Methods

Top Methods Statistical Analysis Results Discussion Author & Article Info References

Participants were consenting patients in the Economy and Efficacy of Medicare Reimbursement of Preventive Services Project [6]. To be eligible to participate, participants had to be continuing care patients at the participating primary care practice sites, be enrolled in both parts A and B of Medicare, be living in the community rather than a nursing home or rest home, be approved by their physician to participate, and not be enrolled in a health maintenance organization.

Questionnaire

Initial in-person, structured, nondirective interviews were conducted between October 1986 and June 1988 at the primary care practice site where the participant was a patient. Follow-up interviews, consisting of the same questions as the baseline interview, were conducted by telephone 2 years after the initial interview between October 1988 and June 1990.

As a secondary research question, we examined the stability of patients' treatment preferences in the event of terminal illness. Selected items from the baseline and follow-up questionnaires included demographic characteristics (age, sex, marital status, and race), education, income, health insurance coverage in addition to Medicare, number of hospitalizations between baseline and follow-up, several self-reported measures of health including Self-Perceived Health Status [7]; the Quality of Well-Being scale and subscales [8-10] (modified for a population over age 65); the Perceived Quality of Life scale [11]; a 10-item (0 to 30 point) shortened version of the Center for Epidemiologic Studies Depression Scale [12]; measures of social support (a 7-item scale with scores ranging from –7 to +7); possession of a living will; and the Treatment Preference Scale.

The Treatment Preference Scale was developed using the following series of questions reported previously [13]: "Now I want you to think about what things would be like if you were diagnosed as having a terminal illness, by this, I mean if you were dying with a disease that could not get better no matter what your doctor does. Do you want to: 1) Go to the hospital? 2) Go to intensive care? 3) Have your heart revived? 4) Have surgery? 5) Be put on a breathing machine? 6) Be fed through a tube into your stomach or your veins?"

The Treatment Preference Scale was not intended to reflect the entire set of conditions (for example, dementia or permanent unconsciousness) that a person might anticipate in planning their future care, nor did the interviewer discuss the patient's responses to further clarify and understand them, as a physician would be advised to do when actually planning future care with a patient. Rather, these research questions provided a measure of desire for life-sustaining treatments and stability of this preference.

Patients were asked to respond "yes" (assigned a value of +1), "no" (assigned a value of –1),or "don't know" (assigned a value of 0) for each treatment choice. These values were summed for each person's six treatment choices, resulting in a range of possible scores from –6 ("no" to all treatments) to +6 ("yes" to all treatments). This scale had a Chronbach of 0.85, indicating excellent reliability. A score for the stability of preferences was calculated by subtracting the baseline Treatment Preference scale score from the follow-up Treatment Preference scale score. The new score could range from –12(a change from all "yes" to all "no" answers) to +12 (a change from all "no" to all "yes" answers). The purpose of this score was to evaluate the general tendency of participants to move toward more treatment or less treatment. Patients who did not complete both a baseline and follow-up interview were excluded.

Statistical Analysis

Top Methods Statistical Analysis Results Discussion Author & Article Info References

We divided the Treatment Preference Scale into three categories: most treatment (+2 to +6), moderate treatment (–1 to +1), and least treatment (–2 to –6).Similarly, we divided the stability of treatment preference scores into three categories: a change to desiring more treatment (score of +2 to +12), no change in choices [–1 to +1], and 3) a change to desiring less treatment ( –12 to –2).We excluded from the analyses patients who responded "don't know" to all of the questions and examined these patients separately.

The McNemar chi-square statistic was calculated to assess stability between baseline and follow-up responses. Summary statistics were used to examine demographic characteristics and changes over time for all the study variables. Subsequently, bivariate analyses (the Pearson chi-square for categorical variables and one-way analysis of variance for continuous variables) were done to determine what factors were associated with changes in treatment choices. Following these unadjusted analyses, we fit an ordinal logistic regression model, using the three categories reflecting stability of preferences as the response. The initial model included all the study variables in the previous bivariate analysis, as well as any potentially important interaction terms. Factors considered were demographic characteristics, changes over time in measures of perceived health and medical care, and changes over time in measures of the Quality of Well-Being Scale and the Perceived Quality of Life Scale. A backward elimination technique was used to remove terms that were no longer statistically significant (P > 0.1). Interactions were tested first, followed by global (multiple degree-of-freedom) tests to remove groups of the least important factors, and then individual terms were examined and removed if their significance level fell below an of 0.1. The selected model included the joint set of factors that were the best predictors of the tendency to change choices. Finally, we used the estimates from the model to calculate adjusted probabilities (reported as percentages) of the three categories of stability for subgroups of each predictive factor.

Results

Top Methods Statistical Analysis Results Discussion Author & Article Info References

Participants (n = 2536) ranged in age from 65 to 99 years, 62% were female, and 70% were white [6]. Eighty percent of the participants had a high school education or less. Only 3% had Medicaid as well as Medicare, and many participants (73%) carried some type of health insurance other than Medicaid or Medicare, although many of these additional policies were limited.

