Dissecting interRAI Instrument Data Using Visual and Predictive

Analytics

Waqar Haque

, Shannon Freeman

and Piper Jackson

Department of Computer Science, University of Northern British Columbia, Prince George, Canada

School of Nursing, University of Northern British Columbia, Prince George, Canada

Department of Computing Science, Thompson Rivers University, Kamloops, Canada

Keywords: Visual Analytics, Predictive Analytics, Geriatrics, Long-term Care, interRAI Assessment Tools.

Abstract: Healthcare data for older adults is often collected through globally standardized instruments and resides in

multiple disparate database systems. For gaining insights into this data, an interactive platform has been

developed which allows visualization of several actionable key performance indicators along multiple

dimensions. The health assessment data was collected from persons receiving community home care services

as well as from persons residing in long-term care facilities. The top-level reports provide aggregations across

geographical regions at the health service delivery area level with capability to drill down to finer granularity

for metrics of interest. By revealing hidden patterns embedded in data, the stakeholders can make informed

decisions pertaining to resource allocation and better patient care. The drill-down and drill-through reports

include demographics, quality of life, medications, health conditions and disease diagnoses, comorbidities,

health service usage, and patient journey across care settings. A predictive model to accurately estimate

resource requirements at the time of admission was also developed for data-driven triaging.

1 INTRODUCTION

In recent years, visual analytics has quickly become a

staple in various data-rich fields, including health

care. This has undeniably led to economic and other

benefits by unveiling hidden information and/or

patterns in data thus leading to informed

administrative decisions, overall better treatment, and

more desirable outcomes for patients. interRAI

(interRAI, n.d.) is a collaborative global network of

researchers and practitioners in over 35 countries who

have developed health assessment instruments. These

instruments have been mandated for use by several

governments, including Canada where the interRAI

MDS 2.0 (Morris, et al., 1990) and the interRAI

Home Care (Morris, et al., 1997) assessment

instruments are widely used for persons receiving

long-stay community based home care services and

for those receiving support in a long-term care

facility. These instruments are based on rigorous

research which allows standardized assessment

https://orcid.org/0000-0002-6921-8097

https://orcid.org/0000-0002-8129-6696

https://orcid.org/0000-0002-7025-5063

protocols and outcome measures. The data used in our

research and the selected key performance indicators

(KPIs) are based on these instruments. This research

also engaged patient partners for valuable input when

selecting relevant metrics and assessing ease of use.

The outcome from this research is twofold. First,

the developed platform demonstrates how data

visualization can inform clinical and health systems

in decision-making for persons receiving care in

home and/or long-term care settings. The reports

present aggregated results based on analyses

conducted on de-identified secondary data. Second, a

predictive model has been developed to predict

resource requirements when an elderly patient is

admitted to a care facility.

2 RELATED WORK

A diverse number of studies where results were

presented using visual platforms revealed patterns

110

Haque, W., Freeman, S. and Jackson, P.

Dissecting interRAI Instrument Data Using Visual and Predictive Analytics.

DOI: 10.5220/0011719100003476

In Proceedings of the 9th International Conference on Information and Communication Technologies for Ageing Well and e-Health (ICT4AWE 2023), pages 110-117

ISBN: 978-989-758-645-3; ISSN: 2184-4984

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

which were otherwise indiscernible. For instance, the

application of visualization techniques to a collection

of home care patient datasets led to the discovery of

twenty-one significant attribute correlations in

admission and discharge data gathered from a total of

988 patients across fifteen distinct home health

agencies. Visualizations selected included histograms

and heatmaps due to their shared ability to

accommodate multiple dimensions and the relative

ease with which one can identify patterns embedded

therein. Some of the hypotheses generated upon

evaluation of these visualizations were later validated

via statistical analysis techniques (Monsen, Bae,

Zhang, & Radhakrishnan, 2016). While the study was

based on a very small number of patients, such a

correlation could be used to improve upon various

aspects of home health care (such as anticipating a

longer episode and/or hospital stay length in patients

with urinary incontinence) ultimately improving

patient quality of life and reducing health care costs.

