Introduction
An algorithm is a stepwise solution used for problem solving. In medicine, the clinical algorithm, most commonly represented as a flow chart, is used in clinical decision-making processes to address diagnosis and treatment ( ). Since the 1990s, many clinical algorithms have emerged as a consequence of evidence-based medicine (EBM), which is defined as the integration of “individual clinical expertise with the best available external clinical evidence from systematic research” ( ). The “EBM movement” took off in the early 1990s after the term was coined by Gordon Guyatt in 1991 at McMaster University, Canada ( ), and several highly-influential texts were subsequently published on the topic by David Sackett and colleagues ( ; ; ). Over the past few decades, there has been a push toward the use of EBM in mainstream clinical practice. While efficacy and cost-effectiveness of practicing EBM is well-established, there is still a challenge to ensure adherence in clinical settings ( ). It is especially difficult for researchers to replicate the results of systematically-designed randomized-controlled trials in real-world clinical practice ( ). The simplicity and ease-of-use of treatment algorithms may present a solution.
Algorithms differ from clinical practice guidelines (CPGs) in that CPGs provide users with a synthesis of up-to-date data and clinical recommendations to direct their decision making for a specific disease or disorder, whereas algorithms apply this information to create a comprehensive step-by-step formula for the user to follow. While CPGs provide general recommendations and information about the safety and efficacy of potential treatment modalities, algorithms are highly specialized and standardized to provide specific strategies and steps for the treatment of a clinical condition based on information patterns ( ). Often, CPGs will include algorithms to synthesize data for ease of use ( ; ; ). Algorithms in psychiatry aim to minimize inconsistencies in prescribing patterns among clinicians, increase likelihood of remission, minimize side effects, and decrease overall costs for the healthcare system.
In certain medical fields, such as oncology, the integration of genetic and other biological markers to identify diagnostic subtypes has enhanced sensitivity and specificity of treatment selection ( ; ; ). However, the heterogeneity of mood disorder etiology and symptomatology makes it significantly more difficult to apply these techniques in the treatment of depression. Failure to identify valid and reliable biomarkers of MDD, that may be used in conjunction with symptom-based features, has hindered the progress of precision psychiatry and therefore the development of reliable treatment algorithms with high sensitivity and specificity ( ; ; ). In the case of treatment-resistant depression (TRD), failure to develop a gold-standard, universal definition of TRD has further hindered this progress ( ). Nevertheless, in the past two decades, several TRD treatment algorithms have been developed, many of which address treatment strategies across varying stages of treatment resistance.
In this chapter, we summarize the major treatment algorithms of the past three decades, including the seminal algorithms of the 1990s and early 2000s, as well as more modern approaches. The potential benefits of using treatment algorithms, both on an individual patient and societal/economic level, are then examined. Further, we will discuss important considerations when developing and implementing treatment guidelines to treat depression in clinical practice. We end the chapter by considering the future of treatment algorithms for depression, particularly the potential role of pharmacogenetics and computational analysis to enhance algorithm sensitivity and specificity in the world of precision psychiatry.
Early treatment algorithms: GAP, TMAP, and STAR*D
The German Algorithm Project
Established in 1990, the German Algorithm Project (GAP) was the first attempt by a clinical research group to systematically create a treatment algorithm for depression ( ; ). The algorithm was developed by a consensus group of psychiatrists in the Department of Psychiatry at the Free University of Berlin ( Freie Universität Berlin ), combining available literature at the time with clinical expertise, and was specifically designed for inpatient treatment. This was achieved through the development of a “standardized stepwise drug treatment regimen” (SSTR; ; ) ( Fig. 6.1 ).
Among 119 patients enrolled in an observational study to test the efficacy, feasibility, and acceptance of this SSTR, 34% achieved remission (Bech-Rafaelsen Melancholia Scale (BMRS) score ≤ 5), and an additional 34% met response criteria (change in BMRS score ≥ 50%) ( ). However, there was no control group to compare the SSTR to treatment-as-usual (TAU), limiting the generalizability of these results. Further, this study was limited by low enrolment and high dropout rates: more than half of patients meeting initial inclusion criteria were not enrolled because their individual treatment needs could not be met by the SSTR, suggesting a reluctance on the part of their physicians to accept algorithm-guided treatment; and one-third of participants dropped-out of the study, most commonly due to intolerable drug side effects ( ). The SSTR was developed and tested at a single university site which may have led to bias due to lack of diversity among the algorithm developers. Finally, the algorithm only provided a single treatment option at each stage, not allowing for adjustments or differences in specific drug choices based on individual differences ( ).
