The Role of Guidelines and Algorithms for Psychopharmacology in 2007

April 1, 2007

Recent issues of Psychiatric Timeshad articles focusing on psychiatricpractice guidelines and algorithms. Dr Michael Fauman examinedthe extent to which they are used,how they are used, and studies that havevalidated their usefulness comparedwith usual care.

Recent issues of Psychiatric Times had articles focusing on psychiatric practice guidelines and algorithms.1-4 Dr Michael Fauman examined the extent to which they are used, how they are used, and studies that have validated their usefulness compared with usual care. This article focuses more on why psychopharmacology guidelines and algorithms are not followed and proposes 7 clinical scenarios in which the recommendations should be followed more often than they are. Major guideline and algorithm projects are summarized in Table 1.

Standardized care driven by evidence-supported algorithms is a model that has attracted the attention of the hospital business community.5 Intermountain Health Care in Salt Lake City has been using standardized treatment for 2 dozen illnesses, such as pneumonia, diabetes, and heart disease in its 21 hospitals and 90 clinics for many years, with robust improvement in care quality and reduction of costs. The business case for their approach is impressive: operating margins are at the very top of the industry.5 Thus, it seems likely that many of us will someday find ourselves working in care systems in which algorithm adherence will be the expectation.

Why are guidelines and algorithms not used?
Fauman discussed the many reasons why physicians object to guidelines and algorithms.1,4 Curiously, physicians often agree with the recommendations when they are presented to them separately.6 Nonadherence may therefore be a problem caused more by a failure of the health care system to provide reminders of the recommendations at a timely moment in the physician's work flow than by disagreement with the recommendations themselves.7 Ideally, they should be in an abbreviated format, but with the option to access the full reasoning and supporting evidence as needed. Clearly, the appropriate vehicle for getting the algorithm advice to the physician is a computerized medical record and order-entry system.8 However, the standard for how best to incorporate the logic and recommendations of guidelines and algorithms into such systems does not exist yet.9

Another reason for differences between what guidelines and algorithms recommend and what physicians do is related to the way practicing physicians make treatment decisions.10 Experienced physicians do not usually think through every decision with a systematic and exhaustive comparison of alternatives: collecting all possible relevant data about the patient and then considering all the pertinent literature. This kind of evidence-based medicine practice is too time consuming to be practical. Instead, physicians do a limited review of the patient's history and mental status, prompted by certain symptoms or historical details. They rather quickly determine the important characteristics of the situation, after which a solution may just "fall into place."10 Such "rules of thumb" are less cumbersome to apply than the "rules" of algorithms that are based on more exhaustive analyses.7

Clinical experience validates these rules of thumb, and they are assumed to exemplify the "art" of medicine. Personal heuristics provide efficient and effective solutions at many decision points, which may be as good as the recommendations of the algorithms. Indeed, research may establish the superiority of some of these other approaches. Physicians tend to oppose recommended practices that are harder or take more time than what they do now. In these situations, the experience-based rules of thumb may fall short of optimal practice.7

When should guidelines/ algorithms be followed?
Most recommendations in guidelines and algorithms probably are followed.4 The following 7 recommended practices, which seem to differ from what many physicians do but may produce better results, are listed in Table 2 and are explained and discussed below. ("Better result" is defined as either a better clinical outcome or the same outcome with equivalent safety but with reduced cost.)

  
Use clozapine after 2 adequate monotherapy trials of other antipsychotics in schizophrenia.
Make 1 medication change at a time, with adequate dose and duration of therapy.
When there is no significant response to monotherapy, switch to a different agent rather than adding a second medication.
When initiating an SSRI, select an inexpensive generic for cost-effectiveness.
Check for potential drug-drug interactions before prescribing.
Use lithium in preference to valproate as first-line treatment for bipolar disorder.
Approach insomnia as a symptom that requires diagnosis and treatment specific to the diagnosis.