Another study found key predictive indicators of

transferring persons with dementia into long term

care from less dependent forms of care (Cepiou-

Martin, Tam-Tham, Patten, Maxwell, & Hogan,

2016). This study involved a meta-analysis of

longitudinal data and was primarily conducted via

statistical analysis, though tables and funnel plots

were used to organize and obtain results. The results

reinforce the potential significance of information

that has yet to be extracted from otherwise dormant

data. The indicators include severity of dementia,

exhibition of specific and/or worrisome behaviours,

general ability to carry out activities of daily living,

and ethnicity.

An ambitious project surrounding a hospital in

India was undertaken with the intention of improving

effectiveness, efficiency, and cost of care (Menon,

Aishwarya, Joykutty, Av, & Av, 2021). The project

aimed to replace the hospital’s existing visualization

tool (Tableau) with a free alternative, constructed

using open-source tools, and capable of data pre-

processing, data visualization, and predictive

modelling via a web portal. Difficulties experienced

by the project team include issues with the sample

data provided by the hospital, such as missing/

corrupted/invalid attribute values, which had to be

dealt with during the data pre-processing stage.

Despite difficulties, the tool was found to be “useful

for … understanding of trends and patterns” (Menon,

Aishwarya, Joykutty, Av, & Av, 2021).

A visualization platform displaying various

metrics of interest related to the spread of CoViD-19

in the state of Indiana, USA, saw not only rapid

development, but also rapid implementation (Dixon,

et al., 2020). Data was retrieved through region-

specific health information exchanges and laboratory

testing sources and employed automated pre-

processing techniques. Figures across both the public-

facing and the closed versions were relatively diverse,

and included stacked bar charts, line charts, tables,

and even geographic heat maps. Figures were also

particularly feature-dense, with many offering the

ability to “drill down” to finer granularity. Extensive

practical usage and feedback suggest that the platform

has “provided essential and otherwise unavailable …

data to inform key decision makers … and enable a

data-driven strategy to a state-wide response” (Dixon,

et al., 2020).

A study with a focus on reducing crowding in

emergency rooms demonstrated the significance of

classification of patients in terms of acuity of care and

adjusting and/or prioritizing resources accordingly

(Khalifa & Zabani, 2016). In particular, the study

selected two key performance indicators: emergency

room length of stay (ERLOS) and percentage of

patients leaving without treatment. The indicators

were chosen for their perceived ability to accurately

convey ER efficiency and effectiveness, respectively.

The application of descriptive (visual) analytics

techniques led to the conception of two procedural

adjustments: fast-tracking patients with relatively less

serious and/or time-sensitive ailments, and the

addition of an internal waiting room to accommodate

those who were able to stand, increasing overall bed

capacity. Ultimately, the study led to significant

reduction in both ERLOS and percentage of patients

leaving without treatment, as well as suggestions for

additional indicators and procedure changes to further

decrease crowding.

3 METHODOLOGY & RESULTS

3.1 Data

The data used in the development of our platform was

provided by two provincial health authorities as two

separate Microsoft SQL Server (MSSQL) databases.

We will henceforth refer to these as databases A and

B. The databases contained data collected and stored

via Cerner (Cerner, n.d.), DAD (Discharge Abstract

Database Metadata (DAD), n.d.), Procura (Procura

Home Health Software, n.d.), and Meditech

(MEDITECH EMR Software, n.d.) systems. Despite

minor differences, the semantics (i.e., the attributes)

remained relatively identical across the two

databases. No personally identifiable attributes were

retained.

Dissecting interRAI Instrument Data Using Visual and Predictive Analytics

111

A set of SQL views was created to “filter” data

without making any permanent alterations to the

tables themselves. This pre-processing step involved

data cleansing such as the removal of patients under

age 50 and patients whose records contained oddities,

such as invalid discharge dispositions and

inconsistent personal information. Some data types

were altered for performance reasons and views were

created for easy access to frequently desired record

subsets. Similarly, when constructing datasets for

chronological metrics in reports, records were

restricted to those dated between 2010 and 2019

because the data outside of this interval was relatively

sparse and could potentially skew reported numbers.

The databases were stored on a secure server local

to the research institution yet on a domain separate

from that of the research lab and disconnected from

the Internet entirely. The secure server was accessed

via a complex authentication protocol. The platform

itself was developed on our research lab domain,

which included various general-purpose client

machines, a file server and a Network Access Storage

(NAS). The software used included Microsoft’s web

development and business intelligence tool stack

consisting of SQL Server (Microsoft, n.d.), SQL

Server Reporting Services (SSRS) (Microsoft, n.d.)

and SQL Server Data Tools (SSDT) (Microsoft, n.d.),

in addition to Visual Studio and the ASP.NET

framework (Microsoft, n.d.). QGIS (QGIS, n.d.) was

used to develop dynamic maps displaying acute care

admissions and assessments by various health

regions.

