The introduction of psychiatric advance directives (PADs) is potentially one of the most significant developments in U.S. mental health law and policy in recent years. Twenty-one states in the past decade have enacted PAD statutes, authorizing mental health advance instructional documents or proxy decisionmakers, thus providing a legal means for competent individuals to refuse or consent to future mental health treatment during periods of decisional incapacity (Swanson, Swartz, Ferron, Elbogen, & Van Dorn, 2006).

Advocates for PADs anticipated that these documents would give persons with severe mental illness greater autonomy and control over their own lives—perhaps indirectly by improving continuity of care and working alliance with mental health professionals and thus decreasing the incidence of involuntary treatment (Appelbaum, 1991; Backlar, 1995; Swanson, Tepper, Backlar, & Swartz, 2000; Winick, 1996). However, the expected widespread benefits of PADs have yet to materialize, due to systemic barriers preventing their use, lack of resources deployed to assist patients in preparing PADs, and lack of “buy-in” and acceptance of PADs by clinicians (Amering, Stastny, & Hopper, 2005; Atkinson, Garner, & Gilmour, 2004; Backlar, McFarland, Swanson, & Mahler, 2001; Miller, 1998; Srebnik & Brodoff, 2003; Swanson et al., 2003; Van Dorn et al., in press).

The problem with PADs is not lack of interest among consumers. A recent study in five cities finds a large latent demand for these instruments. PADs have been completed by 4–13% of mental health consumers sampled from public-sector outpatient treatment, yet large majorities—66–77%—of these consumers indicate that they would like to complete PADs, if given the opportunity and necessary assistance (Swanson et al., 2006). Recent research also finds that, for the most part, clinicians and family members of persons with severe mental illness endorse PADs in concept (Swanson et al., 2003), though they tend to be concerned about certain aspects of PADs and how they will work in practice (Elbogen et al., 2006).

Research on PADs has lately begun to focus on operational barriers preventing PADs from realizing their potential benefits (Elbogen et al., 2006; Srebnik & Brodoff, 2003; Swartz et al., 2005; Van Dorn et al., in press). In that vein, we address here the issue of psychiatrists' decisions to override or follow patient's wishes as documented in PADs—particularly PADs that include advance refusals of hospitalization and medication during a crisis.

While PADs ostensibly are designed to support patients' own decisions about their treatment, PAD statutes in effect give clinicians broad discretion over whether to carry out patients' instructions. Moreover, PADs in every jurisdiction are trumped by civil commitment law. Thus, the patient's decisionmaking via PADs is nested in the (much stronger) apparatus of clinical decisionmaking and involuntary treatment. The degree to which psychiatrists are inclined to override PADs in practice may limit these instruments' effectiveness and diminish their appeal to patients.

How prevalent is the view among psychiatrists that PAD refusals of treatment should be overridden? To what extent are psychiatrists' inclinations to override PADs shaped by their exposure to psychiatric emergencies, perceived legal liability, attitudes about patients' insight and potential violence, and their views about quality of care, access, and the use of coercion in community mental health services? This paper addresses these questions with data from a new survey of N=164 psychiatrists in North Carolina. Specifically, we analyze psychiatrists' responses to a survey vignette about whether they would decide to override a PAD refusal of hospitalization and medication in a hypothetical case involving a psychotic patient in a psychiatric emergency setting. We develop a statistical model of practice characteristics and attitudes associated with this hypothetical decision to override a PAD.

Clinician override provisions in state pad statutes

The 21 states with mental health advance directive statutes all include provisions allowing involuntary civil commitment to override, at least partially, the instructions contained in PADs. Specifically, override is allowed in situations where a patient's PAD instructions conflict with the state's authority to admit and retain the patient in a mental health treatment facility and the patient currently meets the criteria for involuntary admission or commitment. For example, the Pennsylvania statute, enacted in 2004, states that the mental health advance directive shall not be construed to affect the authority to “admit a person to a mental health facility under the voluntary and involuntary commitment provisions of … the Mental Health Procedures Act” (Title 20 of the Pennsylvania Consolidated Statutes, 2004). Similarly, as rendered in the Texas statute (1999; amended):

[The] Declaration for Mental Health Treatment does not limit any authority provided by Chapter 573 or 574, Health and Safety Code: (1) to take a person into custody; or (2) to admit or retain a person in a mental health treatment facility.

