| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||
Correspondence: Address correspondence to Susan E. Fisher, MA, VA Pittsburgh Healthcare System, Behavioral Health Service Line (116A-H), 7180 Highland Dr., Pittsburgh, PA 15206. E-mail: sfisher{at}bama.ua.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
= 0.87 – 0.91. Residents' GPM-M2 scores were stable over time, r =.74–.80, p <.0001, for all comparisons. Implications: Using explicit protocols and reporting procedural data allows researchers and clinicians to better understand and apply results of self-report studies with cognitively impaired elders. Results suggest that many nursing home residents can provide consistent and reliable self-report pain data, given appropriate time and assistance.
Key Words: Clinical assessment • Dementia • Pain • Self-report methodology • Nursing home • Geriatric Pain Measure
Research in the areas of pain (Chu, Schnelle, Cadogan, & Simmons, 2004; Ferrell, Ferrell, & Rivera, 1995; Parmelee et al., 1993; Weiner et al., 1999), quality of life (Brod et al., 1999; Logsdon et al., 1999), health status (McHorney, 1996), quality of care (Simmons & Schnelle, 2001; Simmons et al., 1997), and care preferences and choices (Feinberg & Whitlatch, 2001; Maris, Soberman, Murray, & Norton, 1996; Sansone, Schmitt, Nichols, Phillips, & Belisle, 1998) show that many older adults with mild to moderate cognitive impairment can provide valid and useful information about these important constructs.
However, because of the known complexities of obtaining self-report from cognitively impaired elders, we argue that it is not adequate to present outcome data without using, describing, and reporting the process through which the data were obtained. For example, Simmons and colleagues used and reported objective criteria for excluding cognitively impaired residents in self-report assessments (Levy-Storms, Schnelle, & Simmons, 2002; Simmons & Schnelle, 2001; Simmons et al., 1997); described indiscernible response types (Simmons & Schnelle, 1999); and measured the validity of residents' self-reports by using direct behavioral observation (Simmons & Schnelle, 2001). Using such rigorous methodological criteria, these studies showed that a majority of nursing home residents capable of responding to yes–no questions about their daily care could do so accurately (Simmons & Schnelle, 1999; Simmons et al., 1997) and that objective criteria can readily be used for identifying reliable self-reporters (Chu et al., 2004; Simmons & Schnelle, 2001). Additionally, Kane and colleagues (2003) developed a training manual for interviewing cognitively impaired residents about quality of life. Procedural details included defining how many attempts interviewers could make with Likert-scale response options before switching to binary response options and providing explicit criteria for discontinuing an interview as a result of nonresponse (Kane et al.).
Although these studies represent significant progress in research on self-reporting capacity, we propose that there are still procedural aspects of the complex interview process that must be accounted for in order for such results to be replicated and generalized to the everyday environment of the nursing home. These details include how many and what types of prompts are used in the administration of self-report measures, whether multiple assessors are reliable in the determination of indiscernible and discernible responses, what types of indiscernible responses are obtained and how often, and to what degree resident reports remain stable over multiple (i.e., >2) interviews.
The primary aims of the current study addressed these issues. Our first aim was to use, describe, and report an explicit procedure for obtaining self-report data, including the provision of information on (a) inclusion–exclusion procedures and criteria, (b) discontinuation or truncation of measures or response options, (c) prompting, (d) and interassessor reliability on determination of discernible and indiscernible responses. Our second aim was to evaluate the self-report capacity of residents across a broad range of cognitive status using a multidimensional pain questionnaire. Specifically, we analyzed internal consistency, test–retest reliability, and concurrent validity of residents' responses for the entire sample and for the sample dichotomized by high and low cognitive status.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Of 163 residents surveyed for participation, 116 (71.2%) met inclusion criteria. Of these, we obtained written informed consent and assent directly from residents or by proxy for 63 residents (54.3% consent rate). We dropped 2 additional residents from the study because they died prior to assessment (1) or relocated to another facility (1), resulting in the final sample of 61 residents.
Procedure
Trained research assistants administered the multidimensional pain questionnaire weekly in face-to-face interviews for 4 consecutive weeks. To control for naturally occurring variance in pain experience throughout the day, we scheduled the questionnaire for administration at the same time of the same day for each weekly administration.