An agile development process was followed

using an integrative knowledge translation approach

with a patient-oriented research framework in which

patient partners and representatives of the health

authorities remained engaged at all stages of the

research process.

3.2 The Data Visualization Platform

The reports are grouped by care setting, namely

Home Care (HC) and Long-Term Care (LTC), with a

third group consisting of other reports which fall

outside of these categories. The stakeholder-facing

landing page provides a synopsis of aggregative

statistics, such as the total number of unique patients

across HC and LTC including the number of

assessments, age group at time of assessment, and the

average length of stay (in months) by setting (Figure

1). Additionally, this page allows for navigation to

other reports such as geo-mapping and predictive

modeling interfaces via a user-friendly menu. Various

drill-down and drill-through reports provide

information at finer granularity and details,

respectively. Tooltips are used throughout the

dashboard to popup additional relevant information.

For a meaningful analysis, data is normalized, where

applicable/possible.

Figure 1: The Landing Page.

Figure 2: Demographics Report.

3.2.1 Demographics

The demographics report header displays the number

of total unique patients and assessments for the

selected care type and year interval (Figure 2). A

trend chart demonstrates the number of admissions

over the selected year/interval. Other charts in this

report include patient numbers (and proportions) by

gender, age group and marital status (at first and most

recent assessment dates), reason for care type referral

and assessment, and overall change in care needs.

ICT4AWE 2023 - 9th International Conference on Information and Communication Technologies for Ageing Well and e-Health

112

Tooltips allow viewing of actual patient counts,

where applicable.

An interesting trend illustrated in this report is

the change in numbers of patients over the years. For

instance, a significant (roughly 50%) decrease in HC

admissions was observed in Database A between

2010 and 2013, with a gradual crawl upwards

thereafter. Database B, however, showed the number

of admissions halved from 2010 to 2011, and

remained relatively static thereafter. The aggregate

marital statuses remain relatively stable suggesting a

stable ratio of new and recurring patients. One

potentially actionable trend is the skewing of patient

counts by gender. In both databases and both modes

of care, patient gender ratios skew slightly towards

male in patients aged 80-89 years (especially in the

younger age ranges), only to skew (often

dramatically) towards female thereafter. This is likely

a reflection of the differences in general life

expectancies between the two populations and may

support gender crossover as has been noted in

previous studies of the oldest old (Freeman, Hajime,

Satoru, & Masahiro, 2009)

3.2.2 Home Care: Client Quality of Life

The Quality of Life reports (Figure 3) have the option

to dynamically regenerate the figures using data from

a selected year/interval. These reports are broken

down into four distinct, horizontally divided groups:

Sensory & Cognitive Conditions, Physical

Functioning, Pain, and Mood. The selection of

metrics therein was based upon data released by the

Canadian Institute for Health Information (CIHI),

which showed how “long-term care homes are

performing on nine [specific] indicators”

(McCormick Home, n.d.).

In general, the Sensory & Cognitive Conditions

sections include tables displaying patient counts

divided according to various levels of visual,

auditory, and memory impairment. The Physical

Functioning sections display percentages of patients

divided according to various metrics representing

physical assistance needs and levels of ability to carry

out Activities of Daily Living (ADLs). The Pain

section touches on patient data regarding pain

frequency, presence of pressure ulcers, and frequency

of falls. Lastly, the Mood section displays patient data

regarding mental health related attributes, such as

frequency of feelings of sadness or depression,

patient loneliness, and frequency of verbally abusive

behavioral symptoms. The charts illustrate that the

number of clients with adequate vision was typically

greater than the sum of all clients with any sort of

visual impairment, regardless of chronology or mode

of care. The number of clients with both short term

and procedural memory impairments was also often

greater than the sum of clients with any single type of

memory impairment, or none at all. As expected, the

number of clients needing assistance to perform

ADLs increases significantly from HC to LTC. In

what is likely a testament to quality of care, in

Database A the number of clients reporting feelings

of sadness or depression seems to decrease from HC

to LTC; in Database B, this trend is not seen.

