رفتار تجویز دارو و سیستم های پشتیبانی تصمیم گیری

Adverse drug events plague the outcomes of health care services. In this research, we propose a clinical learning model that incorporates the use of a decision support system (DSS) in drug prescriptions to improve physicians' decisions about the initial drug selection and administration. The model allows for both the analytical investigation of the effects of different DSS features on clinical learning and the estimation of the physician learning behavior given a panel data set. The analytical results suggest that using a DSS to improve physicians' prescription decisions would positively influence their clinical learning. Conversely, without improvements in successful drug selection, the use of a DSS would negatively affect clinical learning. The empirical results provide further evidence on the factors that drive physicians' responses to information sources and the extent to which they rely on clinical experience in prescribing drugs.

Researchers estimate that adverse drug events (ADEs) cause between 700,000 and 1.5 million injuries annually [16], [47], [48] and [73]. A prominent study suggests that 28 percent of the ADEs, most of which are due to prescription errors [41], [46] and [49], are preventable [6]. Mirco et al. [57] find that the most common prescription errors are deficiencies related to choosing the right drug, dosage, frequency, route of administration (i.e., pills, gels, and liquids), drug interactions, and length of therapy. The sheer number of prescription errors has its roots in the challenges that physicians face in keeping abreast of developments in pharmacology. As powerful new drugs and clinical information become available, the need for accurate prescription decisions grows proportionately. Thus, deficiencies in keeping up with new developments in pharmacology unavoidably lead to suboptimal prescription decisions, even though the choice and administration of drugs make up some of the most important clinical decisions in medical practice [69]. Continuous physician learning is arguably the most effective solution to reducing prescription errors. Physician learning involves effectively integrating the clinical experiences with the most recently acquired information and then modifying the prescription behavior accordingly. Physicians regularly update their beliefs and thus learn about the efficacy of drugs from their own clinical experiences [23]. Improving prescription decisions through continuous learning would not only minimize preventable ADEs and provide better treatments for the patients, but also improve patient satisfaction [24] and [28], reduce insurance risks, and lead to superior quality and audit ratings for the physicians [51]. When integrated with clinical, practice guidelines and workflows, decision support systems (DSSs) and computerized physician order entry (CPOE) can help physicians with their clinical learning and thus enhance their prescription decisions. CPOE refers to computerized systems that automate the medication ordering process. Basic CPOE features include verification of typed orders in a standard and complete format, and CPOE systems typically have or interface with DSSs of varying sophistication, although some DSSs are implemented without a CPOE [41]. In general, CPOE and DSSs support two types of decisions: drug selection and drug administration. Drug selection refers to the initial decision of matching a patient with an appropriate drug from a set of alternatives. Computerized decision support on drug selection is provided through drug recommendations, drug–allergy checks, drug–laboratory value checks, and drug–drug interaction checks. Drug administration refers to how the selected drug should be administered in terms of dosage, frequency, route, and length of therapy, and such decisions are supported with appropriate recommendations by the software. The drug selection feature of CPOE has been shown to reduce the rate of non-intercepted, serious prescription errors by more than half [7] and [8]. The use of DSSs has also been shown to reduce the errors associated with drug administration (i.e., decisions regarding medication dosage, frequency, and route). Table 1 summarizes the literature on the effect of DSS use on prescription decisions and outcomes.Because DSSs do not replace physician judgment,1the sustainable positive results can be achieved only through improved physician learning supported with DSSs. Bochicchio et al. [10] also argue that the main benefit of computerized decision support is simply improved pharmacological knowledge. Physicians assume full responsibility of their prescribing decisions with or without using a DSS, and therefore the most successful DSSs are those that best facilitate physician learning. Our objective in this