Codes identify and provide a label for an aspect of the data that is relevant to the research question. In phase one, described above, I examined how individuals wrote about their interactions with the health care staff, the environment of the ICU, and the patient/family member. That examination (open coding) created one source for potential codes. Some examples of the themes that surfaced during open coding include statements of appreciation, comments about ICU not otherwise categorized, before/after experiences (e.g. before/after a DNR order), and care provided at other parts of the hospital or other hospitals.
Central to my codebook creation were categories of care specifically associated with FCC—the first type of code. The FCC codes are emotional support, communication, surroundings and environment, respect and dignity, patient physical comfort,
participation, family physical comfort, coordinated care, and collaboration. Themes from the research literature on family members who had a loved one that was treated (and may or may not have survived) in the ICU provided a second source of codes. Examples of these themes include quality of staff, quality of care, staff availability, family opportunity to say goodbye, and family access to the patient. A third category of codes referred to as codes of special interest to the study were included in the codebook. These
include death (direct mentions), ICU technology, spiritual care, other cultural
considerations, and social work, the code associated with the fourth research question. Finally, several misc codes were added as a fourth source to capture any additional data that seemed like it might be relevant or important as the responses were being coded. This phase of data analysis included the generation of the codebook. There were roughly 9 steps to my codebook development:
1. Labeled content in data. As mentioned above, I unsystematically made notes while reading through a hard copy of the data. I simply named things as I read them, wrote words (labels) in the margins.
2. Labeled topics in literature. I did something similar as I reviewed the literature. I noted categories, themes, and experiences mentioned in the literature that I imagined might be good codes. I made a note on the hard copy of the article if I thought it would be a useful source of codes.
3. Used other relevant sources. I also created a document where I collected concepts from a variety of sources (heard in relevant talks, read in the Boston Globe, other) that seemed like they would be potential codes.
4. Reviewed definitions of family-centered care. I collected a variety of definitions of FCC from the literature. Because FCC is rooted in pediatric medicine, not all
definitions were applicable. Following the recommendation of my committee during my proposal hearing, I chose definitions to use for data analysis and reviewed the selected definitions with my first reader. I met with committee member, Dr. Judith Gonyea, who is particularly knowledgeable about FCC, to review and confirm my selection. These definitions would later be operationalized into codes. To operationalize FCC into coding categories, I used the following three sources that were cited in the theory chapter (Institute for Family-Centered Care, 2016; Shields et al., 2006; Teno et al., 2001;
Wenrich et al., 2003) and an additional two sources from the literature review (Henrich et al., 2011 and Wong et al., 2015).
5. Created the codebook. I aggregated all relevant concepts—as well as their sources-- into one Word document. I then translated the concepts to nodes in NVivo. 6. Created categories and sub-categories. The next step was the creation of families of codes and sub codes or categories within them; these came primarily but not
7. Consulted with NVivo specialist. I next met with the NVivo technical and content specialist at BU so I could download the data set into NVivo and would be able to create case classifications within the data and run queries on these case classifications. During this meeting I also consulted him on analytic strategies. He highly recommended that I use fewer codes and confirmed that I could code for subthemes later in the analytic process. This made intuitive, practical and technical sense to me. Consequently I only used the parent codes (listed above) to code the data.
8. Streamlined the codebook. I then created a streamlined codebook in Word and returned to NVivo to create the nodes I would use to code my data.
9. Established interrater reliability and addressed threats to validity. During this time, I was also establishing interrater reliability practices. See section below labeled Addressing threats to validity for a description of this process.
Addressing threats to validity. Because I was the sole coder, I needed to include additional steps to address validity and reliability. Creswell (2013) identifies several measures to safeguard validity that I incorporated in this study. First, the coding structure and all coding modifications were discussed with my first reader. Second, the coding of multiple responses was discussed with my first reader in order to further refine the coding structure. Then I sent my first reader documents that included random samples of
respondent quotes to enable an independent reading and coding of the data. At other times I sent her a sample of respondent quotes, with the codes I had assigned to the excerpt, for her review. She reviewed all the excerpts. We discussed divergent opinions until we reached agreement. We reviewed over 174 cases via this process. This was greater than 25% of the data. I noted our comments and conventions. This form of review was used to seek alternative explanations to my propositions, and to avoid biases or efforts to force interpretations. I wrote memos (see Appendix C) to capture uncertainties, decisions, indecision, and reactions to the material that may have biased my analysis. As a measure of reliability (Creswell, 2013) and in an effort to reduce bias (Miles,
Huberman, & Saldana, 2014), all notes and memos were sent to my first reader for review. Because of her FCC expertise, I met with an additional committee member to solicit her guidance on the codebook and my efforts to address threats to validity. Once interrater reliability was established, I coded the entire data set with the completed codebook.
As mentioned above, the codes I developed came from multiple sources: some were derived directly from the words used by participants, whereas other codes were derived from the family-centered care literature. For example, descriptions of
participants’ experiences of getting their questions answered were coded with the label, “communication,” a concept addressing the information sharing aspect of experience in FCC theory. Once identified, an extract of data was coded and the text associated with it marked. A portion of data can receive more than one code; in most cases multiple codes were applied to a chunk of data. I discuss this in further detail and provide examples in the section labeled, “challenges in the development of themes,” in this chapter.
Data analysis paid particular attention to what surviving family members said about their experiences with social work. While there is important literature regarding social workers in end-of-life care, little literature focuses on the role of social workers regarding the family members of patients who die in the ICU (Delva, Vanoost,
Bijtterbier, Lauwers & Wilmer, 2002; Rose & Shelton, 2006). In this case the code, “social work,” was simply added to the codebook. Braun and Clarke write that this phase two of data analysis begins the systematic analysis of the data. After the data set was
completely coded, I was in a position to begin what I considered the formal phase of data analysis.