As patients continue to play a more active role in the management of their health, the person-centered model of care has been promoted as a strategy to improve the quality of health care systems [1]. Along with ensuring that all clinical decisions are guided by the patient’s values, the goal of the person-centered model is to respect and respond to the individual’s preferences and needs. This motivates physicians and patients to coordinate their activities, share information, and reach shared therapeutic decisions [2]. This review takes a person-centered approach for the important and challenging case of consultations involving patients with cancer. Patients have come to expect their treating physicians to explain the benefits, as well as the risks, of the therapies recommended to them. Furthermore, patients prefer to be engaged in the therapeutic decision-making process [3,4], except when they are very ill [5,6], rather than permitting their physicians to choose therapies for them. Patients may also want to be given the chance to consider their options and to choose between accepting or refusing a therapy.

Medical consultations in oncology are a multipart process that involves shared decision-making between the patient and the physician. Bomhof-Roordink et al [7] have articulated this process in their model of shared decision-making. A physician starts the anticancer treatment recommendation process by learning about the patient’s preferences, before or during consultations, which they need to consider along with the evidence of efficacy of each potential treatment option. Next, the physician needs to engage the patient in reviewing the potential benefits and risks of the key therapeutic choices available. After collaboratively and carefully examining the situation, the physician provides treatment recommendations. However, the ultimate course of action may be chosen by the patient alone or by the physician when the patient does not want to decide [7].

As the choice of diagnostic modalities and therapies grows, the clinical decision-making process has become extremely complex [8]. Faced with large volumes of fragmented information, physicians must reconstruct, identify, and consider the portion of information that they share with their patients. In addition, physicians need to decide how to best inform their patients and obtain their consent [9]. Hence, physicians need clinical information that is organized and presented in a way that is easy for them to interpret and share in discussions with their patients. Once physicians have determined what they need to share, they need to be able to show the relevant information to their patients in such a way that the patient can understand the meaning of the different benefits and risks of each therapeutic choice [5,10]. When physicians can summarize information that is relevant to patients’ diseases and their survival, explain highly uncertain situations, and manage their interactions with patients well, then patients can more easily understand their physicians’ recommendations and choose their preferred therapy or care pathway. This step establishes the foundation for informed consent in shared therapeutic decision-making.

With the intention to support patients, as well as physicians, in this challenging therapeutic decision-making process, paper-based decision tools have been developed [8]. They have been designed to enhance patient-physician communications and interactions. In addition to the incorporation of research results, for example, evidence from clinical trials, paper-based decision tools inform both physicians and patients of the risks, benefits, and outcomes of the available therapies [6,11,12]. Furthermore, paper-based decision tools have a long tradition in supporting clinical decision-making. They have been shown to improve patients’ knowledge, accuracy of perceived potential risks, understanding of prognosis, treatment goals, and health outcomes [8]. Moreover, in practices where paper-based decision tools are used, they are well accepted [11]. However, paper-based decision tools can be difficult to update when new therapies are rapidly being developed and adopted. Furthermore, increasing the use of genetic testing and the introduction of advanced molecular medicine in routine clinical practice has generated an expanding body of knowledge that increases the complexity of the decision-making process [2]. Thus, it is recommended that physicians and patients use computer-based decision tools to improve the process outlined above [2].

Hunt et al [13] defined a computer-based decision tool as follows: “any software designed to directly aid in clinical decision-making in which characteristics of individual patients are matched to a computerized knowledge base for the purpose of generating patient-specific assessments or recommendations that are then presented to clinicians for consideration.”

Research to create computer-based clinical decision tools has a long history. For example, as far back as 1973, Shortliffe et al [14] published a paper on this topic. Shortliffe [15] believed that with computer-based decision tools, knowledge can be integrated and disseminated to physicians. Similarly, computer-based decision tools may aid in packaging relevant clinical information and therapeutic choices for presentation to individual patients [16]. They may also simplify patient-physician communications [8]. On the basis of these perceived benefits, several computer-based decision tools have been developed to assist therapeutic decision-making during oncology consultations [17].

