Gillman’s Antidepressant algorithm

Summary

This commentary presents my antidepressant treatment algorithm which incorporates concepts from Bayesian reasoning and ‘critical path analysis’; that makes it a ‘multi-lane’ clinical pathway. It emphasises procedures which most algorithms neglect. The appendix discusses problems concerning assumptions about the pharmacology and methodology relating to the derivation of treatment algorithms, especially the mis-leading conclusions produced by ‘stepped’ trials and meta-analysis. The term ‘algorithm’ is now used widely to describe the process of systematic assessment and treatment and there are now hundreds of references to this in PubMed.

Gillman’s AD algorithm

First, my basic ‘fast-lane’ algorithm for more severe cases of biological/melancholic depression (no AD-free periods required for any step).

Step 1. Sertraline (or citalopram/escitalopram) or Nortriptyline: 6 weeks.

Step 2. Sertraline (or citalopram/escitalopram) combined with Nortriptyline 6 weeks.

Step 3). Tranylcypromine.

Cases presenting to a specialist will already be ready for step 2. This is elaborated below.

Background: theory and considerations

I receive many enquiries about minimally effective AD treatment* from unfortunate people who have been on various similar antidepressants for a long time with little regard for what is sensible and reasonable, and little regard for their ongoing partially-treated illness and their suffering and deteriorating life circumstances (like four successive SSRIs, over one or two years — pointless, waste of time (1)). It is of great concern that a proportion of doctors do not think critically and logically about advancing people to more effective treatments in an expeditious manner.

*NB. Opinion in this commentary refers to AD treatment for serious ‘biological’ depression, different criteria may apply to anxiety and mild forms of depression.

This commentary does not consider non-pharmacological interventions which are effective and may be the preferred 1st option for many people, especially those with less severe and persistent symptoms.

Thinking critically and logically about the expeditious progress of treatment benefits from incorporating the precepts of ‘critical path analysis’, which most algorithms do not encompass. A measure of knowledge and resolve are required to achieve that because there is a strong natural reluctance to change a treatment, even if it is only partially effective.

Doctors need to be aware of how the psychology of ‘risk aversion’ and ‘negativity bias’, influence their advice and decisions.

Guidelines and algorithms also have a downside: they promote intellectual laziness; they deal in generalities not individuals; they obfuscate the fact that treatment decisions about individuals are the sole responsibility of the treating doctor; they can stifle innovation and originality and foster a self-fulfilling conformity. Doctors must have a meaningful dialogue with patients about their individual choices.

The forgotten principle of evidence-based medicine

The road to hell is paved with good intentions (Proverb)

It is important to emphasise and adhere to EBM tenets: these require:

“integrating … the best available external clinical evidence from systematic research…[with] the proficiency and judgment that individual clinicians acquire through clinical experience and clinical practice… [without which] even excellent external evidence may be inapplicable to or inappropriate for an individual patient.” (29).

Even excellent external evidence may be inapplicable to, or inappropriate for, an individual patient.

Despite this injunction, the importance of patient experience and preferences, and clinical expertise, are commonly under-emphasized in the care delivered to individuals, and in the synthesizing of evidence for inclusion in clinical practice guidelines (30, 31).

I have written a long criticism of such things elsewhere.

The explanatory justification doctors give for failure of, ‘I followed the guidelines’, is inadequate (morally, intellectually and legally).

The eminent and respected Professor John Ioannidis has recently made an excoriating comment about EBM/guidelines and suchlike publications:

“Despite valiant efforts to make them more evidence-based, guidelines, recommendations and exercise of policy power unfortunately remain among the least evidence-based activities, impregnable strongholds of expert-based insolence and eminence-based innumeracy {Ioannidis, 2019 #21905}.”

‘Impregnable strongholds of expert-based insolence and eminence-based innumeracy’.

For those not familiar with the etiquette of scientific writing and restrained expression, that is an extraordinary statement made in an unprecedented style by someone of exceptional reputation and standing*; and one which my readers will not be surprised to hear, I completely agree with.

