Effectiveness of Computer-Generated Reminders for Increasing Discussions about Advance Directives and Completion of Advance Directive Forms

Methods

Setting and Patients

This study was approved by the institutional review board of Indiana University as expedited research with waiver of informed consent from both patients and physicians. It was conducted in the General Medicine Practice [31], an academic primary care practice affiliated with an urban public teaching hospital. This practice is staffed by general internal medicine faculty, fellows, and residents. Each resident and fellow attends the General Medicine Practice one half-day per week; faculty attend one to four half-days per week. Residents always practice with the same attending faculty physicians. All physicians, except for study investigators, were eligible to participate.

At the time of this study, the General Medicine Practice comprised four separate practices with separate waiting areas, clerks, and nurses. Each practice held eight half-day sessions per week. Each session was attended by two faculty members and two or three residents, each of whom provided primary care to assigned panels of patients. Residents were required to briefly discuss each patient with the attending faculty. Fellows served as faculty and were treated as such. Since 1981, a computerized program has randomly assigned new physicians to the practice sessions [31]. New patients have been sequentially assigned to open appointment slots; this resulted in no important differences in patients or clinical practice among the sessions [29, 33].

We included patients who were at risk for acute deterioration (and therefore might benefit from advance directives) because of advanced age (≤ 75 years, the typical threshold for the “oldest old”) or because they were 50 years of age or older and had one of the following chronic conditions: cardiac ischemia, heart failure, chronic lung disease, cancer other than nonmelanomatous skin cancer, cerebrovascular disease, renal insufficiency, or cirrhosis. We chose 50 years of age as a cut-off to yield sufficient numbers of patients. A computer program identified eligible patients among those with scheduled appointments at the General Medicine Practice by using problem lists and test results stored in the Regenstrief Medical Record System [34].

Eligible patients who kept appointments at the General Medicine Practice were approached by research assistants in the waiting room. The research assistants, who were blinded at all times to the patients' study groups, explained the study to the patients, invited them to participate, and interviewed those who agreed to participate; patients from nursing homes and prisons and patients who were deaf or did not speak English were excluded. The assistants then administered the Pfeiffer Mental Status Questionnaire [35]; patients whose scores indicated cognitive dysfunction were excluded. The remaining patients provided sociodemographic information and stated whether they had previously discussed or completed advance directives; patients who had completed advance directives were excluded. Patients also stated their preferences with regard to six treatments in the event of a terminal illness [36].

Advance Directives

Before the study, we created two separate forms for instruction directives and proxy directives; these forms became the official advance directive documents of the hospital and its outpatient services. The instruction directive allowed patients to indicate whether, in the event of terminal illness and mental incapacity, they wanted or did not want eight types of care: cardiopulmonary resuscitation, mechanical ventilation, surgery, invasive procedures, nutrition and hydration, transfusion of blood or blood products, antibiotics, or noninvasive diagnostic tests. The primary care physician had to sign each completed instruction directive form to indicate that he or she was aware of its contents. The proxy directive designated both primary and secondary health care representatives.

We placed both advance directive forms in a drawer of the desk of each physician in the General Medicine Practice. We also placed the forms in a bin near the door of the staff room along with other forms and requisitions and business reply envelopes for patients who wanted to complete the forms at home. Research assistants entered the data from completed forms into the Regenstrief Medical Record System, where the forms were available for viewing through computer terminals and workstations in all inpatient and outpatient venues [34].

Before the study, the three physician-investigators presented the basic concepts of advance directives at grand rounds. They also had face-to-face meetings with each physician in the General Medicine Practice and explained how to complete and process the forms. We encouraged physicians to discuss advance directives with their elderly and debilitated patients and posted flyers in each practice staffing room suggesting that physicians discuss advance directives with patients who had the target study conditions.

Study Methods

The intent of the randomization scheme was to expose physicians to the same type of reminder or reminders, or no reminders, during all of their scheduled primary care visits with enrolled patients. At the time of this study, 32 weekly half-day sessions took place on the four General Medicine Practice practices. Two sessions attended by study investigators were excluded. We randomly assigned all of the physicians who worked in a particular half-day session to the same reminder category. At the time of randomization, 16 physicians (all of whom were faculty members) practiced in more than 1 session per week (14 practiced in 2 sessions and 2 practiced in 3 sessions). Therefore, we randomly assigned the sessions in a stepped manner by first allocating the 16 physicians and all of their associated sessions to four categories: control (no reminders), computer-generated reminders for instruction directives, computer-generated reminders for proxy directives, and computer-generated reminders for both types of directives. We then randomly assigned the remaining 8 sessions and their physicians to the four categories (Table 1). Each practice contained sessions in all four categories, which were equally distributed between mornings and afternoons.