Similarly, aggregate reports of pain in any capacity

decrease dramatically from HC to LTC, often by

more than 15%.

Figure 3: Quality of Life (Home Care).

3.2.3 Conditions and Diseases/Medications

In the Conditions and Diseases report (Figure 4), a

stacked bar chart displays the top ten most prevalent

diseases. The stacked values represent patient counts,

divided according to whether their diagnosis is

actively being monitored and/or treated by home care

professionals. The other charts on this report display

more specific disease/condition-related patient

counts for incontinence, nutrition, and certain

dermatological issues, each with varying degrees of

information. The top three most prevalent diseases

across various chronologies and modes of care were

Arthritis, Dementia, and Hypertension, though the

Dissecting interRAI Instrument Data Using Visual and Predictive Analytics

113

ordering of these variables varies (e.g., Hypertension

is typically the most common disease across all times

and care types, though, at a notable number of points

in time, Dementia is instead the most common disease

in LTC facilities).

Figure 4: Conditions and Diseases Report.

The Medications report (Figure 5) provide the

average number of medications taken for various

psychiatric diseases together with frequency of

consumption. Two other charts in this report show the

average weekly time spent by patients in different

types of therapy. These charts are subdivided by

various kinds of therapy and/or rehabilitation. Lastly,

two bar charts demonstrate both the numbers and

proportions of hospital, emergency room, and

physician visits.

The average amount of therapy received varied

greatly from year to year, with therapies common in

one year being sometimes almost non-existent in the

next. The average numbers of medications (around

nine) do not trend in any particular direction when

traversing across age groups. Oddly, in years where

average medication counts do not follow this trend,

HC clients in older age groups tend to have fewer

average number of medications than those in younger

age groups.

3.2.4 Acute Care (DAD/MEDITECH)

The Acute Care reports (Figure 6) provide

visualizations of data collected through two

acute/hospital care EHR systems: DAD and

Meditech. These reports include trend charts

displaying the number of admissions by month,

gauges demonstrating the average lengths of stay in

both acute and alternative level of cares, and the

numbers of admissions for each major clinical

category. Other figures found in these reports include

Figure 5: Medications Report.

a bar graph demonstrating the top five facilities by

admissions, and a gauge directly comparing the

estimated and actual lengths of stay (in days). The bar

graph allows one to drill down into a sub-report for

information at a finer granularity (Figure 7).

Additional figures illustrate the most frequently

utilized patient services within the chosen facility, and

the number of transfers from other facilities. Other

metrics reported here are similar to the parent report,

but present values for the selected facility only.

Figure 6: Acute Care.

Perhaps the most intriguing and actionable metric

to originate from these reports were the charts

demonstrating the numbers of admissions by month.

Over the year(s), selected trends that could lead to

improved care by better anticipating acute care

capacity month by month were demonstrated. For

example, regardless of care type, the chart clearly

demonstrates that acute care admissions are lowest in

December, and seem to peak approximately once per

quarter, with differences in peaks and dips consisting

ICT4AWE 2023 - 9th International Conference on Information and Communication Technologies for Ageing Well and e-Health

114

Figure 7: Acute Care drilldown.

of up to hundreds of patients (in HC). In terms of

clinical categories at times of admission, diseases of

the digestive system seem to be the most common

reason.

3.2.5 Other Reports

The dashboard contains several other reports of

interest, including a HC to LTC Transition report, a

Patient Journey report, and a Comorbidities report.

The Transition report (Figure 8) provides a relatively

direct comparison between HC and LTC statistics

including patient and assessment counts, top three

ADLs requiring assistance, and the estimated and

actual lengths of stay in acute care. The report also

illustrates the average duration of wait times when

moving from one form of chronic care to another. An

interesting observation revealed that, on average,

actual acute care times for patients of either care type

are anywhere from 3.5-5 times longer than the

estimated times. The report also contains some

interesting information regarding transfer times

between HC and LTC: although the most common

wait time is 30 to 60 days, nearly 10% of clients wait

a year or more to transfer.

The Patient Journey report is a chronological

representation of events as clients transition from HC

to LTC. The statistics on this timeline include the

average number and duration of service encounters,

ER visits, the most common mode of admission, and

the average length of stay in a specific care type. The

average duration between the last HC assessment and

first LTC assessment was approximately five months.