For example, Shortliffe et al [18] developed a computer-based decision tool to guide physicians treating patients with cancer. The technology consists of a computer user interface that enables physicians to review patients’ historical data and test results, enter new information about patients, and query the computer system for anticancer therapy recommendations. The implemented computer technology was initially based on the IF-THEN rule algorithm: for example, “IF: there is evidence of disease extension THEN: refer the patient to lymphoma clinic” [18]. However, more recently, computer-based decision tools have been redeveloped for oncology consultations by applying artificial-intelligence–based machine learning software technologies to improve the accuracy of the recommended anticancer therapies [16].

It is unclear at what level computer-based decision tools are adopted by oncology physicians. There have been a small number of reviews about computer-based clinical decision tools [19-21]. Pawloski et al [21] reported patients’ outcomes from a treatment delivery viewpoint. Beauchemin et al [20] described decision tools broadly and included nursing care delivery in their study. In contrast, Mazo et al [19] provided an overview of decision tools for breast cancer. However, none of the reviews addressed physicians’ propensity to adopt computer-based decision tools during oncology consultations. The aim of this review is to identify and categorize the factors that influence physicians’ propensity to adopt computer-based decision tools in oncology consultations by using the Clinical Adoption Framework (CAF) [22,23].

The CAF, as shown in Figure 1, is an extension of the Benefits Evaluation Framework (Canada Health Infoway), which was adapted from the DeLone and McLean information system success measurement model, as cited in the study by Lau et al [22].

Conceptually, the CAF is made up of micro-, meso-, and macrolevels. At the microlevel, the focus is on the dimension of quality, which measures success factors such as information completeness, accuracy, relevance and comprehension, system features, performance, security, responsiveness, support services, and leadership; user behavior, intention to use the technology, and user satisfaction; and net benefits, which refer to patient safety, risk, effectiveness, compliance, health outcomes, efficiency and capability, cost and savings, availability and access to services, and patient and clinician participation [24].

The mesolevel dimensions directly influence microlevel users’ propensity to adopt the technology. It addresses people’s characteristics and their expectations, roles, and responsibilities; technology system and organizational fit, strategy, culture, structure or processes, information infrastructure, and return on value; and implementation stages, project management approaches, and technology fit with present and future operations [24].

The macrolevel dimensions directly affect the mesolevel factors, which in turn affect the success of adoption at the microlevel. At the macrolevel, governance, legislations, regulations, and policies; health care and professional practice standards; funding and incentive payments; and trends in public expectations as well as sociopolitical and economic climates with respect to technologies and the whole health care system influence adoption [24].

In addition, as indicated in Figure 1, there is a feedback loop at each level of the CAF. The results of each level are fed back to higher and lower levels of the conceptual model, that is, the outcomes of microlevel factors influence the meso- and macrolevel factors. Similarly, mesolevel factors influence higher macrolevel and lower microlevel factors, and macrolevel factors affect mesolevel factors [24]. Consequently, the CAF represents a technical, social, political, and economic system that must contend with constant internal and external forces that dynamically affect propensity to implement and adopt computerized information systems in health care settings.

The research questions are as follows: (1) What is the extent of adoption of computer-based decision tools in oncology consultations? (2) Is there a difference in levels of adoption by country and period? (3) What factors may have influenced the adoption of the technology? (4) What are the lessons learned to improve adoption of the technology?

This study was conducted in accordance with the Cochrane Handbook for Systematic Reviews of Interventions [26] and followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 checklist, guidelines, and statements [27]. In addition, with the assistance of medical sciences librarians, the search strategy was constructed by applying the PICOC framework [28,29]:

On February 4, 2021, 1 reviewer (AY) used the OvidSP platform (Health First) to search the following databases: Cochrane Database of Systematic Reviews (from 2005 to January 28, 2021), Cochrane Central Register of Controlled Trials (December 2020), MEDLINE (from 1946 to February 4, 2021), and Embase (from 1947 to February 4, 2021). In addition, on the same day, the databases of Web of Science (from 1900 to 2021), Scopus (from 1969 to 2021), and PubMed (from 1991 to 2021) were searched. After relevant articles were selected for inclusion in this review, the reference list and citations of each article were inspected for additional articles. The snowball search was conducted using Scopus and Google Scholar. Further searches for relevant articles were conducted by browsing the BMC Medical Informatics and Decision Making journal website, along with searches of gray literature websites [30-33]. The detailed Boolean expressions of the search strategy are provided in Multimedia Appendix 1.

