Evaluating palliative care: bereaved family members' evaluations of patients' pain, anxiety and depression

Article Outline

• Abstract
• 1. Introduction
• 2. Methods
  • 2.1. Participants and recruitment
  • 2.2. Design and procedures
  • 2.3. Analysis
• 3. Results
  • 3.1. Participants
  • 3.2. Analysis of verbal protocols
  • 3.3. Reliability/validity
  • 3.4. Response difficulties
  • 3.5. Symptom cues and judgments
    • 3.5.1. Pain
    • 3.5.2. Anxiety
    • 3.5.3. Depression
  • 3.6. Symptom severity and frequency judgments
  • 3.7. Recall and reference periods
• 4. Discussion
• Acknowledgements
• References
• Copyright

Abstract 

Palliative care surveys often rely on bereaved family members to act as proxies to provide information on patient care at the end of life, after the patient's death. However, when comparing bereaved family members' assessments with those of the patients, agreement is found to be better for symptoms that are more concrete and observable than subjective aspects such as psychological symptoms and pain. To date, little is known about how proxies actually evaluate these types of symptoms. The present study used retrospective verbal protocol analysis to elucidate the thought processes of 30 bereaved relatives during their evaluations of patients' pain, anxiety and depression. The qualitative analysis raised awareness of the difficulties experienced by proxies when discerning the presence of symptoms. It also provided insights into the cues and strategies used when making decisions, contributing to a fuller understanding of how proxies distinguish symptoms. Recommendations are made to improve the design of retrospective palliative care surveys.

Keywords:  Palliative, symptom evaluation, surveys, family, retrospective

Back to Article Outline

1. Introduction 

Palliative care surveys often rely on bereaved family members to act as proxies to report on behalf of the patient after the patient's death. These retrospective or after-death surveys provide researchers and auditors with valuable information regarding patients' care during the terminal phase of their illness. This is a time when patients are often too ill or unwilling to provide the information for themselves. In addition to this, retrospective palliative care surveys, unlike prospective studies conducted before death, are not limited to diseases such as cancer that have an identifiable terminal phase, and where patients are likely to be known to relevant services.¹ Thus, representative samples can be studied. This approach has been used in a number of influential studies on the care of the dying.^2., ^3., ^4.

However, little is known about the validity of proxies' responses. A review of the evidence comparing proxies' accounts with those of the patients suggests that proxies are better able to provide information on concrete, observable phenomena such as services received and physical functioning, while agreement is poorer for subjective aspects like pain and psychological status.⁵ These conclusions are based mainly on prospective studies that have compared proxies' accounts with those of the patients prior to death.^6., ^7., ^8., ^9. Only four studies have specifically compared retrospective accounts with the proxies' and/or patients' accounts prior to the patient's death.^4., ^10., ^11., ^12. One reason for this is the research design necessary to measure both before and after the patient's death, and the challenges of matching of assessment periods because of difficulties in recruitment, attrition and determining the time of death. These are problems that are commonly encountered in palliative care research.¹³ Consequently, assertions about the validity of the retrospective approach have had to rely to a large extent on evidence from validity studies conducted prospectively. Yet, studies that have assessed proxies' reports before and after the patient's death vary from prospective studies. Overall, proxies' prospective ratings are more in agreement with patients' reports than their retrospective ratings. With respect to pain, in a study conducted by Hinton,¹² pain was rated more severe retrospectively, while Higginson et al.¹¹ found that pain ratings became polarized to the extremes of the rating scale. The findings for mood also contrast with the prospective validity literature. Depression was shown to be under-rated retrospectively^10., ^12. but overestimated prospectively.^6., ^9. Patients' anxiety was also rated as less severe retrospectively and in some instances was more in line with the patients' own ratings.^6., ^11., ^12.

Research so far has focused on measuring levels of agreement without explaining why these discrepancies arise. A better understanding of proxies' perceptions of patients' experiences within a retrospective context is critical, not only for identifying the limitations of proxies, but also for facilitating the design of retrospective surveys used for assessing care for patients at the end of life. The present investigations set out to explore response difficulties to the types of questions asked in retrospective palliative care surveys, to identify symptom cues and the basis of their judgments, and to explore proxies' recall within the assessment periods. The work forms part of a larger set of investigations examining the validity of the retrospective approach in palliative research. The focus of these investigations is on the symptoms pain, anxiety and depression, as this is where congruency between the patient and proxy is poorest. These are important symptoms to evaluate in the care of patients with terminal illnesses because they can negatively affect the patient's quality of life and cause considerable distress to their family members.¹⁴ Furthermore, accurate identification is crucial for the appropriate management of patients' symptoms and evaluation of services.