Addressing several of these issues, two additional phases of the GAP study were completed. The second phase finished in 2000 and compared SSTR to TAU ( ). This study integrated knowledge and insights developed during GAP Phase 1 into an updated SSTR, and also included novel therapeutic agents that had been approved since GAP1, specifically paroxetine and venlafaxine. Participants in the SSTR group achieved remission significantly earlier than TAU (7 vs 12 weeks, on average). However, there were no significant differences in depressive severity scores between the two groups at any time point, neither was there a significant difference between remission rates. A higher drop-out rate in the SSTR versus TAU group (45% vs 16%) was primarily due to physician noncompliance in the SSTR group; withdrawals for physician noncompliance were only applicable to the SSTR group, leading to a biased drop-out rate ( ).
The third phase of GAP was a large-scale, naturalistic study that included participants from 10 psychiatric departments across Germany and tested three different SSTRs against each other and TAU ( ). If the first antidepressant treatment failed, one of the three SSTRs were randomly prescribed: lithium augmentation, dose escalation, or antidepressant switch. Although patients treated according to any SSTR achieved remission more quickly than TAU, the lithium-augmentation group performed the worst of the three SSTRs. However, there was no difference in remission or response rates at any time point between SSTR and TAU. Similar to the previous two phases, drop-out rates were high in the algorithm-treated groups (41%–43%) compared to TAU (19%) ( ).
Texas Medication Algorithm Project
Addressing many of the limitations of the German Algorithm Project, the Texas Medication Algorithm Project (TMAP) began in 1995 and was the first controlled trial to evaluate the effectiveness of algorithm-based treatment in clinical practice ( ; ; ). The developers hypothesized that algorithm-guided treatment would lead to faster and more significant symptom improvement, better functioning, and a lower side effect burden over 1 year, compared to TAU ( ). The TMAP algorithm was developed using a formal consensus, where academic psychiatrists, psychopharmacology specialists, other physicians, mental health consumers, and family members met two discuss the sequence of treatment over a two-and-a-half day conference ( ). Importantly, the TMAP algorithm suggests clinicians take into account individual patient factors, allowing for more flexibility and clinical expertise when determining the next-step for treatment ( ; ; ; ).
A total of 547 participants diagnosed with MDD were enrolled in the original TMAP study ( ). While both algorithm-treated (ALGO) and TAU groups improved after 3 months of treatment, there was a significantly greater reduction in depression severity in ALGO compared to TAU, and this difference remained significant at every follow-up point, up to 1 year (the last timepoint measured). Algorithm-driven treatment was especially effective in comparison to TAU among participants with more severe depressive symptoms and greater functional impairment at baseline.
However, there were limitations to this study design ( ). There was no measurement of physician adherence to the algorithm. The TMAP guidelines required extra and ongoing training of physicians, making algorithm-guided treatment more time consuming for physicians to use compared to TAU (e.g., because of extra paperwork), which may have been detrimental to physician adherence. Further, no randomization techniques were used to assign participants, clinics, or physicians to ALGO or TAU ( ).
The Sequenced Treatment Alternatives to Relieve Depression
Sequenced Treatment Alternatives to Relieve Depression (STAR*D) is an influential study in the field of psychiatry. The goal of this trial was to determine the most effective treatment for individuals with MDD who had an unsatisfactory outcome following initial and subsequent treatment trials ( ; ). An important aspect of this study was the decision to use “remission” (defined as a score of ≤ 5 on the Quick Inventory of Depressive Symptomatology—Clinician Rating 16-item [QIDS-C]) instead of “response” as the primary outcome measure ( ; ). Participants who responded to a treatment trial (defined as a ≥ 50% reduction in QIDS score), but did not achieve remitter status, were combined with the “nonresponders” and moved on to the next treatment level. In other studies where response is used as the primary outcome, responders with clinically significant residual symptoms would be combined with those who achieved remission and not further treated for residual symptoms. STAR*D represents an important shift in psychiatry, particularly in the treatment of mood disorders, where the goal of treatment transitioned from antidepressant response to remission of symptoms and a subsequent return to premorbid levels of functioning ( ; ; ).