*Professor John Ioannidis, full list of appointments:

Stanford Prevention Research Center, Departments of Medicine, Department of Health Research and Policy, Department of Biomedical Data Science, Stanford, University School of Medicine, Stanford, California; Department of Statistics, Stanford University School of Humanities and Sciences, Stanford, California; Meta‐Research Innovation Center at Stanford (METRICS), Stanford University, Stanford, California

The citation stats Professor Ioannidis has clocked up are mind-boggling: H-index 190, total citations >250,000

https://scholar.google.com.au/citations?hl=en&user=JiiMY_wAAAAJ

Algorithms

Anyone who has read much about depression will have heard of treatment algorithms, things like the widely-discussed [STAR*D research]. Algorithms can be useful even if they are, perforce, mainly based on consensus and ‘expert opinion’, because of lack of firm evidence to choose between many of the alternatives.

It must always be remembered that no matter how authoritative such algorithms and guidelines may be perceived as being, that they are not better than the judgement of an experienced scientific clinician. There are significant variations in pronouncements on the same topics by different sets of guidelines (2-4): in my considered opinion this reflects, among other things, the misleading and pseudo-scientific nature of meta-analysis (see below).

It is insufficiently recognised that guidelines constrain the flexibility of clinical practice and produce a self-fulfilling sameness in treatment approaches (3, 5). For instance, this applies to the use of MAOIs: they are only suggested as a ‘last-ditch’ measure, they are hardly ever used, therefore there is virtually no research about them, therefore there is no evidence to support their use.

Not an intelligent way of furthering clinical research.

There is another distinctly negative aspect to guidelines and algorithms: regulators and policy makers may use them to censor and control doctors, and the use of medicines. This is clearly already happening and is exemplified by doctors who refuse to prescribe MAOIs, when they are not ‘in the guidelines’, because the poor little dears are ‘afraid’ — do they understand how a depressed patient feels?

I repeat, it must be remembered that EBM tenets require; “integrating … the best available external clinical evidence from systematic research…[with] the proficiency and judgment that individual clinicians acquire through clinical experience and clinical practice… [without which] even excellent external evidence may be inapplicable to or inappropriate for an individual patient.” (29).

The usefulness of algorithms can go beyond deciding which drug to use, and thereby profit from the precepts of critical path analysis (see below).

In this commentary, I am not going to go into detail concerning the justification for the choices I present, because my reasoning and the evidence is in the general literature, and my scientific papers, and elsewhere on this website.

Complexities and background

It is a complex area, so background explanation is required, especially for people who have not read extensively around the subject, nor read what I have written previously.

First, I am referring exclusively to the treatment of people who have drug responsive illnesses and who exhibit the key central changes of anergia and anhedonia, persistently and consistently, over a period of a month or more, and to a degree of severity which has impaired their social, leisure and work function to a consequential extent.

Such an assessment is best made by an experienced psychiatrist: doctors are giving drugs to many people who do not need them (6), this tendency is contributed to by pressures from society’s expectations and the economics of healthcare, and you have guessed it, drug company advertising. A connected and toxic trio of factors.

There is a compelling logical argument for stratifying the speed of progression through the alternatives in the algorithm (it is ‘multi-path’) depending on the degree of severity of symptoms, and particularly the degree of functional impairment. When that is sufficiently severe to endanger relationships, work and people’s financial security, there is less justification for continuing minimally effective courses of drugs and engaging ‘fine-tuning’ strategies which may be less likely to produce substantial improvement.

This is how and where the focus on functional impairment, and the rapidity and adequacy of improvement, relate to the ‘critical path analysis’ dimension of the algorithm {Habert, 2016 #21763;Lam, 2016 #21765}.

Whilst most doctors would agree that urgent ECT is strongly indicated for a severely suicidal patient who has stopped eating, there seems to be a less clear appreciation of the general need to take account of the medium-term destructive influence of the illness in less serious cases. That is why an algorithm benefits from incorporating a multi-lane pathway with decisions based on the extent and speed of response balanced against the risk of harms from treatment and deterioration of the illness.

A strong indicator of ‘biological’ depression is real improvement in core symptomatology (7-9) within 2-3 weeks of starting an antidepressant. A less decisive response, over a longer time, is not good evidence of a cause-effect relationship between taking the drug and improvement, especially when that improvement is in ‘non-core’ and less illness-specific symptoms. Failure to recognise the key time relationship between drug administration and improvement (or adverse effects) is a failing of much depression research.