Interestingly, the average number of ER visits seems

to drop by 15% between HC and LTC. Similarly, the

average number of service encounters among LTC

patients is roughly 25% lower than of HC patients.

Lastly, the interactive Comorbidities report

(Figure 9) allows a user to select up to three distinct

diseases for which patient counts are generated. The

counts are broken down by both gender and care type.

For instance, one noteworthy comorbidity is

osteoporosis and fractures. A large portion of female

HC clients have osteoporosis and approximately 10%

have experienced a hip fracture while approximately

17% have suffered from another form of fracture

(wrist, vertebral, etc).

Figure 8: HC to LTC Transition.

Figure 9: Comorbidities Report.

Dissecting interRAI Instrument Data Using Visual and Predictive Analytics

115

Figure 10: Resource requirements estimated by predictive model.

4 PREDICTIVE MODELING

Estimating the amount of care needed by a patient is

a complex problem and usually requires

experienced

professionals. Given the recent nursing shortages and

early retirements, it is challenging to find an

experienced nurse to conduct such triaging.

Additionally, errors in judgement can result in

negative impact on both the patient and the healthcare

system. For instance, we noticed variations of over

500% in estimated and actual acute care length of

stay. Thus, it is desirable to have a tool which could

more precisely estimate the amount of care a patient

might need based on their current condition. This

could also result in data-driven resource allocation.

To estimate the amount of care a patient might

need, we use the fields from MDS (Minimum Data

Set) forms such as ADL scores, patient abilities and

pre-existing conditions. A predictive model then runs

in the cloud and renders results through a simple web

interface. The technical details of our model are

beyond the scope of this paper. However, the steps

involved include data pre-processing, cleansing,

performing imputation on categorical variables,

identification of features correlated to target variables

using statistical tests, and training and testing of the

model. To keep the complexity of the model at

manageable levels, twenty features (Figure 10) were

selected using IBM SPSS modeler (IBM, n.d.) feature

selection algorithm. Each feature and its

corresponding values were named in accordance with

the nomenclature provided by MDS forms. A web

form was developed to interact with the model hosted

in the cloud. Once the information about a patient’s

current condition is entered and submitted, the web

form sends a request to an API (Application

Programming Interface) which processes the

information, converts it to an appropriate format and

posts to Watson Studio (IBM, n.d.) cloud. The model

stored in the cloud processes and returns the

predictions to the web form via the same API. The

entire process occurs in real-time in a matter of

seconds. Figure 10 also illustrates an example where

estimated resources (hours per week) are instantly

displayed on the bottom of the web form for the

selected values of the features.

5 CONCLUSION

Even when data is collected via standardized forms,

it poses challenges from quality to comprehension.

To extract value from the collected data, it must be

pre-processed and converted to a form suitable for

visual analytics while maintaining the privacy and

confidentiality of patients and facilities. The data

obtained from two health authorities was extracted,

cleansed, and placed on secure servers. Actionable

KPIs were identified for visual analytics of this data.

For fair comparison, reports were normalized, where

applicable. Displaying data on interactive maps

provide a different perspective.

The visual analytics platform enables

observation along multiple dimensions which could

be used for more informed resource allocation and

improved patient care. The reports reveal hidden

ICT4AWE 2023 - 9th International Conference on Information and Communication Technologies for Ageing Well and e-Health

116

patterns in the data which could be used for informed

decision making and better patient care. For instance,

an important observation was the incorrect estimate

of length of stay and resources in acute care. This

results in unforeseen burden on the healthcare system.

To rectify this problem, a predictive model was built

to estimate the resources more accurately at the time

of admission. This provides a data-driven approach to

resource allocation. The quality of life metrics assist

in early detection of conditions thus affording the

opportunity for addressing situations before they

progress to an unmanageable state. The patient

journey gives a synopsis of the time and resources

that are required by patients as they transition from

homecare to long-term care.

ACKNOWLEDGEMENT

This research was funded by a grant from BC

Academic Health Science Network (AHSN) under

Strategy for Patient-Oriented Research (SPOR)

(Methods Cluster) and resulted in training of more

than a dozen HQPs. In addition, we acknowledge our

patient partners (B. Baker, G. Kramer, I. Muturi, S.

Prior, and C. Zannon) who remained engaged

throughout this research and provided valuable

feedback.