Citalopram was selected as the first treatment (level 1) because of its low risk of discontinuation-related side effects and good representation of the SSRI class ( ; ; ). Individuals who did not achieve remission with citalopram were randomized to receive one of a variety of switching or augmentation strategies during level 2 ( ; ). This pattern was repeated twice more in nonremitters. While the Berlin Algorithm Project and TMAP evaluated the clinical benefit of a specific, predeveloped treatment algorithm, STAR*D set out to determine comparative efficacy of specific agents or treatment strategies, using an algorithm-like approach ( ; ).
Nonremitters during level 1 who tolerated citalopram received augmentation with bupropion, buspirone, or cognitive therapy (CT) ( ; ). The use of bupropion as an adjunct was supported by a survey of 400 psychiatrists who endorsed bupropion as their first-choice adjunctive strategy with an SSRI ( ). At the time of STAR*D’s publication, the evidence for this combination strategy was based on anecdotal reports, case series, and small open-label trials ( ; ). STAR*D presented the opportunity to empirically test the effectiveness of bupropion using a large study cohort. Today (2020), while still widely prescribed as an adjunct antidepressant to SSRIs, the evidence supporting bupropion as an adjunct remains weak ( ). At the time of STAR*D, preliminary evidence supported the efficacy of buspirone as an adjunctive agent in those resistant to SSRIs ( ; ; ) however, in a subsequent metaanalysis, there was no evidence of efficacy for buspirone augmentation ( ). At the time STAR*D was launched, the evidence appeared insufficient to justify adding an atypical antipsychotic agent.
Nonremitters during level 1 who did not tolerate citalopram were switched to bupropion, venlafaxine, sertraline, or CT. These medications were selected to test both intraclass (SSRI-to-SSRI [sertraline]) and interclass (SSRI to non-SSRI [bupropion or venlafaxine]) switching strategies ( ; ). A switch to CT is supported in the literature as a reasonable next-step option following antidepressant nonresponse ( ; ; ). Overall, these stage 2 strategies were selected to test prevalent clinical beliefs and common clinical practices for initial SSRI nonresponse ( ; ).
In treatment level 3, those who failed to remit in the previous treatment stage received either adjunctive lithium or triiodothyronine/T 3 augmentation (among the earliest reported antidepressant augmentation strategies) ( ; ), or were switched to mirtazapine or nortriptyline, in an attempt to study how these treatment options compare ( ; ). Finally, level 4 compared treatment strategies that are typically reserved for more treatment-resistant cases, namely the monoamine oxidase inhibitor (MAOI) tranylcypromine or mirtazapine plus venlafaxine. This combination was expected to have a lower side effect burden and greater safety profile compared to the MAOI ( ; ; ; ). An important limitation of this study is that participants were allowed to select their own treatment options for steps 2–4, which may have biased the outcome. Further, the attrition rate was quite high, at 37% overall (from Level 1 to Level 4) ( ).
While STAR*D did not test a specific algorithm per se, it did provide valuable data on the effectiveness of antidepressants and augmentation agents that could be used for future algorithm development. With no clear “winner,” a definitive “best treatment algorithm” could not be developed. The cumulative remission rate in this study was 67% ( ), with likelihood of remission decreasing significantly with number of failed treatments: 37% and 31% of participants remitted after one or two treatment trials, respectively, but the remission rate dropped to 14% and 13% after three and four antidepressant trials, respectively ( ).
The results from STAR*D have informed many subsequent guidelines, algorithms, and other publications on the treatment of mood disorders ( ; ; ).
