Assessing Quality of Life After Stroke
The Value and Limitations of Proxy Ratings
Background and Purpose Because many stroke survivors have cognitive and communication disorders, self-reported information on a patient’s quality of life (QL) cannot always be obtained. Proxy ratings may be used to prevent exclusion of this highly relevant subgroup of patients from QL studies. The purpose of this study was to evaluate both the value and possible limitations of such proxy ratings.
Methods The patient sample was composed of 437 patients who had suffered a stroke 6 months earlier. QL was assessed by means of the Sickness Impact Profile (SIP). For 108 patients who were not communicative because of cognitive or linguistic deficits, proxy ratings on the SIP were provided by the patients’ significant others. For 228 of the 329 communicative patients, both self-reported and proxy SIP ratings were obtained.
Results When mean SIP scores for patients with both self-reported and proxy-derived data available were compared, the proxy mean scores were generally in close agreement with those of the patients. However, systematic differences were noted for several SIP scales, with proxies rating patients as having more QL impairments than the patients themselves. Intraclass correlations were moderate to high for most SIP subscales (average intraclass correlation coefficient [ICC]=.63), the physical (ICC=.85) and psychosocial dimensions (ICC=.61), and the total SIP score (ICC=.77). The proxy SIP scores were sensitive to differences in patients’ functional health, which supports the validity of these ratings. For all patients combined, more QL impairments were found for patients with supratentorial cortical or subcortical infarctions and hemorrhages than for patients with lacunar infarctions and infratentorial strokes. Although proxy respondents were more frequently needed for patients with the first two types of stroke, we found no evidence of biased results as a consequence of an unbalanced use of proxy respondents across the different types of stroke.
Conclusions These results suggest that the benefits of using proxy ratings for noncommunicative patients outweigh their limitations. The findings stress the need for inclusion of this important subgroup of patients in QL studies. Their significant others are able to provide useful information on these patients’ QL.
Several studies have shown that many stroke survivors experience a decline in their QL in terms of impaired physical, functional, psychological, and social health.1 2 3 4 5 6 7 QL is most often assessed by means of either structured interviews or written questionnaires. However, it has been recognized that these methods of data collection are not always suitable for studies of stroke survivors.8 Given the frequency of serious cognitive, speech, and language disorders, many patients are not able to communicate effectively or to understand what is being asked. The inability of a highly relevant subgroup of patients to participate in such studies may yield results that cannot be generalized to the total patient population of interest.
The use of family members or caregivers as alternative sources of information (ie, proxy respondents) may help to resolve the problem of excluding patients with limited self-reporting capabilities. Although the use of proxy respondents is common in epidemiological research9 and health studies among elderly populations,10 11 surprisingly little is known about this method of data collection in stroke research. In a study of emotional and personality changes following stroke, Nelson et al12 relied solely on information obtained from family members or close companions. In two recent QL studies among stroke survivors,1 2 proxy respondents were used for patients who were not able to communicate because of severe language or cognitive disturbances. De Haan et al1 reported that proxy respondents (most often the partner) were used for 25% of their patient sample, thereby increasing the number of available patients at 6 months after stroke from 329 to 441. However, there are limitations of such proxy reports of patients’ QL, and their impact on study outcomes is not well documented.
Although the use of proxies may be an effective means of obtaining information that might otherwise be lost, it assumes that the proxy can report accurately on several aspects of the patient’s health and QL. The accuracy of proxy reports is most typically determined by examining the extent to which proxy ratings are in agreement with those provided by the patients themselves. To our knowledge, agreement between stroke survivors and their proxies has been examined in only one small study (n=38).13 This study showed good agreement for two instruments measuring frequency of and independence in performing several activities but lower levels of agreement for an instrument assessing perceived limitations in such activities. Evidence from several other studies performed in the elderly and populations of patients with chronic disease suggests that the response agreement between patients and proxies is far from optimal and that the use of proxy respondents may introduce considerable bias.14
By definition, studies comparing patient and proxy ratings can be carried out only among communicative patients. Thus, it is necessary to extrapolate findings of patient-proxy agreement for communicative patients to the subgroup of noncommunicative patients. This can be facilitated by examining trends in the level of agreement as a function of patients’ health status.14 When properly analyzed, examination of patient-proxy agreement can provide important information on the extent and direction of any bias introduced by using proxy respondents. In the current study, we examined, in a subgroup of communicative patients, the level of response agreement between stroke survivors and their significant others (ie, family members and close companions) on a standardized QL questionnaire. To facilitate extrapolation of results to the subgroup of noncommunicative patients, this analysis included determining whether agreement varied across the range of questionnaire scores.