Measures
Multidimensional Pain Questionnaire
The Geriatric Pain Measure—Modified (GPM-M; Ferrell, Stein, & Beck, 2000; Simmons, Ferrell, & Schnelle, 2002) is a 17-item questionnaire measuring pain frequency and intensity, effects of pain on everyday life, and aspects of satisfaction with pain management. We modified the GPM-M in two ways. First, we added a descriptive ordinal scale to all applicable questions so that more thorough information about the frequency and intensity of residents' problems with pain could be obtained. These scaled-response items would also allow us to test the limits of the cognitively impaired individuals in their ability to respond to more complex questions (i.e., yes–no questions compared with scaled-response questions). The addition of the ordinal scale resulted in 15 pairs of questions (2 questions could not be paired with an ordinal scale question semantically). For the majority of questions (13 out of 15), the verbal descriptors were "sometimes, most of the time, and all the time," providing more information about the frequency with which specific problems occurred (e.g., pain when bathing and dressing, pain interfering with social activities, and pain affecting sleep). The remaining 2 questions asked about pain occurring in the moment and if pain is "ever upsetting." For these more general questions, the verbal descriptors of the scale were "a little, moderate, and extreme," providing more information about the intensity of these general states.
Second, we made several modifications so that specific administration details could be quantified. These included (a) how many verbal prompts the resident required to give an identifiable answer, (b) how often nonresponses were given (e.g., unintelligible answers, nonsense answers, or inaudible response) following prompts, and (c) what types of nonresponses were given. We referred to this final modified version as the GPM-M2.
GPM-M2 Administration Protocol
We developed a detailed administration protocol that provided specific instructions on how to introduce the measure, use response cards, provide prompts, and respond to ambiguous responses and nonresponses. The GPM-M2 was introduced with the statement, "I am going to ask you some questions about any pain that you might have." Response cards were 8.5 in. x 11 in. (21.6 cm x 28 cm) sheets of paper showing the response options in large, bold font. For scaled-response questions, we used a vertical scale, with both number and word descriptors, because research shows that these scales are easiest for nursing home residents to use (Ferrell et al., 1995; Herr & Mobily, 1993).
We allowed a maximum of three prompts for each question. Prompts included repeating the entire question or parts of the question for any reason, except when a resident indicated that she or he did not hear the question. Prompts also included summarizing the resident's response if it did not include a direct or identifiable answer to the question. If a response was not obtained after three prompts, the researcher marked "indiscernible response" on the GPM-M2 form, the corresponding Likert-scale question was considered not applicable, and the researcher moved to the next question. If this happened for three questions in a row, the administration was discontinued. If a resident refused to answer any two consecutive questions, the administration was discontinued. If a resident became agitated in such a way that represented a significant departure from her or his emotional state just prior to the administration, the administration was discontinued immediately. If a resident was able to give yes–no responses but unable to provide identifiable answers to the scaled-response questions for three consecutive questions, the resident was only asked the yes–no questions for the remainder of the administration. We trained interviewers to classify indiscernible responses into the following categories: (a) nonresponse (complete lack of response); (b) unintelligible response (verbalizations made, but actual words indiscernible); (c) actual words discernible, but talking "around" an answer precludes discernible response; (d) response impeded by agitation (e.g., repetitive, nonsense verbalizations; restless wandering); and (e) refusal to respond. These categories were not mutually exclusive and we trained interviewers to mark all that apply.
The majority of pain assessments took place during the afternoon hours (e.g., 1 PM–5 PM), subject to researcher availability. Within this timeframe, we determined administration times randomly, so that pain sampling was not systematically biased by any environmental factors in the nursing home. If the resident was unavailable for any of the scheduled administrations, multiple attempts were made within the hour of the originally scheduled administration to obtain the data and, when these attempts were unsuccessful, the administration took place on the following day or the administration was rescheduled for the next week. Ten percent of all administrations were completed independently by two research assistants (i.e., one research assistant administered the GPM-M2; both research assistants simultaneously, but independently, completed GPM-M2 forms based on residents' responses) so that we could evaluate and monitor interassessor reliability.
GPM-M2 Scoring Protocol
We scored the GPM-M2 paired items by first assigning values of 0 (representing "no") and 1 (representing "yes") to residents' responses on the yes–no question (first question in the pair). Second, we assigned values to the scaled-response options (1–3 for the three ordinal options) and summed the values for an item score, ranging from 0 to 4. Finally, we calculated a total score by summing the item scores and dividing by the number of items answered, resulting in a total score ranging from 0 to 4. Because of the frequency of missing data on items that required a specific level of functioning to answer, we used this method to calculate total scores. We did not include 2 of the 15 paired items in total score calculations because these items measured qualitatively different constructs compared with all other items. These 2 items asked about pain communication with nursing home staff ("Do the nursing staff ever ask you about your pain?" and "Do you ever complain to the nursing staff about your pain?"), rather than—as all other questions—about residents' direct experience with pain or direct effects of pain on their daily life.