These provisions specifying the priority of state civil commitment law over PADs are typical of other states' PAD statutes.

State statutes also typically provide broad legal immunity to a clinician who declines to follow PAD treatment instructions, as long as the clinician acts in good faith and according to his or her understanding of clinical standards of care. In the Pennsylvania statute, for example, a qualified mental health professional who:

acts in good faith and consistent with this [statute] may not be subject to criminal or civil liability, discipline for unprofessional conduct or administrative sanctions and may not be found to have committed an act of unprofessional conduct by any professional board or administrative body with such authority as a result of any of the following: … [including] Refusing to comply with a mental health declaration or mental health power of attorney which violates accepted clinical standards or medical standards of care (Title 20 of the Pennsylvania Consolidated Statutes, 2004).

In North Carolina, where the present study was conducted, the Advance Instruction for Mental Health Treatment statute (Statutory Form for Advance Instruction for Mental Health Treatment in North Carolina, 1998) lists five conditions under which an attending physician or other mental health treatment provider is not required to comply with the advance instruction:

  1. 1.

    Compliance, in the opinion of the attending physician or other mental health treatment provider, is not consistent with generally accepted community practice standards of treatment to benefit the principal;

  2. 2.

    Compliance is not consistent with the availability of treatments requested;

  3. 3.

    Compliance is not consistent with applicable law;

  4. 4.

    The principal is committed to a 24-hr facility pursuant to Article 5 of Chapter 122C of the General Statutes, and treatment is authorized in compliance with G.S. 122C-57 and rules adopted pursuant to it; or

  5. 5.

    Compliance, in the opinion of the attending physician or other mental health treatment provider, is not consistent with appropriate treatment in case of an emergency endangering life or health.

In sum, state laws tend to give clinicians broad discretion to override PADs whenever patients' advance instructions conflict with clinical standards of care or the exigencies of emergency treatment. However, the override features of these laws remain controversial and could be vulnerable to legal challenge. A recent U.S. Court of Appeals decision ((Hargrave v. Vermont, 2003); United States Court of Appeals for the Second Circuit) struck down a state law that included a similar exception allowing clinicians to override a patient's advance refusal of psychotropic medications through a general healthcare proxy (Allen, 2004; Appelbaum, 2004; Keefe & Pinals, 2004). Specifically, the court ruled that the Vermont override law—which applied only to persons with psychiatric disorders—was discriminatory on the basis of disability and thus violated the Americans with Disabilities Act, Title 3.

How do psychiatrists view this controversial issue, and how would they apply their discretion to override PADs in practice? This study examines the prevalence of North Carolina psychiatrists' override responses to a PAD treatment–refusal vignette, and statistically compares demographic, attitudinal, and practice characteristics of clinicians who oppose versus favor overriding treatment refusals in PADs.

