Barriers to appropriate prescribing in older adults with multimorbidity: A modified Delphi study

Abstract Objectives We aimed to understand the barriers experienced by physicians when prescribing for older adults with multimorbidity in specialist outpatient clinics in Singapore. Methods A modified Delphi study was conducted via email with 20 panel experts in the field of geriatric medicine. Barriers identified from an earlier scoping review were presented as statements to the panel. Results Eleven barrier statements reached consensus with high importance according to the Delphi panel. Of these statements, seven (64%) belong to the domain of Environmental context and resources in the Theoretical Domains Framework (TDF), while the remaining barriers belong to the domains of skills, knowledge, intentions, and professional/social role and identity. The barriers are further linked to intervention functions in the Behaviour Change Wheel (BCW). Conclusion Linking the TDF domains to intervention functions revealed strategic directions for the development of an intervention to address the barriers and optimize prescribing.

understand factors leading to PIP, which would provide insights into interventions that could reverse those trends.
Many factors contribute to the issue of inappropriate prescribing. Some of the wider, systemic issues include the lack of research in patients with multimorbity, 1 and exclusion of older adults in general from clinical trials. 5 In addition, most clinical guidelines are based on single diseases and offer no clear guidance for application in multimorbidity. 1 As a result, these wider systemic issues manifest in day-to-day clinical practice as barriers, with the lack of evidence-based knowledge to inform practice. A previous scoping review identified barriers to effective prescribing among older adults with multimorbidity at the physician-related, patientrelated, and health-system-related levels. 6 These barriers were mapped onto the Theoretical Domains Framework (TDF), which is an evidence-based implementation framework that identifies factors that impact behavior change. 7,8 It entails 14 validated domains that are based on theories of change and their constructs. 8 The TDF domains could in turn be linked to intervention functions in the Behaviour Change Wheel (BCW) and their associated behavior change techniques 9 to facilitate the translation of knowledge into practical implementation. As these findings were mostly based on studies conducted in Europe and in other countries with differing health systems and cultures, we needed to explore if those barriers exist in Singapore's context.
Hence, the primary aim of this Delphi study is to identify key barriers to appropriate prescribing for older adults in the outpatient care setting in Singapore. The secondary aim is to link the identified barriers with their already mapped TDF domains to the intervention functions in the BCW. This information will provide the evidence base to guide clinical practice and policy improvements through the development of an intervention prototype that aims to optimize prescribing for older adults with multimorbidity.

| Study design
The Delphi technique is commonly used to solicit the opinions of experts and achieve group consensus on a subject matter through a series of structured iterative questionnaires. 10 Unlike traditional Delphi studies, the modified Delphi technique adopted for this study utilizes preexisting literature to develop the initial questionnaire, rather than starting the first round with open-ended questions. 10

| Panel participants
Practicing geriatricians with at least three years of postprofessional qualifications in three of Singapore's public hospitals were invited to participate in the study. As there are no set standards on the number of panelists in a Delphi Study and having 10-15 experts with homogenous background is considered appropriate, 10 we planned to enroll 20 participants to account for chances of attrition. Email invitations were sent to potential participants with an explanation on the purpose of the study, brief results from our scoping review on barriers to effective prescribing, the Delphi process, and the survey period. The invitation continued until we enrolled 20 participants and obtained their informed consent.

| Modified Delphi rounds and the process
This study consisted of two rounds of questionnaires. Statements presented were formulated based on findings from a previous scoping review. 6 The barriers identified in that study were mapped onto 10 of the 14 6 The barriers were further divided based on stakeholders influencing the prescribing process, either directly or indirectly: physician, patient, or healthcare system at large. 6 Most barriers were categorized under the physician perspective. In the current study, we reviewed and consolidated similar barriers, when appropriate, resulting in a total of 98 barrier statements for the round 1 questionnaire. Participants were asked to rate the importance of the barriers as factors impacting their prescribing process.
In addition to the rating, a comment box was added below each statement for participants who wished to comment or explain their decisions further.
Round 2 of the Delphi questionnaires contained statements that have low group agreement, with some minor refinement based on participants' corresponding comments. Prior to the start of round 2, participants also received formal feedback from round 1 results, comprising the group's median rating of each statement, as well as their respective ratings. The purpose of providing feedback is to create an opportunity for the participants to review and reconsider their stance on the statements, 10 which in our study, were those that did not reach consensus in round 1.

