This chapter discusses the methods used to examine the words and phrases cancer patients use to describe their pain. The Instrument study is discussed first to provide context, including its sample selection and data collection. Then, the design, procedure, ethical considerations, and data analysis specific to this study will be outlined. Finally, validity and reliability will be addressed.
Data for this study were originally collected for a larger study entitled, Study of pain in the British Columbia cancer population, phase II: Development of an instrument evaluating cancer pain (Deschamps, Coldman, Bradley, & Baker, 1991), referred to here as the Instrument study. The interviews were transcribed but the data were not analyzed because the study ended when further funding was not obtained. Because I was the person who conducted the interviews, I have received permission to use data for this study (see Appendix B).
As noted in Chapter 1, nursing research in pain at the local provincial cancer treatment facility was focused in two areas: (a) determination of the incidence and prevalence of pain in the BC cancer patient population and (b) the development of a pain assessment tool. This study is concerned with the second area of the nursing research programme (i.e., the Instrument study). The first step of the Instrument study was designed to identify pain descriptors spontaneously used by patients experiencing pain related to cancer or its treatment. The pain descriptors were to be classified into groups based on the categories of the McGill Pain Questionnaire (i.e., sensory, affective, and evaluative). Its second step was designed to determine the intensity level of each pain descriptor and validate the classification scheme obtained in the first step. Patients were to be instructed to assign intensity values to each of the pain descriptors and to indicate which descriptors corresponded to the pain they were currently experiencing. Patients would then be instructed to state whether they agree if pain descriptors in individual categories reflected a similar type of pain. Finally, patients were to be instructed to rank the pain descriptors in ascending order of intensity.
I conducted 39 interviews for the Instrument study and 31 have been transcribed. It was not possible to transcribe all interviews due to poor sound quality (background noise, e.g., fire-alarm testing) and occasional technical difficulties with the audio equipment. Along with the 31 transcribed interviews, other interviews were conducted by a number of undergraduate nursing students participating in a research course exercise. Unfortunately, the quality of those interviews is variable. The students used a slightly different interview guide, and due to the large number of interviewers, inter-rater reliability is a confounding variable. The student interviews have not been transcribed and were not used in this analysis.
Participants were asked to use their own words to describe what their pain feels like. Most interviews lasted approximately one-half hour with none exceeding one hour. The interviews were conducted over a ten-week period during February, March, and April, 1991. The pain descriptors were collected in audiotaped interviews by asking cancer patients to describe their pain with the aid of an interview guide (see Appendix C). Face and content validity of the interview guide were determined by a Nurse-Research Consultant and an Oncology Clinical Nurse Specialist. Patient demographics were recorded at the time of the interview on a data collection guide (see Appendix D).
In addition to the interviews, participants were asked to score the intensity of their pain using a Visual Analog Scale. The Visual Analog Scale used in this study consisted of an instrument with a sliding indicator. On one side of the instrument, the words "no pain" and "worst pain ever" were printed on opposite ends. The other side of the instrument was marked in centimetres (cm) with 0 cm corresponding to "no pain" and 10 cm corresponding to "worst pain ever." Participants were asked to move the sliding indicator to a point between "no pain" and "worst pain ever" that indicated their pain intensity. The instrument was then turned over to reveal placement of the sliding indicator on the centimetre scale for recording of the pain intensity. Visual Analog Scales are usually well understood by patients, are sensitive to variations in pain intensity, and represent a ratio measurement scale enabling quantitative expression of pain intensity (Deschamps et al, 1988).
