Sources of data will include a scoping review of published ethics codes, a narrative review of legal case law relevant to feedback and reporting (completed), and a systematic review of empirical quantitative and qualitative data using previously published methodology for mixed methods meta-synthesis [19]. For the empirical qualitative data, we will use the Grading Recommendations Assessment, Development and Evaluation-Confidence in Evidence from Reviews of Qualitative Research methods for grading the evidence [20]. These reviews will collectively represent the available evidence for consideration by the stakeholder panel.

We will create 8 to 10 IMED scenarios to which the evidence from step 1 may apply. Panelists will be provided with the scenarios as well as the evidence synthesis as a written report. They will be asked to propose modifications to the CANDOR process based on the evidence synthesis and their expert judgment. Participants will respond anonymously.

The study team will compile the responses, and the panelists will then be brought together in a face-to-face meeting (videoconference if necessary). The aggregate proposals will be presented for discussion. During the face-to-face meeting, we will utilize nominal group technique to encourage contributions from all stakeholders and to prioritize recommendations. Through this technique, each participant will have opportunities to share their priority proposals with the group in turn. A facilitator will record and further consolidate proposals as needed. Thereafter, the group will discuss each proposal in turn and further prioritize them using the multivoting procedure. Discussions will be audio-recorded for further analysis.

The research team will then summarize the written and audio-recorded recommendations and deliver them to the stakeholder panel in a written report. Issues of disagreement and areas requiring further elaboration will be highlighted in the report as specific questions. The stakeholders will again be asked to propose answers to the specific questions as well as modifications or revisions to the recommendations generally. Stakeholder responses will be received by email and anonymized.

The study team will compile the responses, and the panelists will be brought together for a second face-to-face meeting. An iterative facilitated process will follow, through which panelists will have an opportunity to provide feedback on the draft recommendations and approve the final recommendation.

The research team will then provide final recommendations of the stakeholder panel in a published report. The expected outcome from aim 1 will be a modified process for the transparent communication and optimal resolution of errors identified between systems—the intersystem CANDOR process (ICANDOR) based on a comprehensive stakeholder analysis. The modified process (Figure 2, adapted from the CANDOR Toolkit [5]) will describe ICANDOR system activation, provide recommendations for feedback and/or reporting, and establish guidelines for disclosure in these scenarios.

CFIR is a meta-theoretical framework that comprises 39 constructs across 5 domains consolidated from published implementation theories to systematically assess contextual factors influencing practice change. Domains include intervention characteristics, outer setting (eg, external policies and incentives), inner setting (eg, implementation team communication), individual characteristics, and implementation process [13]. CFIR was selected because it provides a pragmatic, consistent typology applicable across multiple implementation contexts. Because CFIR focuses on organizational characteristics, we will supplement the survey with constructs from TDF to enable a thorough evaluation of individual behavior change constructs. TDF was developed for implementation research to identify influences on health professional behavior; it consolidates 33 theories of behavior change into 14 domains [15,25]. On the basis of our study and previous studies by others on error resolution [10,11,26], we will select the CFIR and TDF constructs most likely to be the potential determinants of implementation or to have sufficient variation across organizations.

We will purposively sample 5 of the 69 NCI-designated cancer centers in the United States (excluding the centers selected for pilot testing in aim 3) to maximize diversity in site characteristics (eg, size, geographic region, and affiliation with a university medical center; Figure 3). We have chosen cancer centers as the site of testing because identification of errors between facilities is particularly relevant to complex oncologic care. The screening, diagnosis, and multidisciplinary management of cancer requires patients to interface with multiple physicians and facility types with varying levels of integration [27]. In the cancer care environment, consequences of errors can be especially harmful, further complicating the willingness or responsibility for disclosure of the discovering provider [28,29]. Our preliminary work suggests cancer specialists lack consensus on whether or how to communicate about these errors to patients and responsible providers [10,11].

