Routine prescription records from 11 nursing home organizations were used to map prescribing behavior over eight years. The dataset spanned January 2016 to March 2023 and included 17,738 residents who occupied dementia special care units for at least one day during the study window. Investigators reported an aggregate fall in the proportion of residents with at least one regular psychotropic prescription from 48.3% in 2016 to 42.7% in 2023. The decline was not dramatic, but it was consistent across the observation period. Quarterly point prevalences were calculated and confidence intervals derived using the Wilson score method to indicate temporal changes in prevalence.
Quarterly estimates distinguished between regular prescriptions and pro re nata (PRN) options. The source data permitted calculation of point prevalence on the first day of each calendar quarter and the authors provided reference prevalence values for select quarters aligned to regulatory events. The analysis used established statistical packages (SPSS and R) and graphical tools (ggplot2, GraphPad Prism) for visualization and inference.
The five drug classes evaluated were defined by ATC codes: antipsychotics (N05A), anxiolytics (N05B), hypnotics (N05C), antidepressants (N06A), and anti-dementia drugs (N06D). Antidepressants and antipsychotics remained the most frequently prescribed categories throughout the period. Antidepressants showed the largest absolute decline in prevalence over time according to the authors’ descriptive analyses. Anti-dementia drugs declined most in relative terms.
The study excluded anticonvulsants (N03A) from the primary analysis because these agents are commonly prescribed for conditions other than neuropsychiatric symptoms; the authors did report anticonvulsant prevalence in supplemental materials to assess possible substitution effects. Data extraction included ATC code, start and stop dates, administration frequency, dosing fields and whether prescriptions were regular or PRN. When EHR entries did not reliably capture care-type or admission period, those records were excluded, a factor that affected the available sample for some quarters.
The investigators segmented the study period into three analytic intervals tied to Dutch regulatory or guideline changes: baseline (2016 Q1–2017 Q4), post-guideline revision (2018 Q1–2020 Q1), and post-dementia care standard/CCA (2020 Q2 onward). The 2015 Long-Term Care Act preceded the available data and therefore was not modeled. Interrupted time series (ITS) with segmented linear regression was applied to test whether the linear trend across quarters differed between these periods. The authors observed a steeper declining trend in psychotropic prescribing after the combined introduction of the Care and Coercion Act and the Dementia Care Standard in the second quarter of 2020, relative to the period following the 2018 guideline revision.
Analyses compared mean prevalences between the three periods and used interaction terms in regression to evaluate whether slopes changed relative to both baseline and the immediately preceding period. Investigators explicitly used non-overlapping 95% confidence intervals as a conservative heuristic for differences between quarters and stated that these comparisons were exploratory and not adjusted for multiple testing.
The research team separated regular prescriptions from PRN prescriptions in all prevalence calculations. Reference values for PRN-inclusive prevalence in specific quarters were reported in the article. To quantify psychotropic polypharmacy, the authors counted how many of the five ATC categories a resident had prescriptions for at a given quarter, producing a polypharmacy metric that ranged from one to five categories per resident. Inter-organizational variation was assessed by calculating quarterly prevalence of regular psychotropic prescribing per nursing home organization, revealing between-organization differences in prescribing patterns.
Data extraction also captured a wide range of comorbidities by querying free-text medical history fields and mapping common disease names and synonyms to a predefined comorbidity list. The query development was iterative and involved clinicians and a medical informatician, but the authors acknowledge that some diagnoses may not have been recorded or may have been missed by the query.
This work applied routine care data to produce real-world prevalence estimates across a large, unselected population of residents in dementia special care units. The strengths include long follow-up, standardized extraction from electronic prescription systems, and a focus on policy-relevant timepoints. The authors identified several limitations in the source material. Data rows missing information on type of care or admission period were excluded (affecting almost 40% of rows), which may have introduced selection bias. Some residents in the dataset lack a recorded dementia diagnosis because the EHR capture was imperfect; the authors therefore included residents of dementia special care units as a pragmatic proxy. Anticonvulsants were excluded from the main analysis because of heterogeneous indications; supplemental tables reported anticonvulsant prevalence separately. The dataset is pseudonymized and access is restricted; the data cannot be deposited in a public repository because of privacy and contractual constraints.
Unanswered questions left by the routine data include the clinical appropriateness of individual prescriptions and the proximate causes for observed reductions in prescribing. The authors noted that while regulatory changes coincided with a steeper decline after Q2 2020, routine prescription records cannot by themselves establish causation or the degree to which non-pharmacological care alternatives were implemented. Future work linking prescribing data to structured measures of neuropsychiatric symptoms, functional outcomes, and documented decision-making would be necessary to determine appropriateness and patient-level benefit or harm.