Back to Article Outline

2. Methods 

2.1. Participants and recruitment 

Ethical concerns have been raised regarding the use of bereaved relatives in research and ethical guidelines proposed for conducting such research.¹⁵ These concerns were taken into consideration when designing the present set of investigations. Before seeking ethical approval, a number of ethical safeguards were put into effect. These included clinical supervision and social work support from the hospice involved in the care of the patient and their proxy. Ethical approval was then sought and obtained from three local Research Ethics Committees. A sample of thirty bereaved relatives was identified through records from one hospice in southeast London between February and April 2000. The hospice covers a wide geographical area and takes in both urban and semi-rural locations. Selection of participants was convenient and purposeful. The inclusion criteria were: bereavement in the last three months; cognitively intact; contact with the patient in the last month and week before the patient's death.

An invitation letter including information about the study, the rights of participants and a consent form was sent to 86 participants. An opt-in approach was adopted, which meant that only those wishing to take part needed to reply to the letter. All of the participants chose to be interviewed in their homes. Written and verbal consent were obtained to audiotape the interview.

2.2. Design and procedures 

The verbal protocol or think aloud protocol was used to investigate and improve understanding of the underlying mental processes involved in answering questions about patients' symptoms. This technique has been applied in the design of a number of surveys.^16., ^17. The technique requires participants to verbalize their stream of consciousness and thoughts, while answering a question (concurrent verbal protocol) or after they have finished answering a question (retrospective verbal protocol). Like many other cognitive techniques, there is the underlying assumption that,

“The psychological significance of an individual's thoughts and feelings, as well as the underlying cognitive processes, can be examined by content, analyzing the individual's reported thoughts, ideas, images and feelings.” (p.929).¹⁸

Retrospective verbal protocols were used within a semi-structured face-to-face interview. The structured component asked participants to respond to questions regarding the patient's pain, anxiety and depression in the last week and month before death. These questions are drawn from VOICES (Views of Informal Carers-Evaluation of Services), a retrospective semi-structured postal questionnaire used to assess palliative care.¹⁹ Pain, anxiety and depression are evaluated on a five-point Likert verbal rating scale from “most of the time” to “never” for symptom frequency and “severe” to “none” for symptom severity. (These descriptors are from an earlier version of VOICES. Later versions have changed the intensity descriptors significant and severe to severe and very severe.) Following each question from VOICES, participants were encouraged to verbalize their thoughts, for example, after answering a question regarding patient's pain in the last week, they were asked, “Can you tell me what you were thinking about when you answered that question?” (retrospective verbal protocol). The instructions for the retrospective verbal protocol followed those proposed by Ericsson and Simon.²⁰ In conjunction with this, some verbal probing asking why questions was used to make explicit the particular cognitive processes used by the proxy during their verbalizations, such as, “Why did you choose severe to describe their pain?.” Other investigators attest to this technique.²¹ Demographics and characteristics associated with the patient and proxy relationship were also collected. Interviews took between 35–75 minutes.

2.3. Analysis 

The emphasis of the analysis was the exploration of the underlying mental processes involved in answering the question. In addition to this, content based information, such as cues to symptoms or indications of difficulties like “don't know” responses were coded. Content-based information is valuable in understanding response processes as it can facilitate the identification of information the participant uses to base their decision upon.¹⁶

Qualitative content analysis of the transcripts and development of the codes was both inductive and deductive. This is important because restricting the analysis to just fixed categories might exclude information that may be particularly relevant to palliative care research. The coding scheme developed for these investigations was based on one developed by Sudman et al.,²² which was used to assess behavioral and attitudinal questions with participants and their spouses. Categories represent comprehension, retrieval and judgment problems (Table 1). This coding scheme has its theoretical roots in models of the cognitive processes used in answering survey questions.^23., ^24. Codes for response difficulties were from examples given by Fowler and Cannell.²⁵

