Assessment of prescribing decision

Research to date has tended to focus on measuring a proxy for physicians’ decision making such as prescribing quality, prescription loyalty, the improving prescribing, and how prescriptions are made in general.¹²,¹⁹,²³,^{26 27 28}-²⁹They used the number of prescriptions or patients who received it, panel data, clinical measures, and physician reports to measure prescribing decision.¹²,²⁷,³⁰,³¹For example, Joyce et aldetermined the quantity and kind of drugs initially prescribed by each physician.¹²Wensing et alused two indicators: prescription cost and the proportion of patients who received a prescription.³²Venkataraman et alused the overall quantity of prescriptions (newly and formerly diagnosed patients).³³Kersnik and Peklar use the proportion of generic prescriptions to measure the choice of drugs.³⁴Some authors have addressed prescribing behavior as measured by guideline adherence.³⁵Shrank et albased on 1200 random sampling found that 90% of the study physicians select drugs that minimize patients’ out-of-pocket costs.³⁶Khan et alreported that 93.5% of USA physicians corroborate the prescribing drugs that would decrease the personal medical expenses of patients.³⁷The critical fundamentals that a physician needs to take into consideration the drug type (brand vs. generic), guidelines of drug selection and cost/patient’s ability to pay when selecting a drug.³⁸However, it is unclear whether the use of these factors fully captures the prescribing decision. Moreover, these studies do not attempt to fully validate the process of prescribing decision items. Hence, the reliability and/or validity of these studies are doubtful.

Assessment of marketing efforts

There is no consensus on suitable measures for marketing efforts constructs include information, drug brand, and the effectiveness of marketing agents and sales promotion of a drug in prescribing literature. The reviewed studies mostly used panel data of physicians either clinical measure, or focus group discussions or interviews, or measured overall promotions, or administered single factor with single item to address marketing efforts scales in the domain of prescription drugs.¹,²⁸,^{39 40 41 42 43 44}-⁴⁵For example, Zahrani reported that promotion tools did not influence prescribing among 275 Saudi GPs and family physicians.⁴³However, the author did not accurately measure sales promotion, since she only considered sponsored lectures and gifts. Al-Areefi et almeasured the physicians’ acuities of MRs visits, it does not measure the influence of the effectiveness of MRs on physicians’ prescribing behaviour.⁴⁴Parker and Pettijohn investigated the views of MRs and physicians as they relate to the effectiveness of marketing strategies in the USA.⁴⁵The results showed that there is a consensus among both physicians and MRs that the MRs have a minimal influence on the prescribing. However, these results may be doubtful or invalid because they do not indicate causality (both MRs and prescribing were evaluated in the same item).

Other studies endeavored to evaluate the face or content validity of their measurement instruments using either pre-tests or opinions of experts.³¹,⁴⁶Ladeira et alused SEM to analyze factors that affect the prescription of medical drugs in Brazil among 232 physicians.¹⁷The questionnaire includes 18 items that measure available drug information, brand and advertising. Furthermore, Karayanni utilized cluster analysis to measure physicians’ scores of MRs (4 items) and sales promotion (3 items).⁴⁶The resulting seven-item scale proved to be reasonably reliable. However, the study utilized a very small sample size for the analysis.

Assessment of patient characteristics

This study takes into consideration the patient’s request for a particular drug and expectations as well as their subsequent influence on prescribing behavior. Several researchers employed instruments to measure the patients’ drug requests using single-items, which are poor in method “base questions” that were correlated with prescribing.¹⁴,^{14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46}-⁴⁷Parker and Pettijohn developed an instrument to measure patients’ drug requests included only two items that reflect the direct patients’ drug requests.⁴⁵Other studies were based on different questions surveyed assessed patients’ expectations with one item comprising two, three or more categories for answers, i.e., “Yes’’ or ‘’No’’ categories.^{49 50}-⁵¹The items do not require the respondent to state when prescribing a drug, and how much importance was given to the patients’ expectations. Using 250 GPs, Tusek-Bunc et alfocused on four items to assess the physicians’ perceptions of patients’ expectations.¹³However, content validity and reliability were not evaluated in this research.

