This issue of Psychiatrie Annals presents the results of four of the five initiatives of the Harvard Psychopharmacology Algorithm Project. The fifth, the Algorithm for the Pharmacotherapy of Depression, has been presented elsewhere.1 The project's overall goal is to create psychopharmacologic treatment algorithms and then, using the Internet, test their value by soliciting organized feedback about the clinical outcomes achieved.
Recently, algorithms have begun to appear frequently in the psychiatric literature, probably because they respond to some of the major challenges facing psychiatrists. Important among these challenges is information overload. Psychopharmacologic therapy, in particular, is becoming more complex. Because we have a limited capacity to learn, store, and retrieve information, the value of information systems for supporting clinical decision making is becoming more and more apparent.
WE ALL USE ALGORITHMS
Actually, treatment algorithms are nothing new. All of us use treatment algorithms, which we keep in our minds, to select treatment strategies as we see patients. These algorithms are often clear about the initial steps of treatment. However, they tend to be less clearly articulated for uncommon and treatment-resistant situations. Our personal algorithms are also subject to daily change based on our recent experiences, which often draw us away from following even our own best algorithmic thinking. Two such influences are single case experiences (eg, a patient who did unusually well with a new drug) and potentially biased drug-company-influenced information.
The algorithms that follow are not likely to clash with the ones you already use. They are likely to be valuable extenders and clarifiers of your algorithms.
Some psychiatrists fear that algorithms developed by others will restrict their freedom to choose the best treatments for their patients. To address this concern, the algorithms discussed in this issue might be thought of as generic treatment suggestions. It is expected that a psychiatrist will likely consider many additional factors before deciding on treatment for a specific patient. Below are some factors that can appropriately influence the decision to choose a particular drug or strategy.
* Certainty of the clinical observations associated with a specific recommendation in the algorithm
* Potential impact of side effects
* Therapeutic alliance
* Patient, family, and other caregiver's preferences
* Cost, formulary issues
* Treatment setting (outpatient, inpatient, or residential)
* Urgency for relief of symptoms
* Drug interactions
* Clinician familiarity with using particular drugs
* Comorbid medical or psychiatric conditions
* Admonitions from authorities that may be more "political" than scientific
* Factors not considered by the algorithm
HOW THE ALGORITHMS ARE DEVELOPED
Each author approached the task with a commitment to use research evidence as the principal basis for the algorithm development.
Algorithms are a way to bring research findings into a more clinically useful form.2 Because the authors sometimes had to generalize findings to situations for which there are no specific research data, the algorithms occasionally relax the scientific rigor to provide reasonable recommendations for all situations addressed. When more than one approach or drug appears to be equally valid, the choice is left to the user.
The parts of the algorithms that may be most helpful are those that apply later in the therapeutic process after initial steps have been unsuccessful. These typically are areas where research data are less certain. It is here, particularly, that the author's analysis of research, case reports, and expert opinions can help the psychiatrist to make decisions.
These algorithms have been, and will continue to be, offered for external review in a variety of ways. Reviews occur in the process of publication, during presentation at scientific meetings, through specially commissioned review panels, and by the authors' participation in algorithm development groups such as the International Psychopharmacology Algorithm Project.3'4
It is hoped that these efforts to ensure quality will, over time, be outpaced by feedback data from users of the computerized versions. Information coming from users will support or refute the validity of each decision point in the algorithms. The goal is to have algorithms that, although initially created by authors, eventually become structures supported and shaped by outcome data generated by users.
ADVANTAGES OF COMPUTERIZED ALGORITHMS
The first advantage of computerized algorithms is that the Internet version is always the very latest. Second, you can quickly check the "What's New" page to see what has been added recently. Third, extensive discussion and explanation can be included via hyperlinks without overwhelming the user.
If you have a specific question about a patient, you can quickly find the answer in the same way you would from a timely hallway consultation with an expert. The expert answers your specific question, makes a recommendation, and explains the reason. Once you become familiar with the programs, you can quickly answer the prelirninary questions and get to the place relevant to your current patient. You can instantly access ancillary information such as possible cytochrome interactions, strategies for dealing with noncompliance, drug costs, and suggestions for optimizing a maintenance treatment regimen.
