To the Editor:

These investigators have identified a high prevalence of missing data when pediatric oncology nurses were asked to complete health status questionnaires as proxy respondents for children undergoing therapy for acute lymphoblastic leukemia (ALL). These results, based on assessments of 122 patients at 2 institutions, raise important issues and warrant further discussion. The questionnaire used in the study was the interviewer-administered version of the Health Utilities Index Mark 2 (HUI2) and Health Utilities Index Mark 3 (HUI3) questionnaire.1,2 Four important issues are raised by these investigators-the extent to which nurses are knowledgeable about the current health status of patients, the extent to which nurses solicited information from patients and their parents to inform their responses, the roles of generic and specific measures of health status and health-related quality of life (HRQL), and whether the format of the questionnaire used in the study was appropriately matched to the method of data collection. We will discuss each issue in turn.

Clinical investigators are interested in obtaining information on health status and HRQL from the patients themselves. However, both for practical and more fundamental reasons, investigators may also be interested in proxy assessments provided by family members and/or clinicians. In many clinical situations, there are multiple versions of "truth"-multiple valid viewpoints. Patients may have deeper insight into subjective dimensions of health status, such as their emotional state. Family members may see a patient in a variety of contexts and thus have the opportunity to form well-informed impressions of how well the patient is or is not functioning. By virtue of the clinicians' training and experience, they may be better able to distinguish normal from abnormal functioning. Thus, in many situations in pediatrics, it makes sense to include both self-assessment and proxy assessment by family members and clinicians in a study to provide comprehensive information on health status and HRQL.

These arguments about the potential usefulness of proxy assessments are, however, predicated on the assumption that the proxy assessor has adequate familiarity with the health status of the patient. Perhaps the design of the study by Hinds et al artificially limited the ability of the nurses to obtain information from patients and parents. In pediatric settings, it is common for clinicians to ask patients and their parents about various aspects of the child's health, including physical functioning, emotional state, and pain and discomfort. Hinds et al indicate that information was often missing for 5 of the 8 dimensions of health status included in the HUI3 system: vision, ambulation, emotion, cognition, and pain. Vision is seldom affected by ALL and its treatment, so nurses would have had little clinical motivation to obtain information on visual function. Emotion and pain are relatively subjective and might be expected to change over the course of therapy. For these attributes, there may be no substitute for obtaining information from patients and/or parents. Ambulation is observable, especially in out-patient clinic settings. Perhaps cognition lies somewhere between the observability of ambulation and the subjectivity of emotion and pain. It is plausible that nurses in a busy clinical setting might not have adequate opportunity to observe cognitive functioning and will need to reply on information collected from patients or parents. The investigators indicate that nurses completed the proxy Health Utilities Index (HUI) questionnaire after completing a standard nursing clinical assessment of the patient, "independently of family or patient input." However, it is artificial and contrary to routine clinical practice to impede the ability of nurses to ask patients (or their parents) how the patients are feeling with respect to dimensions of health status that are of clear relevance, such as emotion, pain, cognition, and ambulation. The missing data problems reported are, we think, a result of using a questionnaire format not matched to the data collection method (discussed below) and the reliance on respondents who were not well informed and were inhibited from becoming well informed. From our perspective, the missing data problem is not intrinsic to HUI.

The investigators indicate that "our original intent in using the HUI3 was to identify the potential effect of ALL therapy on the HRQoL of a defined patient group." This is a purpose for which HUI3 is suitable. In the context of pediatric ALL, both morbidity and mortality are relevant. Use of HUI3 allows for comparing patients with ALL to other groups of patients and population normative data. Generic instruments are required for making broad comparisons and economic evaluative studies. A disadvantage of generic measures of health status and HRQL is that they may omit dimensions of health status (or levels within dimensions) relevant in a particular context. Indeed, a number of prominent guidelines recommend the use of generic preference-based measures (to permit broad comparisons and cost-utility analyses) and relevant specific measures (to provide greater precision and responsiveness). Hinds et al note that although nurses felt that the dimensions of health status included in HUI3 were relevant and important, other dimensions of health status associated with the process of dealing with illness and its treatment were also relevant and omitted from HUI3.

Health Utilities Index questionnaires are available in 2 basic formats: one designed for self-completion and the other designed for interviewer administration.1 The self-complete version consists of 15 multiple-choice questions with response options that closely match the wording of the HUI attribute levels but does not include "don't know" or "refused." The format for the interviewer-administered questionnaire includes 40 questions and a complex skip pattern. The interviewer-administered format is not appropriate for self-completion, in part because the skip pattern and response options should not be presented to respondents (don't know, refused). The skip pattern and these response options are meant for the use of the interviewer, not the respondent.

There are several other issues raised by these investigators that warrant clarification. Health Utilities Index Mark 3 is suitable for participants aged 5 years or older. Furthermore, the multiattribute utility function for HUI3 is based on preference scores obtained from a representative sample of the Canadian population aged 16 years and older; the HUI3 scoring function is not also based on the preference scores of respondents aged 11 to 13 years, as stated by the investigators. However, in several studies including one involving teenage respondents, there was a close agreement between group mean HUI scores and group mean directly measured standard gamble utility scores provided by patients.3 The computation of mean attribute levels (Table 2) is questionable, as the levels within each attribute are categorical; for 7 of the 8 attributes, the levels are ordinal (the levels for cognition are nominal). Instead, one can readily convert attribute levels into single-attribute utility scores and compute the mean single-attribute utility score for each attribute.2

The investigators are surprised by how high the overall HUI3 scores were based on the proxy reports by nurses. It is interesting to note that results from the use of the self-rated (proxy rated in this case) health question (excellent, very good, good, fair, or poor) also indicate that patients were doing well; it seems that none of the nurses used the fair or poor categories. The high overall HUI3 scores of course may reflect the frequent omission of deficits in ambulation, emotion, cognition, and pain. The scores may also reflect the schedule for assessments outlined in the study protocol. The first assessment was scheduled for 40 days after the initiation of treatment. An assessment at 7 days after the initiation of treatment might well have captured more of the HRQL burden associated with induction therapy. Perhaps 40 days was too late to capture such effects in most patients.

Health Utilities Index Mark 3 has been used in numerous studies of children and adults with a wide variety of health conditions. Missing data have not been a major problem in studies using proxy assessors, including nurses.4-7 In all of these situations, subjective dimensions of health status, such as emotion and pain, are relevant, and proxy respondents (nurses, other clinicians, and family members) seem to have been able to respond. Nurses in the study by these investigators experienced more difficulty. The missing data in these studies are perhaps mainly an artifact of the instructions to, and training of, the nurse proxy respondents who were discouraged from obtaining information from patients and parents rather than the nature of the clinical situation or the characteristics of the HUI questionnaire.

References