All physicians routinely received computer-generated reminders for patients with scheduled visits. They were reminded to give preventive care, note abnormal test results, and avoid drug interactions [29, 34]. These reminders appeared at the bottom of computer-generated printed encounter forms [34] (Figure 1). Physicians routinely reviewed the encounter forms and the practice chart immediately before visiting the patient. As recommended by Litzelman and coworkers [37], the advance directive reminders were followed by a choice list (discussed today, next visit, not applicable, patient too ill, patient refuses to discuss, I disagree with advance directives). Instruction directives were called “advance directives,” and proxies were called “health care representatives” (Figure 1).

After patients were enrolled, research assistants attempted to interview them in the waiting room after each scheduled appointment to assess whether they had discussed advance directives with their physicians that day. Patients who answered “yes” were defined as having had an advance directive discussion. During the first scheduled General Medicine Practice visit in the 5-month period between 11 and 16 months after enrollment, a close-out interview was attempted. If no scheduled visit had occurred by 15 months after enrollment, the close-out interview was attempted by telephone.

Form completion was defined as having occurred if either completed form was received between study enrollment and 30 days after the final interview (this made it possible to receive by mail forms that were completed after the close-out visit). The same advance directive forms were used for all of our inpatient and outpatient services. We included all completed forms regardless of where they were obtained.

Statistical Analysis

We performed all analyses by intention to treat. A two-tailed P value less than 0.05 was considered significant. We used chi-square tests to assess categorical variables and used one-way analysis of variance to assess continuous variables. To account for the expected correlations among an individual physician's patients, we applied generalized estimating equations [38] to logistic regression models that related the outcome variables to the different reminder categories and a preselected list of covariates. These covariates included sociodemographic variables, clinical eligibility conditions, age greater than 75 years, and an interaction term that included age greater than 75 years and eligibility conditions.

We controlled for severity of illness by assigning each patient to 1 of 51 Ambulatory Care Groups (groups of diagnoses) [39]. These groups were generated by using billing codes from the International Classification of Disease, 9th Revision, Clinical Modification [40] recorded during all outpatient visits in the year before study enrollment. Because our study sample consisted of older and sicker adults, patients fell into a limited number of Ambulatory Care Groups. Therefore, two investigators independently collapsed the Ambulatory Care Groups into 6 groups that each contained 8% to 39% of enrolled patients with similar severity of illness; the investigators resolved minor differences between their lists by discussion.

We also controlled for patients' end-of-life treatment preferences, recorded during the enrollment interview, by assigning a score of 1 (“want it”), −1 (“don't want it”), or 0 (“don't know”) to the patient's desire for hospitalization, intensive care, resuscitation, surgery, mechanical ventilation, and nutrition and hydration if he or she was terminally ill [36, 41]. Thus, scores could range from −6 to +6 [36]. Because controlling for Ambulatory Care Group and patient sex, ethnicity, and marital status did not affect the results, the analyses presented do not include these variables.

The study period overlapped the start of an academic year, causing primary caregivers to change for some study patients. We constructed models that used all distinct patient–physician pairs as units of analysis. We repeated the analyses using only data associated with the patient's first physician and obtained identical results. Therefore, we report data from all distinct patient–physician pairs.

Results

A total of 1394 potentially eligible patients kept a scheduled visit in the 9-month enrollment period. Forty-two patients (3%) were missed by the research assistants, and 162 (12%) were found to be ineligible: Eighty-three of the 162 patients were excluded because of their scores on the Pfeiffer Mental Status Questionnaire, 36 were in nursing homes or prisons, 29 were deaf or spoke no English, 9 had previously completed an advance directive form, and 5 were excluded for other reasons. Of the 1190 eligible patients interviewed, 1042 (88%) were enrolled and completed the baseline interview. Patients who were missed or refused to participate were significantly older and had a higher prevalence of cancer.

Ten physicians who changed sessions with the arrival of a new academic year also changed study reminder categories. One physician with multiple sessions was in three reminder categories. For our analysis, we excluded the 33 study patients cared for by these 10 physicians after they changed practice sessions. Results were similar and our conclusions did not change when these 33 patients were included. The remaining 1009 study patients represented 1160 unique patient–physician pairs: During the study, 145 patients had 2 different primary care physicians and 3 patients had 3 different primary care physicians (Table 1). Of the 1009 patients in the final analysis sample, 15 changed during the study to practice sessions with study status that differed from their original sessions. Because these patients had not yet completed an advance directive form, we included their new patient–physician pairs as separate observations in our intention-to-treat analysis. We also included 101 patient–physician pairs (9%) in which the patients kept no scheduled visits after enrollment and 94 patient–physician pairs (8%) in which postvisit questionnaires were not completed because patients refused or were missed by the research assistants. Table 1 shows physicians, patients, and visits in the four reminder categories.