Modified Mini-Mental State Examination
Teng and Chui (1987) modified the Mini-Mental State Examination (MMSE; Folstein, Folstein, & McHugh, 1975) in order to sample a broader variety of cognitive functions and to cover a wider range of cognitive difficulty levels. Specifically, there are two primary differences between the MMSE and its modified version, known as the 3MS: (a) the 3MS uses more graded scoring on most items (i.e., respondents can receive partial credit for nearly accurate answers) and (b) the 3MS has an additional four items that measure cognitive constructs not measured by the MMSE (recall of personal information, category-specific retrieval fluency, abstract-thinking ability, and extended recall of previously learned material). The 3MS yields a total score ranging from 0 to 100. The general cutoff score for presence of cognitive impairment is approximately 77, although, like the MMSE, cutoff scores vary according to age and education level (Jones et al., 2002; McDowell, Kristjansson, Hill, & Hebert, 1997; Tombaugh, McDowell, Kristjansson, & Hubley, 1996).
For both cognitively intact and cognitively impaired elders, the 3MS and the MMSE measure very similar constructs (Tombaugh et al., 1996). Given this similarity, we chose the 3MS for this study because it yields a wider range of cognitive status scores. However, we calculated total scores for both the 3MS and the MMSE items for all residents by using the scoring guide provided by Teng and Chui (1987), because the MMSE is used in most other studies.
Resident Demographic–Diagnosis Information Form
We recorded all resident diagnoses on a detailed tracking form (Allen et al., 2003; Fisher et al., 2002). On the basis of a classification procedure used by Horgas and Tsai (1998), we identified diagnoses associated with chronic pain. For diagnoses not already classified by Horgas and Tsai, we asked two experts in the field of pain assessment (one a clinical psychologist with advanced training and research expertise in pain and pain management, and the other a registered nurse with specialty training and experience in geriatrics) to independently classify diagnoses as either "associated with chronic pain" or not. We identified diagnoses for which both experts agreed as "associated with chronic pain."
Physical Self-Maintenance Scale
We administered the Physical Self-Maintenance Scale (Lawton & Brody, 1969) to residents' primary certified nursing assistants to assess the residents' degree of functional impairment in six domains: toileting, feeding, dressing, grooming, ambulating, and bathing. Higher scores indicate greater impairment.
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
Discontinuation of Interview
Of 244 total possible administrations (61 residents x 4 administrations), we discontinued 5 for 2 residents (3% of residents) because they gave indiscernible responses for the first three questions. In each of the 5 cases, the nonresponse was due to a complete lack of response from the resident, either verbal or nonverbal, after scheduled prompting. Three administrations between 2 residents were refused at the outset. Thus, for all the following analyses, the total number of administrations is 236. We discontinued an additional 4 administrations among 3 residents as a result of agitation, although each of these administrations contains partial data, which we used in the following descriptive analyses. However, we did not use administrations with partial data in analyses of resident self-report characteristics. Table 2 presents a summary of data on residents' ability to answer yes–no and Likert-scale questions.
|
= 0.998; for identifying prompts provided per item, the overall mean kappa was = 0.926 (range = 0.647–1.0).
Characteristics of Resident Self-Report
Evidence of Concurrent Validity
Because of the absence of a gold standard for measuring pain among nursing home residents, we obtained an alternative measure of pain that could generally substantiate or refute residents' self-reports by means of the classification of residents' medical diagnoses as either "associated with chronic pain" or not (Horgas & Tsai, 1998). Spearman's rank-order correlation between total number of diagnoses associated with chronic pain and residents' mean total score on the GPM-M2 (across four administrations) was rs =.37, p <.01, indicating that residents with more chronic pain-associated diagnoses tended to score higher on the GPM-M2.