Participants were a broad range of outpatients, from those with serious chronic illness to those who were generally healthy. Reviews of the medical records of 455 of the first participants from three of the practices showed that only 14% had no chronic illness, 19% had a minor chronic illness, 50% had a major chronic illness, and 15% had a severe chronic illness or advanced cancer.

Of the 2536 patients completing baseline interviews, 2073 (82%) completed interviews at follow-up. Of the 463 patients lost to follow-up, 41% refused the second interview, 48% died, and the remainder either moved or could not be located. The patients lost to follow-up were somewhat older and had less education than patients who completed both interviews. They also tended to be male and unmarried, but the differences were small. The response rate for blacks and whites did not differ.

Stability of Preferences

Variation between baseline and follow-up was substantial in the percentage of respondents desiring the six life-sustaining treatments (Table 1). Of the patients who answered "yes" to any of the questions at baseline, only 18% to 43% answered "yes" to the same question at follow-up. Patients responding "no" to baseline questions were more consistent, with 66% to 75% still answering "no" at follow-up.

Patients who wanted the most treatment at their first interview were as likely to want the least treatment at follow-up as to still want the most treatment (Table 2). Patients whose choices initially fell in the midrange were the most likely to change their overall choices at follow-up; only 18% gave choices in the same range after 2 years. Patients who wanted the least treatment at the first interview were most likely (85%) to make choices in the same range 2 years later.

The distribution of scores on the stability of preference scale shows that the largest number of people had either similar preference scores between baseline and follow-up or they moved in a slightly negative direction (mean, –1.18;median, –1,representing an overall change toward less desired treatment) (Figure 1). Approximately 42% of participants changed to less desired treatment choices ( –12 to –2),38% had essentially no change (–1 to +1), and only 20% moved toward more desired treatment (+2 to +12).

View larger version (28K): [in this window] [in a new window]   Figure 1. Frequency distribution of the stability of treatment preference scores. The stability was calculated by subtracting the follow-up Treatment Preference Scale score from the baseline Treatment Preference Scale score.

The 254 participants (12%) who responded "don't know" to all the treatment questions were excluded from the stability score. These patients were more likely to be married, male, white, and better educated than the remaining patients, but did not differ in age or in any of the health or social support scales. Their characteristics were similar to participants whose scores fell in the "no change" or "less desired treatment" categories.

Stability of Treatment Preferences for Patient Subgroups: Bivariate Analysis

In the unadjusted bivariate analyses Table 3, blacks were more likely than whites to change to either more desired treatment (24% compared with 18%) or less desired treatment (46% compared with 39%). Patients with the least education also were more likely to change toward either more desired (26%) or less desired treatment categories (43%) when compared with patients with more than 12 years of school (15% and 39%, respectively). Patients moving toward more desired treatment also tended to have lower income and were less likely to carry additional health insurance. Age and sex were not significantly related to the stability of choices.

Changes in several of the health measures were associated with the stability of choices of care. Patients who were hospitalized one or more times between baseline and follow-up interviews were more likely to change their choices to more desired treatment (27%) than were patients with no hospitalizations (18%). However, changes in self-perceived health did not seem to affect stability. Of the 103 patients who had some type of accident without injuries between the two interviews, 32% changed their choices to more aggressive care, whereas only 19% of the others did so.

Patients with lower depression scores at follow-up were less likely to change their choice to more treatment (17%) than were patients with increases in their depression scores (26%). Patients receiving much less social support at follow-up (as indicated by a decrease of more than 3 points on a 15-point scale) tended to choose more treatment (29%) than patients receiving much more social support (15%). Loss of mobility, more restricted physical capability, or health-related loss of social activity (as indicated by the Quality of Well-Being subscales) caused shifts in choices toward more life-sustaining treatments. Change in the total Quality of Well-Being score or Perceived Quality of Life Scale did not result in unstable choices.

Multivariable Associations between Patient Characteristics and Stability of Treatment Preferences

Ordinal logistic regression to predict the probability of choosing more, less, or the same amount of treatment indicated that having had at least one hospitalization (23% compared with 18%), an accident with no injuries (29% compared with 19%), worsening of the Quality of Well-Being Mobility Scale (23% compared with 19%), an increase in depression (25% compared with 15%), loss of social support (25% compared with 14%), and loss of access to medical care (30% compared with 19%) were independent predictors of a desire for more treatment at the follow-up interview (Table 4). Those who initially chose less treatment tended to adhere to this choice whether they experienced one of these events (83% still desired less treatment) or did not (87% still desired less treatment). Demographic characteristics, as well as other health measures and scales, were no longer statistically significant (P > 0.10) after adjustment for other variables.

Possession of a Living Will and Stability of Preferences

Baseline and follow-up preferences for patients with and without living wills throughout the study period were examined separately (Table 5). Patients who consistently stated that they had a living will at both interviews were more likely to maintain stable preferences (86% unchanged treatment categories) than patients who consistently did not have a living will (59% unchanged treatment categories). Among patients with a living will throughout the study, 14% initially expressed a desire for more or moderate amounts of life-sustaining treatment during terminal illness. At follow-up many changed their wishes. In contrast, patients who had living wills and chose the least amount of care at their initial interview had extremely stable preferences (96% unchanged).

