Correspondence
Trials of Improved Care for the Elderly
N Engl J Med 1996; 334:665-666March 7, 1996
- Article
To the Editor:
In their study of annual in-home comprehensive geriatric assessments for elderly people living at home (Nov. 2 issue),1 Dr. Stuck and colleagues use multiple imputation to address problems arising from missing data. Multiple imputation is a great statistical innovation for strengthening inferences when data are missing. However, I think that researchers should either describe the multivariate model used to generate imputed values or at least indicate where readers can obtain such information. Dr. Stuck and colleagues include only a general statement that “the imputed estimates were derived from the known base-line and outcome data, with the use of maximum-likelihood techniques and simulations.” What model was used for the maximum-likelihood techniques? What kind of simulation approach was used? The authors give a reference,2 which provides an excellent summary of the general methodology but not the methods specifically used by Dr. Stuck and colleagues.
The authors describe only what happened to point estimates of the intervention effect when imputed values were used. Although this may address the issue of potential bias, missing data also add uncertainty to estimates of intervention effects. Multiple imputation can be used to assess this added uncertainty. It would be appropriate for the authors to discuss not only the direction of results in their sensitivity analyses but also the statistical significance of the results.
2 ReferencesThomas R. Belin, Ph.D.
University of California, Los Angeles, Los Angeles, CA 90095-17721
Stuck AE ,Aronow HU ,Steiner A , et al. A trial of annual in-home comprehensive geriatric assessments for elderly people living in the community. N Engl J Med 1995;333:1184-1189
Full Text | Web of Science | Medline2
Rubin DB. Multiple imputation for nonresponse in surveys. New York: John Wiley, 1987.
To the Editor:
Clinical trials need to define explicitly the follow-up periods. However, in some clinical trials this has ethical implications. People in the experimental or intervention group receive special care, sometimes very intensive care. What happens at the end of the clinical trials? How do the patients feel? After the end of the clinical trial their situation has changed forever. The patients know firsthand a different way of treatment, perhaps a better one that they cannot afford by themselves. Must they go back to receiving their “regular medical care” (if they had any)?
In the three-year study by Stuck et al., the intervention group had both annual comprehensive geriatric assessments performed in their homes by gerontologic nurse practitioners and follow-up visits at home every three months. The nurse practitioners were available by telephone.
In the clinical trial by Rich et al. (Nov. 2 issue),1 the study treatment consisted of intensive education, individualized dietary assessment and instruction, consultation with social-service personnel, analysis of medications, and intensive follow-up after discharge, including home visits and contact by telephone for 90 days after discharge.
Did the patients fully understand that the treatment was experimental and temporary? How were they shifted back to their regular medical care? What impact did the transition have on the patients?
One cannot ignore these ethical questions.2 Participants in clinical trials of improved care should not be considered disposable.
2 ReferencesJuan Gérvas, M.D., Ph.D.
Mercedes Pérez-Fernández, M.D.
Equipo Cesca, 28020 Madrid, Spain1
Rich MW ,Beckham V ,Wittenberg C ,Leven CL ,Freedland KE ,Carney RM . A multidisciplinary intervention to prevent the readmission of elderly patients with congestive heart failure. N Engl J Med 1995;333:1190-1195
Full Text | Web of Science | Medline2
Lilford RJ ,Jackson J . Equipoise and the ethics of randomization. J R Soc Med 1995;88:552-559
Web of Science | Medline
Author/Editor Response
The authors reply:
To the Editor: As stated by Gérvas and Pérez-Fernández, subjects in the intervention group in our study were visited by gerontologic nurse practitioners for a period of three years. Since all the patients continued to receive medical care from their regular physicians, the continuity of medical care was maintained after the end of the project for all subjects in the intervention group. The termination of the project was carefully planned in advance, since most older persons surveyed wished to continue the preventive visits and had developed close relationships with their nurse practitioners. All participants were informed about the duration of the project at the beginning of the study. Before project termination, the project team reviewed each participant's situation to decide whether any special help was needed, such as referral to a community-based agency. In addition, one nurse practitioner continued to be available by telephone (or, in urgent cases, at home) for an additional year after the end of the project, and contact by telephone every six months is being maintained.
Dr. Belin, who helped calculate the imputed estimates for our study, makes interesting points. Multiple imputation has been developed to estimate the effect of missing values on the results of multivariate analyses, since simply excluding missing data from analyses can result in biased estimates because of selection bias. In our study, we had complete three-year follow-up data on survival and health care use, but data on functional status at three years were missing for 12 percent of the subjects. Therefore, the purpose of the imputation analysis was to test whether the finding of a favorable effect of the intervention on functional outcome might be the result of selection bias due to missing information rather than a true intervention effect. To test this hypothesis, three imputed values were generated for each missing functional-status value on the basis of a set of preselected variables. The models used for this multistep analysis are complex and have now been described in detail.1,2 As stated in our paper, the results of the imputation analysis indicated that missing values did not result in an overestimation but rather in an underestimation of treatment effects on functional outcomes.
2 ReferencesAndreas E. Stuck, M.D.
Zieglerspital, CH-3001 Bern, SwitzerlandHarriet U. Aronow, Ph.D.
John C. Beck, M.D.
University of California, Los Angeles, Los Angeles, CA 900241
Ezatti-Rice TM, Khare M, Schafer JL. Multiple imputation of missing data in NHANES III. In: Proceedings of the Annual Research Conference of the Bureau of the Census, Arlington, Va., March 21–24, 1993. Washington, D.C.: Bureau of the Census, 1993:459-87.
2
Schafer JL. Algorithms for multiple imputation and posterior simulation from incomplete multivariate data with ignorable nonresponse. (Ph.D. thesis. Cambridge, Mass.: Harvard University, 1991.)
Author/Editor Response
Drs. Gérvas and Pérez-Fernández raise important ethical questions concerning the fate of patients after the termination of a clinical trial. In our study, the patients were informed at the time of enrollment that the duration of the study was 90 days. Throughout this period, all the patients were followed by their regular physicians, thereby ensuring that the continuity of care was preserved after the study was terminated. For patients assigned to the intervention group, continued follow-up by the study team was offered, but fewer than 10 percent of patients indicated a need for further support.
Despite these precautions, we were concerned about a potential rebound increase in readmissions after the termination of the study intervention. For this reason, all patients were followed for an additional nine months, during which readmissions were carefully monitored. Fortunately, as reported in our paper, not only was there no evidence of a rebound effect, but also there was a strong trend toward a continued reduction in the number of admissions for congestive heart failure during the extended follow-up period. Thus, although we understand the concern of Drs. Gérvas and Pérez-Fernández, we believe that the ethical rights of the participants in our study were fully protected.
Michael W. Rich, M.D.
Washington University Medical Center, St. Louis, MO 63110