If a drug response is going to occur, that will usually be evident to an experienced observer within 2-3 weeks of attaining a therapeutic dose. It is impossible to mention assessment of improvement without noting an ‘elephant in the room’: the problem of the sensitivity and validity of rating-scales generally, and the much mis-used, but ubiquitous, ‘Hamilton’ rating scale (10, 11), on which most drugs trials rely (11-14). It is an inadequate instrument that should have been superseded long-ago (12, 15-17). It was not well-designed to be sensitive to change in severity of symptoms, particularly the key ‘core’ symptoms, à la Parker and colleagues (8, 9, 18).

Results from drug trials using the DSM diagnosis of MDD are prone to include too great a percentage of cases that are unlikely to be ‘biological’ depression — see, for instance: (9, 18-22). Also, bias from sponsorship (a huge proportion of trials are drug company sponsored) and cases recruited in primary care*** both muddy the waters even more — to the point of opacity — yet such trials are all grist-to-the-uncritical-mill of meta-analysis and the formation of guidelines and algorithms.

In summary: we know we are including inappropriate cases in trials and we know we are assessing change with a poor and outdated instrument.

That is not smart.

*** It is inconceivable that such cases are predominantly ‘biological’ depression. That explains why mainly sedative drugs like mirtazapine appear effective: Even a small proportion ‘non-biological’ cases contaminating the sample will invalidate most conclusions.

Meta-analysis

If there is space for another ‘elephant in the room’ it is meta-analysis. This procedure is the 21st century successor to Phrenology and Psychoanalysis.

Meta-analysis is a capricious and deceitful siren.

Some time ago I coined the expression ‘Penrose stairs with drugs’ see: https://www.psychotropical.com/anti-psychotics and Heres et al. (23). They showed ‘olanzapine beats risperidone, risperidone beats quetiapine, and quetiapine beats olanzapine’ and the same illogical circle (of A>B>C>A) can be replicated with almost all the AD drugs. It is clear to anyone of a scientific mind that many trials are poor quality data and that over-interpretation is omnipresent (24).

As I stated in ‘Atypical Anti-Psychotics and Humpty Dumpty’, it is important to keep repeating that no meta-analysis can be better than the original data upon which it depends (25, 26). That is the old computation saying “GIGO”, “garbage in, garbage out”. What would Charles Babbage have thought? well, we know, because he told us.

‘On two occasions I have been asked, —”Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?” … I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question.’

Babbage, Passages from the Life of a Philosopher.

Since it has been my view for a long time that clinical trials (many of which are ‘pseudo-scientific’) have unjustifiably overshadowed good clinical science and clinical experience, I am much heartened to see, in researches for this commentary, that researchers are supporting a re-balancing of that question. The following references are eloquent testimony to the up-welling of the view about the unreliability of meta-analysis, and the appropriateness of putting more weight on scientifically conducted clinical practice (4, 25, 27-32). Several very savvy authors are represented in these references (Altman, Ioannidis), which are well worth checking out.

I will retain a healthy respect for my own judgement based on clinical science and sound Bayesian logic.

Critical path analysis

Critical path analysis is a method for formalising the structure and timing of a task and focussing more clearly on the assessment of progress, and the means of quantifying that, and the timing of key decisions affecting the process. It has not been utilised much in medicine (33).

To incorporate precepts from critical path analysis a treatment plan it needs to focus on:

A pre-defined definition of the desired outcome

Objective assessments of the baseline state (not self-rating scales) including symptoms, signs and functional activity assessment covering work, social and leisure domains.

An assessment of the degree of risk of adverse outcomes engendered by the illness (e.g. loss of job etc.), and by any treatment.

Objective assessments of the intermediate stages of severity of the illness (at defined treatment changes).

Predefined time periods for achieving objectives at any treatment stage.

Both doctor and patient may consider, at each visit:

For how long has the drug been administered?

For how long has the present dose been the same? (Appendix Note 1)

What degree of improvement has so far occurred from baseline or worst state?

What degree of risk of severe or prolonged impairment of work social and leisure activities is there in the current situation (i.e. loss of job, relationship separation, etc.)?

Is that improvement, and the degree of reduction of the risks, sufficient to justify continuation of this treatment (or this dose)?

How expeditiously should we progress on the ‘critical path’?