Table 1. Coding Scheme for the Content Analysis of Proxy Interviews

Back to Article Outline

3. Results 

3.1. Participants 

Nineteen of the participants were women and 11 were men. Their ages ranged from 25 to 90 years (M=71, SD=13.4). All participants were close family members and the majority (n=24) lived with the patient and cared for them (n=21) in their last week life. Sixty percent (n=18) were partners, ten were sons or daughters and two were ex-partners of the deceased. They were from a range of socioeconomic backgrounds. The prevalent ethnic group was European (n=28), with one participant Hispanic and another Afro-Caribbean. The primary cause of death for the patients was advanced cancer: lung (n=12), colorectal (n=4), upper gastrointestinal (n=4), breast (n=3), prostate (n=3), liver (n=1), brain (n=1), and spine (n=1). One patient died from coronary heart disease. The place of death for half the patients was the hospice, nine died at home, while six patients were in either hospital or a nursing home at the time of death.

3.2. Analysis of verbal protocols 

The process began by coding the first three transcripts. The unit of analysis was the response or segments of the response. This involved identifying key words and themes within the text. As new codes and themes emerged, these were integrated into the coding system and used as a framework for further analysis of the transcripts. Table 1 contains all the codes from the analysis. Codes varied depending on the type of question asked. A case in point was the symptom frequency codes, as these were designed specifically to examine how proxies reached a decision on the frequency of a symptom, so were not applied to other questions.

3.3. Reliability/validity 

To ensure that the analysis provided a valid interpretation of what the proxies were saying, the comprehensiveness and appropriateness of the analysis was discussed with another researcher (FA). The codes were then applied and compared across transcripts during the analysis in order to assess the validity and consistency of their interpretation. To fully capture the meanings in what the proxies were saying, additional codes were added to the framework as they emerged from the transcripts. Reliability was evaluated through consistency across the transcripts and between raters. Two coders (CM and FA) independently coded the transcripts in order to examine the reliability of the analysis. Cohen's kappa was used to evaluate the level of agreement. This is a more comprehensive measure of inter-rater agreement because, unlike percentage agreement, it takes into account chance agreement. Kappa ranged from 0.80 (P < 0.01) to 0.93 (P < 0.01). Mean kappa values for each of the twenty-one questions was 0.88 (P < 0.01; SD 0.03). These estimates suggest substantial to almost perfect agreement between coders.²⁶ Where differences between the two raters emerged, the divergent interpretations of the code or response in question were discussed and agreement was reached.

3.4. Response difficulties 

When evaluating patients' symptoms, proxies encountered most difficulties because of a lack of communication between the patient and proxy, an overlap in symptoms, or the patient's decreased level of consciousness. This was particularly evident when the proxy was asked to identify the presence of anxiety (10/30), depression (8/30) and pain presence (4/30) and severity (5/30) in the last week of life. As one participant remarked, “She was unconscious so it was difficult to tell at that stage. She might have been in pain but not able to communicate.” Estimates suggest that sedation of patients nearing death, through opioids or psychotropic drugs, is not uncommon in controlling some symptoms, such as pain, with levels ranging from 16% to 52.5%.^27., ^28.

Lack of communication between the patient and proxy was a notable problem when detecting depression (3/30 week, 4/30 month) and anxiety (3/30 month). This was evidenced in a response about the presence of anxiety, “I can't be completely certain… there were no verbal signs, nothing to indicate that.” These problems were often associated with uncertainty regarding the severity of these symptoms, particularly for depression. Disruptions in communication between patients and their partners are not unusual as they try to cope with the patient's illness and impending death.²⁹ Furthermore, coping strategies such as denial may inhibit the open expression of feelings between the patient and proxy. Unless these symptoms are communicated to the patient via verbal or non-verbal means, the proxy will have no way of knowing how the patient is feeling. References were also made to the effect that patients might have been concealing their feelings so as not to upset or burden significant others. There is some evidence to support this finding. Dar et al.³⁰ found that 60% of patients admitted hiding their pain in order to avoid upsetting their spouses. Psychological symptoms, like anxiety and depression, may be easier to conceal than more overt symptoms such as breathlessness. This could lead to an underestimation of patients' symptoms. However, it was noted that some proxies bear this in mind and infer from the patient's character and the situation when estimating anxiety and depression.

Overlap between the symptoms of depression, the patient's physical condition, and anxiety caused problems for a small number of proxies. Symptoms such as fatigue, weight loss and poor concentration are common in patients with cancer, features which overlap with the symptoms of depression and anxiety.³¹ Since proxies rely on these types of overt expressions when deciding whether the patient is anxious or depressed, any overlap with their physical condition can lead to significant problems identifying depression and anxiety.