Figure 1
Visual Analog Scale
A descriptive exploratory design was used in this study to conduct a secondary analysis of data from the first step of the Instrument study. According to Woods and Mitchell (1988), exploratory studies emphasize identification of factors related to a phenomenon of interest; "these studies frequently address health-related phenomena such as pain, anxiety, or well-being" (p. 150). Accordingly, the focus of this study is the identification of pain descriptors and phrases used by patients in the Instrument study. Further, this study explored the relationship of the pain descriptors and phrases to selected demographic variables (age, level of education, mother tongue, and sex) and present pain intensity. Analyzing the relationship of pain descriptor frequencies with such variables could provide useful information between them and the pain descriptors used by patients with cancer. Therefore, data were analyzed using content analysis techniques and statistical tests.
Participants in the Instrument study were recruited from a cancer treatment facility providing active treatment for people with cancer in British Columbia and the Yukon Territory. The facility has four inpatient hospital units (one of which includes an outpatient chemotherapy programme), large ambulatory clinics (one of which includes a pain clinic), and a radiotherapy unit. Participants were recruited from the hospital units and the pain clinic.
Participants were selected using the following criteria:
Sixteen women and fifteen men were interviewed. There average age was 57.2 years and they had an average of 11.26 years of educational preparation. The participants had a wide variety of primary cancers and they had a number of different mother tongues, though all spoke English. These participants also reported that they were currently experiencing pain with an intensity of 3.95 as rated on a 0 - 10 Visual Analog Scale. The sample is more fully described in Chapter 4.
Participants on the hospital units were interviewed in their rooms (all of which were one or two bed units), and participants in the pain clinic were interviewed in examination rooms. All interviews were conducted in private. Hospitalized patients were interviewed while sitting or lying in their beds and pain clinic patients were interviewed while sitting in chairs in the examining room. Occasionally, family members were present during the interview, however, they did not participate in the discussion. Interviews of hospital patients were occasionally interrupted due to medication rounds or porters arriving to escort patients for tests. Also, there were frequent intrusions due to fire alarm testing that was performed over several months, coinciding with the period that the interviews were conducted.
The Instrument study was granted approval by the Clinical Investigations Committee at the provincial cancer treatment facility (see Appendix E) and by the Behavioural Sciences Screening Committee for Research and Other Studies Involving Human Subjects at the University of British Columbia (see Appendices F & G).
A letter explaining the Instrument study was given to the participant in the morning (see Appendix H), and if he or she agreed to participate, I would return in the afternoon for the participant to sign the consent form (see Appendix I). Prior to consenting, participants were informed that: (a) declining to participate or withdrawing participation in the Instrument study would not affect their present or future treatment; (b) participation in the study would not result in any expected ill-effects either to themselves or anyone else; and (c) the study would not benefit them directly but it was hoped that information collected could be used in the future to help with the assessment of pain. Only one patient declined participation.
To protect the confidentiality of the participants, unique study numbers were used to code the data for each participant and all personal identifying information was removed from the data. No information which could identify individuals is discussed in this report. Only the content experts consulted, members of the thesis committee, and I had access to the data used for this study. Upon completion of this analysis all data were returned to the principal investigator of the Instrument study for safe keeping according to the ethical stipulations outlined in the Instrument study.
The interviews were analyzed using content analysis methods to identify specific pain descriptors and contextual phrases describing pain. The relationship between the frequency of pain descriptor phrases and patient variables noted earlier was analyzed. In addition, the pain descriptors were categorized according to type (dimensions from the Multidimensional Model of Pain). Contextual phrases (similes and metaphors) describing pain were also identified and categorized according to type.
Content analysis is "essentially a coding operation," and may be applied to virtually any form of communication (Babbie, 1986, p. 271). It is described as the application of quantitative research methods to qualitative data (Catanzaro, 1988). It is particularly useful in instances where the participant’s own language is crucial to an investigation (Holsti, 1969; Polit & Hungler, 1991). Content analysis is usually described as consisting of two forms: latent or manifest content analysis. Latent content analysis codes the investigator’s impression or interpretation of the underlying meaning of a communication, whereas manifest content analysis codes the visible or surface content (Babbie, 1986). Concepts emerge from the data when latent content analysis is used. Conversely, data are coded or classified according to a conceptual framework with manifest content analysis (Babbie, 1986; Holsti, 1969). This study applied both methods of content analysis in coding the pain descriptors and phrases used by cancer patients according to dimensions of the Multidimensional Model of Pain (MMP) (surface content) and similes, metaphors, and expressions of difficulty when describing pain (items which emerged from the data).