Moreover, the validity of proxy QL ratings was examined by exploring the relationship between QL scores and the level of patients’ functioning as indicated by the well-known Rankin scale.15 16 This scale is a frequently used handicap index in stroke outcome research that can be viewed as a global functional health index with a strong emphasis on physical disability.15 The validity of proxy QL scores would be supported by a substantial association between these scores and the patients’ Rankin grade. In contrast to the former type of analysis (ie, patient-proxy agreement), this analysis can give a direct indication of the clinical validity of proxy ratings for noncommunicative patients.
In addition to evaluating the value and limitations of proxy ratings, we estimated the impact of using proxy QL ratings for noncommunicative patients on the results of the original QL study from which the data for this article were derived.1 Based on a combined analysis of self-reported and proxy-derived QL data, this earlier QL study reported a significant relationship between stroke type and QL. Specifically, patients with infratentorial strokes were found to have less QL impairment than patients with supratentorial strokes. However, among the patients with supratentorial strokes, those with lacunar infarctions exhibited significantly less QL impairment than patients with cortical or subcortical infarctions and hemorrhages. Thus, it was concluded that patients with lacunar infarctions and infratentorial strokes exhibit less QL impairment than survivors of larger supratentorial strokes (ie, cortical or subcortical infarctions and hemorrhages). In the current analysis, based on a strategy suggested by Semaan17 and Nelson et al,9 we examined the relationship between stroke type and QL with and without substitution of proxy ratings for noncommunicative patients. As in the original study, we first analyzed the combined data, substituting proxy ratings for noncommunicative patients. Subsequently, we performed separate stratified analyses of the same relationship among communicative and noncommunicative subgroups of patients.
Subjects and Methods
The study sample consisted of patients who had suffered a stroke 6 months earlier. They were the survivors of an original cohort of 760 consecutively admitted stroke patients who had participated in a multicenter quality of care study in the Netherlands. Two hundred fifty-eight patients died after the stroke, yielding a 6-month mortality rate of 34%. Of the remaining 502 eligible patients, 17 declined to participate. Of the 485 consenting patients, 112 were not able to communicate because of severe speech, language, or cognitive disorders. For these patients, data were obtained from proxy respondents. In the current analyses, 4 noncommunicative patients with a healthcare provider as proxy respondent were excluded to limit the proxy sample to the patients’ family members and close companions. For 44 of the 373 communicative patients, no QL data were available because the interview was unacceptably lengthy and burdensome to the patients. For 228 of the remaining 329 communicative patients, both self-reported and proxy ratings were obtained. Thus, the patient sample could be divided into three subgroups: communicative patients with self-reported data only (n=101), communicative patients with self-reported and proxy-derived data (n=228), and noncommunicative patients with proxy ratings only (n=108).
