Introduction
Globally, the population group of adults aged ≥ 65 years is growing faster than all other age groups combined. In 2019 one in every 11 persons was 65 years and over, this has been predicted to increase to one in six persons by the year 2050 [
1]. With ageing societies, also come growing numbers of older adults with multiple chronic conditions. Multimorbid patients often use multiple medications and with polypharmacy comes a higher risk of using potentially inappropriate medications (PIMs). PIMs are medications for which the risk of potential adverse events outweighs the clinical benefits, such as when there are more effective and safer alternatives available for use in older adults [
2]. The use of PIMs is associated with increased risk of adverse drug events, falls and cognitive impairment [
3‐
6]. Patients with multimorbidity and polypharmacy often have complex healthcare needs, which in turn lead to substantial health services use and associated costs [
7]. The use of potentially inappropriate medications is high in this patient group [
8]. In this context, the ‘Optimizing PharmacoTherapy in older multimorbid adults In primary CAre’ (OPTICA) trial was launched with the aim of investigating whether an electronic clinical decision support tool can help GPs to optimise medication use of older multimorbid patients with polypharmacy.
Lack of external validity of clinical trials, the extent to which results can be generalised to the wider population, has been cited as a reason that interventions do not get adopted after publication of the study. One factor that can influence external validity is the characteristics of the participants recruited into the trial; that is, whether they are comparable (have similar characteristics) to those found in the real-world population [
9].
Despite societal ageing and widespread multimorbidity, patients with chronic conditions and older adults in general are often underrepresented in clinical research [
10,
11]. Evidence from studies of younger and healthier participants may not be generalizable to the broader older multimorbid population [
12]. The reasons for the exclusion and general underrepresentation of complex older adults in research are multifaceted. On the one hand, studies often have inclusion and exclusion criteria to maximise participant retention and minimise variability among participants [
13‐
15]. On the other hand, even if older multimorbid adults are not explicitly excluded, major barriers to recruiting this type of study participants include limited mobility (e.g. not being able to attend multiple appointments or complete certain tests), and in the case of cognitive impairment, inability to provide informed consent [
14,
16,
17]. Additionally, the person identifying and selecting patients for recruitment (e.g. member of the research team or through healthcare professionals with established relationships) can impact the external validity of participants [
18]. Use of routinely collected patient information to identify participants for clinical trials is a promising method to reduce the labour of recruitment. However, concerns exist about the error rate of using electronic medical records for this [
19].
Not only can the recruitment of older multimorbid patients be challenging, so can the recruitment of GPs [
20]. Previous studies found that time constraints, lack of training, fear of loss of professional autonomy as well as lack of rewards and recognition are barriers to research participation for physicians in general [
21]. Conducting clinical research in the primary care setting comes with additional challenges. For instance, a lack of infrastructure, lack of financial remuneration of practice staff involvement, misunderstandings on how daily clinical work in general practice could accommodate the clinical research, and seasonal changes in workload [
22,
23]. There is the concern that GPs with specific characteristics or attitudes can be motivated more easily to participate in clinical research. If true, it would mean that the results of an interventional study (such as our OPTICA trial) would not be generalizable to even the local context outside of those who participated in the trial. Overall, little is known about whether it is possible to recruit an externally comparable sample of older multimorbid patients and GPs for research in primary care.
Further, past medication optimization interventions in patients with polypharmacy have shown limited effect in changing medication use [
24,
25] and/or clinical outcomes (e.g. mortality, cognitive decline) [
26]. This may be due to patient resistance to medication changes and their unawareness of potentially inappropriate medication use [
27]. It is therefore important to consider not only the characteristics of participants, but their attitudes as well.
The ‘Family medicine ICPC Research using Electronic medical records’ (FIRE) database is the largest Swiss electronic database containing anonymized routine patient data from the electronic medical records in > 10% of Swiss primary care practices. It also contains information about the GPs who regularly export data from their electronic medical records. The FIRE database therefore provides a unique opportunity to examine the likely external validity of the OPTICA study results to the wider Swiss general population in primary care.
The overall aim of this study was to determine if the GPs and patients participating in the OPTICA trial are comparable to the real-world population in Swiss primary care. We hypothesised that our broad inclusion criteria and support provided to participating GPs would result in recruitment of comparable participants. This information is not only important for interpreting the forthcoming results of the OPTICA trial (i.e. the likely external validity of the study findings), but can also inform the ability to recruit complex older adults for clinical trials in primary care.