Assessment of pharmacist factors

Some studies measured the perception of physicians about the influence of pharmacists’ role which may be effective, but may not truly reflect relationship, the influence of expert power on the prescribing decision is not even weighted.⁵²,⁵³For example, Moore et alassessed opinions on expanding the role played by community pharmacists, but they do not accurately measure the influence of pharmacist expert on prescribing.⁵²Tahaineh et aldeveloped the scale for measuring physicians’ expectations of pharmacists’ role among 200 physicians in Kuwait.⁵⁴However, they have weaknesses that compromise their validity and the measure can fully gauge the expert power. Basak et aloperationalized the relative expertise power between pharmacists and physicians, but they do not measure the influence of the actual expertise on the prescribing.²²

Most studies that focus on the physician - pharmacist collaboration were in health care while recognizing that this variable may act independently or as a group. For example, Van et aland Zillich et alattempted to measure attitudes of GPs towards cooperation with pharmacists.⁵⁵,⁵⁶Kucukarslan et alinvestigated the influence of physicians’ perception toward pharmacist collaboration however they do not measure the influence of the actual pharmacist-physician cooperation on the prescribing.²³Liu et alused 750 self-administered surveys to develop a scale collaborative care at baseline.⁵⁷However, this scale was performed with sampling that is concentrated on pharmacists rather than physicians. Overall these measures are more pertinent to interface GP/pharmacist attitudes toward collaboration than to collaboration -prescribing relationship measurement.

Assessment of contextual factors

Even though a considerable number of empirical studies have identified contextual factors, there is a need to develop an effective measure that integrates their influences. For example, Venkataraman and Stremersch empirically implemented the physician- level panel data and clinical trial reports approach to measure the influence of drug attributes on prescribing, however, this measure was not corrected for the test the variables independently.³³Ladeira et alused an instrument to measure drug characteristics and the cost/benefit ratio of the drug.¹⁷However, the development of the adapted measure was questionable.¹⁷Janakiraman et alconcluded that physicians who are more willing to respond to marketing efforts have a lower likelihood of being persistent.²⁷However, this work may have more limitations, given that the insights were gathered through panel data. Abdul Waheed et alexamined brand loyalty with two items among 71 respondents, which is too small for the outcomes of the results to be generalized.²⁶Similarly, the scales reported by Mehralian et alis limited because the study used six measures (compassion) to evaluate the prescription loyalty of physicians.¹⁹

Instrument development process

For physician decision to prescribe the drug, items were measured using the three-item including drug type’s choices (brand and generic), drug guidelines, and the buying power of patients, with context-specific modifications.³⁶,³⁸Information on a drug (4 items) and the brand (4 items) was adapted from Karayanni.⁴⁶Sales promotion was measured using the six items, three items were adapted from Karayanni, and three items were adapted from Ladeira et alThe effectiveness of MRs was measured using six items considered in related studies.¹⁷,³¹,⁴⁶

For patient request the drug, two items were modified from Parker et al., and a single item was developed from the literature. Patient expectations were assumed by four items.¹³,⁴⁵The adapted items from Tahaineh et alwere used to measure pharmacist expert power with some modifications to suit the context of the current research.⁵⁴Collaboration and trustworthiness were measured by four items each.²⁴,⁵⁷Drug characteristics (6 items) and the cost/benefit ratio of a drug (5 items) was drawn from a scale of Ladeira et alwith context-specific modifications.¹⁷Three items used to measure the construct of habit persistence in the current study (Online appendix 1).¹⁹,²⁶This study employed a five-point rating scale (‘strongly disagree’ to ‘strongly agree’).