Objectivity was achieved by conducting the analysis on the basis of explicitly formulated rules (Polit & Hungler, 1991). The MMP provided structure and helped focus the categorization of pain language during the data analysis. For example, words and phrases that address the location, intensity, or quality of pain were coded as Sensory (after the Sensory Dimension of the MMP). Use of manifest content analysis methods enables researchers to make "replicable and valid inferences" about data when empirical and statistical methods are applied to data (Catanzaro, 1988, p. 437).
Steps Involved in Content Analysis
Wilson’s (1985) three basic steps involved in a content analysis were used:
In conducting a valid content analysis, the researcher’s ability to address the following points is important (Wilson, 1985):
The Multidimensional Model of Pain provided the categories for the data: the theoretical dimensions of the model. Pain descriptors were, therefore, coded into the MMP categories of Affective, Behavioural, Cognitive, and so forth. For example, when a patient discussed his or her emotional reactions to pain (such as anxiety or depression), the data were coded as Affective. Finally, coding data according to the theoretical dimensions of the model enabled the realization of two of this study’s purposes which were to determine what pain descriptors were used and the frequency and type of descriptors used (according to dimensions if the MMP).
Besides validating the categories, rater reliability is a concern when content analysis research methods are used (Holsti, 1969). Problems with category reliability occur when the coder is unable to clearly formulate categories in the data (Holsti, 1969). As the dimensions of the model were defined and formed the categories for the data, coding was simplified. The coder determines which dimension is represented by each descriptor or contextual phrase. For example, the phrase "the stabbing pain makes it difficult for me to sit" is coded as: "stabbing" – Sensory descriptor; "difficult for me to sit" – Behavioural phrase.
Inter-rater reliability occurs when two or more coders code data in a consistently similar manner. As a test of my ability to identify and classify occurrences of pain descriptors, content experts in pain management (an Oncology Clinical Nurse Specialist [CNS] and a Palliative Care Clinical Nurse Specialist) coded a random sample of five unmarked transcripts. We independently highlighted sentences that contained a description of pain or a reference to pain description. There was an agreement of 0.75 between the Oncology CNS and me and 0.81 between the Palliative CNS and me. The highlighted sentences were then each printed onto index cards. The same content experts and I independently sorted the index cards (containing pain descriptors and phrases) according to the dimensions of the MMP. The was an agreement of 0.70 and 0.73 between me and each of the CNSs. The consistent agreement between the CNSs and me indicates that as an analyst, I accurately identified and classified descriptions of pain in the selected transcripts.
Once the content analysis was completed, data concerning the frequency of descriptors and phrases in each of the MMP dimensions was available. Selected demographic variables were then split into subgroups of two (male-female, mother tongue of English-Other, etc.), and comparisons were made between the number of phrases used by each subgroup. These comparisons helped with the determination of significant relationships and differences between the variables and descriptor phrases.
The usual statistical tests to utilize in this situation are t-tests (or the ANOVA for three or more groups), correlation coefficients, and Chi-Square tests (Shavelson, 1988). However, the data in this study violated the assumptions normally associated with a t-test (for example, a normally distributed population) because they were based on a convenience sample. The assumption of a normally distributed population is not tenable and the use of a parametric test such as the t-test, therefore, is not applicable. Instead of the t-test, a non-parametric test was used.