For all symptoms, more response difficulties were evident for symptom severity than for frequency. For instance, there were almost double the number of problems for depression severity compared with depression frequency. It is not known why this occurs, although it may be possible to conjecture that the frequency of a symptom can be to some extent observed, whereas the intensity of the symptom is largely a perceptual experience, thus less discernible to the onlooker.

3.5. Symptom cues and judgments 

Proxies' ratings of patients' pain revealed that they perceived their relative to be in a considerable amount of pain, with three-quarters stating that the patient had moderate or greater pain during the last month and week of their lives. Overt expressions of pain were the most frequently used indicators of pain in the last week, and were identified as the most important indicators by the proxies themselves. These included use of medications (n=34), verbal (n=25) and non-verbal behavior (n=29) such as agitation, posture and facial expressions (wincing, grimacing). Disruption to the patient's functioning and normal daily activities was commonly referred to in the week and month reference periods for identifying pain presence and severity. Restricted mobility and reduced participation in their usual activities were observed, from not walking the dog or gardening, to the extent of not wanting to get out of bed. Surprisingly, none of the proxies stated that the patients' level of consciousness was affected in the month, even though there was more emphasis on non-verbal behavior than verbal behavior. A possible explanation for this finding might be that the patient's level of consciousness was less affected in the month than in the week, giving the proxies access to more overt behavioral indicators. Alternatively, patients might not have wanted to bother their significant other by talking about pain so proxies relied on non-verbal indicators.

Fifteen proxies mentioned the affective components of the pain experience. These were evident at intense levels of pain and manifested through aggression, anger, frustration, agitation, tearfulness, quietness and distress. These were seen as a change in the usual character of the patient; as one proxy recalled her husband's behavior, “he became aggressive with it, that made me think it must have been severe [pain] for him, for him to be like that because he wasn't normally like that.” Trait knowledge (knowledge based on the person's characteristics/personality) was important in forming judgments. Knowing their character and what the patient was able to endure was often mentioned when deciding on the presence (6/30 week, 7/30 month), and severity (6/30 week, 7/30 month) of pain. This concurs with previous findings where participants are asked to report about others.¹⁷ What the patient had been like previously during their illness was also important in their decisions on the presence of pain. Norms were sometimes used, “She didn't get the pain that some cancer patients get,” but lack of knowledge and experience upon which to base their judgment, may have limited this as a basis for judgments.

Seventeen participants reported the presence of anxiety in the week and twenty in the month. Overwhelmingly, references to patients' feelings such as anger, frustration and the need for reassurance were repeatedly used in the week (n=28) and month (n=30). This was particularly true for the presence and severity of anxiety. These feelings were expressed in a number of ways. Verbal communication was the predominant cue to the presence, frequency and severity of anxiety in the week (n=25) and month (n=21). Few patients spoke directly about being anxious; instead, anxiety was inferred from their communications, both in its content and frequency. For example, proxies spoke of patients' concerns about ensuring matters were in order and that the family was provided for after their death. Others raised issues about suffering with symptoms, in particular pain, loss of functioning, wanting to be at home, losses associated with life events that they would miss, such as children and grandchildren, and uncertainty about the future. The frequency of these communications and the need for reassurance were indicative of anxiety severity and frequency.

In the month, the patients' functioning, the need for help, and control were cues to the presence and severity of anxiety. Functioning was used more in terms of the ability to function in spite of the illness and its consequences. This was closely associated with coping and medication/help, and was apparent at lower levels, infrequent episodes or no anxiety. Non-verbal behaviors and changes in the patient's demeanor were other cues to their feelings of anxiety in the week (n=16) and month (n=11). These included crying, withdrawing, agitation and hostility, as well as small gestures that were significant to the proxies in suggesting that the patient was anxious. “He grabbed hold of my hand, which wasn't like him 'cause we're not that type of family… That's what made me think he was anxious like.” This example is also illustrative of anchoring the judgment on the patient's usual character. This type of trait knowledge was used by a considerable number of proxies in deciding whether the patient was anxious or not and to judge the severity of anxiety. Another significant basis for judgments was what the patient had been like previously during their illness and how they had coped with events.