Patients’ sociodemographic and clinical information was obtained from the medical and nursing charts by trained research assistants shortly after discharge from the hospital. This information included age, gender, history of stroke, stroke type, and lesion location. CT data were available for 430 patients (98%). The scans were made within 2 weeks after the stroke and were evaluated by local radiologists. Stroke types were divided into supratentorial strokes (subcortical, cortical, and lacunar infarctions and hemorrhages) and infratentorial strokes. A hemorrhage was considered to be present if the CT scan showed evidence of a recent intracerebral hemorrhage. The diagnosis of lacunar stroke was made if there was a clinical picture of a lacunar syndrome and the CT scan was compatible with that diagnosis.18
The patients were interviewed 6 months after their stroke. Patients’ handicaps and disabilities in ADL were assessed by means of the modified Rankin scale15 16 and the Barthel Index.19 The Rankin scale is a six-point index that ranges from no symptoms (grade 0) to severe handicap (grade 5). The Barthel Index consists of 10 items: continence of bowels, continence of bladder, grooming, toilet use, feeding, transfer, mobility, dressing, climbing stairs, and bathing. Scores can range from 0 to 20, with a higher score indicating more ADL independence.
QL was assessed with the SIP.20 The SIP is a widely used, reliable, and valid health status questionnaire that addresses a wide range of health-related QL domains, with a focus on behavior rather than subjective expressions.8 21 The questionnaire, which consists of 136 yes/no statements describing limitations or recent changes in functioning, is organized into 12 subscales: sleep and rest, emotional behavior, body care and movement, household management, mobility, social interaction, ambulation, alertness behavior, communication, work, recreation and pastimes, and eating. Respondents are asked to endorse items that apply to them on a given day because of ill health. The SIP is scored by summing the weights attached to all endorsed statements and is expressed as a percentage of the maximum possible score. Thus, scores can range from 0 to 100, with a higher score representing more impaired QL. An aggregated score can be obtained for the total SIP as well as for physical and psychosocial dimensions. The subscales that make up the physical dimension are body care and movement, mobility, and ambulation. The subscales included in the psychosocial dimension are emotional behavior, social interaction, alertness behavior, and communication. Because most of the patients were retired at the time of their stroke, the work subscale (10 items) was excluded from all analyses.
Proxies completed a slightly modified version of the SIP in which the wording of individual items was changed so that each item clearly referred to the patient. Proxies were instructed to endorse only those statements that they were sure applied to the patient on a given day because of ill health. The proxies were also asked to provide information on their own age and gender and the nature of their relationship with the patient.
Differences were examined between the three subgroups of patients (communicative patients with self-reported data only, communicative patients with self-reported and proxy-derived data, and noncommunicative patients with proxy ratings only) in terms of stroke type, lesion location, history of stroke, Rankin and Barthel Index scores, age, and gender. To determine the extent to which these patient characteristics were associated with the need to use proxy respondents, differences between communicative (subgroups 1 and 2 combined) and noncommunicative (subgroup 3) patients were tested with χ2 tests. To examine whether the patients used in the direct patient-proxy agreement analyses were representative of the total communicative patient sample, subgroups 1 and 2 were compared as well.
Patient-proxy agreement was evaluated in the 228 communicative patients with both self-reported and proxy-derived SIP data (subgroup 2). For both sources of SIP data, mean scores, standard deviations, and the range of scores on the specific SIP scales were calculated. Internal consistency reliability for both patient and proxy scores, as indicated by Cronbach’s coefficient α,22 was established as a frame of reference for interpreting patient-proxy agreement. This is relevant because the level of agreement between patient and proxy ratings is dependent, in part, on the reliability of the instruments used.9 23
Three analytic strategies were used to examine patient-proxy agreement. First, agreement at the group level was evaluated by comparison of group means. For each SIP scale, we calculated the mean difference score between patient and proxy ratings (proxy minus patient score). Mean difference scores significantly different from zero, using paired Student’s t tests, were interpreted as providing evidence of systematic bias.24 25 To examine the statistical magnitude of any observed systematic bias, the mean difference score was standardized by relating this score to its standard deviation. Given the similarity to effect size (d) calculations for paired observations,26 a standardized difference of d=0.2 was taken to indicate a small bias; d=0.5, a moderate bias; and d=0.8, a large bias.