Demographic and clinical characteristics of the enrolled patients are shown in Table 2. The mean age was 65 years, two thirds were women, and slightly more than half were African American. Only 3% had discussed advance directives with their physicians before enrollment. Of the various baseline characteristics, being African American, having cardiac ischemia, and having previously thought about advance directives differed significantly among the four study groups. Therefore, these three factors were included in all analyses. Although 17% of the patients were 75 years of age or older, only 5% were eligible solely because of their age.

After enrollment, 83% of the study patients were interviewed during at least one scheduled visit to the General Medicine Practice during the study period (mean number of visits ±SD, 2.6 ± 1.7 during 14.4 ± 3.6 months in the study). The number of patients, number of scheduled visits, and length of time in the study did not differ among study groups. Twenty-eight percent of patients were hospitalized during the study.

Although the unit of intervention was the physician, the analysis was performed at the level of the unique patient–physician pairs. We used generalized estimating equations to account for the correlation of patient outcomes within physicians [38]. All patients with no contacts were assumed to have had no advance directive discussions. Patients in all three reminder groups reported more advance directive discussions than did patients in the control group (Table 3). The odds of having advance directive discussions was eight times higher among patients whose physicians received reminders for both types of directives than among patients whose physicians received no reminders. Although we found no significant interaction between the two types of reminders, patients whose physicians received reminders about both types of directives reported significantly more discussions than did patients whose physicians only received reminders for proxy directives. When discussions did occur, they were initiated by the physician in 86% of instances in the group that received reminders for both types of directives, in 77% of instances in the instruction directive reminder group, in 43% of instances in the proxy directive reminder group, and in 38% of instances in the control group.

Patients whose physicians received reminders for both types of directives completed substantially more forms than did control patients and patients whose physicians received only proxy directive reminders (Table 3). Patients whose physicians received only reminders for instruction directives also completed more forms than did control patients. Seventy-eight of the 1009 study patients completed at least one of the two types of advance directive forms during the study period: Forty-one patients completed both forms, 34 completed an instruction directive only, and 4 completed a proxy directive only. Four patients whose physicians received both types of reminders returned completed forms within a week after the close-out interview; we included these forms in the analyses.

Multivariable analysis found that receiving both reminders was the strongest predictor of advance directive discussions and receiving only instruction directive reminders was the second strongest. Of all other covariates considered, only age 75 years or greater was significantly correlated with advance directive discussions (odds ratio [OR], 2.3 [95% CI, 1.5 to 3.6]). Similarly, receiving both reminders was the strongest predictor of completing an advance directive form, followed by receiving instruction directive reminders only. The only additional covariates that were significantly associated with form completion were age 75 years or greater (OR, 3.0 [CI, 1.7 to 5.1]), presence of chronic lung disease (OR, 1.7 [CI, 1.1 to 2.7]), and treatment preference scores of 2 or greater at the time of enrollment (OR, 0.4 [CI, 0.3 to 0.8]). Overall, 45% of advance directive discussions were associated with completion of an advance directive form by the patient. When a variable indicating the occurrence of an advance directive discussion was introduced into this analysis, it was highly predictive of subsequent form completion (adjusted OR for discussion, 44 [CI, 25 to 78]).

Discussion

In this randomized, controlled trial, simple computer-generated reminders given to primary care physicians during scheduled outpatient visits increased the number of advance directive discussions, almost half of which were followed by the patient completing an advance directive form. The importance of such discussions was affirmed by the investigators of the Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatments (SUPPORT) [42], who found that 36% of patients who had discussed end-of-life care with their physicians before hospitalization had completed a form compared with 20% of patients who had not had such discussions. Our intervention-unobtrusive printed reminders-was effective, whereas the intensive and expensive SUPPORT intervention did not increase the number of completed advance directives (40% of patients who received the intervention and 37% for control completed advance directives) [43]. The key, we believe, is that our intervention was delivered during regularly scheduled primary care visits, when patients are usually not acutely ill and are visiting a physician whom they know and probably trust. In SUPPORT, the intervention was delivered in the intensive care unit, where the term advance directive is, essentially, a misnomer. Our computer reminders were also effective when the Patient Self-Determination Act [11, 14, 15, 17] and intensive patient education [20-24] had little or no effect. We found it heartening that a highly technical aspect of modern medicine-reminders from an electronic medical record system-enhanced the oldest and least technical aspect of care: the patient–physician relationship.