Thirteen proxies reported that the patient had depression in the last week and slightly more in the month (17/30). Again references to the patients' feelings in the week (n=27) and month (n=35) were most often used as indicators for presence, frequency and severity of depression. Among these were sadness, anger, flatness of mood and boredom. In contrast to anxiety, non-verbal communication was frequently mentioned in the week of the patient's death. The patient's appearance of unhappiness was a significant cue, while others included lack of interest, inability to concentrate, tearfulness, irritability, tiredness and withdrawal. These feelings were seen to affect patients' ability to fully function in usual activities and to disrupt activities of daily living such as sleep patterns.

In the month reference period, verbal communication (n=19) was mentioned slightly more than non-verbal communication (n=15). Of the 13 proxies who verbally communicated with the patient, in four of these cases the proxy said that the patient was not depressed. Only two proxies explicitly talked with the patient about feelings of depression. Similar to anxiety, most judged the symptom on what the patient talked about, for example, fears about dying and unhappiness with their circumstances. One proxy recalled her mother's depression, “Well, she kept saying she wanted to die.” Two patients were being treated for depression with antidepressants.

Again, judgments regarding the presence of depression were often founded on trait knowledge about the patient's personality and characteristics (11/30 week, 11/30 month). An actual change in personality was observed in a few instances. As one participant remarked, “She changed. Her whole personality just seemed to change.” Others based their decisions on what the patient had been like previously during their illness. Noteworthy were judgments based on norms, a finding not observed for pain or anxiety. As the following example illustrates, “It's [depression] a natural thing really, because you've got cancer.”

3.6. Symptom severity and frequency judgments 

Analysis of the symptom severity ratings showed the variable nature of the symptom experiences. Proxies used various strategies in making their decisions. These included basing their response on the most severe level experienced, even if this was infrequent, and most frequently experienced level of intensity (though some proxies were unable to say why they had decided on a particular response, particularly when judging anxiety and depression).

Strategies used to judge symptom frequency varied according to the symptom and the period being assessed. A third of proxies stated that the patient had pain all the time during the last month and week of life, implying that there was no estimation of frequency in these cases. Recall and count of episodes (episode enumeration) was not often used in the week (4/25) or month (4/24), except for infrequent episodes of pain, supporting previous research that has found that this strategy is more likely to be used when there is a short assessment period and infrequent episodes.³² Almost equal numbers used either rate-based estimation or heuristic-based estimations. Rate-based estimations are a strategy whereby the participant assesses the frequency over a shorter period and multiplies the frequency over the period being assessed.³³ Precision relies on the regularity of occurrence. Proxies tended to use this strategy when there was some regularity in the analgesic regimen or the pain was more salient at particular times (such as during the night). Heuristic-based estimations (i.e., rule-of-thumb) vary widely from averaging to guessing. This strategy is prone to inaccuracies because of biases in the way the information is processed. For example, the ease with which information is recalled can lead individuals to believe that an event had occurred more often.³⁴ Heuristic-based estimations expressed by proxies in this study characterized the irregularity of the pain experienced by some patients.

In contrast to pain, rate-based estimations were seldom used in reporting depression or anxiety frequency. This might have been due to the variable nature of these symptoms. Instead, heuristic-based estimations were used for anxiety (7/17 week, 6/20 month) and depression (7/13 week, 6/17 month). For example, “It [depression] was now and again … it changed and became worse over that month but overall it was sometimes.” Furthermore, in some instances proxies recalled and counted actual episodes when the anxiety or depression was reported as “often” or “most of the time.” This contrasts with other survey appraisals of frequency where episode enumeration tends to be used for infrequent episodes.¹⁷ In instances where this occurred it was for prolonged episodes of depression and anxiety over hours or days, unlike the discrete episodes such as doctor's visits, or alcohol and food consumption usually studied in surveys. Heuristic-based estimations included averaging over the reference period and judgments based on the proxies' innate sense of frequency. However, without extensive probing it was not possible to distinguish all the types of heuristics used.