Second, to determine whether agreement varied across the range of SIP scores, the pattern of agreement was visually examined by means of a scatterplot. That is, for each patient, the difference between the patient and proxy scores (proxy minus patient score) was plotted against the average for each pair of scores (patient plus proxy score divided by 2).23 24 When depicted graphically, using the y axis to show difference scores and the x axis to show average scores, perfect correspondence would be represented by a horizontal line through an ordinate of zero. Any observed differences between the patient and proxy scores as a function of the range of their average scores (eg, comparable scores at high levels of functioning but diverging scores at low levels of functioning) were taken as evidence of scatter bias.25
Third, the ICC was used as an indicator of chance-corrected agreement between patient and proxy ratings at the individual patient level.24 27 Guidelines for the ICC as a measure of the strength of agreement were labeled as follows: ≤0.40, poor to fair agreement; 0.41 through 0.60, moderate agreement; 0.61 through 0.80, good agreement; and 0.81 through 1.00, excellent agreement.28
The clinical validity of proxy QL ratings was determined by examining differences in mean physical, psychosocial, and total SIP scores between patients with different Rankin grades. ANOVA was used for statistical testing of the differences. The relationship between proxy-derived SIP scores and patients’ Rankin grades was examined in two patient subgroups, communicative patients with both self-reported and proxy-derived SIP data (subgroup 2) and noncommunicative patients with proxy ratings only (subgroup 3). The availability of both self-reported and proxy-derived SIP data in subgroup 2 also allowed for a head-to-head comparison of the between-group differences in SIP scores based on patient and proxy ratings in one sample.
The effect of using proxy data for noncommunicative patients on the observed association between stroke type and QL was examined by performing both combined and stratified analyses of this relationship. Specifically, we examined differences in mean total SIP scores between patients with cortical or subcortical infarctions, intracerebral hemorrhages, lacunar infarctions, and infratentorial strokes. Statistical testing for differences between the four stroke types was conducted as follows. ANOVA was applied to test whether the mean total SIP score of at least one group (ie, stroke type) differed from one other group. Additionally, the Newman-Keuls procedure was used to test the statistical significance of any observed differences between each pair of groups.29 The between-group differences were tested for all patients combined, as well as for the communicative and noncommunicative patients separately. Within the subsample of communicative patients with both self-reported and proxy-derived SIP data (subgroup 2), we also made a head-to-head comparison of the between-group differences in total SIP scores based on patient and proxy ratings in one sample.
The characteristics of the patient sample are summarized in Table 1⇓, both for the total study sample and broken down by the source of information on the SIP. Three hundred thirty-one patients had suffered a supratentorial stroke (200 subcortical and cortical infarctions, 49 intracerebral hemorrhages, and 82 lacunar infarctions), and 61, an infratentorial stroke. For 45 patients the stroke type was unknown or incompletely described. In terms of lesion laterality, 172 patients had right-hemisphere and 191 patients had left-hemisphere lesions (for 74 patients, lesion laterality was undetermined, or they had infratentorial strokes). For 65% of the patients, it was their first stroke. Moderate to severe handicaps (Rankin grades 3 to 5) were noted in 59% of the patients, but 77% of all patients were nevertheless considerably ADL independent (Barthel Index scores of 15 to 20). Fifty-nine percent of the patients were older than 70 years of age (mean age, 70 years; range, 20 to 94 years), and 55% were male.
As expected, the characteristics of the noncommunicative patients (n=108) differed significantly from those of the communicative patients (n=329, subgroups 1 and 2 combined). Noncommunicative patients more often had supratentorial cortical or subcortical infarctions and hemorrhages (P<.001), left-hemisphere lesions (P=.01), and moderate to severe handicaps (P<.001) were more frequently ADL dependent (P<.001), and were older (P<.001) than the communicative patients. Among the communicative patients, no significant differences were noted between the subgroup of patients for whom both patient and proxy SIP data were obtained (n=228) and those with self-reported SIP data only (n=101).
The proxy respondents most often were the patient’s spouse or partner (65%), with the remainder being family members or friends. The mean age of the proxies was 61 years (range, 18 to 90 years), and 73% were female.
