Aged care health status algorithms - FAQs

Q: What do the algorithms do?

The two algorithms will identify health conditions using medication administration data and free text fields from electronic health records.

Q: What type of free text data should be used as input data?

The algorithm that uses free text was developed to read text fields in residential care electronic health records (EHR) where notes about the residents’ conditions and needs appear. The extract field will differ depending on which EHR software a facility uses.

Q: What data sources can the algorithms be used with?

These algorithms were developed for use with electronic data (EHR and medication administration records) from residential aged care facilities (also called nursing homes). They may not be appropriate or reliable to use with data from other care settings and also not appropriate to use for younger populations.

Q: How is the algorithm that uses medication administration data different from other algorithms that use medication data to identify conditions?

Our macro was developed specifically for older adults living in residential aged care, whereas other algorithms were developed for different care settings and populations, for example Rx Risk was developed for a broader adult population. Certain medications may be used for different indications depending on age, and you need to consider whether a general or geriatric-specific algorithm is most appropriate when planning your study.

Q: Why would I use EHR data and medication administration data to identify health conditions in people who use residential aged care if I already have conditions data from aged care assessments (e.g. Aged Care Funding Instrument (ACFI))?

Using all available sources of data will provide more complete identification of health conditions compared to using aged care assessment data alone. ACFI assessments are more likely to underreport certain conditions. This bias may not be a substantial issue if you are conducting a study on conditions that are associated with a high degree of functional impairment (with substantial incentives to record in aged care assessments).However for many common chronic conditions, for example hypertension, these conditions are much less likely to be recorded in aged care assessments despite their importance for resident care. To better understand which conditions are likely to be underreported in ACFI assessments compared to our algorithms please read our paper.

Q: What do I need to do in order to use the macros?

You will need to complete the access request form on this page and agree to the terms and conditions of use. Please note that the macros are freely accessible to researchers; if you work at a private for-profit company and would like to use the macros, please contact us directly at acer.chssr@mq.edu.au.

Q: What are the terms and conditions of use for the macros?

The macros are to be used only for research and quality improvement projects. You must cite the use of these macros when reporting results generated with their use. If you modify the macro you still must still cite the macro and note that you are using a modified version. The macros are not to be used to generate a profit. You are solely responsible for checking your data.

Q: Who can help me if I encounter a problem while using the SAS macros for the algorithms?

We expect that all users will be comfortable using SAS software. If you are a proficient SAS user and encounter issues with the macros or if you find ways to improve the algorithm (e.g. if you identify new false positive text strings in your data which did not appear in the data we used for developing the macro) then please contact us at acer.chssr@mq.edu.au.

Q: What data did you use to develop these algorithms?

We used electronic health records and electronic medication administration records. These records came from a single aged care provider with over 70 facilities in New South Wales and the Australian Capital Territory. For most of the facilities, the records spanned greater than four years.

Q: How did you select which text strings to flag for each condition?

A member of the research team with previous clinical experience in aged care parsed and classified every unique text string recorded as notes in the “special needs” (health conditions and needs) field of residents’ electronic health records. We then developed a SAS program that would flag the shortest part of each string (rather than using every single variation). While developing this program we manually reviewed records that had been flagged for a given text string to ensure that the strings we were flagging were in fact relevant. We checked for false positive text strings (if false positive strings were identified, the algorithm was updated to a more specific string or a subsequent data step was added to remove specific false-positives).

Q: How did you select which medications to flag for each condition?

A member of the research team with a background in pharmacy identified medications that have specific indications (particularly in the residential aged care setting) and are unlikely to be used off-label.

Q: What is the sensitivity and specificity of this algorithm for identifying each of the conditions?

We believe that the algorithm works well at identifying conditions where there is recorded evidence of a condition, but it cannot identify conditions that go unrecorded and untreated. Currently we don’t know the sensitivity or specificity of the algorithm. We believe that the specificity is likely to be high since the text algorithm has been checked for false positives, and we think it is unlikely that the medications in the medication algorithm are used off-label. We think that sensitivity varies depending on the condition; for example, we suspect that osteoporosis is unreported in the electronic health record and undertreated with medications since the prevalence of osteoporosis in our sample was less than half of what we expected based on other international studies in this setting. Conducting a validation study for each of the conditions would require obtaining all medical records for a large sample of aged care residents and carrying out full chart reviews, which we have not done to date. Despite this uncertainty, we are confident that this algorithm provides more complete identification of conditions compared to using ACFI assessment data alone, and we think this is an important advance in maximising the use of data in this setting to generate information on health status that is severely lacking for Australians who use residential aged care.