Internal Consistency
Of the 13 paired items that comprised the total score, 3 additional items were excluded in the internal consistency analyses because they were not applicable for 25% or more of the residents in this functionally impaired sample. These items asked about pain that individuals experience while (a) walking, (b) going to the dining room for meals, and (c) sitting on the toilet. The number of residents with complete data for the remaining 10 pairs of items ranged from 42 to 48. (Several residents had missing data for 1 or more of the 10 pairs of items as a result of individual nonapplicability or inability to provide a discernible response.) Cronbach's alpha coefficients for the 10 pairs of GPM-M2 items ranged from = 0.87 to = 0.91 across the four administrations.
Temporal Stability
We measured the stability of residents' responses across the four administrations in two ways. First, we calculated Pearson's product–moment correlations for residents' total scores, with the total score on Administration 1 compared with total scores for each successive administration to show the overall degree of relatedness among residents' responses over time. Pearson's r ranged from.74 to.80, p <.0001, for all three comparisons. Second, we used paired samples t tests as a quantitative measure of amount of actual change in total scores from Administration 1 to Administration 2, Administration 2 to Administration 3, and Administration 3 to Administration 4. GPM-M2 total scores did not change significantly over time (t = 1.28, p =.21; t = –0.62, p =.54; t = 0.04, p =.97, respectively).
Influence of Cognitive Status
To describe the influence of cognitive status on residents' self-report characteristics (Table 3), we grouped residents into two cognitive status categories by the median split procedure (median 3MS score = 49; median MMSE score = 14). We dichotomized cognitive status for these analyses because the corresponding variables of interest (internal consistency and temporal stability) were aggregated constructs, precluding the ability to use correlational or regression analyses that require data at the level of the individual. This dichotomization of cognitive status is a common analytical procedure in similar studies (Kane et al., 2003; Levy-Storms et al., 2002). In the low-to-moderate cognitive status group, the mean 3MS score was 26.5 ± 13.4 (MMSE: 8.4 ± 4.1); in the moderate-to-high cognitive status group, the mean 3MS score was 68.7 ± 14.5 (MMSE: 20.6 ± 4.4).
|
The mean alpha coefficient (across the four administrations) for the low-to-moderate cognitive status group was = 0.82 (n = 18–24) and the mean alpha coefficient for the moderate-to-high cognitive status group was = 0.93 (n = 24–26). Regarding temporal stability, the mean correlation between successive total scores was.60 (p <.01, all values) for the low-to-moderate cognitive status group and.86 (p <.001, all values) for the moderate-to-high cognitive status group. Paired-samples t tests showed that, for both cognitive status groups, GPM-M2 responses did not change significantly over time.
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Eighty-nine percent of residents in our sample completed all four GPM-M2 interviews. This completion rate is similar to that found in a large-scale study of nursing home residents interviewed with a four-item GPM-M pain questionnaire (83% completion rate; Chu et al., 2004). Of the 7 residents in our study (11%) who did not complete all four interviews, only 2 were dropped as a result of persistent (i.e., three consecutive) indiscernible responses. Seventy-five percent of residents provided discernible responses to all yes–no questions, requiring an average of 2.8 prompts per administration. Across an average of eight Likert-scale questions asked per administration, 82% of residents provided discernible responses, requiring an average of 2.4 prompts per administration. It is noteworthy that our completion rates (and those of Chu et al.) are considerably higher than those reported in similar studies (i.e., 45–65% completion rates; Weiner et al., 1999; Ferrell et al., 1995; Simmons et al., 1997). It is possible that this difference is attributable to our administration protocol, which required interviewers to provide up to 3 verbal prompts when a discernible answer was not given and to continue the administration until three consecutive indiscernible responses or nonresponses were obtained. Previous research has not reported whether verbal prompts are provided (and if so, how many) to elicit discernible responses. For example, it may be that other studies allowed no prompts or fewer prompts or allowed for discontinuation after one or two indiscernible responses. It is our position that, in the absence of such procedural information, it is difficult to explain differing results and to integrate results across studies. Different completion rates among studies also may be influenced by response format options, content and structure of questions, and cognitive status inclusion criteria.
Given the expansive nature of our protocol and the mean cognitive status score of the sample (see Table 1), it is important to note that the average time of administration was only 13 min. The brief administration duration combined with the multidimensional nature of the questions and objective administrative protocol present a potentially attractive research and clinical tool for pain assessment in the nursing home. Perhaps more important, however, is the general implication that researchers can capture the voices of many cognitively impaired elders in a relatively brief period of time by using standardized screening and administration protocols (Chu et al., 2004). Given that nursing home researchers are calling for the inclusion of resident self-report in assessing nursing home quality and resident quality of life (Kane et al., 2003; Rubinstein, 2000; Schnelle, 2003; Simmons & Schnelle, 2001; Simmons et al., 1997), these data are encouraging in their implications for clinical use. Clearly, more research is needed to determine the extent to which a 13-min pain questionnaire is clinically feasible in the everyday milieu of the nursing home, whether shorter measures can capture comparable information, and whether indigenous nursing home staff could appropriately administer the measure.
