instead may signal to researchers’ important areas for future falls prevention areas in RACFs. As this study identified environmental factors in falls, practical considerations of the design of RACFs should take place. Adoption of best practices in the design of facilities would help to minimize potential factors causing falls. Such best practices could include planning or designing of new facilities or refurbishment of current ones to consider choice of floor finishes, selection of furniture, familiar layouts, and appropriate lighting [27]. An outcome of this study was the demonstration that it is feasible to applying the ICPS to learn from incident data within RACFs. One of the main barriers to reporting incidents is the time it takes to report [28] and so adding extra ICPS classification fields for all falls may be counter-productive. Additionally, training all staff on the use of the classification would be a significant burden. However, aged care organizations could classify serious incidents such as falls resulting in hospitalization and a representative sample of lower-harm incidents to characterize the main contributing factors and to inform and devise preventive strategies.

Conclusion

Understanding of fall mechanisms is essential to drive needs-based practices of falls prevention and interventions. In this application of the modified ICPS, we demonstrated that personal factors were the most common contributing factors to falls assessed as requiring hospitalization and that other common reasons including environmental and organizational factors are also critical in affecting fall events. While some personal factors (e.g. pre-existing medical conditions) are not immediately modifiable to prevent a fall, the results from this study could be used to target falls prevention interventions. Further research is needed to develop a more comprehensive understanding of the mechanisms of falls in RACFs and how to prevent them. This could also include the linkage of aged care facility data with hospital data to allow a more comprehensive examination of fall-related factors and outcomes.’

Acknowledgements

The authors would like to thank the partners and collaborators of this project including Anglicare, Northern Sydney Local Health District, Sydney North Primary Health Network, the Deeble Institute for Health Policy Research, and the Australian Aged Care Quality and Safety Commission.

Author contributions

K.S., A.N., and P.H. conceptualization the present study. K.S., I.M., S.S., P.P., and A.N. completed formal analysis. The writing—original draft by K.S., I.M., S.S., P.P., and A.N. The manuscript Writing—review & editing by J.W., G.H., and P.H. Funding acquisition was secured by J.W.

Supplementary data

Supplementary data is available at INTQHC online

Funding

This work was supported by the National Health and Medical Research Council in partnership with Anglicare (APP1170898). J.W. was supported by an National Health and Medical Research Council Elizabeth Blackburn Leadership Fellowship (1174021).

Data availability statement

The data underlying this article cannot be shared publicly due to the sensitive and personal nature of resident incident information.

Ethics and other permissions

Ethical approval for data use was obtained from the Macquarie University Human Research Ethics Committee (reference number: 52019614412614).

Appendix 1: International classification for patient safety

The following incident types and contributing factors are based on the ‘Operationalising the World Health Organization’s International Classification for Patient Safety: Data Dictionary Version 2’ with modification added to support the application to aged care context.

Appendix 2: Level 3 incident types and contributing factors

Appendix 3: Correspondence analysis

References

Author notes