Specifically, the aims of this manuscript were to:
1.
Describe the baseline characteristics of participants (GPs and older patients with multimorbidity and polypharmacy) recruited to the OPTICA trial.
2.
Compare the characteristics of GPs and patients participating in the OPTICA trial with those in the FIRE database.
3.
Compare the characteristics of the patients recruited for OPTICA from random screening lists generated from electronic medical records with patients recruited through GP identification of eligible patients.
4.
Describe the patients’ willingness to have medications deprescribed.
Discussion
To inform the likely external validity of the results of the OPTICA trial, we compared the characteristics of our participating GPs and patients to a Swiss real-world reference cohort. We also examined the characteristics of patients recruited based on random screening lists (created from electronic medical records) and those recruited outside of these lists by their GP to see whether a bias in the selection may exist. Finally we explored the willingness of patients in OPTICA to have medications deprescribed which allows us to reflect on the possible impact that this may have on the outcomes of the trial and compare them to previously studied populations. From our analyses we have some confidence that the findings of the OPTICA study will be generalizable to the broad Swiss population of GPs and patients. We found that the GPs in the FIRE project and OPTICA were similar in terms of sociodemographic characteristics and their work as a GP (e.g. age, experience as GP, employment status, and GP practice type). We also found that patients participating in the OPTICA trial and patients in the FIRE database were comparable in terms of age, median number of certain lab and vital data measurements (e.g. BMI, lipid profile, GFR measurements) and certain clinical characteristics (e.g. systolic blood pressure, BMI). For the variables that differed between the two groups according to the statistical tests, the absolute standardized differences were generally around 0.2 (or 20%), with an imbalance of the two groups having previously been defined as > 0.2. Patients who participated in the OPTICA trial reported a high level of willingness to stop one or more of their medications.
Overall, our study results showed that GPs who participated in the OPTICA trial and those who participated in the FIRE project were comparable in most of the variables examined. Previous research showed that high performing physicians are more likely to participate in research [
45]. When looking at the patient data, we observed that OPTICA patients had more chronic conditions, but less medications. The absolute standardized differences indicate some imbalances between the groups on these variables. While one can argue about whether the differences are clinically relevant, this observation could indicate that GPs in the OPTICA trial may have been more proactive in reviewing the medications of their patients than other GPs. If the latter was the case, this would mean that the intervention of the OPTICA trial may be limited in its effect (i.e. if the patients had little room for further optimisation of their medications). We also found differences in sex, location and self-dispensing between GPs in both groups. These differences may have stemmed from the recruitment strategy used in the OPTICA trial, which in the context of difficulties of recruiting GPs for clinical research focused (and therefore needing to optimise GP recruitment) did not specifically recruit based on their baseline characteristics. The sex composition of the OPTICA GPs could affect the final results, since female physicians have been found to be less likely to make deprescribing decisions [
46].
We found that the multimorbid older patients who participated in the OPTICA trial were comparable to those in the FIRE database in terms of sociodemographic variables, health services use and clinical characteristics. For the variables were there was a statistically significant difference between the groups, most had standardized differences close but not passing the ASD threshold of 0.2 for meaningful differences between the groups (e.g. number of medications: OPTICA median = 6, FIRE median = 7, ASD = 0.23, number of consultations OPTICA median = 16, FIRE median = 13, ASD = 0.216). There is a lower proportion of female participants in the OPTICA trial than in the reference FIRE cohort. However, since no difference in willingness to deprescribe according to sex has been identified [
39,
47], we do not anticipate that this sex imbalance will affect the results of the OPTICA trial.
We found that the trial participants recruited from the random screening lists (around two thirds of patients) and those who were recruited outside of these lists (around one third of patients) were comparable. While systematic differences in recruitment behaviour (i.e. differential recruitment [
48]) has been reported previously in the context of a cluster-randomized controlled trial in primary care (UK BEAM trial) [
49], we did not find evidence for a bias in the selection of participants in the OPTICA trial. The UK BEAM trial reported, for example, that patients in participating practices were experiencing milder back pain (which the intervention targeted) than those in the control group and thus highlighted the potential for the recruitment process to bias study results [
49]. The use of random screening lists helped to standardize patient recruitment but, in light of the imperfect nature of the screening lists, we also allowed GPs to recruit patients who were not on these lists. We assumed that giving participating GPs some flexibility in the recruitment process would allow them to better integrate recruitment into their regular practice and would therefore optimise recruitment.