For key informant interviews, we will contact institutional risk management offices through publicly available data center websites or professional contacts. We will introduce our research and identify and recruit key hospital personnel at each site who led CANDOR process implementation, if applicable, or are involved in patient safety and error resolution. These may be risk managers, patient safety and quality improvement personnel, legal counsel, or clinical ethicists (3-5 interviews per site). We will use snowball sampling [30] supplemented with information-rich informants to ensure representation of diverse perspectives. Interviews (60 min approximately) will be recorded, transcribed, and imported into a software that supports qualitative/mixed methods analyses. We will analyze data using framework analysis in the following steps: (1) immerse in the details of each transcript, (2) use CFIR constructs as key themes, (3) code transcripts with CFIR framework to identify recurrent subthemes, (4) summarize data in a matrix, and (5) synthesize data by comparing across cases. Preliminary analysis will be performed iteratively with interviews to assess sample size for appropriate information power up to a total of 25 individuals. Information power is an approach to estimate maximum sample size in qualitative studies that considers salient study characteristics affecting the amount of relevant information a sample is likely to provide. This maximum sample size takes into account the focus of the study aim, the specificity of the sample to personnel with experience in the topic, our prior experience with the quality of dialogue with hospital personnel [10,11], and the theoretical framework that will structure both the interview guide and analysis [31].

A random sample of clinicians (50 per site; medical, radiation, and surgical oncologists) will be recruited to complete the survey at each of the 5 sites (n=250). A roster of physicians and their contact information will be obtained from Web-based Find a Physician registries of each cancer center. A research assistant will confirm accurate mailing addresses by phone before mailing. Physicians will be recruited by letter with an attached survey and a nonconditional cash incentive [32,33]. We will use the Dillman method of survey administration [34] to achieve optimal response rates with a goal of 65% [27,35-37]. The survey will be developed to determine barriers and facilitators to ICANDOR implementation, using CFIR constructs as a guide for selecting survey items, augmented by the TDF to fully explore individual behavior. The questionnaire will be pilot-tested for face validity, clarity, and stability over time with 15 participants representing all study sites, who will be excluded from subsequent participation. Face validity will be established by qualitative assessment via field notes with pilot respondents and other nonparticipant stakeholders. Clarity will be established by observing pilot participants completing the survey, asking them to think out loud as they are completing the survey, and identifying any items requiring further clarification. Finally, stability will be tested using test-retest methodology [38]. Likert scale responses will be dichotomized as follows: (1) likely/very likely versus not sure/unlikely/very unlikely. Summary statistics will include sample size, mean, median, SD, and range for continuous variables, and counts and percentages for categorical or ordinal variables.

Summary statistics for clinician surveys at each site will be imported to the qualitative data analysis software and linked to the qualitative analysis of key informant interviews at the site level (ie, each of the 5 sites will be analyzed as a case). This will enable mixed methods analysis by examining potential patterns in the data among the 5 sites.

We will measure 5 implementation outcomes from the Proctor et al [41] taxonomy of outcomes—acceptability, appropriateness, reach, adoption, and feasibility (Table 1). To assess initial acceptability, appropriateness, and feasibility, we will conduct key informant interviews (n=10 at each site, or until appropriate information power is achieved [31]) in the planning phase. Following the clinician training in the dissemination and implementation phase, we will conduct short posttraining surveys to measure adoption. Finally, at month 12, we will invite all cancer specialists participating in the training sessions (estimated n=100) and the error resolution staff (n=20) during the study period to complete surveys to reassess acceptability, appropriateness, and feasibility. To gain a greater understanding about the quantitative survey findings, we will then purposively sample 2 or 3 respondents within each stakeholder type at each site (n=30 total, or until appropriate information power is achieved [31]) with very high or very low scores to participate in semistructured interviews. We will also measure reach by determining the number of unique providers who report an ICANDOR event, triggering use of the implementation toolkit, during the study period.

Given this study’s focus on acceptability and feasibility, our analysis will be primarily descriptive. For analysis of the quantitative data, we will calculate descriptive and bivariate statistics on survey responses (acceptability and appropriateness). We will then generate mean acceptability and appropriateness scores for each component of the implementation toolkit. For key informant interviews, all meetings will be recorded, transcribed, and analyzed using rapid assessment. Using a joint display organized by prespecified implementation activities, we will visually merge findings from qualitative and quantitative data analysis, presenting quantitative scores with representative qualitative quotes. Monthly ICANDOR adoption rates in the 6 months after pilot implementation will be measured. We will present findings to the stakeholder panels and refine the implementation toolkit to include a detailed description of the implementation planning process, advice about toolkit use, and improved strategies and tools.