The Mann-Whitney U test is a non-parametric equivalent of the t-test (Shavelson, 1988) and was used here to compare relationships between grouped variables and pain descriptors (e.g., the difference between older and younger participants and the mean number of affective phrases used, etc.). Unlike the t-test, the Mann-Whitney U test is not based on the actual scores of participants. Instead, it is based on the rank order of raw scores from lowest to highest. The rank scores for two groups are used to compute a U value which is compared with values referenced in any statistics text to determine whether or not the U value is significant. In the case of SYSTAT/SYGRAPH® (described below), the probability value for the U statistic is automatically generated, obviating the need for comparison with a reference table. Mann-Whitney U tests were used to test for relationships among grouped variables and the frequency of phrases used to describe pain.
Correlations test for relationships among continuous variables. For example, correlations were computed among phrases coded according to the dimensions of the MMP and variables such as level of education. The Chi-Square was used to test for differences between nominal variables based on the number of responses that fell into certain categories. Thus, for example, the Chi-Square test was used to test for differences between women and men who were younger or older.
Three personal-computer programs were used to assist with the data analysis. The first program, The Ethnograph©, manages computer files associated with qualitative data. The program assists with the mechanical tasks of managing data (such as interview transcripts) (Seidel, Kjolseth, & Seymour, 1988). It generates print-outs with line numbers alongside the data. In this case, I hand coded directly on the print-outs, highlighting words and phrases and placing codes alongside the highlights (e.g., S for Sensory data, A for Affective data, etc.). Once this coding was completed, I entered the codes into The Ethnograph© for each transcript according to the line numbers associated with each code. The program was then used to generate summaries of codes (e.g., the occurrence of affective phrases across all interview transcripts). All affective phrases were then saved into a separate file, labeled "affect.txt," for example.
The second program, RightWriter®, is intended to assess grammar, style, usage and punctuation in word processor files (RightSoft, 1988). However, an additional feature of the program can generate word frequency lists from word processor files (a feature not available in The Ethnograph©), which allows detection of every occurrence of a descriptor.
After print-outs of the transcripts were coded, The Ethnograph© was used to compile files containing coded phrases (affect.txt, behav.txt, etc.). Once the content analysis identified phrases used by participants to describe pain, word frequency lists based on those phrases made it possible to accurately determine the frequency of individual pain descriptors. Word frequencies were tabulated from these files using RightWriter® which provided the frequencies for all words in the files. As a result, I manually culled extraneous words (e.g., "a," "an," "my," "the," etc.). The remaining descriptors were then scrutinized within transcripts using the search feature of a word-processor. These words were examined in their original context to determine if they were used as descriptors.
The third program, SYSTAT/SYGRAPH®, is a compilation of statistical and graphing programs designed for use with parametric or non-parametric data. SYSTAT/SYGRAPH® was used to compute Chi-Square tests, Pearson r Correlations, and Mann-Whitney U tests, and to generate illustrative tables and graphs.
This study employed a descriptive exploratory design to identify descriptors and phrases used by selected cancer patients to describe their pain. The data were collected as part of a larger study to develop an instrument to evaluate cancer pain, using a convenience sample of patients. Thirty-one existing transcribed interviews were analyzed for occurrences of words and phrases that describe pain. Content analysis methods were used to categorize pain descriptors and phrases used by the cancer patients, and the relationships between the descriptors and participant variables are reported in the next chapter.
The coding process borrowed categories (dimensions) from the Multidimensional Model of Pain which provided the theoretical framework for this study. Coding involved sorting pain descriptors and phrases used by participants according to the theoretical dimensions of the model. Reliability was established by the use of two content experts who independently coded the same randomly selected transcripts and data. Although the use of a convenience sample limits the generalizability of the results of this study, these patients with cancer were articulate in describing their pain.
Chapter 4 presents characteristics of the sample,
lists the pain descriptors used by the participants, and examines the relationship of
participant variables and use of descriptor phrases.
| Thesis: | Intro. Abstract Contents Acks. Refs. Apps. | HealthPro |
| Chap. 1 Chap. 2 Chap. 3 Chap. 4 Chap. 5 |