A flowchart of the study design is depicted in Figure 1. This is a study of personal patient stories published on websites hosted by 3 established colorectal disease support organizations: Bowel Cancer Australia, Bowel Cancer UK, and Bowel Cancer NZ. These countries were chosen as they are all English-speaking and have universal health care access. This project will access the public domain sections of these websites in which people post accounts of their experiences under banners such as “real life stories” or “your stories.”

There is no established ethical stance relating specifically to research involving unsolicited web-based narratives. We have obtained permission from each of the organizations to analyze these personal accounts and to use deidentified excerpts and quotes in reports of findings from the study. The organizations that host the websites have agreements with individual patient contributors regarding the use of their information and narrative material. These contributors are not direct participants of our study. However, the study of unsolicited autobiographical narratives is a unique research space with particular ethical issues relating to recruitment [47]. To establish the ethical position of this study, we refer to the Australian National Statement on Ethical Conduct in Human Research (2007-Updated 2018) [48], which indicates that privacy concerns arise when the proposed access to, or use of, the data or information does not match the expectations of the individuals from whom this data or information was obtained or to whom it relates. Therefore, we were granted ethical and scientific approval for this project from the Macquarie University Human Research Ethics Committee (MQ HREC Reference No:52020666115757). In publishing their personal accounts on the selected colorectal support organization websites, these contributors agreed that their stories would be made available for public access and used to raise awareness of young-onset colorectal cancer. This study meets the expectations of the contributors. Moreover, this study does not place burdens of active research participation on these potentially vulnerable contributors [49,50]. Additionally, unsolicited accounts enact the values of patient-reported data.

We defined inclusion and exclusion criteria to identify the types of personal accounts published on these sites that would be relevant to the study’s aims. We will include personal accounts that are written by people diagnosed with colorectal cancer (self-reported disease state, including but not limited to cancer of the colon, cancer of the rectosigmoid junction, and cancer of the rectum) [13]; before their 50th birthday; published in the public domain spaces of websites hosted by the 3 prominent colorectal disease support organizations, under agreement for the public dissemination and republication of the material; written by people aged 18 years or older at the time of submitting their personal accounts for publication on the website; and autobiographical, first-person accounts of experiences and outcomes relating to care for colorectal cancer. We will exclude personal accounts from the study if they solely comprise feedback on, or criticism of, a named institution or clinician or substantially describe someone else’s experiences and outcomes relating to care for young-onset colorectal cancer. We are not including serialized narrative material published as ongoing weblogs or blogs. We will take a random sample of 30 personal accounts from the eligible selection of personal stories using the Microsoft Excel (2011) random function, comprising 10 samples from the patient stories published on each of the 3 websites.

Narrative accounts will be downloaded from websites and collected and analyzed using the qualitative analysis software NVivo 12 Plus (QSR International) [51]. To avoid identification of individuals, each story will be deidentified and assigned a unique identifier code. Our framework method for qualitative analysis [24] identifies a priori what features to account for in our research reporting [16]. We detail our process for establishing the analytical framework in the following section on PRE and PRO domain coding. The PROD approach involves 4 key steps in creating a new structure for the data, as shown in Figure 2: (1) classifying attributes of the contributors, their disease states, their routes to diagnosis, and the clinical features of their treatment and posttreatment; (2) coding each line of each narrative into PRE and PRO categories and domains according to phases of the patient pathway; (3) thematic analysis of content within each domain; and (4) quantitative analysis of the narrative content.

Manual coding and analysis will be undertaken by the process of line-by-line attention to the content in a series of iterative readings. Consistent with the principles of qualitative research, each step of the data extraction and analytical process will be undertaken by at least two researchers [18], as qualitative work with narrative data is interpretive, even when coding to a framework of categories and domains established a priori.

With research questions to guide their choices, 2 researchers working together and constantly comparing their findings can arrive at agreement on the significance of the narrative content and the conclusions that can be drawn from it [18]. The third researcher will validate the findings of the thematic analysis, the fourth researcher will undertake the quantitative analysis, and the team will collaborate to reach consensus on the significance of the findings in relation to triangulation with extant colorectal PRE and PRO data.

The first step of the PROD analysis is to identify and classify the key demographic characteristics of the patient contributors, their disease states, the features of their diagnostic pathways, and the clinical features of their treatment and posttreatment phases. We will organize these data in a framework of yes/no and multiple choice categories. The sets of selections are based on conventional research participant attributes and adapted to the level of detail obtainable from unsolicited narratives. In these accounts, attributes such as age, gender, relationship status, and current disease status information may be unknown from the basic information provided in a source website. These characteristics may only be identifiable with close attention to both content and language in a narrative [52], and even then, may only be inferred from implicit clues [40].

To develop a set of domains and subdomain items for the a priori analytical framework, we reviewed literature on core outcome sets for PRE and PRO measures [2,23,53] and mixed-methods approaches for analyzing PRE and PRO data [12,54-57]. Our conceptual approach to PRE and PRO domain coding is presented in Figure 3. The domains and subdomain items comprised in our analytical framework are presented in Table 1 and also described below.

From patients’ perspectives, experiences of health services and outcomes of care occur as a continuum of patient journey within and across the phases of the patient pathway [40]. We will undertake a patient pathway analysis of the PRE and PRO data by coding for 3 key phases of the patient pathway: diagnosis, treatment, and survivorship/palliative care (Figure 2) [36,68,69].

There are 4 steps in our thematic analysis: coding for concepts, categorizing codes into groups, detecting patterns across categories, and interpreting themes within and across these patterns [70]. This process transforms the text into a narrative dataset, moving from highly descriptive findings to highly interpretative findings [16,20].

We will investigate opportunities to interpret the data quantitatively by means of network analysis [71] and quantitative text-based analysis, which uses automated natural language processing to analyze topics across different documents [12] and can measure sentiments within texts. This method may draw out aspects that contextualize other findings [12,13]. Quantitative approaches to analyzing unstructured text are emerging; however, as yet, there is little consensus on optimal strategies [12,46].

Our methods will have limitations, including that we will be dealing with text not written for research purposes, not all text will map to our framework, and the data are subjective and will require interpretation. Additionally, data reported in different health systems will need to be seen in the light of those structural and contextual differences. Further, regardless of validity, there are limitations to standardized instruments and these limitations will be reflected in the a priori domains and dimensions that are the foundation of our analytical framework.