Dignity

To translate the categories of D&A identified in the review [12 ] into measurable domains, investigators from two USAID-TRAction funded projects (in Kenya and Tanzania) met to harmonize and contextualize the working definitions of D&A during childbirth. The team discussed research methodologies and developed common definitions of D&A in a Construct Map. A detailed description of the definitions is published separately, focusing on normative and experiential building blocks [13 (link)]. The focus of the current measurement is based on experiential building block that took account of women’s experiences of disrespect and abuse. These were a specific set of behaviors or conditions agreed by all stakeholders to constitute disrespect and abuse. The basis of this definition is that if the goal is to promote women’s dignity in childbirth, then it matters if a woman experiences her treatment as disrespectful and abusive. Such an experience is likely to influence future decisions about where to deliver and whether to recommend that facility to others [13 (link)].
The second dimension of definition of D&A includes the normative building block which comprise codes of behavior or infrastructural standards, where departure from these standards could be considered violations constituting D&A. The normative block has four key dimensions: human rights law, domestic law, ethical codes and local consensus on behaviors [13 (link)].
The experiential building block, refers to events or conditions considered as D&A, regardless of patient experience or provider intention and classified into three dimensions: 1) subjective experiences whereby women experience D&A even if it does not result from actions observed; 2) objective events or conditions that are observable actions experienced or intended as such; and 3) intentionality, whereby a woman does not interpret an action as D&A, but the provider actually intends it as disrespectful or abusive [13 (link)]. Subjective experience of D&A was measured through the client exit survey described in this paper. Table 1 outlines the normative and evidentiary building blocks and provides examples of actions and behaviors that may be experienced as disrespectful and how they link to the building blocks.
With a set of definitions, measurement instruments were developed and validated through qualitative interviews with clients to identify potential gaps in the Construct Map. A client exit tool was developed and validated through an exit survey conducted among 75 respondents. In order to check the reliability of the exit tool in estimating the prevalence of D&A, we further conducted follow-up case narratives two weeks later among 25 participants who reported any form of D&A in the exit survey and 25 others how did not report any form of D&A. The outcome of this analysis enabled us to refine the tools for measuring the prevalence of D&A.

The machine learning approach had two components: (1) pre-processing, in which data from patient comments are split into manageable units to build a representation of the data [17 ], and (2) classification, in which an algorithm decides which category each comment falls into. A consistent set of methodologies was applied in our machine learning process, including a “bag-of-words” approach, “prior polarity”, and “information gain”.
In the “bag-of-words” approach, the total body of words analyzed (known as the corpora) is represented as a simplified, unordered collection of words [18 ]. For this analysis, unigrams (single elements or words) and bigrams (two adjacent elements in a string of tokens, in this case, a 2-word phrase) were used as the basic units of analysis. We extracted 5695 n-grams in total. Higher n-grams (longer phrases) could have been used, but the constraints were computer power and processing time. We also included our own classification of certain words in the machine learning approach, known as “prior polarity”. The 1000 most common single words, and the 1000 most common 2-word phrases were extracted from the complete set of comments in the corpora. Two researchers independently rated the sentiment of each as positive, negative, either, or neutral separately for each of the three domains under consideration: (1) overall recommendation, (2) cleanliness, and (3) dignity. Where disagreements occurred, the sentiment was discussed and resolved between the 2 researchers. Kappa statistics for overall ratings were .76 for 1 word and .71 for 2-word phrases. For rating of dignity, they were .71 for 1 word and .70 for 2 words. For rating of cleanliness, they were .52 for 1 word and .48 for 2 words. For all of these calculations, P<.001.
A technique called “information gain” was used to reduce the size of the bag-of-words by identifying those words with the lowest certainty of belonging to a given class, and then removing them—this is an approach to feature selection [19 ]. This improved the computation time and also demonstrates the words with highest predictive accuracy.
A number of different technical approaches can be taken to classification in machine learning. We applied four different methods, to see which gave the quickest and most accurate results: (1) naïve Bayes multinomials (NBM) [20 ], (2) decision trees [21 ], (3) bagging [22 ], and (4) support vector machines [23 ]. Decision trees and bagging were carried out with REPTree in the Weka package. Support vector machines used an RBF Kernel. The accuracy of the prediction was compared with the patient’s own quantitative rating by calculating, for each method, the accuracy (the percentage of correctly predicted observations from the total number of observations), the F measure (the harmonic mean of precision and recall), the Receiver Operating Characteristic (ROC), and the time taken to complete the task were calculated. To reduce computing processing time of the classification, we limited the words in the learning process to the top 10,000 words by frequency. All text was converted to lower case, and we removed all punctuation. Typographical errors and misspellings were not corrected.

