Methodology

What was the data source for this tool?

We used data from the Clinical Practice Research Datalink GOLD and Aurum databases, which are anonymised, nationally representative primary care databases in the UK. These were linked with Hospital Episode Statistics data to obtain an accurate dementia diagnosis date and Office for National Statistics death records to obtain an accurate end of follow-up. This study was approved by CPRD’s research data governance team (protocol 23_002757).

What was the study population?

The study population included older adults aged 65 years or older at dementia diagnosis. Dementia diagnosis date was defined as the earliest of a diagnosis in primary care, a diagnosis in secondary care, or a prescription for a dementia medication.

What was the study period?

The study period ran from 1 January 1998 to 31 March 2021. Medication use was examined before and after dementia diagnosis, where follow-up data were available.

How was node prevalence calculated?

Node prevalence describes how common each medication group was among people in the study. For each network, we calculated the total amount of time that people were observed in the data and the total amount of that time during which they were prescribed each medication. The prevalence shown in the tool is therefore the percentage of observed patient-time covered by that medication group.

For example, if statins had a prevalence of 40% in a given year, this means that statin treatment episodes accounted for 40% of all observed follow-up time in that year. It does not necessarily mean that exactly 40% of people received a statin at one point during the year. Values can also be shown separately by sex and age group.

In simple terms: node prevalence provides a patient-time-based measure of medication burden over time, rather than simply counting whether a person received at least one prescription during the year.

How was co-medication prevalence calculated?

Co-medication prevalence describes how often two medication groups were prescribed at the same time. For each pair of medications, we identified people who had overlapping treatment episodes of at least 14 days during each year relative to dementia diagnosis. The co-medication percentage was then calculated using the number of people with that overlap as the numerator and the total number of people observed in that year as the denominator.

For example, if the co-medication prevalence for statins and beta-blockers was 20%, this means that 20% of people observed in that year had overlapping treatment episodes for both medication groups. To protect privacy, small-cell results were suppressed and shown as missing.

Which medications were considered?

The chronic medication network includes medication groups such as statins, antihypertensives, antiplatelets, anticoagulants, proton pump inhibitors, opioids, non-steroidal anti-inflammatory drugs, antidiabetic medications, respiratory medications, antihistamines and CNS medications. The CNS medication visualisations include antipsychotics, antidepressants, benzodiazepines, Z-drugs, anticonvulsants, gabapentinoids, dementia drugs and Parkinson’s disease medicines.

How was this built?

Analysis was conducted in R. MINT-D was built using D3.js, a JavaScript library for creating interactive data visualisations in the browser.