The algorithms on schizophrenia and depression can be viewed on the Internet and downloaded (without charge) onto your computer from www.mhc.com / expert.html. The other algorithms are being computerized and will be placed there as they become available.
COLLARORATION TO IMPROVE THE ALGORITHMS
When a psychiatrist reaches a recommendation in the algorithm and decides what intervention to make, this decision can be transmitted through the program to the author via the Internet. In a few weeks, when the result of the intervention is clear, that information can also be transmitted. This two-step process prevents the reporting bias associated with retrospective outcome reports. Because reporting is easy, large numbers of cases can be collected. Such data can be used to answer questions that have remained unanswered for many years because we lacked a practical way to collect such data. The figure is a flowchart of how Internet-enabled algorithms can use feedback.
It may be necessary to develop a system for ensuring the trustworthiness of those submitting clinical outcomes. The system must not be vulnerable to manipulation of reports to make one recommendation or drug appear better than another because of biased reporting.
Patient Confidentiality. Psychiatrists use an identifier (not the patient's name) that they choose when reporting a case. The algorithm author's database will have only that identifier and specific pharmacologic data without any psychosocial history. Thus, the information sent over the Internet is unlikely to compromise patient confidentiality. More elaborate confidentiality protection will be required if we are to correctly recognize reports on one patient from more than one psychopharmacologist.
Figure. How computerized algorithms can be used to improve clinical outcomes.
THE IMPACT OF ALGORITHMS ON QUALITY
Research in other fields where complex decision making is required shows that consistent adherence to a specific plan or algorithm yields better outcomes than not doing So - even if the plan is constructed without rigorous testing.5,6 Thus, it seems plausible that implementation of psychopharmacology algorithms will improve the quality of care. However, to date, there have been practically no demonstrations of this. An exception, with positive results, is the effort by Katon et al. to improve the treatment of depression in primary care by encouraging the use of a practice guideline.7 The Texas Medication Algorithm Project is an ongoing comprehensive effort to implement algorithmdriven psychopharmacologic care for mood disorders and schizophrenia in a public system.8"10 It will measure the impact of algorithm-assisted treatment compared with "no algorithm" control treatment. Their major work in progress has added credibility to the algorithm movement within psychiatry.
The Harvard Psychopharmacology Algorithm Project hopes to use its technologic innovations to make carefully thought-out algorithms available to psychiatrists working in a wide range of clinical settings.
1. Osser DN, Patterson RD. Algorithms for the pharmacotherapy of depression: part one. Directions in Psychiatry. 1998;18:303-317. (Part two. 1998;18:319-334).
2. Rush AJ, Prien RF. From scientific knowledge to clinical practice of psychopharmacology: can the gap be bridged? Psychopharmacol Bull. 1995;31:7-20.
3. Jobson KO, Potter WZ. International Psychopharmacology Algorithm Project report. Psychophannacol Bull. 1995;31:457-507.
4. Trivedi MH, DeBattista C, Fawcett J, et al. Developing treatment algorithms for unipolar depression in cyberspace: International Psychopharmacology Algorithm Project. Psychopharmacol Bull. 1998;34: 355-359.
5. Dawes RM, Faust M, Meehl PE. amicai versus actuarial judgement. Science. 1989;243:1668-1674.
6. Dowie J1 Elstein A, eds. Professional Judgement. Cambridge, England: Cambridge University Press; 1988.
7. Katon W, Von Korff M, Lin E, et al. Collaborative management to achieve treatment guidelines: impact on depression in primary care. JAMA. 1995;273:1026-1031.
8. Rush AJ, Crismon ML, Toprac MG, et al. Consensus guidelines in the treatment of major depressive disorder. / Clin Psychiatry. 1998; 59(suppl 20):73-84.
9. Gillbert DA, Altshuler KZ, Rago WV, et al. Texas Medication Algorithm Project: definitions, rationale and methods to develop medication algorithms. / Clin Psychiatry. 1998;59:345-351.
10. Chiles JA, Miller AL, Crismon ML, Rush AJ, Krasnoff AS, Shon SP. The Texas Medication Algorithm Project: development and implementation of the schizophrenia algorithm. Psychiatr Serv. 1999; 50:69-74.