Agreement between patient and proxy SIP scores was examined in the subgroup of 228 patients for whom both sources of information were available. Table 2⇓ presents patient and proxy mean scores and reliability coefficients for 11 of the 12 SIP subscales (work was excluded as mentioned above), the physical and psychosocial dimensions, and the total SIP. For most subscales and dimensions, there was substantial variation in scores, with both the patient and proxy scores spanning a relatively large segment of the possible range of scores. However, for the total SIP, the observed scores in this subgroup of communicative patients covered only half the possible score range (0 to 50.3 and 0 to 58.3 for patients and proxies, respectively). Internal consistency reliabilities surpassed or approached the .70 criterion for group-level comparisons30 for 9 of the 11 SIP subscales. Based on both patient and proxy ratings, the internal consistency of the sleep/rest and eating subscales was relatively poor. High-reliability estimates were noted for the physical and psychosocial dimensions as well as for the total SIP score.
Differences between patient and proxy mean scores were statistically significant for 7 of the 11 SIP subscales, the physical and psychosocial dimensions, and the total SIP score (Table 3⇓). Except for the eating subscale, the mean differences were all in the same direction, with the proxies rating the patients as having more functional limitations than the patients themselves. The magnitude of the differences between patient and proxy mean scores was low to moderate (d=−.04 to .45).
Evidence of scatter bias was found for 6 SIP subscales, the physical and psychosocial dimensions, and the total SIP. Specifically, the tendency of proxies to rate patients as having more functional limitations than the patients themselves was most notable among patients with more impaired levels of functioning. Fig 1⇓ illustrates this pattern of bias for the total SIP score. This plot shows that both the magnitude and direction of the differences between patient and proxy total SIP scores are dependent on the patients’ level of functioning. Thus, although the overall bias for the total SIP score is moderate (d=0.44), this bias is much more pronounced at more impaired levels of functioning.
At the individual patient level, chance-corrected agreement between patient and proxy scores for the SIP subscales ranged from moderate for the eating scale (ICC=.47) to excellent for the ambulation (ICC=.80) and body care and movement scales (ICC=.82). Good to excellent agreement was noted for the physical (ICC=.85) and psychosocial (ICC=.61) dimension scores and for the total SIP score (ICC=.77).
Clinical Validity of Proxy Ratings
The relationship between patient and proxy SIP scores and patients’ Rankin grades is depicted in Table 4⇓. The first two columns show a head-to-head comparison of the patient and proxy SIP scores in the same sample of communicative patients (n=228). The mean patient- and proxy-rated SIP scores on the physical and psychosocial dimensions, as well as the total SIP, were significantly associated with the patients’ Rankin grade. As expected, based on the results described above, the mean proxy SIP scores were systematically higher than the mean patient SIP scores, with this difference being most pronounced for patients with Rankin grade 4 (grade 5 was not observed among the communicative patients). Importantly, the third column of Table 4⇓ shows that the mean physical, psychosocial, and total SIP scores for noncommunicative patients, based on proxy ratings, were also significantly associated with the Rankin grade (ranging from grade 2 to 5 among these patients). The between-group differences in mean scores were larger for the physical dimension than for the psychosocial dimension, as expected given the emphasis of the Rankin scale on physical disability.
Impact of Proxy Data on Study Results
Table 5⇓ shows the results of both the combined and stratified analyses of the association between stroke type and the total SIP score. Overall, the stratified analyses demonstrated substantial differences between communicative and noncommunicative patients in their level of QL impairment, with the mean total SIP score of the noncommunicative patients (36.0±14.3) being almost twice as high as that ofthe communicative patients (18.6±11.8). Because the results described above indicated that for the communicative patients proxy-derived SIP scores were systematically higher than patient-derived SIP scores and that this bias was dependent on the patients’ level of functioning, the observed difference may partly be due to the source of SIP information (see footnote in Table 5⇓).