Subsequent treatment algorithms: 2015–20
Since the seminal TMAP and STAR*D studies, interest in treatment algorithms for depression seems to have waned, with attention shifting to CPGs. Guidelines incorporate more detailed descriptions of clinical evidence and grade recommendations according to efficacy and safety considerations, while allowing clinicians to use these data to create their own treatment plans. However, these are often biased toward depression in the primary-care setting and do not provide progressive steps to deal with treatment-resistance.
Among treatment guidelines published since 2015, several have included algorithms to synthesize data presented in the CPG. The most influential algorithms include the Royal Australian and New Zealand College of Physicians (RANZCP), Canadian Network for Mood and Anxiety Treatments (CANMAT), and the Maudsley Prescribing Guidelines (MPG) algorithms ( ; ; ). Other treatment algorithms published since 2015 are summarized ( Table 6.1 ), as well as CPGs without algorithms ( Table 6.2 ) ( ; ; ; ; ; ; ; ; ; ; ).
| World Federation of Societies of Biological Psychiatry ( ) |
| Choose initial AD based on individual patient factors: (a) intolerance: switch to another AD with evidence of better tolerability; (b) inadequate response after 2–4 weeks: increase dose (where appropriate); (c) inadequate response: augmentation strategy* if initial antidepressant is an SSRI, first try combining with a presynaptic autoreceptor inhibitor (e.g., mirtazapine) before trying an augmentation strategy. Inadequate response: AD switch or consider ECT * 1st choice: lithium, quetiapine, or aripiprazole; 2nd choice: T 3 , T 4 , or olanzapine + fluoxetine. Note: Consider adding psychotherapy at any time during treatment |
| Department of Veterans Affairs and Department of Defence ( ) |
| (a) Mild/moderate MDD: monotherapy or pharmacotherapy + psychotherapy; (b) severe/complicated MDD: refer to speciality care or pharmacotherapy + psychotherapy; (c) remission: continuation, maintenance treatment, and relapse prevention; (d) no remission: provide referral to higher level of care/speciality care |
| Florida Medicaid Drug Therapy Management Program ( ) |
| Psychotherapy*, SSRI, SNRI, vortioxetine, bupropion, or mirtazapine for 4 weeks: (a) partial response: continue for another 2–4 weeks or treat as no response; (B) no response: dose optimization, evaluate adherence, AD switch, monotherapy + psychotherapy, or combine current AD with aripiprazole, brexpiprazole, or another AD. No response and/or poor tolerability: evaluate adherence, seek psychiatric consultation, and/or try TCA, MAOI, ECT, TMS, or (SSRI or SNRI) + (quetiapine or lithium or T 3 or l -methylfolate or SAMe). No response and/or poor tolerability: reevaluate diagnosis, switch to MAOI, l -methylfolate augmentation, other neuromodulatory approach, intravenous ketamine, or one of the following triple-drug combinations: (SSRI or SNRI) + mirtazapine + bupropion (SSRI or SNRI) + mirtazapine + lithium (SSRI or SNRI) + bupropion + atypical antipsychotic * Evidence-based, i.e., CBT, IPT, or behavioral activation |
| Korean Society for Affective Disorders ( ) |
| (a) Mild/moderate MDD: AD monotherapy; (b) severe MDD: AD monotherapy or AD + atypical antipsychotic. No response: AD switch, AD combination, or augment with atypical antipsychotic. No response: AD combination, augment with or switch to atypical antipsychotic, or try another augmentation strategy |
| Psychopharmacology Algorithm Project at Harvard South Shore Program ( ) |
| Nonmelancholic depression sertraline, escitalopram, or bupropion (if not previously tried). No/partial response: switch to another of the above ADs, a dual-action agent (venlafaxine, mirtazapine), TMS, SAMe, or St. John’s wort; or augment with omega-3 fatty acid, l -methylfolate, SAMe, light therapy, quetiapine, risperidone, aripiprazole, bupropion, mirtazapine, lithium or T 3 . No/partial response: switch to TCA or venlafaxine + mirtazapine, augment with ECT, or try another of the above options. If the patient has atypical features, first try an MAOI (selegiline or phenelzine) or SSRI + aripiprazole. Severe melancholic depression. Determine if there is an urgent indication for ECT (a) urgent indication: try ECT, or ketamine if ECT is refused or unsuccessful; (b) no urgent indication: try venlafaxine, mirtazapine, or TCA. No/partial response: try another of venlafaxine, mirtazapine, or TCA; or augment with lithium or T 3 |