| Defining consensus
A 7-point Likert scale was used by participants to rate their level of agreement with each statement: 1-Not at all, 2-Low, 3-Slightly, 4-Neutral, 5-Moderately, 6-Very, and 7-Extremely. A priori criteria adapted from previous studies 11,12 were used to define consensus, with median and interquartile range used as measures of central tendency and dispersion, respectively. 13 Only those statements that had low group agreement in round 1 were rerated in round 2. The following criteria were as follows: • Median ≥6 and IQR ≤1 = High group agreement on being very and extremely important →Item is included, • Median <6 and IQR ≤1 = High group agreement on being of moderate or low importance →Item is removed, • IQR >1 = Low group agreement on level of importance (nonconsensus) →Item is refined and continued to round 2 for rating.
Due to the large number of barriers derived from the scoping review results, 6 we made an a priori decision to only consider barrier statements that fulfill the criteria for high importance in the subsequent intervention design. Hence, the statements with consensus on moderate to low importance and statements that did not reach consensus were not analyzed.

| Mapping barriers onto intervention functions
Barriers that were regarded as highly important by the panel were mapped to intervention functions to characterize the types of intervention elements that would best address them.

| Delphi rounds
The Delphi study took place from September 2019 to December 2019. All 20 participants responded to both rounds of the survey. Among the participants from the Geriatric Medicine clinics of three public hospitals, 10 were males (50%) and 10 were females (50%). Among them, 17 (85%) were qualified geriatricians while the remaining 3 were senior resident physicians with extensive experience in geriatric medicine. Of the 98 statements that were presented in round 1 of the survey, 9 reached consensus, while 41 were deemed to be of lower importance and were not pursued further. The remaining 48 statements that did not reach consensus (IQR > 1) were refined and included in the second round of the survey for repeat rating. Among these 48 statements, three that belonged to the domain of skills (physician-related) were split into two statements each to further clarify the concepts. Out of the 51 presented statements in round 2, two reached consensus with high importance, whereas 20 reached consensus for moderate to low importance. Consensus was not reached for the remaining 29 statements. Figure 1 shows the flowchart of the Delphi process and results.

| Consensus barrier statements with high importance
Based on our a priori criteria, statements with an interquartile range (IQR) of 1 or less and median of 6 or above were considered to have reached consensus for high importance. Accordingly, 11 statements were identified as highly important barriers to appropriate prescribing as shown in Table 1. Of these statements, seven (64%) were previously mapped onto the domain environmental context and resources in TDF, 6  I feel more importance (sic) given to disease management, rather than age-specific issues and taking into consideration interaction of medication in different age group (Participant 02, site 1).
Although the remaining barriers from the physicians' perspective cluster were mapped onto different domains, namely in environmental context and resources, intentions, and professional/social role and identity, 6 they shared similar sentiments in reflecting uncertainty and aloneness in the prescribing decision-making process. This leads to reluctance in making changes to medications that have been prescribed for long periods or by other physicians. In other words, maintaining the status quo in prescribing is easier. One of the participant's comments sums up the observation: As patient's medication list become (sic) more complex, there are many specialty drugs that not all physicians will be familiar with. Hence, there will be hesitancy in changing medications prescribed by another prescriber. Also, sometimes, the patient has a long term rapport with the other prescriber, or has (sic) the impression that they are doing well on the current regimen, thus, patient will be reluctant to have the medication/dosage change (Participant 17, site 2).
Only one barrier identified is related to patients' perspective and that is their lack of knowledge of the medications that they are taking.
However, there is indication that the lack of knowledge is due to inadequate communication with the patients:

| Remaining barrier statements
At the end of the round 2 survey, 61 barriers were deemed to be of lower importance based on our criteria and 29 barrier statements did not reach consensus. Although the results were not further analyzed, the TDF domains found among these statements were: beliefs about capability; memory, attention, and decision processes; emotion; social influences; beliefs about consequences. The barrier statements with consensus on low importance and nonconsensus can be found in Tables A1 and A2 respectively in the Appendix.

