Organizational Context and Quality Indicators in Nursing Homes: A Microsystem Look

The association of organizational context with quality of care in nursing homes is not well understood at the clinical microsystem (care unit) level. This cross-sectional study examined the associations of unit-level context with 10 unit-level quality indicators derived from the Minimum Data Set 2.0. Study settings comprised 262 care units within 91 Canadian nursing homes. We assessed context using unit-aggregated care-aide-reported scores on the 10 scales of the Alberta Context Tool. Mixed-effects regression analysis showed that structural resources were negatively associated with antipsychotics use (B = −.06; p = .001) and worsened late-loss activities of daily living (B = −.03, p = .04). Organizational slack in time was negatively associated with worsened pain (B = −.04, p = .01). Social capital was positively associated with delirium symptoms (B = .12, p = .02) and worsened depressive symptoms (B = .10, p = .01). The findings suggested that targeting interventions to modifiable contextual elements and unit-level quality improvement will be promising.


Supplemental File S1 Risk adjustment process for the computation of risk-adjusted unitlevel quality indicators
The TREC program calculated risk-adjusted unit-level quality indicators (QIs) following the technical guide of the risk adjustment methodology developed by the Canadian Institute for Health Information (CIHI). 1 While the CIHI technical guide describes the risk adjustment process for creating facility-level QIs, the same method applies to the calculation of indicator results at other levels, such as corporations, regions, provinces and territories.The TREC program has used this method for the computation of unit-level QIs.The risk adjustment process compares the risk profile of the resident population on an individual care unit with the profile of a standard reference population and then modifies the QI results for that care unit, so it is relative to the standard reference population.
The process for calculating risk-adjusted unit-level QIs involves first determining the appropriate numerator and denominator at the unit level for each QI.Following this, four steps are undertaken: stratification, indirect standardization via logistic regression, direct standardization, and outlier trimming.

Stratification.
For each care unit, this step stratifies the unit population (if facility-level QIs were calculated, this would involve stratifying facility population) into three risk groups, or strata: high, medium and low.The risk group for each indicator are based on either a RAI-MDS 2.0 outcome scale (such as the Cognitive Performance Scale, Activities of Daily Living Long Form Scale) or the Case Mix Index (CMI).The observed (unadjusted) QI score for each risk group on the unit is calculated.
Indirect standardization using logistic regression.This step involves calculating an expected QI score for each risk group on the unit using a logistic regression model that adjusts for multiple resident-level risk covariates that are aggregated to the unit level.The parameters for the logistic regression models (one for each risk group) are calculated from the standard reference population and then applied to the data for each risk group from the unit.A performance ratio for each risk group is subsequently calculated by dividing the observed QI score for the risk group by the respective expected QI score.Finally, the adjusted QI score for each risk group is calculated by multiplying the performance ratio by the QI score from the standard reference population.
Direct standardization and creation of a single adjusted QI score.As each unit has its own unique distribution of residents across the three risk groups, this step modifies the adjusted QI scores to treat each unit as though it had the same distribution among the three risk groups (and the same as the standard reference population).The adjusted QI scores for each risk group are combined to create a single adjusted QI score for that unit.
Outlier trimming.The final step of the risk adjustment is to check the distribution of the adjusted QI scores.If the adjusted score for a specific unit is above (or below) the maximum (or minimum) unadjusted QI score across all the units that are being risk-adjusted, the unit's adjusted QI is "trimmed" to within 10% of the standard deviation of the unadjusted QI.
Supplementary Table 1 provides comprehensive details, including the numerator, denominator, stratification variable, covariates used for calculating each risk-adjusted unit-level QI included in this study.S1

Supplemental Table S2 Associations between quality indicators and Alberta Context Tool (ACT) scales: Results (regression coefficients and 95% CI) from two-level random intercept linear regression
For each QI, we used 2-level mixed-effect linear regression to examine the relationship of the QI with eight of the Alberta Context Tool scales, controlling for the clustering of care units nested within the same facility.We also controlled for covariates including facility size, ownership model, services/care programs, quality of improvement activities, total care hours per resident day, percent of care hours per resident day provided by care aides.Marginal R 2 was calculated, which focuses on variance of the outcome explained by fixed factors.2WeusedVariancenull -Variance model /Variance null to calculate marginal R 2 .Variance model is the residual variance of the full model that includes fixed effects of predictors and the random effect.Variance null is the residual variance of the null model that includes the intercept and random effect only with the random effect being constrained to be the same as that in the full model.Coefficients and 95% CIs of the covariates are presented in the Supplemental TableS2.