The direct effect of our reminders is evidenced by the fact that physicians who received reminders not only had more discussions but initiated them more frequently than did control physicians. Other researchers [1, 44] have shown that although both patients and physicians believe that discussions about advance directives are important, each believes that the other should initiate them. Our reminders broke this impasse and motivated physicians to take the first step; this confirms the suggestion of Miles and colleagues [28] that clinician counseling most effectively catalyzes completion of advance directives.

Our study also produced two secondary findings. First, reminders to discuss proxy directives had small to negligible effects, whereas reminders about instruction directives were routinely effective. Sending both types of reminders was the most effective strategy. Although it is understandable that physicians who receive two sets of reminders would comply more frequently, it is not clear why physicians preferred instruction directives over proxy directives. Perhaps physicians or patients prefer to discuss specific aspects of care or perhaps barriers to completing proxy directives were encountered, such as not knowing the potential proxy's address and telephone number or not having that person's consent to be listed as a proxy. Second, in our sample of elderly and seriously ill outpatients, only advanced age was associated in multivariable analysis with both more discussions about advance directives and completion of forms. This is consistent with other researchers' findings [45-47] that completed advance directives and do-not-resuscitate orders were more prevalent among patients of advanced age. Of note, after we controlled for demographic and clinical data, higher treatment preference score (that is, wanting more care at the end of life) was associated with fewer completed advance directive forms. This finding is consistent with previous work [28] and suggests that advance directives are used predominantly to limit rather than ensure care at the end of life.

One requirement for implementing computer-generated reminders is the presence of a computerized medical record system. The promise of these systems for improving the quality of care while controlling costs [48] is sufficiently great that the Institute of Medicine has considered them an essential technology for health care [49]. Although a computerized medical record system is not necessary for attaching advance directive reminder notices to patient charts, targeting selected patients and avoiding reminders for patients who already have completed forms would be very time intensive and prone to error. This is the type of work computers do best [50]. Perhaps the greatest advantage of electronic medical records for advance directives is that completed forms can be stored so that they are accessible to providers in all inpatient and outpatient venues rather than being buried in outpatient charts or physicians' offices. This fact may have augmented the compliance of our physicians with the advance directive reminders.

The long history of using computer reminders in our practice [29-31] may also have augmented physician compliance. A recently completed negative study of computer-based inpatient preventive care reminders [51] suggests, however, that physicians can and do discriminate between reminders that they do and do not want to follow. This study used a state-of-the-art medical information system with capabilities that are not currently available to most health care providers. However, similar reminder systems are basic components of almost all office clinical information systems currently being marketed [52] and are a major factor supporting the Institute of Medicine's contention that electronic records are essential to medical practice [49]. We believe that systems with reminder capabilities will soon be widely available [50].

We limited our study to outpatients who were at least 75 years of age or were at least 50 years of age and had serious illnesses. The rates of advance directive discussion and form completion would probably be lower among younger, healthier patients. Our study was also conducted in an inner-city practice with many poor and minority patients, which may explain the low rate of advance directive discussions among control patients. Other researchers [43, 53] have found that minority patients have fewer advance directive discussions with their providers than do patients with less education and lower socioeconomic status [28]. Advance directive reminders helped our physicians overcome these barriers. We hope that increasing the number of advance directive discussions between patients and their primary care physicians may reduce the distressing number of end-of-life decisions that are made without patient involvement [54].

Although establishing advance directives does not overcome all of the difficulties involved in making end-of-life decisions [28, 55], advance care planning can generate important, memorable, and therapeutic discussions among patients, providers, and family members [28]. However, many questions remain. Did physician and patient communicate effectively about end-of-life care? Will either the discussion about advance care planning or the completion of an advance directive actually affect patient care? What types of written directives are most useful? Future research needs to show that enhanced communication about advance care planning not only extends the autonomy of the individual patient but also improves patient care at the end of life.

From Indiana University School of Medicine, The Regenstrief Institute for Health Care, and Richard L. Roudebush Veterans Affairs Medical Center, Indianapolis, Indiana; and the School of Public Health, St. Louis University, St. Louis, Missouri.

Dr. Wolinsky: 3663 Lindell Boulevard, St. Louis, MO 63108-3342.

Dr. Gramelspacher: Program in Medical Ethics, Wishard Memorial Hospital (M200), 1001 West 10th Street, Indianapolis, IN 46262.

Mr. Eckert: Division of Biostatistics, RR 135, 699 West Drive, Indianapolis, IN 46202-5119.