paper is to understand the interaction between physician learning and the use of a DSS and the corresponding impact on prescription decisions. We also aim to understand which type of decision support is more critical for physician learning. To this end, we develop a model of physician prescription behavior supported by two types of DSS features. One category of DSS features supports the decisions regarding when to prescribe a focal drug (drug selection), and the other category supports the drug administration decisions for the focal drug. Using the DSS features can potentially reduce the variances and uncertainties behind drug selection and administration decisions and influence physicians' learning, with the objective that prescription behaviors are in line with the clinical guidelines established for the focal drug. The proposed framework provides both an analytical model to investigate the effects of these two DSS capabilities and an empirical model to estimate the physician prescription behavior given a panel data set (for other similar empirical models, see [1], [23] and [46]). The model accounts for the following two factors: (1) physicians may be subject to different patient profiles and experiences, and (2) they may arrive at different clinical conclusions, even after observing the same evidence, because of their prior clinical experiences [46]. Using the proposed model, we ask the following research questions: How are the two types of DSS features related to physicians' clinical learning about a focal drug? What are the salient physician characteristics that affect clinical learning? What are some of the important physician-level factors that facilitate the adoption of DSSs? We use a hierarchical Bayesian estimation technique that captures the individual, physician-level uncertainties and learning behavior. Thus, the proposed model can be used to analyze, compare, and contrast different physician responses to the use of computerized decision support in the prescription process. Previous research in information systems has shown the importance of combining individual-level learning behavior and user environment [38]. A contribution of this study is that it combines physician learning and the use of information technology in modeling physician behavior. The analytic modeling approach combined with the empirical analysis of clinical learning behavior provides a powerful framework for capturing the impact of DSS on physician learning. The analytical results emphasize the importance of computerized support for drug selection decisions and highlight both the benefits and the risks associated with designing and implementing DSSs. When DSSs lead to superior drug selection decisions, patient-level observations are better integrated into the prescription behavior, which improves physician learning. An implication of this result is that proper design and use of DSS may help in enforcing compliance with treatment protocols and reducing prescription errors. Thus, the model provides an explanation on when and how the use of a DSS would allow us to observe physician decisions similar to those of an expert panel [67]. We also find that, without improvements in the accuracy of drug selection decisions, the use of a DSS negatively influences the physicians' clinical learning because they attribute less importance to the information they gather from patients than to their established expectations of the drug. Consequently, improper design and implementation may lead to negative outcomes [22] and [49]. The empirical results provide further evidence on the role of the information acquired through clinical experience. We find that physicians differ substantially with regard to their responses toward the information sources and clinical experiences. Physician specialty and location have significant effects on the overall physician responses to new information about a focal drug. General practice physicians (i.e., generalists) and physicians located in high-income areas rely more on their clinical experiences than specialists and physicians located in low-income areas, respectively. Accordingly, our analysis suggests that computerized decision support for drug selection benefits specialists and physicians located in low-income areas relatively more. These results provide further evidence on the importance of specialty and location on the success of DSS use. We organize the rest of the paper as follows: We first present an analytical model that captures the physician prescription and learning behavior in Section 2. Then, we describe our data and empirical methods in Section 3. The empirical results on salient physician characteristics and how they are related to DSS usage are then presented in Section 4. The paper concludes with a summary and discussion in Section 5.