A single review author (AY) removed duplicates and screened the titles and abstracts of all retrieved articles for relevance in accordance with the criteria of the research questions. Similarly, another 2 review authors (JK and TS) independently assessed the eligibility of a randomly selected sample of articles from a subset of the retrieved articles to judge their eligibility for inclusion or exclusion in the review. Disagreement among the 3 review authors was resolved through discussion.

First, guided by the evidence-based medicine pyramid [34], articles that used a study design within the categories of randomized controlled trials, cohort studies, case-control studies, and case series or reports were included for review, whereas articles that were published as conference papers or abstracts, protocols, commentaries, editorials, letters, or opinions were excluded because of their perceived low quality. No limitation on language was imposed. For articles that were not published in the English language, attempts were made to translate them into English by using a web-based translator [35]. Second, studies that met the following key criteria were included: (1) the study was conducted in an oncology consultation setting, (2) it involved distinct real-world computer-based decision tool use by oncology physicians, (3) a computer-based decision tool assisted patient-physician communications to share information and to agree on an anticancer therapy; and (4) the elements of the effectiveness of a computer-based decision tool in oncology consultations were reported.

A data extraction spreadsheet to capture study information was developed a priori by 3 reviewers. The selected studies were then screened by 1 review author, and relevant qualitative data were extracted. The spreadsheet was populated in accordance with the requirements of the review questions. As more experience was gained with data extraction, the review authors iteratively adjusted the required variables in the spreadsheet. The final set of data variables required to answer the review questions was as follows: study; study design and participant sample size; computer-based decision tool versus comparator; clinical setting context and country; primary objective; and study outcomes (Table 1).

Using the Cochrane risk-of-bias tools for randomized controlled trials and nonrandomized studies, 1 review author assessed the risk of bias of the included studies [26]. The tool for randomized controlled trials [46] assesses studies on each of these 6 domains: (1) randomization processes, (2) identification or recruitment of participants into clusters, (3) deviations from the intended intervention, (4) missing outcome data, (5) measurement of the outcome, and (6) selection of the reported result. The tool for nonrandomized studies [47] assesses studies on each of these 7 domains: (1) due to confounding, (2) selection of participants into the study, (3) classification of intervention, (4) deviations from the intended intervention, (5) missing data, (6) measurement of outcomes, and (7) selection of the reported result. Finally, the judgment in each domain is carried forward to an overall risk of bias for each study. The tools highlighted some risk of bias in all the selected studies.

The articles included in this study reported a high diversity of functionalities and features of computer-based decision tools. Therefore, the reported outcomes of the studies were grouped according to the dimensions of the CAF [22]. The results within each group were subsequently assessed and combined into a common set of factors that directly affect physicians’ propensity to adopt computer-based decision tools in oncology consultations.

The initial searches in the aforementioned databases retrieved 6407 articles (Figure 2). Browsing searches and inspections of reference lists and citations identified 3 additional articles. Of the 6407 articles retrieved through database search, 1979 (30.89%) duplicates were removed. Of the remaining total 4431 articles, 4368 (98.58%) were excluded after titles and abstracts were screened. Next, the full-text articles were assessed for eligibility, and of the 63 articles, 53 (84%) were excluded. A total of 10 studies were thus included in this review.

When the 10 selected studies for review were assessed by using the Cochrane risk-of-bias tools, they all exhibited some level of risk of bias. Of the 10 studies, 3 (30%) were randomized controlled trials [38,44,45], and 1 (10%) was a mixed-methods randomized study [41] (Multimedia Appendix 2, Table S1 [38,41,44,45]). All (4/4, 100%) the randomized studies included a high risk of bias because of the practices observed when assigning participants, adhering to the intervention, and accounting for missing outcome data. Of the 10 studies, 6 (60%) were nonrandomized studies (Multimedia Appendix 2, Table S2 [36,37,39,40,42,43]). Of these 6 nonrandomized studies, 1 (17%) [39] included a moderate risk of bias, whereas the remaining 5 (83%) [36,37,40,42,43] included serious risk of bias due to confounding [36,37,40], bias in selecting participants [43], bias in accounting for missing data, and measurement of outcomes [42].