Use clozapine after 2 adequate monotherapy trials of other antipsychotics in schizophrenia. Numerous lines of evidence support this recommendation, found in all schizophrenia algorithms including the International Psychopharmacology Algorithm Project (IPAP), the Texas Medication Algorithm Project (TMAP), and the Psychopharmacology Algorithm Project at the Harvard South Shore Department of Psychiatry (PAPHSSDP). The latest Clinical Antipsychotic Trials in Intervention Effectiveness (CATIE) data confirm this recommendation.11 Yet clinicians prefer to try many additional monotherapy trials, various combinations of antipsychotics, and other polytherapy.

The resistance to clozapine is likely a consequence of the fact that it is a much more arduous treatment to implement for the physician and patient. There is fear of adverse effects, additional time and effort involved to get consent, and the need for appropriate medical monitoring. Patient resistance can often be overcome if the physician presents an appropriately positive assessment of this treatment option. Certainly, the evidence supporting clozapine versus alternative medications should be part of the discussion.

Make 1 medication change at a time, with adequate dose and duration of therapy. All guidelines and algorithms stress this. The studies of algorithm-driven treatment versus treatment as usual clearly show that organized, diligent, consistent, measurement-based care that gives adequate time for the medication to be dosed properly produces better and faster results than treatment as usual.12,13 Using this one-thing-at-a-time approach may be more important than using the specific drugs favored in the algorithms.

Managed care may be the chief source of opposition to this approach; daily changes in the pharmacotherapeutic regimen of inpatients are often demanded in order to justify "active" treatment. Approval of reimbursement for additional days in the hospital or outpatient visits may be withheld unless the physician complies. The other major opponent of the methodological approach is clinical experience, which often seems to support the various add-ons and premature switches. Patients often improve, or continue to improve, after these changes are made; however, placebo effect and the passage of time may explain much of this improvement.14

When there is no significant response to monotherapy, switch to a different agent rather than adding a second medication.Most physicians switch medications when the patient does not improve after a reasonable period. The key difficulty is in the evaluation of a partial but unsatisfactory response. How much of this partial response is due to the non-drug-related aspects of care? Has there in fact been no significant response to the medication itself? The Figure shows hypothetical data representative of the findings in hundreds of studies of different medications for mood and anxiety disorders.

With both active drug and placebo, patients show gradual improvement over 12 weeks but the active drug starts to separate from placebo after 2 weeks, and the effect size (difference from placebo) increases gradually over the 12 weeks; but the placebo also does moderately well at each time point. If a patient improves 20 or 25 points on this hypothetical rating scale by week 8 or week 12, this is a partial response, and most physicians (and patients) would attribute it to the active drug. However, a 20 to 25 point improvement is, in these hypothetical data, right at the mean of what is expected from placebo.

In clinical practice, there is no placebo but there are therapeutic elements, including the alliance and expectations set up by the diagnostic process, supportive follow-up meetings with the patient, and investigator bias (ie, the physician's belief and expectation that the medication will work). So, is this 20 to 25 point improvement a placebo effect? Active awareness of this issue, preparation of the patient for this possibility before treatment starts, and avoidance of premature drug add-ons before completing a full trial of a single agent could allow a more objective collaborative assessment of this question. It may prevent unnecessary polydrug therapy resulting from "augmentation" of placebo-related changes.

When initiating an SSRI, select an inexpensive generic for cost-effectiveness. The difference in acquisition costs among SSRIs is up to 60-fold in health care systems with bulk purchasing power, such as the Veterans Affairs Department (Table 3) and Medicaid programs. The least expensive choices right now are citalopram and fluoxetine. Sertraline has been available as a generic since July 1, 2006, but it is still very expensive, though its price is expected to come down. Many physicians have personal heuristics favoring the expensive SSRIs, probably resulting from biases induced by pharmaceutical company marketing. Patients may come to the office with their own preferences based on advertising, negative media coverage of certain products, and the recommendations of peers. However, the aggregate evidence suggests no significant differences in efficacy15,16 or adverse effects in adults, children/adolescents, or geriatric patients-although there may be more weight gain with paroxetine.17 On the other hand, evidence does not support favoring paroxetine if insomnia is an initial symptom.18