| American Psychiatric Association |
| Based on literature review, developed by psychiatrists, and reviewed by 15 organizations and an Independent Review Panel Provides guidance for acute, continuation, and maintenance treatment, as well as treatment discontinuation Addresses pharmacotherapy, somatic therapies, psychotherapy, and complementary and alternative medicine Guidelines for optimization, switching, augmentation, and combination strategies Guidelines for special populations |
| British Association for Psychopharmacology ( ) |
| Based on literature review and consensus meeting with experts in depressive disorders Provides guidance on acute treatment with antidepressants, psychological and behavioral treatments, physical treatments, and complementary and alternative medicine Guidelines for treatment nonresponse and relapse prevention/treatment Guidelines for special populations |
| American College of Physicians ( ) |
| Based on randomized-controlled trial data Reviews psychotherapies, complementary and alternative medicine, exercise, and second-generation antidepressants Discusses briefly switching and augmentation strategies |
| The National Institute for Health and Care Excellence ( ) |
| Recommends tailoring care and treatment based on depressive severity Discusses general principles of care and first-line treatments Guidelines for addressing poor response to treatment, electroconvulsive therapy, and relapse prevention |
| Chinese Society of Psychiatry ( ) |
| Based on recent literature and other, international guidelines Discusses diagnosis and acute, continuation, and maintenance treatment |
| French Association for Biological Psychiatry and Neuropsychopharmacology ( ) |
| Specifically developed for treatment-resistant depression Based on scientific evidence and expert clinicians’ opinions Discusses identification of treatment resistance, indications for hospitalization, and relapse prevention Guidelines for antidepressant medications, switching strategies, adjuvant treatments, combination strategies, psychotherapy, and brain stimulation |
The Royal Australian and New Zealand College of Physicians clinical practice guidelines for mood disorders (RANZCP)
The RANZCP guidelines are based on both literature review and expert consensus ( ). Their treatment algorithm recommends psychosocial and psychological interventions as first and second treatment steps, primarily addressing community primary-care strategies for MDD. Further, these guidelines reflect regional differences in practice patterns in treating more resistant depression. For example, the authors include an MAOI as a third-line monotherapy before considering augmentation strategies, in contrast to other recent guidelines, where adjunctive atypical antipsychotics are recommended before an MAOI, due to the tolerability and safety issues, as well as the general lack of experience among prescribers in using MAOIs ( ; ; ; ).
Canadian Network for Mood and Anxiety Treatments
While known primarily as a comprehensive CPG, the CANMAT guidelines also include an algorithm for pharmacotherapies ( ; ; ). These guidelines were developed for use by psychiatrists and other mental health professionals to span the spectrum from primary-care physicians to general psychiatrists and mood disorder specialists dealing with more complex cases and treatment methods. Recommendations are based on published data, specifically metaanalyses and systematic reviews, where available and include expert opinion and consensus when study data are insufficient ( ). CANMAT addresses the use of adjunct treatment strategies and suggests circumstances when it is more appropriate to switch or apply adjunctive therapy ( ). For example, if a patient experiences side effects with an initial antidepressant, switching is recommended over adding an adjunct agent. Further, if a patient has already had two or more antidepressant trials, adjunct medication is recommended over switching. Finally, CANMAT details how long to wait for response and remission before trying an optimization, switching, or adjunctive strategy: 2–4 weeks for response and a further 6–8 weeks for remission ( ). While CANMAT includes comprehensive, evidence-based guidelines for nonpharmacological treatments, there are no strategies to integrate or sequence nonpharmacotherapies into a comprehensive algorithm ( ).