3.7. Recall and reference periods 

Event cues tended to be used to signify the presence of pain, anxiety and depression rather than its intensity and frequency. Places were usually mentioned, particularly in the month reference period. These consisted of references to periods before or after admission or input from a hospital or the hospice. This is in line with the memory literature where event cues prompt recall of events.³³ Several proxies talked of escalating pain and the hospice's involvement in controlling the patient's pain. This was not surprising given that uncontrollable pain and difficulty coping were the main reasons for admission to the hospice in the last week of life for ten patients. Events were particularly salient when proxies talked about the patients' anxiety in the week before death (7/17) as these were often associated with a significant event in the lives of the patient and proxy, such as admission to the hospice, the patient being given their diagnosis, and a deterioration in the patient's condition and symptoms. Whereas events such as being in hospital, deterioration in the patient's condition and loss of functioning were associated with depression in the week and month prior to death.

There was a tendency for some proxies to report on periods extending to the reference periods asked about, for the presence of a symptom. The numbers were relatively small for pain in the week and month and anxiety in the week. In contrast, several proxies spoke of periods from weeks up to several months outside the reference period when evaluating anxiety in the month (5/20) and depression (5/13 week and 6/17 month). A possible explanation for this was that the symptom was associated with significant events that occurred outside the reference periods, adding support to that notion that sequences of events in memory are organized within a personal time frame.³⁵ This could lead to inaccuracies in the evaluation of symptoms within a specified period.

Back to Article Outline

4. Discussion 

The aim of the study was to explore what proxies are recalling when they report on patients' levels of pain, anxiety and depression, as a means of understanding possible biases and errors. Although the study identified areas where these might occur, such as heuristic strategies and reference periods, it is not possible to state from this study that these actually biased proxies' responses. It is only possible to infer their likely effects. In addition to this, relations between the cues used to judge symptoms and the validity of proxies' reports cannot be established, as there is no comparison with the patients whom they are representing. Despite these limitations, analysis of proxies' verbalizations revealed that they are drawing on different sources of knowledge when evaluating patients' symptoms, some of which may be reliable indicators. These included signs commonly associated with the symptom such as analgesic use, facial expressions and verbalizations of pain; expressions of fear and agitation signifying anxiety; to sadness and a reduction in functioning indicating depression. Comparisons between patients' and proxies' expression and understanding of symptoms would help elucidate the reliability of these cues as symptom indicators.

Further to this, the study showed that proxies' symptom evaluations are grounded in the context of their general knowledge particularly for affective components of pain, and for anxiety and depression. Of significance is knowledge of the patient's character before illness and during the course of their illness. Close relatives and carers in close contact with the patient are likely to have better access to this type of information. It is possible to surmise that significant others may be better proxies for some aspects of patients' experiences than health professionals who have limited contact with, and information about the patient.

The study also revealed a number of factors that could affect the quality of the information provided. These included response difficulties, varying judgment strategies and recall outside the reference periods. In relation to response difficulties, we found that proxies were willing to give a response even though they were unsure. Under these circumstances, proxies' responses are likely to be inaccurate. Therefore, within the survey there should be an option stating that the respondent is unable to judge the patient's symptoms. Researchers may also want to consider the reference period as symptoms in the last week of life were particularly difficult to evaluate because of the patients' deteriorating health and level of consciousness.

Other design recommendations would be to phrase the question in such a way that the strategy is evident. For example, proxies are asked to assess a symptom by averaging episodes over a specified reference period. This would introduce some consistency into the responses and might improve congruency between the patient and proxy. The variable nature of the symptoms suggests that in addition to this, reference periods should be kept fairly short. We found that despite the fact that the reference period was given in each question, participants still went outside the period. This highlights the need to clearly define the period and continually remind participants. A more appropriate method would be to encourage participants to recall events within a personally significant time frame. Here, proxies could report on events such as the patient's experiences following admission to a hospice or hospital, a common event cue used by proxies in this study. This would give variable reference periods, but would be useful for evaluating care within a particular setting. Alternatively, using their personal framework, proxies could first be asked to recall events within context and then asked to place events within the reference periods.

It should be noted that verbal protocols are based on the assumption that individuals have conscious access to relevant cognitive processes. Critics have challenged this assumption, suggesting that individuals' cannot be relied upon to provide direct verbal reports to some types of cognitions.³⁶ Refuting this claim, Ericsson and Simon²⁰ cite several studies where individuals have been very adept at verbalizing cognitive processes. However, verbal protocols only access conscious thought processes, which are only part of the processes used in decision making. For example, we found that participants were not always able to verbalize their thought processes, particularly in relation to the heuristic strategies they used. Recognizing this limitation, Ericsson and Simon²⁰ affirm that individuals may not have access to all the processes involved, but they state that what individuals are saying has meaning and importance to them and is likely to be relevant when making decisions.