Table 1 includes significant details gathered from the reviewed studies. Of the 10 studies, 6 (60%) were conducted in the United States, and 1 (10%) each was conducted in Canada, the Netherlands, Norway, and the United Kingdom. In all, 8 different computer-based decision tools were used across the 10 studies.

A summary of the identified computer-based decision tools from the review is provided in Table 2. The details include the name of the computer-based decision tool, country where each evaluation was conducted, categories of disease that were handled, types of decision that were settled, number of studies that were conducted for each computer-based decision tool, and bibliographical references. Of the 8 computer-based decision tools, 4 (50%) were evaluated for breast cancer consultations; 1 (13%) each for colorectal, prostate cancer, and cancer pain; and 1 (13%) for breast or colorectal cancer.

The results of this review showed that of the 8 identified computer-based decision tools, 4 (50%) were developed and studied in the United States, as shown in Table 2 [36,37,40,42,44,45]. Next, to determine whether a study was positive, negative, or neutral, the greater than or equal (≥) 50% rule, as cited in the study by Lau et al [22], was adopted. Consequently, of the 10 studies, 6 (60%) reported positive results for computer-based decision tools [37,38,41,43-45], whereas only 1 (10%) reported negative results [42]; 3 (30%) were neutral [36,39,40].

The CAF was extended to accommodate factors that influenced physicians’ propensity to adopt computer-based decision tools in oncology consultations. Of the 83 factors at the microlevel, 20 (24%) were identified as influential (Multimedia Appendix 3). Of these 20 factors, Textbox 1 reports 11 (55%) that were identified as positively affecting physicians, Textbox 2 reports 7 (35%) that negatively affected physicians, and Textbox 3 reports 2 (10%) that had no effect on physicians.

The studies did not explicitly provide evidence of meso- and macrolevel factors that influenced physicians’ propensity to adopt computer-based decision tools.

This review has 3 aims: (1) to understand the different levels and periods of adoption of computer-based decision tools during oncology consultations across the world, (2) to identify the factors that influenced the adoption of the technology by physicians, and (3) to learn how to guide future implementation and adoption of the technology in the context of shared therapeutic decision-making during oncology consultations [48].

This review showed that the development and studies of computer-based decision tools were primarily conducted in North America and Europe in the last 16 years. Although 10 studies were specifically selected for review based on the topic of computer-based decision tools that were used by physicians in oncology consultations, only 60% (6/10) of the studies addressed some aspects of the perspectives of physicians. Most of the studies focused on patients’ views. Our findings of low adoption of computer-based decision tools converged with similar patterns in previous studies [49].

In all, 2 computer-based decision tools—Adjuvant! and an in-visit decision aid—were used across 40% (4/10) of the studies. Adjuvant! provided the strongest evidence of user satisfaction, information-sharing, care quality, and productivity measures. The in-visit decision aid was assessed for users’ perception, knowledge retention, and treatment decision. A summary of the 8 identified computer-based decision tools is provided in Table 2.

By extending the CAF to computer-based decision tools in oncology consultations, these findings suggest that of the 20 factors, there are 11 (55%) that can facilitate physicians to adopt the technology and 7 (35%) that can stifle adoption, whereas 2 (10%) may have no effect on physicians’ propensity to change and adopt the technology.

Along with helping physicians to understand their patients’ treatment preferences, computer-based decision tools enable physicians to refer to information and to provide treatment information and recommendations that are related to their patients’ survival. Some physicians used the technology in routine practice in academic and community practices to review information with patients during consultations. They believed that the technology is useful for their patients because their patients become more engaged in discussions and understood the information. Thus, the conversation between the physician and the patient was facilitated during consultations, and the patient-physician communication about preferences and values improved.

In contrast, some physicians perceived that computer-based decision tools made patients more anxious and added 5 minutes to their total consultation time. The study by Siminoff et al [44] gave the impression that an additional 5 minutes was insignificant. The effect, however, was subjective, depending on each physician’s expectation. For a 1-hour consultation, an additional 5 minutes may be acceptable. However, the impact of adding 5 minutes to a 10-minute consultation in usual care may become objectionable. Furthermore, when the technology does not provide all the information that physicians want, is not readily available on their desktop, or is not integrated with the electronic health record, then physicians are not able to take advantage of the information conveyed through the technology. Consequently, they are not able to share information and treatment alternatives with their patients.