TABLE 3 Antidepressant procurement costs in the Department of Veterans Affairs (February 2007)*
 AntidepressantDose (mg)Monthly cost ($)
Citalopram400.69
Fluoxetine200.89
Nortriptyline1001.89
Mirtazapine303.30
Paroxetine209.00
Bupropion SA30026.00
Nefazodone40029.00
Zoloft (Sertraline)10037.00
Lexapro (escitalopram)2041.00
Bupropion XR6052.00
Cymbalta (duloxetine)15060.00
Effexor SA (venlafaxine)15066.00
Venlafaxine15078.00

There may be more drug interactions with some of the SSRIs. The risks differ with SSRIs, and none are free of risk. Among non-SSRIs, the costs of mirtazapine and the tricyclics are low; however, the generic versions of venlafaxine and bupropion SA are very expensive, and their price is much higher than even the branded SSRIs. Their unique mechanisms, along with a small market share that discourages competition, has enabled the generic manufacturers to keep prices up.

Check for potential drug-drug interactions (DDIs) before prescribing. Computerized, frequently updated drug interaction information is available. It is impossible to remember all the known interactions much less keep up with all of the new evidence being published. It is particularly difficult when the patient is taking multiple drugs.

DDI information will someday be built into electronic order entry systems. For now, if you have access to fast Internet service, there are excellent applications that enable you to examine the entire regimen at once and the potential impact of adding or subtracting drugs. They are available on-line by subscription from www.genelex.com (GeneMedRx) and www.micromedex.com (Drug REAX). However, the time it takes to do these checks is a barrier. In addition, it should be noted that, though recommended by many19 and despite high face validity, the value of formal checking for DDIs is unproven. Adequate studies with sufficient numbers of patients have not been done.

Use lithium in preference to valproate as first-line treatment for bipolar disorder. Marketing influence, personal heuristics, and deficiencies in training20 have resulted in an unwarranted preference for agents other than lithium in the United States.21 Fear of the adverse effects of lithium and the inconvenience of monitoring for them is the usual objection. However, most guidelines and algorithms suggest these concerns do not outweigh the clinical advantages of this product. Psychiatrists may also be overlooking the significant risks associated with the alternatives to lithium. For example, valproate has exceptional teratogenicity, resulting in a recommendation by epilepsy treatment experts that it be avoided as a first-line treatment for all women of childbearing potential.22 Lithium is also a cost-effective choice that may have the best chance of sustained effectiveness as a monotherapy.21

Approach insomnia as a symptom that requires diagnosis and treatment specific to the diagnosis.Many of the algorithms and guidelines discuss insomnia as a common symptom in patients with the disorder under discussion. However, the guidelines consistently recommend diagnosing the cause of the insomnia first, then treating. Often the cause is multifactorial. This evaluation requires extra time, and the impulse is to go to the faster solution of adding a favored hypnotic; however, patients may be using excess caffeine or other stimulants, or they may be waking to have a cigarette because of nicotine dependence.

In addition, some patients have symptoms suggestive of sleep apnea or restless legs syndrome. Conditioned insomnia is a common component, often involving preoccupation with watching the clock, worrying about whether one will be able to sleep, and inability to redirect one's thoughts to more restful topics. Cognitive-behavioral approaches can be very helpful for this. Patients may be on polydrug therapy that includes activating medications that might best be switched to less stimulating alternatives, rather than adding one more drug to the regimen.

Comments and conclusions
Guidelines and algorithms will be of more practical value when their most important advice-specifically the advice that differs from usual practice and may give better results-can be provided efficiently and just at the point of decision making. Input of this kind may someday be considered a necessary contributor to, but never a sufficient basis for, clinical decision making.