An important aspect insufficiently prominent in some algorithms (and treatment plans) is a clear sense of purpose and urgency about achieving full remission and avoiding potentially irreversible life changes (i.e. before peoples’ lives fall apart).

There is a too easy and uncritical acceptance of partial improvement and of continuing in that state with unchanging treatment (Appendix note 2).

A particular impediment to the expeditious progress of effective treatments concerns widespread misunderstandings among doctors about which antidepressant drugs are safe in combination and whether there is a necessity for washout periods when changing from one drug to another (most aspects of such interactions are covered in the relevant sections on this website).

Another element of many of these algorithms is deficient discussion of combination treatments and statements about drugs which are possibly less effective. These are matters that an experienced and knowledgeable clinical psycho-pharmacologist is well positioned to advise on.

Difficulties with algorithms are an inevitable consequence of schemes devised by committees, whose members often have an insufficient depth of knowledge about pharmacology (see below).

Some academics will be miffed by that comment (cf. Prof Ioannidis comments); but it is true, witness the extensive mis-understandings about MAOIs and ST that still endure. Many academics should be ashamed by their poor level of pharmacological knowledge.

Gillman’s AD algorithm

This algorithm is a ‘multi-lane’ choice*. The traffic-lane analogy helps people conceptualise what approach to treatment suits them best, depending on their attitude to life and risk** (do you usually drive in the fast lane or the slow lane?). Clearly, if functional impairment is producing substantial risk of (potentially irreversible) changes in family, work or financial status (as it so often does) then the a faster-lane pathway will be preferred over a slow-lane.

** I often used to say, ‘Slow and steady wins the race, or, nothing ventured nothing gained’, your choice.

Step 1

Sertraline (or citalopram/escitalopram)‡ or Nortriptyline*. In a great majority of cases an SSRI, or SNRI, will already have been utilised in primary care.

See: ‘When to consider avoiding SRIs as first choice’ in: https://www.psychotropical.com/anti-depressants/tcas

And, re Sertraline as 1st choice SSRI, rather than (citalopram or escitalopram) (34, 35) and: https://www.psychotropical.com/sertraline

‡It is unlikely to be useful to exceed the recommended dose with any SSRI, with the possible exception of Sertraline, its weak DA re-uptake potency may confer an advantage at doses around 200 mg or more. Sertraline also has the useful advantage of linear pharmaco-kinetics even at high dose levels (36, 37).

*NTP is the pharmacological ‘gold-standard’ TCA (38),

Step 2

An SNRI strategy. This may utilise a single drug, e.g. venlafaxine, as has been fashionable for some years now, or a combination. Gillman’s algorithm strongly favours two separate drugs (for reasons explained), as in step 2a).

Step 2a

Sertraline (or citalopram/escitalopram) and nortriptyline (NTP)*/reboxetine or 2b) clomipramine (CMI)** (SNRI strategy). Or (des)venlafaxine*** or duloxetine etc.

*Sert 50 mg + NTP 50 mg may be sufficient but full doses of Sert 200 mg and NTP 150 mg may be used providing due caution is exercised about NTP levels (in view of mild 2D6 inhibition of NTP metabolism by high levels of Sert). Therapeutic drug monitoring (39, 40) of NTP levels is increasingly desirable if Sert in increased beyond 200 mg daily.

**See note on CMI: https://www.psychotropical.com/clomipramine-potent-snri-anti-depressant

***Venlafaxine has several disadvantages (relatively more toxic, especially in the elderly (41)) and I remain unconvinced that it is a) any better, or b) is a fully effective SNRI, and the evidence for its superiority is poor, for evidence and refs see: https://www.psychotropical.com/venlafaxine

Step 2b

An SNRI strategy with addition of lithium and tryptophan. This may be preferred if factors relating to situation, illness, side-effect considerations, patient preference etc. militate against proceeding to step 3 with an MAOI.

Step 3a)

From Step 2a) cease sertraline, continue nortriptyline*; after one week start tranylcypromine** (TCP); or b) cease CMI, then after washout 1–2 weeks start TCP***

*For detailed discussion of why it is perfectly safe to combine tranylcypromine with nortriptyline see section on this website re MAOIs and my published papers. MAOI/TCA interaction mis-understandings are the prime exemplar of the poor pharmacological knowledge of psychiatrists.