The present study provides insightful findings on the processes underlying proxies' judgments of patients' pain, anxiety and depression. It also raises awareness of the complexities of the mental processes used by proxies and the importance of social processes in human decision making. Although the research is based on the VOICES survey, the results are applicable to the retrospective approach in general and may be useful in identifying ways to improve existing survey designs. Future research should focus on identifying other potential biases that might affect proxies' responses, such as their own feelings during bereavement. Even though this was not examined here, proxies frequently made reference to their own emotions. The timing of retrospective interviews and proxies' memory for events also needs to be examined as surveys are often conducted several months after the patients' death.

Back to Article Outline

Acknowledgements 

The authors would like to acknowledge the generosity of the bereaved family members who took part in the study. In addition to this, they would also like to thank Fiona Aspinal (FA) for her help coding the transcripts with CM (Christine McPherson). The research was supported by a grant from King's Health Care Trust Joint Research Committee, King's College, London, UK.

Back to Article Outline

References 

1. Morris JN, Mor V, Goldberg RJ, et al.  The effect of treatment setting and patient characteristics on pain in terminal cancer patients: a report from the National Hospice Study. J Chron Dis. 1986;39:27–35
   • View In Article
2. Addington-Hall J, McCarthy M. Regional study of care for the dying: methods and sample characteristics. Palliat Med. 1995;9:27–35
   • View In Article
   • MEDLINE
   • CrossRef
3. Lynn J, Teno JM, Phillips RS, et al.  Perceptions by family members of the dying experience of older and seriously ill patients. SUPPORT Investigators. Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments. Ann Internal Med. 1997;126:97–106
   • View In Article
4. Cartwright A, Seale C. The natural history of the survey: An account of the methodological issues encountered in a study of life before death. London: King's Fund; 1990;
   • View In Article
5. McPherson CJ, Addington-Hall JM. Judging the quality of care at the end of life: can proxies provide reliable information?. Soc Sci Med. 2003;56:93–107
   • View In Article
6. Field D, Douglas C, Jagger C, et al.  Terminal illness: views of patients and their lay carers. Palliat Med. 1995;9:45–54
   • View In Article
   • MEDLINE
   • CrossRef
7. Sneeuw KC, Aaronson NK, Sprangers MA, et al.  Value of caregiver ratings in evaluating the quality of life of patients with cancer. J Clin Oncol. 1997;15:1206–1217
   • View In Article
   • MEDLINE
8. Sneeuw KC, Aaronson NK, Sprangers MA, et al.  Evaluation of the quality of life of cancer patients: assessments by patients, significant others, physicians and nurses. Brit J Cancer. 1999;81:87–94
   • View In Article
9. Spiller JA, Alexander DA. Domiciliary care: a comparison of the views of terminally ill patients and their family caregivers. Palliat Med. 1993;7:109–115
   • View In Article
   • MEDLINE
   • CrossRef
10. Ahmedzai S, Morton A, Reid J, et al.  Quality of death from lung cancer: patients' reports and relatives' retrospective opinions. In:  Watson M,  Greer S,  Thomas C editor. Psychosocial oncology. Oxford: Pergamon Press; 1988;p. 187–192
    • View In Article
11. Higginson I, Priest P, McCarthy M. Are bereaved family members a valid proxy for a patient's assessment of dying?. Soc Sci Med. 1994;38:553–557
    • View In Article
    • MEDLINE
    • CrossRef
12. Hinton J. How reliable are relatives' retrospective reports of terminal illness? Patients' and relatives' accounts compared. Soc Sci Med. 1996;43:1229–1236
    • View In Article
    • MEDLINE
    • CrossRef
13. McWhinney IR, Bass MJ, Donner A. Evaluation of a palliative care service: problems and pitfalls. BMJ. 1994;309:1340–1342
    • View In Article
    • MEDLINE
    • CrossRef
14. Pasacreta J, Pickett M. Psychosocial aspects of palliative care. Sem Oncol Nurs. 1998;14:110–120
    • View In Article
15. Parkes CM. Guidelines for conducting ethical bereavement research. Death Studies. 1995;19:171–181
    • View In Article
    • MEDLINE
    • CrossRef
