Low-value Chronic Prescription of Acid Reducing Medication Among Dutch General Practitioners

A randomized controlled interventional study was conducted and evaluated using data derived from a subset of practices participating in the Nivel Primary Care Database (Nivel-PCD). The Nivel-PCD contains care data routinely collected from the electronic medical records from 529 GP practices throughout the Netherlands, representing approximately 2 million registered patients. Furthermore, the database contains longitudinal information regarding patient characteristics such as age, sex, GP consultations, diagnoses, and drug prescriptions. Socioeconomic status (SES) scores (on the level of Dutch postal codes) were obtained from the Central Statistical Office (CBS). Patients were assigned to one of five categories (lowest, below average, average, above average, highest) based on quintiles. Age categories were defined based on the available GP guidelines. Diagnoses are recorded using the International Classification of Primary Care version 1 (ICPC-1). Prescriptions are recorded using the Anatomical Therapeutic Chemical classification system (ATC). This study was approved by the relevant governance bodies of the Nivel-PCD (nr. NZR00322.017) and by the Research Ethics Committee of the Radboud University Medical Centre (dossier number 2022-13579).

Intervention and recruitment The intervention consisted of the distribution of a poster for the waiting room and flyers to be given to patients aiming to inform both patients and GPs with respect to the correct indications for treatment of dyspepsia. After signing up, practices assigned to the intervention group received a package containing 60 flyers and one waiting room poster to use during consultations. The flyer and poster provide a short description of the correct indications for treatment of dyspepsia. Additionally, both the flyers and posters contained a QR-code linking to a decision aid explaining the correct indications and causes of dyspepsia.

Recruitment of the practices took place from July until October of 2022. Practices were recruited from the pool of practices that provide their registration data to the Nivel. The recruitment involved sending several e-mails asking for their participation. The responding practices were randomly assigned to either the intervention or control group. Throughout the recruitment process, no information regarding the exact research topic or the intervention was provided, ensuring the blind participation of the practices.

Sample size calculation Based on a z-test sample size calculation using the proportion of patients that received an inappropriate chronic ARM prescription observed in an earlier assessment in the Netherlands (88% of chronic ARMs users do not have an indication), an alpha of 0.05, power of 0.80 and an expected reduction of 10%, a minimum number of 28 GP practices (with a mean of 328 patients that are inappropriately using a chronic ARM) were required to achieve significance.

Randomisation The participating general practitioners were recruited in a blinded manner from the Nivel-PCD. Meaning that the GPs were approached by the Nivel-PCD without receiving information regarding the purpose of the study. After having consented to participation, GPs were randomly assigned to either the intervention or control group. When a GP was assigned to the intervention group, the entire practice was seen as being exposed. GPs assigned to the intervention group received the poster and flyers, to be shared with the patients suffering from dyspepsia. GPs assigned to the control group received nothing. However, it is important to note that the access to the decision aid was not limited to the GPs of the intervention group and their patients, it was freely accessible to anyone through the website Thuisarts.nl.

Assessment of the low-value chronic prescription of acid reducing medication. Our assessment of the amount of ARM users was conducted using a patient-indication lens. Implying that all patients that were chronic ARM users were included in our denominator and all patients without indication for chronic use in our numerator. Individuals were considered chronic ARM users when they had received acid reducing medication for at least 180 days in the previous year. A patient's chronic prescription was defined as being of low-value when for at least 75% of all prescription days there was no clear indication for chronic ARM prescription present. This part of the analysis was performed using STATA 16. [32]

Outcomes and statistical analysis of the difference in prescribing over the two periods To assess the differences in ARM prescriptions the incidence rate of (inappropriate) chronic ARM prescriptions was compared in the same 6 months before and after the intervention (i.e. last 6 months of 2021 and last 6 months of 2022). The primary outcome therefore would be the odds ratio (OR) of patients receiving a low-value chronic ARM prescription between the pre- and post-intervention periods. For this purpose, a multilevel binomial model was built with an interaction term between both the indicator of cohort (i.e. 2021 vs 2022) and an indicator indicating whether a patient was part of a practice belonging to the intervention or control group. The aim was to include random effects for both the patient and practice level when possible. However, models were only used including a practice level because of the limited number of observations on the level of the patient. Generalised variance inflation factors (GVIF) were calculated to test for collinearity among the included variables before multilevel analysis was conducted. Patient age, socioeconomic status (SES) and sex were included as case-mix variables in the models, since previous research has shown they could affect the amount of care a patient requires, receives or has access to. Patients for which either the age or socioeconomic status was unknown were excluded from the multilevel analysis, but were included in the table showing the general description of both cohorts. Following our analysis of the baseline characteristics of the included population, patients above the age of 80 were excluded from this analysis while no cases of low-value care provision were present, which would result in too little variation on the practice level. Patients aged 80 and above were therefore excluded from our analysis, prioritising the recognition of clustering at the practice level over the inclusion of this age group in our model. The pre-intervention period (2021) was taken as reference period. A P-value smaller or equal to 0.05 was considered statistically significant for all analyses, based on two-sided testing. Data analysis and visualisation was performed using R (version 4.4.2)