Patients who had change in their lives (as indicated by any one of the changes listed in Table 4 were more likely to increase their desire for treatment whether or not they stated that they had a living will (Table 6).

Discussion

Top Methods Statistical Analysis Results Discussion Author & Article Info References

Our analysis was designed to examine the stability of elderly patients' preferences for life-sustaining treatments and the factors that influence this stability. The results show that, on average, the population tended to choose to receive one less treatment at follow-up than it did initially. The results also suggest that a choice to forego a treatment was approximately twice as stable as a choice to accept a treatment. Persons who were likely to want more treatment when questioned the second time were those who had had an accident or hospitalization, had become more immobile, had become more depressed, or had lost social support. Patients possessing a living will generally had more stable preferences, but this was not the case when they had had various difficulties.

Several limitations of the study must be acknowledged. First, patients did not receive any information about treatments before they were asked about their treatment preferences; thus, we cannot know how informed their choices were. We believe that because this was true during both interviews, it does not impede our ability to examine stability of preferences. It does prevent us, however, from learning if a change in choices was caused by patient acquisition of new information. Second, the format for inquiring about life-sustaining treatment choices through closed-end questions in a research questionnaire is not the approach that would be used by clinicians and patients in discussing advance directives because a clinician might explore the reasoning and guide the choices stemming from patients' preferences. In particular, the question we use about preferences during terminal illness leaves much ambiguity about what "terminal" means. If left unclarified, the interpretation of "terminal" may vary from implying a diagnosis that is ultimately fatal but associated with several potential years of life to implying an illness that is associated with imminent death. Ideally, when a patient and physician discuss treatment options, they develop an explicit, shared understanding of the words they use. Although we have not given such an explicit definition, we believe that the study design permits us to examine the stability of life-sustaining treatment choices.

The general tendency of the group was to desire less treatment at follow-up. One possible explanation is that as the group grew older, it chose to forego life-sustaining treatments. This explanation is not tenable, however, because age did not correlate with willingness to forego treatment when comparing younger and older respondents either at the first or the second interview. Another possible explanation is that the historical events experienced by this cohort of persons influenced their attitudes. A survey of the popular press for articles using the keywords "right to die," "advance directives," "living will," "resuscitation orders," or "life support care" over the duration of the study shows an increase from 94 articles when the study began in 1986 to 266 articles in 1990 when the study ended. We speculate that public discussion may have shifted individual preferences.

Instability was substantial during the 2-year period in patients' preferences for a given treatment. Everhart and Pearlman [4] showed that among patients in an intensive care unit, 80% had stable preferences for life-sustaining treatments when asked at the end of their intensive care stay and 1 month later. Silverstein and colleagues [5] showed that patients with amyotrophic lateral sclerosis had less stable preferences for cardiopulmonary resuscitation when asked twice at a 6-month interval. When combined, these results suggest that as the interval between questioning lengthens, patient preferences become less reproducible.

The factors that were associated with a change toward more desired treatment may share a common theme. Having an accident or hospitalization, becoming more immobile or more depressed, or losing social support are all experiences that increase vulnerability. Being more vulnerable may make a person more aware of or threatened by his or her mortality. This is consistent with other research indicating that patients with life-threatening illness have strong desires for life-sustaining treatments [11].

It is interesting that patients who had a living will did not uniformly choose the least treatment when initially interviewed. Over time, however, those persons who initially chose more treatment moved toward desiring less treatment. Those with living wills who chose to forego treatments at the outset generally remained steadfast in their choices.

What are the implications of these findings for the clinician who relies on written advance directives or other forms of previously expressed wishes to guide the care of incompetent patients? This study and others show that most patients do not write advance directives [3, 14, 15]. We have shown that those without written directives had less stable preferences than those with them, although as more patients are encouraged to sign advance directives, this difference may diminish. The data also suggest that written or verbally expressed wishes are not entirely stable, particularly as a patient's status changes. Thus, it is important to talk to patients about their desires regarding life-sustaining treatments whether or not they have written directives. The Patient Self-Determination Act requires Medicare-funded organizations to inform patients at the time of admission about their right to have an advance directive. One might anticipate from our results that treatment choices elicited at the time of hospital admission are likely to be influenced by the vulnerable feelings a person may have during acute illness and hospitalization and thus might differ from choices elicited at a less traumatic time. Clinicians should be aware of this possibility and should review patients' wishes over time.

If some patients' wishes are unstable, one might argue that this undermines the validity or usefulness of the advance directive as a prediction of future wishes. However, perhaps the advance directive should be looked at as an instruction for future care rather than a prediction of future wishes. In this light it may be important to routinely ask patients how closely they wish their health care provider to adhere to their advance directives because patients have been shown to differ in how strictly they want prior directives to be followed [16].