Appropriate covariates were chosen from the sociodemographic, functional, and clinical items available from the interRAI-AL assessment and from the linked administrative health data.
40
Certain facility-level variables relevant to medication oversight were also considered for descriptive purposes. Based on previous publications from the ACCES study as well as insight from the literature, items considered for inclusion in the multivariable models included age, sex, comorbidity, number of distinct medications used, and history of health system use (hospitalizations, Emergency Department visits) in the year prior to the date of baseline assessment. The findings of the descriptive analyses were used to inform which covariates were considered for inclusion in the model.
There has been some evidence from previous publications that age greater than 90 years112 and male sex149,150 may be associated with risk of hospital admissions. Based on these findings, and on the importance of these variables for most outcomes in older individuals, age and sex were included as covariates in the models. As in a previous ACCES publication,112 age was coded into the following categories: 65-79, 80-85, 86-89, and ≥90 years.
Comorbidity has been shown to be associated with hospitalization risk in previous ACCES publications,112 as well as other studies of AL148and LTC 144,151 residents. In addition, comorbidity is highly associated with frailty status, and it is likely that comorbidity may be independently associated with the medication exposures of interest. Given the fact that comorbidity is a risk factor for hospitalization and is also associated with frailty status, multivariate models considering frailty as the exposure of interest (objective 2a) included adjustment for comorbidity level. Comorbidity was measured as the sum of the diagnoses documented on the interRAI-AL instrument. Consistent with previous ACCES
publications,112,120 the comorbidity score considered 49 possible diagnoses, and was coded into the following three groups: 0-3, 4-5, and ≥6 chronic conditions.
41
For multivariable models in which medication use was the exposure of interest (objectives 2b and 3), specific diagnoses relevant to the use of the medication class were adjusted for, rather than a measure of comorbidity. Inclusion of specific diagnoses, rather than comorbidity, allows for more direct adjustment for conditions relevant to both the medication use and the risk of hospitalization. In particular, this is relevant for insulin and oral antidiabetic agents, since only those with diabetes are eligible to use these medications. Additionally, these steps avoid the risk of masking the impact of frailty as an effect modifier, as the Full FI includes many of the same diagnoses included in the comorbidity measure. However, there was adjustment for medication number, which can be considered a proxy measure for comorbidity number, and is described in more detail below.
Previous publications from the ACCES study112 and elsewhere149,150,152 have identified an association between higher numbers of medications used and risk of hospitalization. There is also likely an association between use of specific medications and overall medication use. For models in which the exposure of interest was the use of certain medication classes (objectives 2b and 3), there was adjustment for ‘number of medications’, subtracting the medications which fall into the medication class of interest. The medication number variable was coded into the following four categories: 0-6, 7-8, 9-10, and ≥11 medications, consistent with a previous ACCES publication.112
Previous hospitalizations have also been identified as associated with risk of hospitalizations in AL112 and LTC 146,150,152 residents. History of hospitalizations may also be associated with the exposure of interest, use of high-risk medications (e.g., if previous hospitalizations were due to adverse drug events, or if new medications were prescribed during the hospital stay). Thus, the multivariate models included adjustment for previous hospitalizations. History of health system
42
use was determined using the linked health service utilization data for the year prior to baseline assessment. As in a previous ACCES publication,112 hospitalizations in the past year were coded into the following three groups: 0, 1 or ≥2 inpatient hospital admissions in the past year.
Additional variables which have been explored as predictors of hospitalization from DAL in previous ACCES publications were also considered as potential covariates, including marital status; strength of social relationships; fatigue; time involved in activities; and cognition and function, as determined by the interRAI Cognitive Performance Scale (CPS) and Activities of Daily Living (ADL) score. However, these variables were not included as covariates in the final multivariable models due to their importance as components of frailty measures (e.g., strength of social relationships, activity involvement, and cognition in the Full FI; function in CHESS and the Full FI; fatigue in CHSrel).
For descriptive purposes, associations between facility factors and frailty/medication use status of residents were investigated. In particular, factors related to skilled care and oversight of medication use were considered in order to explore the distribution of frailty and high-risk (HR)/antipsychotic medication use based on these measures of resident oversight. In previous ACCES studies112,120 level of skilled staffing was represented by the presence of Licensed Practical Nurse (LPN) or Registered Nurse (RN) staff on site, and coded in the following three categories: neither on site, LPN/RN on site <24/7, and LPN/RN on site 24/7. Similarly,
affiliation of a physician with the facility (especially, presence of a physician office on site) was considered, coded into the following three groups, as in a previous ACCES study:112 no physician affiliated with site, affiliated physician with no office on site, and affiliated physician with office on site. Lastly, involvement of a pharmacist in the facility over the past month was
43
considered as the following three categories: no pharmacist involvement, pharmacist consultant, and pharmacist on staff.