Maudsley Prescribing Guidelines
The MPG was introduced in 1994, and by 2018 it is in its 13th edition ( ; ; ). Not only are these guidelines used as a reference in the United Kingdom, but they are also influential in other countries, including China, Japan, and the United States ( ). The MPG was developed by combining data from previously published guidelines with expert opinion and clinical experience, and is heavily influenced by recommendations from the National Institute for Health and Care Excellence’s (NICE) clinical guideline for depression in adults, which are integrated into the MPG text ( ). NICE works closely with the United Kingdom’s National Health Service to provide UK physicians with treatment recommendations that are both evidence based and influenced by economic factors.
The MPG is presented as a comprehensive textbook of over 800 pages addressing treatments for MDD, bipolar disorder, and schizophrenia ( ). Drug treatments for TRD are summarized in a single algorithm. The algorithm first addresses strategies for initial treatment nonresponse, which would likely be useful for primary-care physicians. This is followed by treatment options for refractory depression, which contain more complex and investigational treatments, and therefore, we recommend them for use by a psychiatrist or other physician specialized in mood disorder treatment ( Fig. 6.2 ).
Similar to the RANZP, the MPG is limited by the fact that it does not suggest adjunctive strategies until later stages, instead suggesting changing the antidepressant three times before trying an adjunct medication. Further, the MPG algorithm does not outline treatment steps for partial response; in cases of partial response, an adjunct agent is typically recommended over a full switch, whereas a switch may be recommended in cases of complete nonresponse or poor tolerability ( ). However, unlike the RANZP and similar to CANMAT, the MPG algorithm provides an extensive list of potential treatment options for refractory depression, including less studied or more controversial treatment modalities, such as pramipexole and estrogen, that may be tried when other options have been exhausted. Interestingly, the MPG suggests intravenous ketamine as a second-line option for treatment-refractory depression. This drug has shown robust antidepressant effects in pilot studies ( ; ; ). However, the use of intravenous ketamine for the treatment of depression remains “off-label.” At the time of publication of the MPG 13th edition, intranasal esketamine had not been approved for use; the guideline therefore states that “intranasal ketamine may become available and supplant the intravenous form” ( ). Since publication, intranasal esketamine has been approved in Europe, the United States (2019) and Canada (2020) for treatment-resistant depression.
The benefits of treatment algorithms
The emergence of CPGs, followed by treatment algorithms, allows busy practitioners, who realistically cannot integrate all the emerging data, to implement best practices ( ; ). Over the past four decades, there has been a significant increase in the number of treatments available for MDD, as well as potential treatment combination and augmentation strategies ( ; ). It is a challenge for physicians to make sense of constantly emerging data and understand which of the numerous treatment strategies is best for their patients. Algorithms are designed to aid physicians with decision making by synthesizing available data and presenting them in a practical format, thus enabling physicians to consistently practice evidence-based medicine.
It is assumed that the implementation of treatment algorithms will decrease prescribing variance among physicians, and increase the appropriateness of prescribed medications, thereby improving the quality of patient outcomes and reducing the prevalence of treatment resistance ( ; ; ; ). In turn, by increasing response/remission rates and decreasing likelihood of treatment resistance, algorithms are expected to improve the cost efficiency of clinical practice and manage burgeoning health care costs ( ; ; ).
Further, algorithms can minimize the frequency of “pseudo-resistance,” where nonresponse is a result of poor quality treatment or poor treatment adherence ( ). Common causes of “pseudo-resistance” are inadequate dosing, early discontinuation of treatment, or insufficient use of the wide variety of available treatment options for depression ( ). Treatment algorithms not only provide guidance on which antidepressant to prescribe, but also at what dose and for how long. Further, algorithms provide guidance on when a physician should consider, in cases of apparent nonresponse, either increasing the dose of the current antidepressant; adding an adjunctive medication; or switching to a new antidepressant altogether ( ).
Increased visit frequency and positive outcomes
While several studies have found algorithms to be superior to TAU in improving patient outcomes ( ), explanations for these findings remain unclear—are algorithms effective because of the specific treatment recommendations, or is it because of the increased frequency of, and structured approach to, clinical visits? ( ). In a review of electronic health records and administrative claims data of 250 million people treated for depression, diabetes, or hypertension conducted by “Observational Health Data Sciences and Informatics,” 11% of individuals with depression followed a treatment pathway that was not shared by any other patient in the database ( ). The authors of this report suggest that this heterogeneity may represent ineffective differences in clinical practice, trial-and-error prescribing patterns, and/or failure among clinicians to agree on a most effective treatment pathway ( ). Therefore, the specific treatment recommendations provided in algorithms may reduce this heterogeneity in prescribing patters, and, in turn, improve patient outcomes.
