Coding: a two-phased inductive approach

While the project has a clear aim of understanding how practitioners encounter (potential) critical moments and how they avoid getting into critical situations, there is no a priori determination of which particular issues may be relevant to study further. There is an interest in negative and positive feedback mechanisms within teams, but no decisive knowledge exists of what exactly defines a positive or negative feedback mechanism at a concrete, practical level within this field (see chapter 1). Thus, particular cases of interest are suggested empirically from the data. An important remark on identifying these cases will be made.

As coaction relies on direct action and perception, it consequently operates with processes without inherent starts and ends. Interestingly, several approaches to spoken language and cognition treat their object of investigation as a fixed and demarcated object.

However: “a special sense impression clearly ceases when the sensory excitation ends, but a perception does not. It does not become a memory after a certain length of time. A perception, in fact, does not have an end. Perceiving goes on.” (Gibson, 1979/86:253).

Gibson continues to argue that verbal tenses bias our understanding of time and perception.

An epistemological distinction is simply confused for an ontological process (Gibson, 1979/86:253). When activities are disrupted into situations or events, we need to define a beginning and an end with the use of external criteria (Gibson, 1979/86:101).

The coding process is an interpretive technique that helps organise, structure and prepare the video-data for detailed analysis based on such external criteria (Seale, 2012). In this case, the overall coding process is an iterative process that uses a two-phased inductive approach, only framed by an external criterion of function in relation to work tasks: how cognitive events are managed in relation to the overall task.

The educational staff at the ward explained to me how they, in the ward, work with well-defined hypotheses of what develops as erroneous and successful activities. At the same time the healthcare practitioners explicitly uttered a frustration of not getting past the surface of what happens beyond their immediate level of reflection. For instance, they had the feeling that working in ad hoc teams was a huge challenge and often caused time delays, frustration etc. However, the enabling conditions or the feedback mechanisms within the system, were less easy to identify. First, this calls for thorough investigations of the pico-scale dynamics that lead to errors and successes beyond that which teams experience and are able to explain. Second, to get beyond pre-established hypotheses of which situations are relevant to scrutinise further, I started with an exclusively data-driven approach to coding. The first categorisation I made represented - as objectively as possible - what happened and this organised the data into large categorical chunks labelled with categories that were used within the dataset rather than theoretical labels. This process gave an overview of the dataset with minimal interpretation. The coding process established an idea of the general patterns across the dataset with notifications of relevant general pivots that needed to be investigated further.

Roughly, the coding process is divided into two main phases. However, the process is fluid and iterative and can be illustrated in the following figure, inspired by Seale (2012).

The table and its theoretical concepts are elaborated subsequently.

Two-phased inductive coding process

1) coding for concepts/categories, tasks and subthemes via open coding 2) linking task-based events and subthemes to emerging hypotheses

1

- Immersion in the data (notes and comments)

- Identify recurring and important categories (making connections within the dataset)

- Indexing in coding scheme (Simplification through categorising. Enlisting the data chunks into similar task-based events and subthemes. Defining topics for further analysis to narrow down the amount of data remarkably)

2

- Charting, hypotheses generation and final selection (hypotheses emerge when cross-relating the most salient task-based events and subthemes in the coding scheme. The thematically-developed hypotheses serve as an organising principle for analyses)

Initially, the first phase included a basic and rough overview of the enormous corpus of raw data material. As a starting point, I viewed all recordings several times and from the various angles they were recorded. Most recordings were recorded with three cameras, and at a minimum with two. One of the cameras was handheld. First, the data were trimmed down by a criterion of overall usefulness: for instance acceptable visual and audio quality.

The data were initially reduced to include 14 treatment situations, which amounted to 13

hours times two or three depending on the number of cameras used. The recordings varied in length, degree of emergency, practitioner experience, and team or individual performance.

After the rough viewing, I identified and connected recurring and salient categories within the dataset. The 14 treatment situations were analysed into small, demarcated task-based events that were inductively defined from watching how multiple elements in diagnostic processes and pre-treatment contribute to achieving the overall goal. In other words it is a task-based criterion that determines an activity as a local nested event within the overall diagnostic situation. Second, a criterion of cognitive function determines how a task includes subthemes: for instance as tasks are permeated with interruptions, hesitation etc.

To begin with, the coding scheme was broad and the coding categories were rather incomparable across the dataset as it used a data-specific language via open coding⁹ to label chunks of data. After coding all 13 hours by this approach, the chunks and codes were interpreted and reduced by the identification of recurring task-based events and subthemes, for instance physical examination and documentation tasks. The categories were boiled down to a few overarching and relevant tasks to simplify and limit the number of thematic focus points. Also, this simplification made comparisons across the dataset possible. Specifically, 22 categories represented a specific medical or interpersonal task or related activity. The categories were used for framing episodes into delimited task-based events or subthemes within the overall interactivity trajectory that defines an overall and shared project: diagnostic practices and pre-treatment of patients. Each category is numbered from 1-22. For an overview, see table 4.1¹⁰ below:

2 History taking The healthcare practitioner (often the doctor) asks for information, (medical history etc.)

28 3 Physical

examination The healthcare practitioner examines the patient 27 4 Summarising (to

  9 Essentially open coding means to break data into conceptual components (Strauss and Corbin, 1990). This process allows data to be grouped and organised.