Content Validity

Three pre-tests were carried out to evaluate the quality of the 56 items. In pre-test 1, a brief questionnaire was subsequently sent to 8 academics and physicians using e-mail. Academics experts in pharmaceutical sciences, prescription marketing, and health assessment were examined the quality of face validity of the survey in terms of clarity, simplicity as well as the ambiguity of the scale items. While physicians in a different area were asked to comment on the suitability of the questions for doctors whether the instrument relates to their practice or difficult to answer. The experts also delivered several recommendations resulting in the elimination of three items (one item of drug characteristics) and the addition of three (drug requests, habit persistence, and MRs’ effectiveness). Lastly, seven items, one item in the information available on a drug, two in pharmacist collaboration, two items in pharmacist expert power, and two items of habit persistence were recast to reflect the dimensions. This process expanded the items to 52. The 5-point Likert scaling techniques were suitable for performing tests.

In pre-test 2, a pre-test of the novel instrument was carried out with a small sample of 20 respondents comprising diverse physicians in both public and private hospitals. The respondents were asked to identify and eliminate potential problems. The pre-test participants recommended that the entire items for one construct should be measured in the same direction. Another issue wording in the items of habit persistence, the items should reflect the phrase “prescribing the same drug”.

In pre-test 3, based on pre-test results, fifty drop-off questionnaires were piloted in Yemen amongst physicians in private and public hospitals in Sana’a City. Out of the 50, 30 questionnaires were returned and valid for the pilot study analysis used. The observations gathered from the pre-test were used to revise the questions of the brand, expert power, trustworthiness, patient expectations, and habit persistence. Cronbach’s alpha was adopted for the reliability test of this pilot study using the SPSS software v22.0 (Table 1).

Data collection

In Yemen, most doctors hold two positions simultaneously; the central in public hospitals and the other in the private sector (hospitals/clinics). In this situation, the sample was calculated regardless of their place of employment or their affiliation with the public or private sector.⁴,⁴⁴Furthermore, there is no statistically significant differences were found between physicians at public and private hospitals regarding attitude toward marketing efforts such as MRs. Therefore, they may have not a different prescribing pattern between categories.

The physicians who participated in the study were purposively sampled and were from both specialists and GPs working in private and public hospitals and were all practicing the prescribing drugs. However, since the referral chain was carried out to expand the sample, the snowball sampling technique was utilized as a type of non-probability sampling approach, thus increasing the response rate in the data. In June 2017, 420 of 705 questionnaires were completed, however, 393 (59%) of the questionnaires were usable. While 71% and 28.4% of the respondents are males and females. The age of the majority of the respondents varies between 35 or less and 35-45 years. Over half of the respondents have a postgraduate degree (63.6%), and 53.4% of the participants worked at general and private hospitals. Many of them are consultants (38.4%) and GPs (27%). More specifically, the study participants comprise GPs (27.5%), and gynecologists (17.3%). Approximately 51.1% of the respondents had 5 to 10 years of experience in practice, 42.8% of the respondents see 1-15 patients per day and 52.7% of the respondents see <5 MRs per week.

Test for non-response bias

An independent sample t-test was employed mainly to check whether any form of a discrepancy between these two groups by comparing their means.⁵⁸The present study divided the respondents into two main categories: those who responded within 30 days (291 early respondents) and those who returned after 30 days (102 late respondents) following Armstrong and Overton’s approaches.⁵⁹The findings reported that the equivalent variance significance values for all the variables were >0.05 significance level of Levene’s test for equality of variances. Therefore, it can be deduced that the theory of equal differences between early and late respondents has not been infringed.

Common method bias test (CMB)

A single-factor test was employed to identify the existence of this bias by using principal component analysis (PCA) in SPSS (v22.0). All items of understudy constructs were entered into factor analysis and run the PCA. The unrotated primary components analysis yielded 10 factors with eigenvalues>1, accounting for 63.7% of the variance (Table 2). Given that a single factor solution did not emerge, (the highest one accounted for 26% of the difference) and an overall factor not did account for the majority of the variance, CMB was not considered a major problem in this research (see Table 2).