Data regarding the association between visit frequency and response to antidepressant medication are scarce. In psychotherapy studies, more frequent visits with a health care practitioner are associated with greater improvement in symptom severity and higher rates of recovery. Further, an adequate number of follow-up visits after initiation of antidepressant therapy is associated with sufficient duration of antidepressant treatment ( ). In studies of late-life depression, there is a positive relationship between visit frequency and response rate among patients receiving placebo treatment, but not active antidepressant therapy ( ). This may suggest that more frequent visits enhance the positive placebo effect with antidepressant therapy ( ; ). Nonetheless, data show that individuals receiving antidepressant treatment, on average, receive much less follow-up monitoring than is recommended ( ; ; ). For example, in a study of nearly 85,000 patients receiving antidepressant therapy, only 15% of patients received the recommended level of follow-up, according to the U.S. Food and Drug Administration, during their first month of treatment ( ).
While it is most likely that both of these factors explain the superiority of algorithms over TAU, this presents an interesting avenue for future research. Perhaps, the efficacy of algorithms does not lie in the practice of evidence-based medicine per se, but the improved patient-clinician relationship that results from this practice via the increased frequency and quality of clinical visits.
Cost efficacy of algorithms
In the United States, the annual economic burden of MDD exceeds $200 billion, with 55% attributable to indirect costs, specifically lost work productivity ( ). Algorithms are expected to decrease both direct and indirect costs of depression on society by increasing rates of remission and time to recovery.
In Phase 2 and 3 of the German Algorithm Project, algorithm-guided treatment was associated with significantly lower costs than TAU ( ). Using a cost-effectiveness analysis, which compares direct treatment costs with clinical outcome (i.e., average direct treatment cost divided by remission rate), cost per remission was twice as much in the TAU group compared to the ALGO group ( ). This was mainly attributable to less treatment days and lower treatment costs in the ALGO group than in the TAU group ( ).
In contrast, algorithm guided treatment was actually more expensive than TAU in TMAP. Over 1 year, the average cost per patient receiving algorithm-guided treatment was over $10,000 (USD) more than TAU ( ). Unlike GAP, TMAP combined medication algorithms with clinical care coordinator visits, patient education sessions, enhanced clinical documentation, and expert consensus ( ; ). These additional aspects of treatment incur many more charges than standard of care, which likely explains the stark difference between costs associated with TMAP versus GAP algorithms ( ). Further, GAP was completed in Germany and TMAP in the United States, where healthcare systems are administered quite differently. Therefore, differences in costs between studies may be explained by regional differences ( ).
Unfortunately, the indirect costs of algorithm-guided treatment versus standard of care have not been compared, likely due to feasibility issues.
Algorithm development
An effective algorithm should provide users with guidance to decide which treatments should be used and at what dosages and frequencies, and in what order different treatments should be prescribed in cases of treatment-resistance ( ). In order to achieve this, researchers should have a predefined goal and scope for their algorithm ( ; ). First and foremost, based on the standards of 2020, the desired outcome of any algorithm should be achieving symptomatic remission and functional recovery ( ). Exceptions to this include patients with persistent or chronic depression; in these cases, a chronic disease management approach may be more realistic and effective, where improved functioning and quality of life are the goals of treatment ( ; ).
Treatment algorithms should adhere to the principle of greatest efficacy and lowest side effect burden—i.e., the more effective and safer a treatment, the earlier in the algorithm it should be tried ( ). One must also consider what stage(s) of disease progression the algorithm will address (e.g., initial diagnosis, early treatment, treatment resistance) ( ; ). Finally, prior to algorithm development, researchers must determine how outcome(s) will be measured in order to test the efficacy of an algorithm: both clinician-rated depression scales and patient-reported outcomes should be included ( ).