Previous information systems research in health care has examined the business value of information technology [26] and its adoption within the sector [9], [13], [14], [35], [36], [44], [55], [58] and [64]. We contribute to this literature by investigating when and how DSSs can improve physicians' clinical learning and thus improve their prescription decisions in terms of choosing and administering the right drug. The results have implications on how to increase the perceived usefulness of the technology and facilitate adoption [20] and [60]. Bates et al. [8] indicate that physicians tend to be more pragmatic in their acceptance of the technology. The literature suggests that physicians value the usefulness of information technologies much more than their ease of use. Keil et al. [43, p. 89] note that “no amount of ease of use will compensate for low usefulness.” Usefulness is typically operationalized as increasing physicians' productivity, improving the quality of care, and enhancing their effectiveness. Because facilitating clinical learning is perhaps the most critical benefit of DSSs for physicians, we can argue that drug selection features are of paramount importance in the adoption of DSSs that are used in the prescription process. Usefulness of DSSs is also critical from an educational perspective. Teich et al. [70] question whether DSSs really help the medical education of physicians-in-training. They acknowledge that DSSs improve care in the hospital, but they also suggest that it is not known how physicians perform in other settings without computerized decision support after having been trained with it. One possibility is that physicians learn some facts less well because of their growing dependence on the computer to supply important pieces of information. Another possibility is that clinical learning is enhanced because guidelines and recommendations are frequently re-presented and reinforced at crucial moments. By focusing on the benefits of DSSs on clinical learning in this study, we tend to support the latter argument in that physicians should be able to carry over their improved skills to settings that lack a DSS. Our learning model provides analytical justification for the existing empirical results in the literature that associate DSS adoption with reduced ADE rates. Differences in clinical use of DSSs have been documented empirically. For example, Grant et al. [34] find that primary care physicians are associated with greater use of DSSs. The results of this study provide implications on which types of decision support offer more potential for which categories of physicians and, correspondingly, on which DSS implementations are more likely to fail. Despite the documented benefits and the mandates, the widespread clinical acceptance of DSSs has been lacking, and this has been a concern for researchers and medical informaticians [4], [40] and [42]. A recent study estimates that under current conditions, computerized order entry adoption in urban hospitals will not reach 80 percent penetration until 2029 [32]. There is a clear need to help facilitate the adoption, and our results can help policy makers design better incentives and mechanisms so that they can identify and target physicians who stand to benefit the most from computerized decision support. Our empirical estimations highlight the importance of physician-level differences and salient physician characteristics that affect clinical behavior and DSS use. Coscelli [22] argue that clinical DSSs should be viewed as socio-technical systems in which an individual physician's social background and demographics also play a role in the success of the adoption of the system. In a somewhat related vein, our results show that physician specialty and location have significant effects on the overall physician response to information sources and that the specialists and physicians in low-income areas are likely to benefit more from decision support on drug selection than general practice physicians and physicians located in high-income areas. Ray et al. [63] report that more than one-quarter of office-based Tennessee physicians mis-prescribed an antibiotic (i.e., tetracycline), which is associated with permanent discoloration of developing teeth, to young children. The authors show that rural family and general practitioners faced a high risk of prescribing these and other agents (e.g., chloramphenicol) in a potentially unsafe manner. Such findings illustrate that general practitioners may have more “room” for learning about pharmacology than specialists, in which case they may benefit more from computerized decision support over time, in line with our results. However, because DSS use lengthens the duration of a physician–patient encounter ([68] has shown that it takes 245 and 113 seconds to make a decision with and without the DSS, respectively), such benefits may be difficult to realize in low-income areas that typically exhibit relatively high demand for physician services. Thus, DSS developers should incorporate their products into physician workflows well, especially when targeting physicians who work in low-income areas. This study offers several future research directions. An immediate application of the learning model presented herein is the individual identification of the physicians who would benefit the most from decision support and those who may be distracted by it. Such a targeted approach may facilitate the diffusion of DSS adoption and may provide new avenues to overcome the adoption difficulties [61]. Researchers can also apply our learning model to specific types of DSSs to understand the role of more detailed aspects of these systems on physician learning. For example, researchers can examine, both analytically and by observing actual physician behavior, which types of DSSs (e.g., optimization systems, expert systems, and data mining tools) and which form of recommendation systems (e.g., those that employ collaborative-filtering versus content search through machine learning) are most promising from a learning perspective. Additionally, a similar methodology can be used to investigate the role of DSSs used for physician training. The dynamic nature of our model makes it a suitable framework for capturing the relative importance of external information sources (over time) in supporting physicians' prescribing decisions. A similar methodology can also be used to explore the optimal recency, frequency, and amount of training needed for each physician. Finally, with suitable modifications, the methodology developed here can be applied to understand and improve the professional learning of other knowledge workers.