Combined analysis of patient-derived scores for communicative patients and proxy-derived scores for noncommunicative patients yielded significantly higher total SIP scores (ie, more QL impairment) for patients with supratentorial cortical or subcortical infarctions and hemorrhages than for patients with lacunar infarctions and infratentorial strokes. When performing stratified analyses, this pattern of results could not be confirmed. Among the communicative patients, the highest mean total SIP score was observed for patients with cortical or subcortical infarctions, and the lowest for patients with lacunar infarctions, with intermediate scores for patients with intracerebral hemorrhages and infratentorial strokes. A statistically significant between-group difference was noted between patients with cortical or subcortical and lacunar infarctions only. For noncommunicative patients, the highest mean score was observed for patients with intracerebral hemorrhages, and the lowest for patients with infratentorial strokes, but no statistically significant differences were found between the four stroke types.
Table 6⇓ shows the distribution of mean total SIP scores across the four stroke types within the subsample of communicative patients for whom both self-reported and proxy-derived SIP data were available (n=228). Although the magnitude of the patient and proxy scores differed, the pattern of mean scores across the stroke types was similar for self-reported and proxy-derived SIP scores. Regardless of the source of SIP data, statistically significant differences were noted between patients with cortical, subcortical, or lacunar infarctions, with intermediate levels of QL impairment for the remaining stroke types. This suggests that the use of proxies would not change the observed relationship between stroke type and the total SIP score.
The primary aim of the current study was to examine the value and limitations of proxy ratings in evaluating the QL of stroke survivors. Proxy ratings of patients’ QL may be needed for patients with cognitive or communication disorders who would otherwise be excluded from study participation. Exclusion of such patients can compromise the validity and generalizability of study outcomes. However, this type of selection bias must be weighed against the bias that may be introduced when significant others are used as proxy respondents for noncommunicative patients.
Comparison of the patient characteristics of the noncommunicative patients, representing one quarter of the total sample, with those of the communicative patients yielded several important differences. Not surprisingly, the noncommunicative patients more often had supratentorial cortical or subcortical infarctions and hemorrhages and left-hemisphere lesions and were more severely handicapped, more frequently ADL dependent, and older than the communicative patients. This finding supports empirically the argument that patients who are unable to provide self-reported data should not be excluded from QL assessments, as has been the case in several earlier studies (eg, Niemi et al6 and Viitanen et al7 ).
The quality of the proxy ratings was evaluated by comparing patient and proxy responses to the SIP in a subsample of patients for whom both sources of information were available. This patient subgroup, including more than two thirds of the communicative patients, was representative of the total communicative patient sample with respect to a number of relevant clinical and sociodemographic characteristics. Importantly, this subsample of patients was highly heterogeneous in terms of stroke type, disease severity, and QL, thereby facilitating the examination of trends in the pattern of patient-proxy agreement across the range of SIP scores. In turn, this enabled us to estimate the potential degree of bias introduced by using proxy respondents for the noncommunicative patients.
The comparative analyses of the patient self-reported and proxy-derived SIP data yielded somewhat conflicting results. Correlations between the patient and proxy ratings on the SIP subscales ranged from moderate to excellent. For some subscales (ie, sleep and rest, eating), the lower correlations may be attributed, in part, to lack of scale reliability. The patient-proxy correlation was high for the aggregate physical dimension score and moderate for the aggregate psychosocial dimension score. This finding is at odds with the observations of an earlier study using the SIP,31 in which the correlations between patients and their significant others were high for the physical dimension but poor for the psychosocial dimension. In this latter study, among more severely impaired patients (ie, as indicated by SIP scores), the psychosocial dimension score was found to be more closely associated with the proxy’s own level of psychological distress and caregiver burden than with the patient’s psychosocial health. Finally, for the total SIP score, we also observed a fairly strong correlation between patient and proxy ratings. Again, the observed correlation was higher than that reported in a study of more severely impaired patients (ie, chronically or terminally ill homebound patients) and their caregivers.32
Encouraging results were also noted when we evaluated the comparability of mean scores at the group level. Although the proxies systematically rated patients as having more functional impairments than the patients themselves (a finding in line with a consistent trend in the proxy literature),10 14 these differences were relatively small in magnitude. This suggests that, at the group level, only a modest degree of bias would be introduced when substituting patients’ self-report of their QL by ratings provided by significant others.