The findings of this review advance our understanding of the extent to which computer-based decision tools have been successfully adopted in oncology consultations. The evidence suggests that there have been very few studies that address physicians’ propensity to adopt computer-based decision tools in routine oncology consultations. This review provides a starting point and direction for further investigations to incorporate computer-based decision tools in usual oncology consultations. This review also provides a guide and key lessons—as shown in Textboxes 1, 2, and 3—for the design and development of new computer-based decision tools. In addition, the review highlighted some important areas that need to be improved in future computer-based decision tools, such as integrated access with electronic medical records (Textbox 2). Some studies have reported negative outcomes with computer-based decision tools [50,51], whereas others have shown benefits [52]. In our review, of the 10 selected studies, 6 (60%) were positive, with only 1 (10%) being negative, whereas 3 (30%) were neutral. Consequently, the impact of computer-based decision tools on oncology consultations is unclear. Taken together, our findings and the findings of similar past studies [19-21,53-56] point to the need for further research in several dimensions of the CAF to uncover the value of computer-based decision tools in oncology practice.

Looking at Figure 3, it is obvious that the studies included in this review have addressed only a small set of factors among the numerous factors that could influence the adoption of computer-based decision tools in oncology consultations. Therefore, future studies will need to address additional dimensions at the meso- and macrolevels to gain a better understanding of what factors lead to successful implementation and adoption of computer-based decision tools in oncology consultations.

This systematic literature review includes some limitations. First, only 10 studies were included in this review because of the dearth of studies that addressed the issues with computer-based decision tools from the perspectives of physicians. Second, the literature search was conducted by only 1 reviewer, which could have introduced bias and limited the findings. Third, the selected studies for review included a high risk of bias. Furthermore, most of the studies were conducted at nontraditional cancer centers or at health care organizations affiliated with academic institutions, which limit generalization. Fourth, our review covered a wide range of health information systems’ issues, which might not have been explored sufficiently and fully explained. Future researchers should refine the search strategy to identify additional potentially relevant studies that may have been missed and allocate more reviewers to search the literature databases to minimize potential biases.

In this review, we investigated the extent to which computer-based decision tools have been adopted in oncology consultations and physicians’ propensity to adopt the technology. The results of the investigation suggest that the adoption of computer-based decision tools in oncology consultations remains low. Of our 10 reviewed studies, 6 (60%) showed positive outcomes, whereas 1 (10%) showed negative outcomes, and 3 (30%) were neutral. To date, improvements have been made in communication and information-sharing between patients and physicians. However, unavailability of the information that physicians need, lack of access to the technology on physicians’ desktops, and lack of integration with existing electronic health record systems are some of the findings that stifle successful adoption. Therefore, this review shows that, in addition to improving communications between physicians and patients, technology is needed to streamline the flow of information that physicians need to better inform patients. Notwithstanding the 5 minutes that would be added to the overall time of consultations, this review indicates that it is possible to create leaner oncology practices by adopting computer-based decision tools. The technology would eliminate the need to track paper-based information, making the decision-making process more streamlined and eliminating the risk of missing hard-copy paperwork. Hence, in the long run, physicians would have more time to dedicate to their patients. As a result, patients may engage more in discussions during consultations, may be better informed, and they may be more apt to provide consent for treatment.

The CAF provides the capacity to make sense of complex multidimensional factors that influence the adoption of computer-assisted decision-making in oncology consultations. Furthermore, it provides a starting point as well as a sense of direction for research in the design and development of new computer-based decision tools. Thus, this review provides a set of key factors that need to be addressed to enhance the possibility of successfully implementing and adopting computer-based decision tools in oncology consultations. However, although the review shows that it is possible at the microlevel for patients and physicians to improve their communication by using computer-based decision tools, the effects of meso- and macrolevel factors remain understudied. It is therefore important to conduct additional studies in real-world oncology consultations to understand the impact of higher-level factors on physicians’ propensity to adopt computer-based decision tools.