Practical Application of Content Analysis

Content Analysis stems from media and social research and has been used in a wide range of disciplines. The purpose of content analysis is “to organize and elicit meaning from the data collected and draw realistic conclusions” (Bengtsson, 2016, p. 1). Even though all qualitative research involves interpretation to varying degrees and levels of abstraction, content analysis provides a highly credible and rigorous quantitative approach to extracting meaning from text.

Downe-Wambolt's (1992) definition of content analysis as “a research method that provides a systematic and objective means to make valid inferences from verbal, visual, or written data in order to describe and quantify specific phenomena” emphasises two aspects. Firstly, it presents content analysis as a technical and objective methodology that is reliable and repeatable, thus precluding the researcher’s personal authority. Secondly, it establishes content analysis as a latent and qualitative method that attempts to understand relationships in the text without ignoring the initial context. This is appropriate to this study because I am attempting to elicit factors that have a causal relationship to ART patients falling out of care.

To this affect, Content Analysis can be used to determine the presence of certain items (i.e., words, concepts, constructs, and themes) in a given text. Subsequently, researchers are able to quantify, interpret and extract meaningful insights, patterns and relationships of these items in the text. It is a process that identifies meanings of interest and ensures that the researcher’s interpretation of this meaning ‘stays true’ to the text and achieves trustworthiness. Content analysis also enables us to contextualise data into different ecological contexts (i.e., individuals, community, social, political etc.) or domains of measurement (i.e. clinical, biological, demographic, and behavioural measures).

The following is an outline of the stages of the content analysis, as described by Bengtsson (2016), that this study will pursue:

Analysis Method: The study will utilise qualitative content analysis and deductive reasoning approach. Qualitative content analysis emphasises a latent interpretation of the text and it is appropriate as I aim to uncover causal factors which may not be explicitly stated in the text (explicitness as measured against my unit of analysis) (Bengtsson, 2016; Shava et al., 2021). The analysis method will also slightly extend into quantitative content analysis (which emphasises the frequency of meaningful insights) because causal factors that emerge more frequently imply a strong consensus for a particular factor. Deductive reasoning stems from the study’s hypothesis that there are multiple causal factors related to falling out of care. Through examining and comparing the collected research literature, I will look for these predetermined relationships, themes and patterns.

Aim of Content Analysis: To elucidate the causal factors of ART patients falling out of care.

Unit of Analysis: This can also be interpreted as a unit of meaning (Bengtsson, 2016). It is the smallest unit of text that contains some of the insight that the research seeks and answers the question set out in the aim (Rose, Spinks, & Canhoto, 2014). In this step, I will determine which type of units in the selected literature would indicate a causal relationship with falling out of HIV care.

Proposed units of analysis:

• Phrases or textual themes that indicate a causal relationship with falling out of care (non-adherence, non-retention, or lost-to-follow-up).

  
  

• Phrases/words indicating causality: “factor”, “positive association” “co-relationship”, “link”, “relationship”, “tie”, “affiliation”, “tie”, “correlation”, “influence”, “causal relationship”, “contributing/determining factor”, “determinant”, “predictor”, “influence”, “risk factor”, “consequence”, “indication” and “association”.

  
  

• Phrases/words indicating impediment to HIV care: “hinders”, “obstacle”, “prevents”, “inhibit”, “barrier”, “challenge”, “hurdle”, “block”, “deterrent”, “disruption”, “setback”, “inhibition”, “hamper”, “inhibition”, “constraint”, and “impede”.

  
  

• Concepts related to target outcome – falling out of ART care: “non-retention”, “attrition”, “non-adherence”, “lost-to-follow-up”, “missed/missing appointment”, “disengagement”, “treatment interruption”, “suboptimal adherence”, “early discontinuation”, “loss of viral suppression”, and “ART/ARV/medication/treatment fatigue”.

  
  

• Data located under the following sub-sections: “results”, “findings”, “discussions”, “literature review”, “background”.

A sample of this in the selected research literature will look like this:

(target causal factor) “factor”, “positive association” “co-relationship”, “link”, “relationship”, “tie”, “affiliation”, “tie”, “correlation”, “influence”, “causal relationship”, “contributing/determining factor”, “determinant”, “predictor”, “influence”, “risk factor”, “consequence”, “indication” and “association”) . . . (“hinders”, “obstacle”, “prevents”, “inhibit”, “barrier”, “challenge”, “hurdle”, “block”, “deterrent”, “disruption”, “setback”, “inhibition”, “hamper”, “inhibition”, “constraint”, and “impede”) . . . falling out of ART care (“non-retention”, “attrition”, “non-adherence”, “lost-to-follow-up”, “missed/missing appointment”, “disengagement”, “treatment interruption”, “suboptimal adherence”, “early discontinuation”, “loss of viral suppression”, and “ART/ARV/medication/treatment fatigue”).

Following this process, the relationships identified will be condensed into usable input variables or attributes. Therefore, a coding scheme will be developed iteratively from the data. This process will inevitably lead to both manifest (observable) and latent (unobservable) variables. The accuracy of the prediction model relies not just on the input data's quality but also on the representativeness of the model. Data containing outliers, incomplete sets, or errors significantly restrict the predictive capabilities of the model (Hara, Piekutowska, & Niedbała, 2021). For this reason, a Delphi survey will draw the sentiments of panel experts about the fit of input data and model features for a prospective predictive model.