Setting and study populations
Blekinge is located in the south eastern corner of Sweden and is one of the smallest counties with approximately 153 000 inhabitants in 2011 and 2013. Almost all inhabitants are registered to a primary care centre in Sweden. The majority of funding for primary health care comes from a specific county council tax, both public (operated by the county council) and private care centres. Both public and private primary care centres were included in the study. We included two cohorts for comparison in this registry based repeated cross-sectional study. For the cohorts, we included individuals aged 75 or older listed at a primary care centre in Blekinge on the 31st March 2011 for the first cohort and, for the second cohort, individuals listed on the 31st December 2013. The information campaign to improve care of the population of older adults was active between 2010 and 2014. However due to possibility of access to data of medication data the cohorts were chosen for a slightly shorter period. This is because the 31st March 2011 was the earliest date that we had access to a three-month period of medication data within the local register described below. And due to changes in how medication data was encrypted in 2014 the 31st December 2013 was chosen to ensure quality of data for the second cohort.
Data source and measurements
Data on chronic conditions, age and gender in the study were based on anonymized registry information obtained from the County Council of Blekinge from both primary and secondary care.
Use of medications was identified from the county council’s register of dispensed medicines for all inhabitants in Blekinge. Data in this register was received by the County Council from the Swedish eHealth Agency. It contains the same patient level data on prescribed medicines as the national Prescribed Drug Register at Swedish National Board of Health and Welfare, but the coverage is restricted to the residents in the county [19, 20].
In Sweden, prescribed medicines are prescribed for use at most three months within the high cost threshold for medicines . Therefore a three month period was used to construct a medicine list on both regularly used and as-needed medicines . If the same drug was dispensed more than once it was still counted only once. Since the county council’s register of dispensed medicines does not contain exact dose, we used Defined Daily Doses (DDD) to calculate the duration of the drug exposure. We assumed 0.9 DDDs for regularly used medicines based on calculations for regularly used medicines in an older adult population [22, 23]. Medicines were classified according to the anatomical therapeutic and chemical (ATC) system . A constructed medication list was determined for each individual in the cohorts; 31/3 2011 for cohort 1 and 31/12 2013 for cohort 2. From this constructed medication list, polypharmacy and use of PIM were identified, according to specified definitions.
We used indicator 1.1, ‘Medicines that should be avoided unless there are special reasons’ from the Swedish National Board of Health and Welfare report ‘Quality indicators for good drug therapy in elderly’ as the definition of PIM . As the title ‘Medicines that should be avoided unless there are special reasons’ states, it is medicines that should be avoided in patients, 75 years and older, unless there are special reasons because of the higher risk of side effects. If prescribed, the prescriber should have a well-founded indication and the treatment should be evaluated at regular and frequent intervals. This definition of PIM was used in the national information campaign to follow-up the effects of the campaign and therefore it was used in this study also . The following drug groups and substances are included in this definition of PIM: long acting benzodiazepines, tramadol, propiomazine and medicines with anticholinergic effect. The use of PIM was identified from the constructed medication list by the medications ATC-codes.
Multimorbidity was defined as number of chronic conditions. It was determined by using a validated assessment tool that captures chronic conditions grouped in 60 different diagnoses categories . All information about diagnoses for a two-year period prior to 31/3- 2011 (cohort 1) and 31/12- 2013 (cohort 2) were included.
All variables were used as categories in the analyses. Gender was categorised as male or female and use of PIM; use or no use of PIM. Age was categorised into four groups: 75-<80, 80-<85, 85-<90 and ≥90 and number of chronic conditions was divided into five groups or strata: none, one, two to four, five to seven and eight or more, chronic conditions. For the descriptive analysis of the cohorts, use of medications were divided into three strata; no-medication, use of 1 to 4 and use of five or more. A first descriptive analysis of the two cohorts in the different strata of the variables age, gender, use of PIM, number of chronic conditions and number of medications was performed. The differences were analysed using chi-square test. A significance level (α) of 0.05 and 0.001 was used. Since polypharmacy is a known risk factor for ADEs we wanted to analyse the prevalence of polypharmacy in the different strata . Therefore, number of medications was divided into two strata for the rest of the analyses; no use to use of four medications (<5, no-polypharmacy) and use of five or more (≥5, polypharmacy).
We then described the cohorts from use of PIM in different strata of the variables age, gender, number of chronic conditions and polypharmacy and analysed the changes between the 2011 and 2013 cohorts. The cohorts were compared using chi-square test.
The cohorts were then described and analysed from use of polypharmacy in different strata of the variables age, gender and number of chronic conditions. The cohorts were compared using chi-square test. A significance level (α) of 0.05 (*) and 0.001 (**) were used.
Logistic regression was then used to analyse how the different strata of the variables from 2011 were associated to the use of PIM 2013. Here only individuals present in both cohorts were included. We created five models; model A adjusted for use of PIM, model B adjusted for PIM and age, model C adjusted for PIM, age and gender, model D adjusted for PIM, age, gender and number of chronic conditions, model E adjusted for PIM, age, gender, number of chronic conditions and polypharmacy. To analyse how the different strata of the variables from 2011 were associated with decreased use of PIM 2013 a logistic regression was performed. We created five models; model A adjusted for age, model B adjusted for age and gender, model C adjusted for age, gender and number of chronic conditions, model D adjusted for age, gender, number of chronic conditions and polypharmacy. A significance level (α) of 0.05 (*) and 0.001 (**) was used.
We used STATA version 14.0 (Stata Corporation, Texas, USA) for statistical analyses.