10 The table is a simplified overview of the categories used in the detailed coding scheme that encompasses explicit references to the video-data and detailed information about the involved practitioners.

progression, diagnosis etc.

7 Pre-treatment The healthcare practitioner initiates a pre-treatment (medication, oxygen etc.)

discussion Healthcare practitioners discuss the clinical and medical situation. This includes evaluation and hypothesis-generation about the patient’s medical condition

12

10 Nursing The healthcare practitioner (often the nurse) handles clinical and interpersonal trifles

(undresses the patient, brings water and food to the patient etc.)

9

11 Patient handover The medical team receives the patient from paramedics and ambulance personnel

13 Small talk Small talk within the medical team and with the

patient 8

14 Technical issues Problems related to technical issues (the computer freezes, the phone is low on battery

16 Reporting The healthcare practitioner reports a patient to a specific medical speciality at another hospital unit

3

17 Interruption The medical team or the healthcare practitioner is being interrupted

app. 30**

18 Closing The healthcare practitioner finishes the

interaction with the patient and leaves 13

20 Hesitation The healthcare practitioner hesitates 1*

21 Explicating The healthcare practitioner informs the patient 2

procedure about what he is doing (for instance, that he is going to key information into the electronic medical record)

22 Patient or relative interruption

A patient or his relative(s) interrupt(s) the healthcare practitioner

4

Table 4.1 Coding scheme

*Often practitioners briefly leafs through the medical record or hesitates during task performance, however, the annotations in the coding scheme cover the main activity that the practitioner/team performs. In cases of, for instance hesitation or team performance, situations of noticeable instances are annotated.

**The number of instances in row 15 and 17 are approximate annotations as I lost access to the data (see preface).

The coding scheme does not contain information on whether or not the doctor and nurse are novices or highly experienced, the degree of emergency, duration of each event and overall treatment situation as well as particular professional roles a team constellation includes. This information is added in the field-notes and it is explicated where relevant in the analyses.

In total 333 occurrences were identified in the coding scheme. The prevalence of some tasks is related to obligatory procedures that need to be followed. For instance, no patient can enter the ward without having relevant medical values measured or without providing a medical narrative. Other tasks are tokens of optional or rare instances. For instance:

‘reading and interpretation of documents’ (row 8) or ‘technical issues’ (row 14).

Apparently, there is no pattern in the way activities are completed and the number of practitioners involved except for ‘professional discussion’ (row 9) and ‘team performance’

(row 19). In 55.3% of all cases, one practitioner alone completes a task. In 44.7% of the cases the task is performed with two or more practitioners present at the same time. The original coding scheme further includes qualitative comments, which gives a solid basis for classifying diagnostic situations into procedural events alongside more informal and interpersonal events. Additionally, it gives an indication of relevant pivots within the overall flow of interactivity that need to be scrutinised further in the analysis. As the coding scheme opens for an overview of ‘what happens,’ it unveils that some tasks are procedural and highly frequent but rather unproblematic, e.g. ‘measuring medical values’

(row 1) even though several interruptions occur during this task performance - whereas other procedural tasks are managed with more difficulty, e.g. ‘documentation in the electronic medical record’ (row 6). Finally, other unique instances (e.g. row 17 and 22) encompass interesting information on how the unexpected is dealt with in the situation.

Thus, the coding of various tasks brings forth an overall pattern of procedural and particular episodes within diagnostic events and it shines light on some aspects that need to be investigated further in analysis.

4.3.1 Hypotheses and event identification

Interpreted on the basis of the comparisons between the comments and annotations across the coding scheme, the most salient tasks that relate to the overall research question were scrutinised further. During this process, hypotheses about how cognitive and interactional aspects related to successful and less successful events emerged.

The hypotheses relate to how cognitive and interactional aspects in emergency situations constrain or support the emergence of error cycles. To illustrate this point, an example is given. For instance, in situations where material tools provide measurements that diverge from explanations coming from the patient, different strategies for handling this discrepancy were identified in the dataset. Some practitioners became biased by objective representations and they immediately prioritised medical guidelines over the patient’s subjective first-hand explanations. Others contained the inconsistency and adapted flexibly as they balanced hard facts with “hard narratives” and sorted out what the discrepancy was about. This difference led to a hypothesis about medical expertise: guidelines, medical procedures and material artefacts both scaffold and constrain diagnostic processes. The crucial point in successful diagnosis and treatment is to know when, and when not, to rely on rules and artefacts as aiding resources.

Multiple hypotheses were generated and related to positive or negative feedback mechanisms that needed further attention. The analytical chapters thus, investigate possibilities and challenges in diagnostic processes and relate these results to the hypotheses generated. Thus, based on the initial coding process, the following six themes were investigated further as hypothetical organising principles of the analyses: (a) medical visual systems; (b) interruptions; (c) diagnostic procedures; (d) medical cultural dynamics;

(e) sense-making in teams and (f) writing the electronic medical record.

Each chapter, thus, focuses on a thematic aspect related to the systemic function of cognition and at an overall level, the analytical chapters contribute to the overall investigation of the overall research question. Each chapter shows the diversity of enabling conditions in the interactive social practice of human errors in emergency medicine.