REFERENCES

Cepiou-Martin, M., Tam-Tham, H., Patten, S., Maxwell, C.

J., & Hogan, D. B. (2016, Apr.). Predictors of long-term

care placement in persons with dementia: A systematic

review and meta-analysis. International Journal of

Geriatric Psychiatry, 31(11), 1151-1171.

Cerner. (n.d.). Retrieved March 21, 2022, from

https://www.cerner.com/

Discharge Abstract Database Metadata (DAD). (n.d.).

Retrieved March 21, 2022, from https://www.

cihi.ca/en/discharge-abstract-database-metadata-dad/

Dixon, B. E., Grannis, S. J., Tachinardi, U., Williams, J. L.,

McAndrews, C., & Embi, P. J. (2020). Daily

Visualization of Statewide COVID-19 Healthcare Data.

2020 Workshop on Visual Analytics in Healthcare, 1-3.

Freeman, S., Hajime, K., Satoru, E., & Masahiro, K. (2009).

Understanding the Oldest Old in Northern Japan: An

Overview of the Functional Ability and Characteristics

of Centenarians. Geriatrics & Gerontology

International, 10(1), 78-84.

IBM. (n.d.). SPSS Modeler. Retrieved from http://www-

01.ibm.com/support/docview.wss?uid=swg27050406

IBM. (n.d.). Watson Studio. Retrieved from

https://cloud.ibm.com/catalog/services/watson-studio

interRAI. (n.d.). Speaking One Language for High Quality

Care Worldwide. Retrieved 4 2, 2022, from

https://interrai.org/

Khalifa, M., & Zabani, I. (2016). Utilizing health analytics

in improving the performance of healthcare services: A

case study on a tertiary care hospital. Journal of

Infection and Public Health, 9(4), 757-765.

McCormick Home. (n.d.). CIHI Quality Indicators.

Retrieved March 21, 2022, from https://www.

mccormickcaregroup.ca/mccormick-home/resident-

care/cihi-quality-indicators-nov-2016/

MEDITECH EMR Software. (n.d.). Retrieved March 21,

2022, from https://ehr.meditech.com/

Menon, A., Aishwarya, M. S., Joykutty, A. M., Av, A. Y.,

& Av, A. Y. (2021). Data Visualization and Predictive

Analysis for Smart Healthcare: Tool for a Hospital.

IEEE Region 10 Symposium, 1-8.

Microsoft. (n.d.). ASP.NET Overview. Retrieved May 19,

2022, from https://docs.microsoft.com/en-us/aspnet/

overview

Microsoft. (n.d.). SQL Server 2019. Retrieved May 19,

2022, from https://www.microsoft.com/en-ca/sql-

server/sql-server-2019

Microsoft. (n.d.). SQL Server Data Tools. Retrieved May

19, 2022, from https://docs.microsoft.com/en-

us/sql/ssdt/sql-server-data-tools?view=sql-server-

ver15

Microsoft. (n.d.). What is SQL Server Reporting Services.

Retrieved May 19, 2022, from https://docs.mic

rosoft.com/en-us/sql/reporting-services/create-deploy-

and-manage-mobile-and-paginated-reports?view=sql-

server-ver15

Monsen, K. A., Bae, S. -H., Zhang, W., & Radhakrishnan,

K. (2016, July). Visual analytics for pattern discovery

in home care. Applied Clinical Informatics, 7

(3), 711-

730.

Morris, J. N., Fries, B. E., Steel, K., Ikegami, N., Bernabei,

R., Carpenter, G. I., . . . Topinková, E. (1997).

Comprehensive Clinical Assessment in Community

Setting: Applicability of the MDS-HC. Journal of the

American Geriatrics Society, 45(8), 1017-1024.

Morris, J. N., Hawes, C., Fries, B. E., Phillips, C. D., Mor,

V., Katz, S., . . . Friedlob, A. S. (1990, June). Designing

the National Resident Assessment Instrument for

Nursing Homes. The Gerontologist, 30(3), 293-307.

Procura Home Health Software. (n.d.). Retrieved March

21, 2022, from https://www.alayacare.com/procura

QGIS. (n.d.). Retrieved May 19, 2022, from

https://qgis.org/en/site/

Dissecting interRAI Instrument Data Using Visual and Predictive Analytics

117