Study design and sample characteristics

A sample of N=164 psychiatrists practicing in North Carolina was surveyed by mail with a structured self-administered questionnaire. Information about the psychiatrists was obtained through their professional organization membership roster. Clinicians' names, work addresses, professional setting, age, race, and gender were included in these lists. These lists were then used to draw a random sample of participants, using a computer spreadsheet for randomization. The questionnaire, the Clinician Attitude Survey, was mailed to 500 randomly-selected psychiatrists. We used several techniques shown to increase survey response rates for physicians and other elite groups: a personalized cover-letter with the correct salutation, first-class mailing with return postage, a financial incentive/honorarium, and a follow-up reminder letter (Kasprzyk, Montano, St. Lawrence, & Phillips, 2001; Olmsted, McFarlane, Murphy, & Hill, 2005). These efforts yielded a response rate of 32%—somewhat lower than average for physician mail surveys in the US (54%; Asch, Jedrziewski, & Christakis, 1997) but in the middle of the 10–75% range reported in the literature for psychiatrist surveys (Addington et al., 2002; Blanco, Carvalho, Olfson, Finnerty, & Pincus, 1999; Colenda, Pincus, Tanielian, Zarin, & Marcus, 1999). The length of the questionnaire—estimated to take 30–45 min to complete—may have been a factor in the response rate. Nevertheless, responders did not differ significantly from nonresponders with respect to age, gender and practice setting. All participants received a $50 gift certificate for completing the survey. The study was conducted according to ethical principles of research. The study was determined to be exempt from human subjects research review by the Duke University Medical Center IRB.

The survey was mailed with a two-page introductory letter and a PAD “Toolkit,” which included basic background information about PADs, blank forms and instructions on how to complete them, and an explanation of the relevant North Carolina PAD statutes—the advance instruction for mental health treatment (G.S. 122C-71 through 77) and health care power of attorney (G.S. 32A-15 through 25). The letter also described how, and in what kinds of situations, psychiatrists are legally permitted to override written instructions or requests from a health care agent or from the advance instruction document. Specifically, the letter stated that advance instructions can be overridden when the requests conflict with “generally accepted community practice standards,” when requests for treatments are infeasible or unavailable, or when any emergency treatment is needed, including the need for involuntary commitment.

Measures

Dependent variable

Advance directive override. Participants were asked to respond to a scenario involving a patient's use of a psychiatric advance instruction for refusal of hospitalization and medication. Specifically, the scenario read:

Family members of an individual with serious mental illness bring their relative to the emergency department and request that he be admitted to the hospital. The patient has psychotic symptoms and impaired decision-making capacity, but is not dangerous to others or overtly to himself, although he could, by statute, be involuntarily committed for his grossly bizarre behavior. You think he can benefit from hospitalization and is insured for hospitalization. Previously, while competent, this patient completed an advance directive refusing hospital admissions and treatment with antipsychotic medications.

Respondents where then asked, “In this situation, would you probably try to follow the advance directive and not hospitalize this patient involuntarily, even if you personally thought inpatient treatment would be in the patient's best interest?” (0=not override [reference]/1=override).

Independent variables

Clinician demographics. Demographic variables included the participants' age, which was divided into quartiles (1=30–44; 2=45–52; 3=53–59; 4=60–88), gender (reference=female), and race (reference=nonwhite).

Practice characteristics. Three variables were used to capture clinicians' practice characteristics. First, we compared psychiatrists who worked mostly with privately-insured patients versus those who spent at least 20% of their time in clinical settings serving public-sector patients, i.e., generally uninsured and publicly insured patients. Next, we compared those physicians whose caseload consisted of fewer than 10% patients with diagnoses of psychotic disorder to those with a caseload containing more than 10% of patients with a psychotic disorder. Finally, those physicians who did not work in an emergency room or crisis center were compared to those who did.

Concerns about patient decision making. Survey participants rated the importance of five patient characteristics as factors they might consider when making a “decision to follow a patient's preferred choice of treatment.” The patient characteristics were: (1) current cognitive functioning; (2) current insight into illness; (3) history of psychosis; (4) history of suicide attempt(s); and (5) history of violence. The response categories were: 1=not important; 2=somewhat important; 3=important; 4=among the most important. For the analysis presented here, the scale was recoded into a dichotomous variable with 4 (“among the most important”) compared to all other responses.

Legal defensiveness. Participants were asked, “When you consider a decision to start or change a course of treatment for a patient with serious mental illness, how often do you worry about being sued for malpractice?” This item was also recoded into a dichotomous variable, with “often,” “very often,” or “always” compared to the other responses.