** The need for 1–2 week washout is a powerful reason for using SERT+ NTP, rather than Ven or CMI as the ‘SNRI’ strategy.

*** Or, in some cases phenelzine

Step 3b

To either to 2a or 2b) Use preferred augmenting strategy. There is not strong evidence favouring a particular choice, as yet. Add Li† + L-tryptophan, quetiapine, olanzapine, aripiprazole or T3 (lamotrigine may have a place here, especially if BPD is known or suspected).

†I think there is a strong case for ceasing Li augmentation within 4 weeks if there is no clear response. Logically a similar approach applies to aripiprazole and other augmenting strategies.

Step 3c

Selegiline patch, potentially with the same augmenting agents as TCP. If the high cost of this is not a consideration it may be quite effective in high doses, but then its advantage vis a vis tyramine is diminished, and it is almost certainly not as effective as TCP (see below).

Step 4a

ECT, then after course finished immediately re-introduce the most effective previous drug strategy, + Li if not previously used.

Step 4b.1)

Augment the TCP with:

Li‡ +/- L-T +/- NTP*, or bupropion (lamotrigine may have a place here, esp. if BPD known or suspected)

*either just continue NTP from step 2a) or cease it after stabilisation on TCP and only re-introduce it if needed.

Step 4b.2)

Or, TCP + methylphenidate (or methylphenidate alone, [see separate note]).

Notes on Gillman’s algorithm

At no stage is an AD free washout needed in my algorithm. However, if one chooses venlafaxine or CMI in step 2b) one then has a gap to cover (e.g. by dove-tailing doxepin* or NTP).

*Doxepin is an H1 sedative, not an AD (see (38).

My preference after step 2a) was to go straight to TCP, rather than expend time on strategies that my experience indicated were less likely to lead to decisive results, i.e. with various augmentation options, e.g. lithium (see below). If TCP (or phenelzine) was not satisfactory then augmentation can be undertaken at that stage, or after ECT.

A critical path analysis from my experience suggests that the pay-off with augmentation* may not be worth the delay in reaching remission: this a where a careful assessment of the potentially irreversible aspects of functional impairment (work, marriage, finances) usefully guides the speed of progress through the algorithm. Hence the reduced emphasis on Bupropion (42, 43), which is widely preferred (in USA) especially in BPD. It also highlights that augmentation is often continued, despite the problems with medium-term SEs, long after it has been seen to fail.

*The term augmentation can be held to mean the addition of a drug not usually regarded as an AD in its’ own right, as opposed to combination which usually refers to addition of a drug that is regarded as an AD in its own right.

In step 2a) sertraline + nortriptyline is preferred (especially because it facilitates smoother transition to TCP, but also because it enables separate adjustment of SRI and NRI components (which cancel out each-others’ SEs to a useful extent, e.g. 5-HT is GI pro-kinetic which opposes the anti-kinetic effect of TCAs & NRIs).

It has been correctly noted that various combinations risk substantial increases in the SE burden for patients (44), and possible pharmacokinetic interactions. However, as observed elsewhere, it is not valid to generalise about heterogeneous classes of drugs (like TCAs and ‘atypical anti-psychotics’, see other notes about [neuroscience-based pharmacology]). Mixing many of the old TCAs with some SSRIs is not a good idea at all, but nortriptyline (and sertraline) is quite different. It is a question of knowing your pharmacology, and most aspects relating to this are discussed elsewhere on this website (e.g. see menu heading ‘Interactions’, and in my published papers (38, 45-54).

There has been an ebb and flow of opinion about combinations over the last three decades. Unfortunately, much of the opinion expressed about combinations has been unduly influenced by doctors whose knowledge of pharmacology is inadequate (e.g. see Charpeaud below*).

I find it hard to suppress a wry smile when I see the latest opinion about combinations, which champions a combination of an SSRI with reboxetine (55). I have written about this elsewhere (56), but in brief I started using combinations of sertraline with both nortriptyline and reboxetine back in the 1990s. Most patients seemed to prefer for the former, because they found reboxetine made them feel hyped-up and agitated (similarly to an amphetamine effect), whereas nortriptyline did not. See also:

https://www.psychotropical.com/snri/

Step 3b) When using ones preferred augmenting strategy, the critical path requirement is that an objective assessment of pre- and post-augmentation clinical state is recorded and that a pre-defined degree of improvement is attained in a pre-defined time-period, or that the augmentation is ceased.

