Statistical methods
All medicines were aggregated per patient according to the fifth level
codes (substance) in the Anatomical Therapeutic Chemical (ATC)
Classification System.9 Based on these aggregated
data, the medicine changes were computed from the differences in the
prescribed medicines at the different time points (baseline, after first
visit, follow-up after 4 months, and follow-up after 13 months). If a
medicine was still discontinued or prescribed with precisely the same
dosage after the first visit and at a follow-up visit, then the
medicine change at the first visit was described as persistent. Overprescribed medicines were defined as medicines that were
discontinued or reduced in dosage at the first visit in the outpatient
clinic. Underprescribed medicines were defined as medicines that
were prescribed or increased in dosage at the first visit in the
outpatient clinic. Rebounds were defined as medicines that were
prescribed again following discontinuation.
The number of medicines prescribed at baseline was summarized according
to the therapeutic subgroups (second level ATC). The number of
underprescribed and overprescribed medicines were then plotted as a
function of the total number of medicines at baseline with trend lines
using loess regression.10 The medicine changes per
group at the first visit were summarized descriptively along with the
persistence of these changes. To identify the medicines that were more
often discontinued during the medication review, we calculated the
absolute difference in the proportions of discontinuations per medicine
between groups. Only medicines prescribed to at least ten patients at
baseline were included in the calculation. Rebounds were summarized for
medicines with at least five discontinuations during the first visit. To
compare the proportion of discontinuations and rebounds between groups,
we plotted the ratio of the number of medicines prescribed at each time
point to the number of medicines prescribed at baseline for
pharmacological subgroups (third level ATC). Only subgroups with at
least 40 medicines prescribed in both groups at baseline and at least 10
discontinuations in the medication review group during the first visit
were included.
The reason(s) for discontinuations in the medication review group were
registered prospectively, and based on these the primary reason
for discontinuation was determined using the following hierarchy: 1)
Treatment not indicated; 2) Treatment with no or poor effect; 3)
Safety-related issues; 4) Patient preferences and circumstances; and 5)
Unknown reason.
Lastly, to identify factors related to the number of overprescribed
medicines, we created two exploratory models using all the subjects from
the medication review group. One model included all patient baseline
characteristics (to identify patient-related factors) and the other
included all medicine groups (ATC fourth level, chemical subgroup)
prescribed at baseline (to identify medicines that were associated with
overprescribing). As the predicted variable was a count of
overprescribed medicines, we fitted generalized linear models with a
quasi-Poisson distribution (log link) using R version
3.6.311 with the tidymodel12 and
poissonreg13 packages. For both models, we first
fitted a full model using all variables excluding variables with
near-zero variance. The statistically significant variables (defined as
P < 0.05) from the full models were then further explored in
univariate models. The models were purely exploratory and confidence
limits and P-values were not adjusted for multiple
comparisons. To illustrate the results
from the models, model predictions using estimated marginal
means14 were plotted for all patient baseline
characteristics that were statistically significant in both the full and
univariate models.