Clinician's perception of quality of services for SMI patients in the community. Participants were asked how strongly they agreed with the following statements: (1) “I think the services that are offered for patients with severe mental illness in this community are generally good,” and (2) “I think the services available for patients with severe mental illness who have a lot of trouble complying with treatment are generally good.” Respondents with average responses across both of these items in the range of “agree” to “agree strongly” were coded positive on this indicator.

Clinician's attitude regarding the need for coercion given high-quality mental health services in the community. Participants were asked how strongly they agreed with the following statement: “If my seriously mentally ill patients had access to high quality community mental health services, they would rarely require involuntary hospitalization, outpatient commitment, or similar coercive treatment.” While the statement could be open to various interpretations, it is meant to capture the view that coercion is not an intrinsically necessary element of community mental health care, but, rather, is a sort of adverse side effect of an inadequate or inaccessible service system; that involuntary treatment is, thus, primarily a system problem, not a patient problem per se. Some psychiatrists agree with this view; others disagree with it. Those who answered “agree” or “agree strongly” were coded as having low support for coercion, i.e., as a routine, widespread practice in a high-quality mental health services delivery system.

Results

Analysis methods

Simple descriptive analyses and multivariate logistic regression analyses were used to examine the relative significance of potential factors associated with psychiatrists' willingness to override a patient's PAD. Independent variables were regressed on whether or not psychiatrists were willing to follow a patient's expressed preferences refusing psychiatric hospital admission and medication. Variable reduction was accomplished using stepwise selection with a 0.20 probability inclusion level. Variables thus selected comprised a main-effects model of characteristics associated with PAD override. Odds ratios estimated by this technique estimate the average change in the odds of a predicted outcome (e.g., likelihood of overriding PAD) associated with exposure to independent variables (e.g., legal defensiveness). The log likelihood chi-square tests the overall significance of a specified logistic regression model.

Finally, a model testing relevant interaction effects was specified. Two-way interactions between relevant dichotomous predictors were tested using dummy variables with reference-class coding (Hosmer & Lemeshow, 2000). Specifically, the method involved coding four mutually-exclusive dummy variables defined by the four combinations of two main effects, X1 and X2: (a) X1 only versus other, (b) X2 only versus other, (c) both X1 and X2 versus other, and (d) neither X1 nor X2 versus other. The last group, the “neither X1 nor X2” group, was omitted from the logistic regression model as a reference category and the odds ratios were calculated for the other three groups versus the comparison group. The relative size of the interaction effect may be expressed by the ratio of the odds in the “both X1 and X2” group relative to the odds for the “X1 only” and “X2 only” groups. An interaction shows that the predictive strength of one variable differs as a function of another variable. When dummy coding is used, the most straightforward occurrence of an interaction may be seen when the odds ratio for the group with both conditions combined is significantly different than 1.0, while the odds ratios for the groups with only one of the two conditions alone are not significantly different than 1.0.

Sample description

One hundred and sixty-seven participants responded after random selection from the roster of psychiatrists registered to practice in North Carolina; of these, 164 provided a valid response to the index PAD override question. The mean age of the sample was 52 years with a range from 33 to 88. Most of the sample was male (71%), and white (87%). Almost half of the sample (44%) reported working in public sector practice settings more than 20% of the time; 21% worked in an emergency room or crisis center; and 66% of respondents' caseloads included at least 10% of patients with a psychotic diagnosis.

Regarding the factors influencing psychiatrists' decisions to follow patients' preferences in treatment, a majority of respondents selected safety factors—history of violence (55%) and history of suicide (51%)—as “among the most important” determinants. A smaller, but still substantial proportion of the psychiatrists selected competence-related factors—cognitive functioning (43%), insight into illness (36%), and history of psychosis (29%)—as “among the most important.”

About one quarter (24%) of participants indicated a high degree of legal defensiveness associated with clinical decisions to initiate or stop a course of treatment for a patient with severe mental illness.

Considering the coercion question, 48% of respondents endorsed the view that involuntary treatment would rarely be necessary if quality services in the community were available and accessible to persons with severe mental illness. A similar proportion (48%) of participants indicated that the mental health services currently available in their community for persons with SMI and problems with adherence were “generally good.”