Californian rocket-fuel

I am aware of the popularity of the ‘Californian rocket-fuel’ combination (Venlafaxine + Mirtazapine) which some might choose to use in step 2b), however my experience of the superiority of sertraline + nortriptyline never encouraged me to try that combination very often, especially because of my scepticism about the claimed properties of both of those drugs and the greater toxicity of venlafaxine. It is also important to appreciate the extent of the deception concerning the bogus pharmacological data about Mirtazapine, see:

https://www.psychotropical.com/mirtazapine-a-paradigm-of-mediocre-science

*Charpeaud reported from France (57) that ‘augmenting SSRI/SNRIs with mirtazapine/mianserin has become the most recommended strategy of antidepressant combinations. Augmenting SSRI with tricyclic drugs is now a less recommended strategy of antidepressant combinations given the increased risk for the occurrence of pharmacokinetic drug–drug inter-actions and adverse effects’.

This is yet another example of ‘lumping’ (i.e. failing to recognise that the TCAs are a markedly heterogeneous group) and ignorance of pharmacology and interactions.

I meam requiem doleat.

Lithium augmentation

The history of lithium augmentation goes back further than most people seem to remember. Its use requires caution and care and a discussion of relevant considerations is in most good standard texts. We were using it extensively in London in the 1970s and I must have been involved in treatment of hundreds of patients using this technique by the time I had been in Australia for several years in the mid-1980s. Once I was in private practice, I found myself using it less and less frequently, partly because decisive improvement was less common and partly because it was a relatively costly and troublesome procedure (58). The trend, over more than three decades, is of a weakening of the evidence for benefit (59), but Nelson’s & Bauer’s reviews’ (based on low numbers) are more positive (60, 61), for those sweet innocents who still retain faith in meta-analysis.

It is possible that a significant contributing factor to my observation of the poor effect of Li augmentation, was the fact that I was using clomipramine in a great majority of patients, rather than one of the other tricyclics. It may be that lithium is working partly, or mainly, through a ‘serotonergic’ mechanism. In that case adding it to a non-serotonergic TCA like amitriptyline might produce a more marked improvement than adding it to clomipramine or tranylcypromine. It is interesting that the evidence for augmenting of SSRIs is also modest (62).

‘Atypical’ antipsychotics

Evidence has led many to favour augmentation strategies with so-called ‘atypical’ antipsychotics, which are now in most guidelines for partial and non-responders, at the same stage of the algorithm as switching or combination strategies (63, 64).

Atypical antipsychotics* are heterogeneous as a group, like the TCAs, so it is meaningless to lump them together. I have no 1st-hand experience of the degree of benefit from any of them. I suggest there are good reasons for using them with caution** and for regarding this to be a questionable idea for several reasons, not the least of which is that neuroleptics reduce DA, when much evidence indicates that increasing DA is what is needed in depression.*** Perhaps low doses of weak DA antagonists, like quetiapine, increase DA through preferential blockade of pre-synaptic receptors? There is a paucity of data and a great deal of uncertainty about both mechanism and benefit.

There is the seriously under-recognised problem that once antipsychotics have been started, they tend not to be ceased, even when a lack of benefit is clear. Patients may thereby be exposed to major side effect problems for no reason. That is a serious error.

Questions like: which one? what dose? and for how long? need to be answered, and promptly, because of these significant SEs, especially in the medium to longer term, not to mention the enormous expense.

*most so-called ‘atypical antipsychotics’ (e.g. risperidone) have tenuous claim to that ill-defined epithet.