Concerning patients’ willingness to deprescribe, we found that the OPTICA study participants had a high involvement in their medication use and > 80% were willing to stop one or more of their regular medications if their doctor told them this was possible. These findings are in line with previous research. Another study conducted in Switzerland found that 77% of older adults would be willing to stop one or more of their medications [
39] and similar proportions were found in studies in other countries (88% in Australia [
47], 92% in the United States [
50], 83% in Singapore [
51]). While these numbers have to be interpreted with caution (e.g. social desirability bias, not medication specific, hypothetical nature of the question), it shows that older patients may be open to optimizing their medication use through deprescribing. We also found the factor scores to be comparable to the results from a study in Australian older adults [
47]. This information is crucial for implementing medication optimization interventions, and in the context of the OPTICA trial, it shows that patients’ attitudes towards deprescribing may not be a barrier to implementation of deprescribing.
While the results presented in this manuscript are primarily Swiss-specific, we can draw a more broadly applicable conclusion; it appears to be possible to recruit a sample of study participants in primary care trials that is comparable to real life cohorts.
Strengths & limitations
The OPTICA trial had a low number of exclusion criteria, which facilitated broad recruitment of study participants. However, the analyses in this manuscript have several limitations. First, in Switzerland there are no complete GP or patient registries. The FIRE project maintains the only primary care database in Switzerland of this size, but it does not include all GPs in Switzerland, and in turn, does not include all patients in Switzerland. In Switzerland, not all GPs use electronic health record programs. The use of electronic health records in Switzerland increased from around 40 to > 70% from 2012 to 2020 [
17,
34], but remains lower than in other high-income countries. Furthermore, not all GPs who fulfil the eligibility criteria self-select to participate in the FIRE project. This raises the question of the representativeness of the GPs in the FIRE database. However, two recent assessments of the Swiss GP workforce showed that the GPs in the FIRE project are comparable to the entire GP workforce in terms of age, sex, experience as GP and work percentage (eTable 2 in the
supplement). These similarities between OPTICA, FIRE and all GPs in Switzerland signify that the recruitment of an externally comparable sample of GPs is possible in randomized clinical trials in the Swiss primary care setting. This confirms previous evidence from the UK, which showed that achieving good levels of external validity was possible in clinical trials in primary care [
52]. However, due to the lack of patient registries, we cannot comment on the comparability of patients in the FIRE project and Swiss patients in general. While the analyses presented in this manuscript do not confirm external validity of the forthcoming OPTICA trial results, they do facilitate future interpretation of our findings.
Next, inherent to routine medical databases, like the FIRE database, is a certain risk of information bias and missing data as information is only collected when it is clinically relevant [
53]. Since we used data from before the OPTICA study intervention started, we assume that both our groups would have been affected by the same potential sources of bias. Despite the similarities found between the FIRE and Swiss GP workforce in terms of sociodemographic and work-related characteristics, we were unable to compare other important characteristics between the two groups (e.g. quality of care, relationship and trust between doctor and patient). Our finding that the patients included in the OPTICA trial had less medication but more chronic conditions than the reference cohort could reflect the selection of “good performers” which may bias the findings of the OPTICA trial. Our analysis of patients’ willingness to deprescribe was limited to patients in the OPTICA trial and could not be compared directly to the reference cohort and this questionnaire is not used in regular clinical care. Other limitations related to the rPATD are that it asks hypothetical questions, it is not specific to certain medications, and it might be subject to social desirability bias. Furthermore, for the purpose of the OPTICA trial the rPATD was translated from English to German; back-translation and piloting was conducted to increase the validity of the translation, but other measures of validation and reliability of the translation in the local context were not conducted (e.g. test–retest reliability). Finally, due to the uncertainties surrounding the absolute standardized differences, we decided to present both
p-values and ASD. While there may be debate of the cut off to use for ASD, we used > 0.2 as this has been recommended by Yang et Dalton [
42,
43]. If we considered a smaller threshold, such as > 0.1, it would not have changed our conclusions about the groups being comparable.
Acknowledgements
Thanks to Heinz Bhend, Pius Bürki, and Sven Trelle for their support in conducting the OPTICA trial. We would also like to thank all GPs and their patients for participating in the OPTICA trial for all their efforts to help us conduct this trial. And finally, we would like to thank Stéphanie Giezendanner and Andreas Zeller for providing information on the Swiss GP workforce.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.