Skip to main content
. Author manuscript; available in PMC: 2021 Oct 1.
Published in final edited form as: Pharmacotherapy. 2020 Sep 2;40(10):992–1001. doi: 10.1002/phar.2452

Table 4:

GEE model output on all-cause 30-day acute care utilization post-index by Beers criteria compliance

Characteristics* Adjusted RR** 95%CI P-values

Any PIM combination (ref: no PIM) 1.23 1.02 1.48 0.032
Female sex 1.01 0.89 1.15 0.881
Age group  
65–74 (ref)  
75–84 1.19 1.02 1.40 0.031
85+ 1.57 1.26 1.95 <.001
Charlson Comorbidity Index  
3 (ref)  
4 1.17 0.96 1.42 0.113
5+ 1.76 1.47 2.10 <.001
History of mental health conditions  
Mood disorders 1.13 0.88 1.46 0.335
Psychoses 1.65 1.26 2.15 <.001
Behavior disorders 0.93 0.64 1.35 0.698
History of ER visits** 1.64 1.22 2.20 0.001
History of admission** 1.51 1.00 2.28 0.052
Hydrocodone (ref)  
Oxycodone 0.94 0.70 1.27 0.687
Codeine 1.05 0.89 1.23 0.554
Others 1.28 0.99 1.66 0.059
Every 10 mg OME increase 1.01 1.00 1.02 0.027

Abbreviations: GEE: generalized estimating equations; RR, relative risk; CI, confidence interval; PIM, potentially inappropriate medication; ER, emergency room; OME, oral morphine equivalent.

*

Due to low event per category in dental procedure variable, we did not include the dental procedures into the model to avoid sparse data bias and its resulting considerable upward bias.

**

Adjusted RR using Proc GenMod – Modified Poisson regression model Poisson regression without robust error variances may result in a conservative CI (i.e., wider CI). A “modified Poisson” method has been proposed to estimate the RR using a robust error variance (Zou 2004). This method leads to the robust error variance estimation and produces 95% CIs with the correct coverage.

Using AR(1) correlation structure