Descriptive analysis

Online appendix 2 presents the mean values, SD, and correlations between the study constructs. The correlation between the variables was in the predicted direction and significant at p<0.01. The correlation values were well below the threshold of 0.80.⁶⁰It shows no problem of high correlation among the variables and provides evidence that multicollinearity is not an issue in the present study.

Conceptual model

The conceptual model specifies the marketing efforts (available drug information, drug brand, sales promotion, and MRs’ effectiveness); patient characteristics (patients, requests and patients’ expectations); pharmacist factors (trustworthiness, expert power, and collaboration); and contextual factors (drug characteristics, cost/benefit ratio of a drug, and physicians’ habit of persistence). Therefore, to assess the prescribing decision of physicians’ model and to measure the factors affecting their choices, we posit that all factors are significant influence physician decision to prescribe the drug (see Figure 1).

Figure 1. Research model 

PLS-SEM approach

To estimate the prescribing decision model, the study utilized component-based PLS-SEM because it is a more accurate method for studies that focus on prediction and is ideal for handling complex frameworks.⁶¹,⁶²Thus, PLS-SEM was applied to establish the measurement models of the latent variables, leading to several modifications to improve the model and its fit. Having confirmed to the individual item reliability and validity, a bootstrap procedure with 1000 (one-tallied, 0.5) bootstrap re-sampling with 399 cases was used to estimate the significance of the interactions.

Assessment of the outer measurement model

The analyses comprise two stages. The first stage involves an assessment of the outer measurement model, which includes reliability and validity tests of measurement properties. As shown in Online appendix 3, the results of the 44 items that were retained in the model had loadings between 0.709 and 0.912 (see also Table 3). A total of 8 items fall below 0.708 could still be fairly weak (shown in italics in Online appendix 1).⁶⁰,⁶¹Online appendix 4 provides that AVE values exhibited higher loadings (> 0.50) on their respective constructs, showing adequate discriminant validity.⁶¹The composite reliability (CR) coefficient of the entire potential constructs which range from 0.838 to 0.904, with each exceeded the minimum level of 0.70 and not above 0.95, indicates adequate internal consistency reliability of the measures. Further, all the values passed the heterotrait-monotrait ratio of correlations (HTMT) 0.90 and the HTMT 0.85 indicating that discriminant validity has been ascertained (Online appendix 5).⁶³

Assessment of the overall parameters

The model of research revealed a coefficient of determination (R²) = 40 percent of the total variance in the prescribing decision of the physician can be regarded as moderate.⁶⁴The standardized root means square residual (SRMR) of 0.061 which less than 0.08 indicates that the degree of misfit is not substantial and suggests that the data support a standard factor model.⁶⁵A GoF value of 0.488 was obtained for the overall factors influencing the prescribing scale, which overtaken the value of cutoff 0.36 for large effect sizes of R²⁶⁶Q²value was 0.248 which considerably higher than zero and is slightly sizeable predictive relevance for the endogenous construct (that is PPD).⁶⁷Consequently, it is concluded that the scale of the PPD has substantial predictive validity.

The assessment of the structure model

Table 4and Online appendix 6 show that the beta, t, f²values meet the criterion for evaluating the model. The results confirm that brand (beta= 0.182, t= 3.588, f²=0.043), sales promotion (beta= -0.108, t= 1.672, f²=0.008), MRs effectiveness (beta= -0.087, t= 1.341, f²= 0.006), patient request drug (beta=0.223, t=3.363, f²=0.043), patient expectations (beta=0.218, t=3.370, f2=0.045), pharmacist expert power (beta= -0.269, t=4.141, f²=0.063), pharmacist-physician collaboration (beta = 0.311, t=4.801, f²=0.089), and trustworthiness (beta=0.165, t=3.038, f²=0.034), which explained 40% of overall prescribing decisions.