| Linking barriers to intervention functions
For our secondary aim, the 11 barrier statements identified as highly important were linked to the intervention functions in the BCW, 9 via their mapped TDF domains. 6 Table 2 shows results of the linkage.
The Behaviour Change Wheel has its core in capability, opportunity, motivation, and behavior (COM-B), interacting with each other. 9 All 9 intervention functions in the BCW were mapped onto the 11 barrier statements. Some TDF domains are linked to more than one intervention function, which means that for those respective domains, there are various ways to address them. For example, the barrier on hesitancy in changing medication, which belongs to the

| D ISCUSS I ON
Developing an intervention in a dynamic healthcare setting is a complex process that requires a systematic approach. Using the Medical Research Council's framework on developing and evaluating complex interventions, the development phase consists of identifying evidence base and theory for an intervention. 14 As such, this study forms part of the evidence base for a wider project to develop an intervention to optimize prescribing for older adults with multimorbidity. To our knowledge, this was the first Delphi study conducted to understand barriers to appropriate prescribing experienced by geriatricians in the outpatient clinics of Singapore's public hospitals. The Delphi panel reviewed a list of barriers that have been identified from a scoping review 6 of studies conducted in other countries and categorized into physician-, patient-, and healthcare system-related perspectives. The 11 top barriers experienced in our local context are consistent with the themes found in the literature. Besides physician-related factors, patient-related factors such as their resistance and ambivalence towards changes, 15 nonadherence with medications and visits, 16 and lack of understanding of the medications 17 were found to affect the prescribing process. With respect to wider healthcare system-related factors, work practice and medical culture 15,16 and difficulty navigating current evidence-based guidelines specific to single diseases [16][17][18] were also found to hinder the prescribing process. In addition, prescribing for an older adult with multimorbidity is a complex process, due to a clear knowledge and skills gap, 15,17,19 as well as the complex interprofessional relationship from having more than one physician involved. 18 The latter leads to a barrier identified by our Delphi panel on hesitancy and reluctance to change prescriptions by others, 18,20 which might set a precedence for devolving of responsibility. 15 Due to the large number of barriers identified from the scoping review, 6 an arbitrary cutoff point was determined in the current study, as it would be impossible to address all identified barriers with one intervention. It is also unlikely that an intervention could directly address some of the pervasive healthcare system-related problems that require broader system or policy changes. Hence, the advantage of having separate stakeholder perspectives is that they can help point to the barriers that are most feasible to address. It would also disentangle the individual-level barriers that were tied to system issues, which help direct a way to circumvent those issues.
In  Increased specialization in healthcare (ie, focus on subspecialty-based care instead of overall management). 28 Fragmentation of care, lack of a specific or unified physician to follow up with. 23 Lack of coordination or communication between transitions and various levels of care across healthcare settings. 21,29 Exclusion of multimorbid older adults in clinical trials. 21 Lack of ownership to assume responsibility for optimizing a specific patient's care plans. 24 prescribing. Likewise, the tendency to maintain the status quo in prescribing due to one's hesitancy in changing medications 20 and the reluctance to interfere with medications prescribed by other physicians 21 may also be addressed by the intervention functions of environmental restructuring and enablement, through medication reviews by on-site clinical pharmacists. By providing recommendations based on their reviews, pharmacists provide support to the prescribing process, which makes prescribing for this complex group of patients less lonely and uncertain.
Overall, we demonstrated that it is possible to link barriers experienced by physicians to intervention functions via the TDF and BCW framework. The BCW in turn sheds light on behavior change techniques that could be implemented and tested in a feasibility study for a physician-pharmacist care collaborative multimodal intervention. Following this, the objective is to scale up and adapt an effective intervention to multiple sites for broader implementation across hospital ambulatory care and primary care in Singapore.

| Strength and limitations
A major strength of our study is the utilization of the modified Delphi technique to calibrate the previous scoping review findings 6 to the local context. As opposed to other group consensus methods like focus group discussions or conferences that require face-to-face meetings, questionnaires in a Delphi study can be disseminated and completed by the participants independently via an online platform.
On the other hand, our study has limitations that should be acknowledged. We made the a priori decision to stop at two survey rounds, resulting in 29 barrier statements not reaching consensus.
Along with consensus statements that were found to be of lower importance, they were excluded from further analysis. This may have led to overlooking barrier statements that are important. 13 A way to mitigate this risk of overlooking important barriers is to include post-hoc considerations with justification, 25 perhaps by considering some of the excluded statements that were close to our set criteria.

CO N FLI C T O F I NTE R E S T
We have no conflict of interest to declare.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that supports the findings of this study are available in the supplementary material of this article.