This study shows that about one-fifth of primary care patients aged 75 years or older are treated with at least one PIM listed in the Swedish indicator ‘Drugs that should be avoided in older people’. The kappa values suggest strong agreement regarding the identification of patients with at least one PIM according to the studied indicator, but no agreement regarding physician assessments of their clinical relevance. In addition, the consensus assessments between the two physicians reveal that relatively few of these PIMs merit action from a medical perspective.

The area under the ROC curve shows that the studied indicator does not outperform chance in distinguishing between patients with adequate and inadequate drug treatment. An indicator based on the number of drugs in the medication list, on the other hand, performs better than chance. In this context, it must not be forgotten that the number of drugs reflects the burden of disease [22]. With greater complexity, the risk of overlooking something during a patient consultation is arguably higher. It may be worth noting that the optimal cut-off point in our study, to identify patients where action related to drug treatment could be medically justified and prioritised, is ≥ 6 drugs in the medication lists. Nonetheless, this threshold has low specificity.

The drugs included in the studied indicator have been determined by an expert panel to be generally unsuitable for older people [8]. However, like any other treatment, these medications require an individual assessment by the prescribing physician. Our results may thus illustrate the difference between what can be considered adequate/inadequate treatment at the population and individual levels, respectively; general treatment recommendations are based on an average individual, but no such individual exists in clinical practice. Still, the indicator may be a valuable tool to increase physicians’ attention to problematic medications and support them during the prescribing process. Indeed, the use of PIMs listed in the studied indicator has decreased in Sweden over time [14]. Nevertheless, our results suggest that there are still patients for whom these medications could be discontinued. Two drugs worth highlighting in this context are hydroxyzine for the treatment of anxiety and propiomazine for insomnia. In all cases, withdrawal of these drugs was considered medically justified, either immediately or in the long term.

In our study, 9% of the patients were prescribed drugs with anticholinergic action. The use of such drugs in those aged 75 years or older has increased over time [23]. Due to the side effects, however, such use may be particularly problematic. Indeed, the studied indicator, the Beers criteria [5], the EU(7)-PIM list [6], and the STOPP criteria [7] all include drugs with anticholinergic action (Table 5). Nevertheless, direct comparisons between criteria sets may be hampered by the fact that both drug-specific and diagnosis-specific criteria exist. Thus, knowledge of differences in criteria definitions is crucial when interpreting research results. Indeed, sets with drug-specific criteria may be suitable as screening tools and for prevalence studies, whereas complex diagnosis-specific criteria may be useful for prescribers in clinical practice and for educational purposes. Nevertheless, drug-specific criteria may have validity problems, supported by our finding that the studied indicator could not reflect the adequacy of drug treatment. Regarding complex diagnosis-specific criteria, on the other hand, both reliability and validity have been shown to be problematic as a substantial proportion have been reported not to be concordantly identified [24], and they also have limitations when it comes to their ability to reflect the adequacy of the drug treatment management [25].

The complexity of medical assessments becomes evident with the level of inter-rater agreement between the physicians. In the original study, 167 (55%) out of 302 patients were assessed similarly as regards overall adequacy of drug treatment (kappa: 0.33) [16]. The primary care setting, where the physician considers the patient’s overall situation and where more than one drug and disease often have to be taken into account, could be a contributing factor. In the hospital setting, on the other hand, acute cases have to be managed; thus, more targeted medical priorities are likely to be made. Therefore, it may not be surprising that assessments of the appropriateness of drug treatment at hospital admission show high inter-rater agreement [26].

The main strength of this study is that it facilitates interpretation of results that are used in health care and research, both regarding the Swedish indicator per se and drug-specific PIMs in general. Indeed, although sometimes labelled as ‘inappropriate drugs’, they are in many cases not inappropriate for the specific patient when their individual circumstances are taken into account from a medical perspective. Furthermore, the results may illustrate the difference between adequately managed drug treatment according to general recommendations and what clinicians consider adequately managed treatment for a specific individual. However, the limited number of patients included in the study may be regarded as a limitation. Nevertheless, the consecutive inclusion of patients from two large primary care centres in both urban and rural settings [16, 17] may contribute to the generalisability of the results. Another limitation is that the data are from 2017; drug treatment changes over time, and three drugs within the current indicator have been withdrawn from the market in Sweden. Nonetheless, the entire criteria set of the Swedish indicator is still in use and has not been updated since 2017. Indeed, the studied indicator is incorporated in the electronic medical records of many primary health care centres and hospitals in Sweden, generating an alert during the prescribing process. Furthermore, almost all PIMs in the studied indicator are also present in the EU(7)-PIM list [6] and the Beers criteria [5]. A significant limitation is the absence of a reference standard to assess the quality of drug treatment. However, our reference standard relies on rigorous assessments of patients’ drug treatment and can be considered medically relevant, in particular as they were carried out first independently, preceded by the application of extensive explicit criteria sets, and then in consensus by two experienced physicians with relevant specialist expertise. Finally, it may be worth noting that the percentage of patients treated according to the studied indicator is higher in our study than in a national register study (19% versus 7%) [15]. Selection bias may contribute to this finding; our patients represented a consecutive sample attending a primary care centre. Furthermore, the medication lists in the current study were based on comprehensive information in the medical records, and drugs used as needed may not be captured when medication lists are estimated on filled prescriptions during a restricted time period [15, 27].