The analysis was conducted in accordance with Merriam and Tisdell (2016 ) and their description of the “step-by-step process of analysis”. First, the interviews were sorted based on the three care contexts—residential care, home care, and specialized palliative care—and we all, separately, read the interviews, one care context at a time. The division into three contexts was based on where the participants mainly lived their daily lives and received long-term care—in a residential care facility, in the person’s home with approved assistance from the care community, or in the person’s home with approved specialized palliative care. Within each context, all texts that we agreed concerned existential loneliness were extracted. The extracted narrations covered, for example, limit and vulnerable situations, while narrations about how existential loneliness was eased, or when the interviewees wanted to be alone, were not included. After having read and discussed the material in the group several times, we all agreed that existential loneliness in different care contexts among older persons was linked to different circumstances and situations. For some, existential loneliness was connected to illness, while for others, existential loneliness was linked to the care provided and/or to life itself. During the readings, recurrent in the narrations about existential loneliness was the notion of suffering and this led us further into the theoretical framework by Katie Eriksson (1994 /2006) about the suffering human being. According to Eriksson, suffering of illness involves experiences in relation to bodily concerns and/or shame and guilt caused by illness and treatment. Suffering of care involves experiences in the caring situation such as uncertainty and waiting, condemnation, and reduced dignity. Suffering of life involves suffering related to what it means to live, and to be a human being among other human beings. We decided to use her three concepts in the next reading of the narrations—suffering of illness, suffering of care, and suffering of life. The choice to use Eriksson’s theory was thus initially based on the readings of the interviews. We realized that we needed a structure to organize the extensive interview material not only according to three different care contexts but also according to a structure relevant to the descriptions of existential loneliness. These readings revealed similarities between suffering (Bergbom et al., 2021 (link); Eriksson, 1994 /2006) and existential loneliness. In addition, we had found in a previous empirical study that existential loneliness could be triggered in encounters with health care (Sjöberg et al., 2018 (link)), what Eriksson refers to as suffering of care. The next step was therefore to construct an analytic grid, described by Merriam and Tisdell (2016 ) as analytical coding, for clustering the extracts about existential loneliness from the three care contexts into Eriksson’s types of suffering; see Table 2. The first author made a first draft where texts in the grid were put together into categories and were given preliminary codes. We thereafter all met again several times to discuss the categorization and the preliminary codes. To be able to distinguish patterns in the material, the categories were compared between the three care contexts. These discussions resulted in several changes, and discussions continued until consensus was reached.
Table II.

The analytic grid.

	Care contexts:
Existential loneliness related to:		Residential care	Home care	Specialized palliative care
Suffering of illness
Suffering of care
Suffering of life

Ethical clearance was granted by the Research Ethics Committee of the University of Zululand (ref. no. 171110-030 PGM 2019/127); this step was followed by a request for permission to conduct the study from the Free State Department of Health Research Unit and the School of Nursing in the Free State. The participants were given information letters explaining the purpose of the study, and those who were interested to participate in this study signed the informed consent voluntarily, without being coerced. Their rights to autonomy and justice were upheld throughout the study by allowing them to participate voluntarily without risking prejudicial treatment and by treating them with respect and dignity (Polit & Beck 2018 ). Confidentiality, anonymity and privacy were also adhered to by using codes such as P1 during data analysis process, and all transcripts (including audio-recordings) were kept in a computer, which will be password protected in a locked office for a period of 5 years.