Table 1 displays the sample characteristics and percent of respondents with each characteristic who indicated that they would override the PAD refusal in the survey vignette. Overall, 47% of the psychiatrists surveyed indicated that they would override the PAD refusal as described in the case vignette; 53% indicated that they would not override the PAD.

Table 1 Sample characteristics and percent indicating that they would override a PAD refusal of treatment
Full size table

Table 2 displays the results of multivariable analyses of effects on PAD override. The main-effects model in the second column shows that participants were significantly more likely to report that they would override the PAD if they: (1) worked in a hospital emergency department or crisis center (OR=3.07, p < .05); (2) considered patients' insight to be among the most important factors to consider in decisions about whether to support a patient's own preferences for treatment (OR=2.04, p < .05); and (3) were highly legally defensive (OR=2.35, p < .05).

Table 2 Bivariate and multivariable models of characteristics associated with PAD override for psychiatrists
Full size table

Notably, the bivariate main effect of violence concern on the likelihood of override was rendered statistically nonsignificant in the multivariable main-effects model. We tested each of the significant covariates as potential mediators (Baron & Kenny, 1986), finding that legal defensiveness was the only variable meeting the conditions for a mediator. Specifically: (1) the effect was significantly related both to the putative indirect predictor (violence concern) and the endogenous dependent variable (PAD override response); (2) when the putative mediator (legal defensiveness) was entered in a two-predictor model with the putative indirect predictor (violence concern), the former was significant OR=2.15, p=.05) while the latter was rendered nonsignificant (OR=1.76, ns); and (3) the model with the mediator fit the data significantly better than the nested model without the mediator. Thus, we concluded that legal defensiveness mediated the direct effect of violence concern on likelihood of overriding the treatment refusal PAD.

Fig. 1
figure 1

Predicted probability of PAD refusal override, by psychiatrists' legal defensiveness and perceived quality of services for SMI in community.

Note. Estimates derived from the logistic regression interaction model shown in Table 2, for subgroup with low support of coercion

Full size image

Finally, the interaction model shows that PAD override was significantly more likely among psychiatrists who were concerned both about patient insight and violence (OR=3.37, p < .05); those who were concerned only about violence but not insight, or only about insight but not violence, were not significantly more likely to override the PAD refusal. Also, PAD override was significantly less likely (OR=0.28, p < .05) among participants who indicated both that (1) coercion would be largely unnecessary given high-quality accessible services in the community, and that (2) such services did in fact exist in their own community. These interaction effects were significant controlling for the other significant main effects—working in a hospital emergency department or crisis center (OR=2.94, p < .05) and legal defensiveness (OR=2.48, p < .05).

Figure 1 displays graphically the model's predicted probabilities of PAD override for two interacting variables—legal defensiveness and perceived quality of community care for SMI—specifically for the subgroup of psychiatrists who view the need for involuntary treatment as largely a function of the lack of accessible community-based services for persons with SMI. As the graph illustrates, in this group of psychiatrists, the probability of PAD override varied from 70% in legally-defensive respondents who believed that community services for SMI patients were poor, to 26% among their counterparts who were not legally defensive and who perceived community services for SMI patients to be “generally good.”

Discussion

The dilemmas that arise in implementing PADs are defined by a tension between the principle of respect for individual autonomy in healthcare decisions and the public-health imperative and social responsibility to provide appropriate care to persons with SMI. In a statewide survey of 164 psychiatrists in North Carolina, we find evidence of sharply divided opinion on the important question of whether mental health professionals should follow or override PADs that refuse treatment. Specifically, in response to a hypothetical vignette, about half (47%) of psychiatrists surveyed indicated that they would override a valid, competently-executed PAD that refused hospitalization and medication in the case of a nonviolent psychotic patient in a hospital emergency department. In effect, these clinicians reported that if they were faced with such a situation in their practice, they would use the discretion the state's PAD law allows and would admit the patient involuntarily. The other 53% of psychiatrists surveyed indicated that they would follow the patient's documented advance wishes, and would not admit the patient.