Implications for theory

First, this study extends the literature on prescribing behaviour by developing and validating the influence of marketing efforts, in addition to providing procedures for new constructs such as “pharmacist expert power” “pharmacist-physician collaboration” and also the patient characteristics into two categories, namely, “patient request for the drug” and “patient expectations” influence prescribing decision. By encompassing the explanatory power of each factor, this research expands knowledge on what influences physicians’ decisions to prescribe the drug in the context of a physician-patient/pharmacist relationship.

Second, it categorizes a wide-ranging set of items that aid the prediction of physician-pharmacist relationship (i.e., expertise power, collaboration, and trustworthiness), and their associated impact on prescribing decisions. These factors cannot be easily identified and separated and have received scant attention in the prescribing behaviours literature. Since these measurements were not reported in any study on prescribing behavior, this current study will contribute to the development of the subject of prescribing behavior in the context of a theory extension. These constructs can be exploited by policymakers to measure physicians’ perceptions of their relationships with pharmacist’s in general health care settings. Hence, this study provides a basis for researchers who are interested in this field to further test the relationships among these constructs, especially in the pharmacy setting. Third, the study frames the final decision of the physician when prescribing the drug as a crucial finding of studies on prescribing behavior, which has not been examined prior to research on prescribing behavior. Based on an analytical perspective, this study ultimately models factors influencing the prescribing decision for the first time as a reflective model using SmartPLS using SEM.

Implications for practice

The findings enhance the understanding of marketing managers regarding how physicians assess the effectiveness of MRs and sales promotion when making prescribing decisions. In particular, such findings suggest that managers should focus on encouraging their MRs to comply with ethical marketing and self-regulation with one outcome being better prescribing, which can be achieved by educational evidence to physicians. Likewise, the cost of a drug is crucial in determining what drug they choose. For policymakers, there is a need to look into the drug cost policy to facilitate patient’s access to medicine, hence the need to address the cost/benefit of the drug issue. A powerful pharmacist should be motivated to influence physicians in ways that are believed to be effective and further produce a positive outcome. It is vital to pay more attention to pharmacist collaboration in ensuring improving prescribing behaviours. Trustworthiness is also an excellent choice for health policymakers while targeting the improvement of particular dimensions such as trust, bi-directional communication, and commitment at different levels. Also, the professional association of physicians should build awareness of the physicians that their decisions on best evidence will consider patients’ requests and expectations could positively influence the prescribing decisions. Overall, the prescribing decision model proposed in this study may help policymakers to achieve and ensure appropriate drug for the patient which consequently help them in performing rational prescribing.

Limitations and future studies

First, the items that make up the instruments are reflective of marketing efforts on prescribing could be considered as a sensitive issue and thus could raise the issue of social desirability bias. Physicians may feel not comfortable indicating their involvement in marketing efforts, rather than stating their effects, in a circumstance in which the promotion was a human being. However, the risk of bias from the self-administrative survey cannot be ruled out. Thus, a longitudinal study could be employed to assess the influence of marketing efforts on the perceptions of physicians towards prescription drugs over time. Second, three items captured only the prescribing decision of the physician, which may not be an adequate proxy for actual prescribing decision behaviour in all circumstances. Third, to gain a deeper understanding, the distinction between GPs and specialist physicians should be considered, which would enable an improved comparative analysis against other studies in this subject. Future research might look at combine patient request and expectation measures in regards to their influence on prescribing drugs. The measuring of pharmacist-physician relationships may vary in salience for different classes of a pharmacist that is clinical, community, etc., and across healthcare facilities. Fourth, replication of this study by assessing the pharmacist-physician relationship from two perspectives may be a worthy effort. Further studies are required to promote an advanced perception of the nature of the relationships projected in the multifaceted model.