In this study, we also sought to more specifically identify the aspects of the interview process that are affected by cognitive status. Not surprisingly, our results suggest that residents with a lower cognitive status require significantly more prompts and provide significantly more indiscernible responses to Likert-scale questions than do residents with a higher cognitive status. On yes–no questions, the residents' cognitive status did not have a significant effect on prompting or indiscernible responses. It is interesting to note that the additional prompting given to residents with lower cognitive status for Likert-scale questions did not result in a significant increase in completion time, nor did the increase in indiscernible responses on Likert-scale questions result in a difference in total scores on the GPM-M2. One contributing factor to completion time similarity is that, among all residents, the most common indiscernible response was "talking around" answers. According to anecdotal researcher notes, such vague or tangential responses, which appeared to be given by more cognitively intact residents, seemed to take more time than simply getting no response or getting an unintelligible response from less cognitively intact residents.
As we expected, Likert-scale questions required significantly more prompts than yes–no questions for the overall sample. It is important to note, however, that—when scaled questions were asked—the prompting appeared to be generally successful for eliciting discernible responses, as the rate of nonresponses for scaled items was quite low (18%) overall for a sample with a mean MMSE score of 15.
Regarding characteristics of obtained self-report, our data show that residents provided preliminarily valid (correlation with total number of chronic pain-associated diagnoses) and generally consistent responses both within (alpha coefficients) and between administrations (Pearson's product–moment correlation between total scores). In terms of validity, we acknowledge that these findings are preliminary, as more data have to be collected in order to validate a subjective, widely variable construct such as pain, particularly when most of the respondents have some degree of cognitive impairment, a diagnosis of dementia, or both. However, given that there is no widely accepted and empirically validated measure for assessing pain among nursing home residents, this study used "number of chronic pain diagnoses" as an approximate validation measure to at least address the inevitable question of validity of residents' reports. Other researchers in this area also have used similar approximations in validation studies (Horgas & Dunn, 2001; Horgas & Tsai, 1998). To reconcile the seemingly contradictory findings about validity of residents' reports in this study (i.e., significant association between number of pain-related diagnoses and pain reports in the overall sample, but not when the sample was separated by cognitive status score), we offer the possible explanation that we were underpowered to show an effect due to smaller sample sizes (i.e., cut in half through median split procedure) in the two groups compared with the overall sample.
Analgesic medication use also has been used to validate resident self-reports of pain (Fisher et al., 2002; Snow et al., 2004), and an analysis of these data, focused primarily on the relation between pain self-report and analgesic medication use, will be undertaken separately. Prior to the latter analyses, however, we believe that it is crucial to address the methodological and psychometric issues posed by obtaining such self-report data from cognitively impaired residents.
The findings of temporal stability of resident responses and interassessor reliability among researchers recording resident responses support those found in the study by Chu and colleagues (2004) in which residents' yes–no responses were significantly similar over a 48-hr time period and interrater reliability among researchers was very high, regardless of resident cognitive status. Taken together, these results build a strong case for directly assessing nursing home resident pain through self-report interviews to improve the ongoing problem of underdetection and undertreatment of pain in the nursing home. However, because of the complex nature of face-to-face interviewing with cognitively impaired elders, it is imperative that researchers and clinicians follow a replicable, standardized process and report this process in sufficient detail. Otherwise, integrating and applying the results of such studies is difficult, and skepticism about residents' self-reporting capacity is likely to remain.
|
Footnotes |
|---|
Portions of this article were presented at the 55th annual scientific meeting of the Gerontological Society of America, Boston, MA, in November 2002.
1 Department of Psychology, The University of Alabama, Tuscaloosa.
2 Center for Mental Health and Aging, The University of Alabama, Tuscaloosa.
3 Department of Information Systems, Statistics, and Management Science, The University of Alabama, Tuscaloosa.
Decision Editor: Linda S. Noelker, PhD
Received for publication April 4, 2005. Accepted for publication August 1, 2005.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
| ||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