On the one hand, it may not be surprising that nearly half of the surveyed psychiatrists would override a PAD that refuses hospitalization and medication for such a patient. In the view of many clinicians, a reasonable risk-benefit calculation would not favor a decision to release an acutely psychotic patient without treatment, notwithstanding the patient's own competent advance instructions. Releasing the patient would seem to require quite strong support for the value of patient autonomy over beneficent parentalism, in a situation where these principles conflict. Such a robust endorsement of patient autonomy is surely not universal among physicians and may be the exception rather than the rule. Moreover, some clinicians might believe that releasing the patient would violate accepted clinical standards of care. In such cases the North Carolina PAD statute provides legal immunity for the psychiatrist who, in good faith, declines to follow the PAD. Indeed, the more surprising finding might be that 53% of psychiatrists indicated that they would follow the PAD instructions refusing hospitalization and medication.

On the other hand, it is striking to realize that a competently-prepared legal PAD—prepared with the intent of avoiding conventional psychiatric intervention during a future mental health crisis—would be overridden by nearly half of psychiatrists. Clearly, this is a controversial issue.

The experiential and attitudinal differences between these two groups of psychiatrists may contain clues about some of the daunting clinical, ethical, and practical dilemmas involved in implementing PADs for persons with severe mental illness. It is informative that the psychiatrists who were actually working in hospital emergency departments and crisis centers were more than twice as likely to say they would override the PAD refusal (OR=2.26, p < .05), compared to their counterparts who did not work in these kinds of emergency settings on a regular basis. Emergency psychiatrists are, after all, the clinicians who presumably have the most experience with difficult-to-treat patients and the kinds of decisions represented in the survey vignette; they also may have greater exposure to the (real or perceived) risks involved in these decisions.

In that vein, two other findings from the study seem to fit together: First, psychiatrists who were most concerned about patients' violence risk and lack of insight into illness—i.e., ranked these “among the most important” factors limiting clinicians' support of patients' own choices—were significantly more likely to say they would override a PAD refusal of treatment (OR=3.37, p < .05). Neither lack of insight alone, nor violence risk alone, were significant factors in the override response; rather it was the combination of these two factors that seemed to tip the balance toward overriding the treatment refusal PAD. Second, psychiatrists who worried most about being sued for any adverse outcomes of their treatment decisions were significantly more likely to override the PAD (OR=2.48, p < .05). Finally, psychiatrists most inclined to override PAD refusals of treatment were more resigned to the necessity of involuntary treatment and had a more negative opinion about the state of community-based services that are available to patients with SMI—particularly those patients who are reluctant or unable to adhere to prescribed treatment in the community.

In contrast, the profile of a psychiatrist who would honor a PAD refusal of hospitalization for a psychotic patient is that of a clinician who is: (1) less likely to actually work in a hospital emergency department or crisis center; (2) less concerned about patients' lack of insight into their illness or the potential for patient violence; (3) less legally defensive; and (4) less inclined to see coercion as intrinsically necessary given high-quality services, while being more sanguine about community-based alternatives to hospitalization for such patients.

One limitation of this study is that the survey vignette did not specify many of the clinical details and history that a psychiatrist would ideally wish to consider in making a decision about hospitalizing a patient involuntarily. It is possible, on the one hand, that the respondents mentally supplied the missing details in the case—imagining real patients they had seen resembling the vignette patient. In effect, then, different participants may have been imagining and responding to “different cases.” For example, the vignette description specifies that the patient is not dangerous and provides virtually no information about the patient's baseline cognitive capacities and insight into illness; nevertheless, some of the psychiatrists probably imagined this patient as someone with poor insight and potentially violent. Otherwise, it would be hard to explain the significant association between the PAD override response here and general concern about these issues measured elsewhere in the survey.