However, although overall agreement was generally quite high when viewed in terms of correlations and mean differences, the magnitude of patient-proxy agreement was found to be clearly associated with the patients’ level of functioning. The tendency of proxies to rate patients as having more limitations than patients themselves was most pronounced among patients with more impaired levels of functioning. Because our patient sample was not as functionally impaired as those of earlier studies using the SIP,31 32 this finding may explain the higher rates of agreement observed in our patient sample as compared with those reported in the latter investigations. In line with an earlier study among patients with brain cancer,33 extrapolation of these findings suggests that lower levels of agreement and more biased ratings can be expected among noncommunicative patients.
Additionally, the validity of proxy QL ratings was determined by examining the association between the SIP scores and the level of patients’ functional health as indicated by the Rankin scale. The results provided support for the clinical validity of the proxy QL ratings. Although the proxy SIP scores were consistently higher than the patients’ own scores (being in line with the results described above), the proxy ratings were clearly sensitive to differences in patients’ functional health. This was also true for the proxy SIP scores for noncommunicative patients, giving a direct indication of the validity of proxy QL ratings for those patients for whom proxies are really needed.
To illustrate the potential effects of combining patient and proxy ratings of patients’ QL on relevant study outcomes, we performed a more detailed analysis of the relationship between stroke type and QL. The results of analyses that combined patient SIP scores for communicative patients with proxy SIP scores for noncommunicative patients indicated more QL impairment for patients with larger supratentorial strokes (ie, cortical or subcortical infarctions and hemorrhages) as compared with patients with lacunar infarctions and infratentorial strokes. When limiting the analysis to the communicative patient group, only the difference between cortical or subcortical and lacunar infarctions could be confirmed. This suggests that the other apparent differences in the combined analysis were due mainly to differences among the noncommunicative patients who were rated by proxy respondents. In turn, the observed differences in the combined analysis might be attributed, to some extent, to the more frequent use of proxy respondents for patients with larger supratentorial strokes. As suggested by our findings, the level of QL impairment of the latter patients may be overestimated because of the reliance on proxy ratings for a substantial percentage of these patients.
There are, however, several reasons to believe that the observed differences in levels of QL impairment across the four stroke types represent real differences in QL rather than an artifact of an unequal distribution of noncommunicative patients and, consequently, an unbalanced use of proxy respondents. First, patients with larger supratentorial strokes are likely to have more severe cerebral dysfunction. It can logically be expected that these patients also experience a more impaired QL than patients with lacunar and infratentorial strokes. Second, within the patient group for whom both self-reported and proxy ratings were available, the pattern of mean QL scores across the stroke types was not dependent on the source of information used. Although this subgroup of patients is not representative of the total patient sample, it suggests that the observed relationship between stroke type and QL would not be altered by the use of proxy-derived information. Together, these findings suggest that the use of proxy respondents for the noncommunicative patients did not affect the observed relationship between stroke type and QL.
In conclusion, although this study provides encouraging results on the validity of proxy ratings of patients’ QL, researchers need to exercise some caution in interpreting their data when using proxy respondents for noncommunicative patients. Most likely, proxies will overrate the level of QL impairments for these patients. However, the impact on the study outcomes is not likely to be of clinical significance. Furthermore, the current findings stress the need for inclusion of this important subgroup of noncommunicative patients in QL investigations. If the analyses of the larger QL study1 had been based on the communicative patients only, the level of QL impairment would have been underestimated, and, perhaps more importantly, questionable conclusions would have been drawn about the impact of several stroke types on the patients’ QL. These findings indicate that noncommunicative patients should not be excluded from study participation and that their significant others can be used, with the necessary caution, as proxies to rate these patients’ QL.