While there are many different methods to determine which treatments are the most favorable, most consistently through the use of consensus groups ( ; ; ), it is also essential that treatment algorithms projects rely on explicit scientific evidence from high-quality, randomized controlled clinical trials ( ). However, high-quality evidence does not always exist. In these situations, expert consensus and data from smaller, open-label or observational studies may be used to guide recommendations; it is important to inform users that these recommendations are based on weaker evidence. In situations where recommendations are based solely on the opinions of a development group, bias can be reduced by forming a diverse algorithm development group that involves different mental health and medical professionals (e.g., primary-care practitioners, psychiatrists, and allied health care practitioners), community stakeholders, patients, and family members, all from a wide variety of geographic and socioeconomic regions ( ). Further, clear guidelines on how expert opinions will be assessed and incorporated into algorithm development should be outlined before starting the process, if the group decides to use expert opinion at all ( ).
Unfortunately, even if extensive clinical data existed for every treatment option, it would still be impossible to create a completely objective treatment algorithm free of all bias and opinion, since members of the algorithm development group must interpret and synthesize the data. Therefore, transparency regarding the data used and how they are interpreted is crucial. Not only does this increase the credibility of the algorithm, but can also prevent misinterpretation on the part of the clinician.
There are several ways to ensure transparency and decrease bias when interpreting and synthesizing data in algorithm development. Firstly, there should be preestablished rules to rate quality of evidence in support of a particular treatment (i.e., level of evidence). Based on the level of evidence combined with other factors, the strength of recommendation for each treatment can also be established ( ). Typically, the higher the quality of evidence, the stronger the recommendation; however, certain medications with high quality evidence for efficacy may not be highly recommended for various reasons, such as side effect burden or risk of drug-drug interactions.
Secondly, the development group must consider how benefits and drawbacks/harms of a treatment will be weighed ( ). As previously mentioned, algorithms should first recommend treatments with the highest efficacy and lowest side effect burden—but what if a treatment demonstrates a very high efficacy and a significant side effect burden? As an example, this could apply to some “atypical antipsychotic” drugs with good efficacy but high side effect burden. Further, where treatment algorithms also aim to enhance cost efficiency, this could result in a treatment with high efficacy and high cost being downgraded in favor of a less expensive therapy with a greater side effect burden. Group members must establish beforehand how they will deal with these important issues as they will inevitably arise.
Typically, it is better to provide more than one treatment option at each level of an algorithm. This allows clinicians to follow a personalized approach to treatment, while working within the confines of an algorithm. A specific treatment may be chosen based on factors such as patient preference, medication availability, and tolerability issues. Since algorithms are meant to assist with clinical decision making, rather than provide rigid instructions, it is important that clinicians can adapt algorithm-guided treatment to fit the specific needs of their patients.
Another dimension that developers may consider including in their treatment algorithm is an evidence-based approach to medication-related side effects or patients with specific concerns ( ). For example, an algorithm may suggest that patients who experience sexual dysfunction during treatment with an SSRI are switched to, or augmented with, bupropion ( ; ; ; ; ; ; ), or that patients with MDD and comorbid sleep disturbances receive mirtazapine ( ).
It is also important to consider the specific obstacles and constraints that clinicians face and that may affect their ability to follow the treatment algorithm, such as socioeconomic constraints, time restrictions, and availability of required equipment and/or medications ( ; ; ). Developers may provide recommendations to the clinician regarding organizational restructuring and/or improvements to clinical practice that would allow for algorithm implementation to be successful; this may be achieved by standardizing processes and procedures used in clinic, or developing protocols for algorithm implementation ( ; ).
Finally, in order to keep up to date with emerging literature, algorithms should be updated periodically, typically every few years ( ; ; ) ( Table 6.3 ).
Related posts:
Defining treatment-resistant depression, difficult-to-treat depression, and staging treatment intensity
Clinical and epidemiological predictors of treatment-resistant depression
The economic burden of treatment-resistant depression: Cost-of-illness perspective
The neurobiology of treatment-resistant depression
Tools to aid precision treatments to prevent or manage treatment-resistant depression (TRD): Pharmacogenomics, machine learning, and artificial intelligence
Modeling treatment-resistant depression in the preclinical setting
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