On the other hand, a particular strength of using a vignette without much detailed clinical information is that this does, in fact, resemble many psychiatric emergency cases in which clinicians are asked to make decisions about patients they do not know, and about whom they have very limited information. (Indeed, this very problem is often described as one of the reasons why PADs could be useful.)

Another limitation of this study is that the findings are based on responses to a hypothetical vignette, which may not correspond to psychiatrists' real decisions in practice. However, given that so few patients to date actually have PADs, a vignette study is perhaps the only feasible way—at present—to gauge psychiatrists' attitudes about this emerging phenomenon.

Also, the 32% response rate could have limited generalizability to all psychiatrists in the state—a limitation common to virtually all mailed-survey studies of physicians and other elite populations. However, we were able to determine that responders and nonresponders did not significantly differ with respect to age, race, gender, and a final limitation is that a cross-sectional survey such as this cannot establish causation between associated variables; we cannot conclude, for example, that psychiatrists' legal defensiveness causes a greater inclination to override PADs—only that these two attitudes are significantly related.

Our major finding, in a nutshell, is that psychiatrists as a group are divided about honoring PADs that include treatment refusals in serious mental health crises. There are probably sound reasons for this, including psychiatrists' genuine concern about their patients' and others' wellbeing, uncertainty in discerning their patients' authentic wishes, and legitimate considerations relating to the psychiatrists' professional integrity. However, there are probably also many psychiatrists who would override PADs for inappropriate reasons—such as lack of knowledge about clinicians' legal obligations and patients' rights with respect to PADs, or a misplaced legal defensiveness, an exaggerated perception of patients' violence risk, or merely a parentalistic skepticism regarding the capacities of mental health consumers to make their own decisions about treatment.

We speculate that some psychiatrists may even act on a motivation similar to that of the surgeon who ignores a DNR and resuscitates a terminally-ill patient, when a medical error results in the patient's cardiac arrest during surgery (Casarett & Ross, 1997; Casarett, Stocking, & Siegler, 1999). The mental-health equivalent of the “iatrogenic override” scenario might be a patient who becomes alienated from the treatment system because of adverse experiences with older antipsychotic medications, stops all treatment, receives no follow-up or outreach in the community—then has a serious relapse and presents to hospital emergency department with a PAD refusing hospitalization and medication. Here the hospital clinician may feel obligated to override the PAD and to provide optimal treatment to the patient, as if to compensate for the past failure of treatment, believing that the poor outcome during the previous course of illness was primarily a “system problem,” not a “patient problem.”

What are the implications of these findings for the challenges of successfully implementing PADs in the real world? Given the complexities and resistance underlying many psychiatrists' responses to PADs, the question arises: How can these documents actually work in ways that will support patients' self-determination and autonomy while also assisting mental health professionals to provide more effective, informed, acceptable, and appropriate treatment to persons with severe illness—especially those who may otherwise be difficult for clinicians to engage?

Perhaps the most important lesson is that outreach and education about PADs should ideally be focused on clinicians as well as on patients, with an understanding of the divergent and common interests of both. Mental health professionals could benefit from receiving systematic information about the legal and ethical underpinnings of PADs, as well as consultation regarding when to invoke the exception to the rule that PADs must be followed. On the patient side, efforts to facilitate PAD completion should help patients to use these documents as authentic expressions of their treatment preferences, but also as vehicles of communication with clinicians. Ideally, PADs should enhance therapeutic relationships, which are a key to achieving the patient's goals of recovery and greater autonomy.

The long-term success of PADs may depend on stakeholders with some inherently different interests coming together on whatever common ground a PAD may provide—not only in that rare moment of an incapacitating crisis when a PAD legally goes into effect, but perhaps more importantly at the time the PAD is originally prepared. To the extent that PADs are seen primarily as adversarial documents designed to protect patients from doctors, they are probably doomed to fail. If PADs are based on mutual understanding and collaboration between clinicians and patients, they may enhance patient autonomy while improving treatment effectiveness over time.