Selected Abbreviations and Acronyms
|ADL||=||activities of daily living|
|ICC||=||intraclass correlation coefficient|
|QL||=||quality of life|
|SIP||=||Sickness Impact Profile|
This study was supported by grants from The Netherlands Heart Foundation (NHS 40.004), Developmental Medicine (OG 1991-037), and the Dutch Cancer Society (NKI 93-139). Dr Limburg is a Clinical Investigator of The Netherlands Heart Foundation.
Reprint requests to Neil K. Aaronson, PhD, Division of Psychosocial Research and Epidemiology, The Netherlands Cancer Institute, Plesmanlaan 121, 1066 CX Amsterdam, Netherlands.
- Received February 25, 1997.
- Revision received May 12, 1997.
- Accepted May 13, 1997.
- Copyright © 1997 by American Heart Association
De Haan RJ, Limburg M, van der Meulen JHP, Jacobs HM, Aaronson NK. Quality of life after stroke: impact of stroke type and lesion location. Stroke. 1995;26:402-408.
Tuomilehto J, Nuottimäki T, Salmi K, et al. Psychosocial and health status in stroke survivors after 14 years. Stroke. 1995;26:971-975.
Angeleri F, Angeleri VA, Foschi N, Giaquinto S, Nolfe G. The influence of depression, social activity, and family stress on functional outcome after stroke. Stroke. 1993;24:1478-1483.
Johansson BB, Jadbäck G, Norrving B, Widner H, Wiklund I. Evaluation of long-term functional status in first-ever stroke patients in a defined population. Scand J Rehab Med. 1992;26(suppl):105-114.
Åström M, Asplund K, Åström T. Psychosocial function and life satisfaction after stroke. Stroke. 1992;23:527-531.
Niemi ML, Laaksonen R, Kotila M, Waltimo O. Quality of life 4 years after stroke. Stroke. 1988;19:1101-1107.
De Haan R, Aaronson NK, Limburg M, Hewer RL, Van Crevel H. Measuring quality of life in stroke. Stroke. 1993;24:320-327.
Nelson LM, Longstreth WT Jr, Koepsell TD, Van Belle G. Proxy respondents in epidemiologic research. Epidemiol Rev. 1990;12:71-86.
Magaziner J. The use of proxy respondents in health studies of the aged. In: Wallace RB, Woolson RF, eds. The Epidemiologic Study of the Elderly. New York, NY: Oxford University Press; 1992:120-129.
Segal ME, Schall RR. Determining functional/health status and its relation to disability in stroke survivors. Stroke. 1994;25:2391-2397.
De Haan R, Limburg M, Bossuyt P, van der Meulen J, Aaronson N. The clinical meaning of Rankin “handicap’ grades after stroke. Stroke. 1995;26:2027-2030.
Semaan S. Impact of proxy-reported data on the relationship between income and severity of functional impairment among impaired elderly. J Appl Gerontol. 1994;13:341-354.
Bamford J, Sandercock P, Jones L, Warlow C. The natural history of lacunar infarction: the Oxfordshire Community Stroke Project. Stroke. 1987;18:545-551.
Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;8476:307-310.
Marshall GN, Hays RD, Nicholas R. Evaluating agreement between clinical assessment methods. Int J Methods Psychiatr Res. 1994;4:249-257.
Cohen J. Statistical Power Analysis for the Behavioral Sciences. 2nd ed. Hillsdale, NJ: Lawrence Erlbaum Associates; 1988:19-74.
Snedecor GW, Cochran WG. Statistical Methods. 7th ed. Ames, Iowa: Iowa State University Press; 1980.
Nunnally JC, Bernstein IH. Psychometric Theory. 3rd ed. New York, NY: McGraw-Hill; 1994.
Sneeuw KCA, Aaronson NK, Osoba D, et al. The use of significant others as proxy raters of the quality of life of patients with brain cancer. Med Care. In press.