16. Bickart B, Felcher EM. Expanding and enhancing the use of verbal protocols in survey research. In:  Schwarz N,  Sudman S editor. Answering questions: Methodology for determining cognitive and communicative processes in survey methodology. San Francisco: Jossey-Bass Publishers; 1996;p. 115–142
    • View In Article
17. Blair J, Menon G, Bickart B. Measurement effects in self versus proxy response to survey questions: An information processing perspective. In:  Biemer PP,  Groves RM,  Lyberg LE,  Mathiowetz NA,  Sudman S editor. Measurement errors in surveys. New York: Wiley; 1991;p. 145–166
    • View In Article
18. Cacioppo JT, von Hippel W, Ernst JM. Mapping cognitive structures and processes through verbal content: The thought-listing technique. J Consult Clin Psychol. 1997;65:928–940
    • View In Article
    • MEDLINE
    • CrossRef
19. Addington-Hall J, Walker L, Jones C, et al.  A randomized controlled trial of postal versus interviewer administration of a questionnaire measuring satisfaction with, and use of, services received in the year before death. J Epidemiol Commun Health. 1998;52:802–807
    • View In Article
20. Ericsson KA, Simon HA. Protocol analysis: Verbal reports as data. Cambridge, MA: MIT Press; 1993; (Rev. ed.)
    • View In Article
21. Willis GB, Royston P, Bercini D. The use of verbal report methods in the development and testing of survey questionnaires. Appl Cognitive Psychol. 1991;5:251–267
    • View In Article
22. Sudman S, Bradburn NM, Schwarz N. Thinking about answers: The application of cognitive processes to survey methodology. San Francisco: Jossey-Bass; 1996;
    • View In Article
23. Oksenberg L, Cannell CF. Some factors underlying the validity of response in self-report. International Statistical Bulletin. 1977;48:325–346
    • View In Article
24. Tourangeau R. Cognitive science and survey methods. In:  Jabine TB,  Straf ML,  Tourangeau R editor. Cognitive aspects of survey methodology: building a bridge between disciplines. Washington DC: National Academy Press; 1984;p. 73–100
    • View In Article
25. Fowler FJ, Cannell CF. Using behavioral coding to identify cognitive problems with survey questions. In:  Schwarz N,  Sudman S editor. Answering questions: Methodology for determining cognitive and communicative processes in survey research. San Francisco: Jossey-Bass; 1996;p. 15–35
    • View In Article
26. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics. 1977;33:159–174
    • View In Article
    • MEDLINE
    • CrossRef
27. Ventafridda V, Ripamonti C, De Conno F, et al.  Symptom prevalence and control during cancer patients' last days of life. J Palliat Care. 1990;6:7–11
    • View In Article
    • MEDLINE
28. Fainsinger R, MacEachern T, Hanson J, et al.  Symptom control during the last week of life on a palliative care unit. J Palliat Care. 1991;7:5–11
    • View In Article
    • MEDLINE
29. Baider L, Sarell M. Couples in crisis: Patient-spouse differences in perception of interaction patterns and illness situation. Family Therapy. 1984;9:115–122
    • View In Article
30. Dar R, Beach CM, Barden PL, et al.  Cancer pain in the marital system: a study of patients and their spouses. J Pain Symptom Manage. 1992;7:87–93
    • View In Article
    • MEDLINE
    • CrossRef
31. Bukberg J, Penman D, Holland JC. Depression in hospitalized cancer patients. Psychosomatic Medicine. 1984;46:199–212
    • View In Article
    • MEDLINE
32. Burton S, Blair J. Task conditions, response formulation processes, and response accuracy for behavioral frequency questions in surveys. Public Opinion Quarterly. 1991;55:50–79
    • View In Article
    • CrossRef
33. Bradburn NM, Rips LJ, Shevell SK. Answering autobiographical questions: The impact of memory and inference on surveys. Science. 1987;236:157–161
    • View In Article
    • MEDLINE
34. Tversky A, Kahneman D, Availability . A heuristic for judging frequency and probability. Cognitive Psychol. 1973;5:207–232
    • View In Article
35. Brown NR, Shevell SK, Rips LJ. Public memories and their personal content. In:  Rubin DC editors. Autobiographical memory. Cambridge: Cambridge University Press; 1986;p. 137–158
    • View In Article
36. Nisbett RE, Wilson TD. Telling more than we know: verbal reports on mental processes. Psychological Rev. 1977;84:231–259
    • View In Article