**A suggestion of a causal link to increased mortality in older patients (65-72).

***For discussion about this point see here

I do not want to seem too much like a picky scientific purist here, but it is very important to understand that the drugs in question, these so-called atypical antipsychotics, cannot be meaningfully grouped together and many of them are probably not significantly different to the older neuroleptics like CPZ and thioridazine, despite the vociferously made claims for them. For instance, risperidone is a pure D2 antagonists, and a very potent one at that, which is related to the old classic antipsychotic haloperidol. On the other hand, quetiapine is closely related to the prototypical tricyclic sedative promazine, and like promazine is a potent antihistamine with very weak D2 antagonism. I could go on …

https://www.drugbank.ca/structures/DB01224/image.svg

https://www.drugbank.ca/structures/DB00420/image.svg

and, needless-to-say, good old chlorpromazine (73, 74). And, one could say vis a vis promazine> chlorpromazine> quetiapine, ‘the wheel is come full circle’ (Edmund, King Lear):

https://www.drugbank.ca/structures/DB00477/image.svg

Quite how it is possible to produce a convincing, or even a vaguely plausible, theoretical explanation of how all these different drugs could possibly be effective augmenting agents escapes me.

Bayesian reasoning indicates that the weak evidence of efficacy is very likely to be wrong (30, 75, 76).

Is anyone else getting a sense of Déjà Vu? Remember thioridazine in depression? e.g. see review of Robertson & Trimble (77).

The theory is weak, the evidence is weak, but the money is good, hugely good. Even Zuckerberg’s eyes would water. $$$, tens of billions, indeed, probably hundreds of billions by now. I do not think we need Einstein’s help on this one.

‘A triumph of hope over experience’. Johnson: Boswell’s Life of Samuel Johnson, 1791

Further considerations about algorithms

The issue of ‘knowing the pharmacology’ is where some of the experts producing these algorithms have a little catching-up to do. There are some unfortunate errors.

Prominent among the various ‘guidelines’ or ‘algorithms’ are: American Psychiatric Association, British Association for Psychopharmacology (78), Canadian Network for Mood and Anxiety Treatments (79), National Institute for Health and Clinical Excellence (80, 81), Texas Medication Algorithm Project (82), and World Federation of Societies of Biological Psychiatry (83).

One example of a serious mistake is the suggestion that it is OK to combine imipramine with MAOIs, and moclobemide with SSRIs (84) — that has a risk of inducing fatal serotonin toxicity. Yet others lump all the TCAs together as if they are inter-changeable; then there is recommending the worst of all the SSRIs, fluvoxamine, as a 1st line drug: if fluvoxamine (or, indeed, fluoxetine) was submitted for registration by the FDA now it would be unlikely to be approved, and for good reason (46). Such criticisms can hardly be over-emphasised, since this concerns expert panels of specialists. Nevertheless, I can assure readers that if some of the above advice was followed and it led, as it could do, so easily, to serious adverse outcomes for patients, then the doctor involved would find themselves on the losing side in an expensive malpractice suit.

I have seen the legal argument advanced that such errors make the producers of the guidelines vicariously liable.

Lastly, recommending moclobemide as a 1st or 2nd line AD is absurd; moclobemide has little useful AD effect. For a drug that has the most benign side-effect profile ever (one would therefore expect it to be very popular), its use has shrunk, almost into oblivion. This is a prime example of a mis-placed faith in meta-analysis.

The ‘CANMAT’ document (79); ‘3.6. How Do second-generation antidepressants compare in efficacy?’ among other things, reports ‘mirtazapine*** is more effective than SSRIs and venlafaxine’ — we are back in ‘Penrose stairs’ territory here. As with moclobemide, anyone who finds mirtazapine to be an effective AD is not treating biological depression and research that claims it is more effective than SSRIs and venlafaxine is not credible. This is an example of where ones clinical experience and judgement must over-ride the artefacts generated by pseudo-scientific trials and the statistical legerdemain of meta-analysis — see these refs on this key issue (3, 5).

***Calling mirtazapine a ‘2ndG’ drug is a bit of a stretch — it was ‘invented’ 50 years ago! ‘2ndG’ is a marketing term and has no basis in pharmacology.

Committees do produce some strange ideas and conclusions, and intellectual excellence less-frequently prevails in the results.

There has long been great confusion among psychiatrists concerning what is, and is not, safe to combine. Speaking with my pharmacologist’s hat on, I regard with despondency some of the suggestions and recommendations that have been made, and with dismay at the poor standard of knowledge exhibited in supposedly informed academic work (glaring examples of this are scattered throughout my commentaries).

From the many possible examples of poor knowledge I will cite, from the recent literature, something that coincides with my experience (e.g. (53)) of the commonly expressed errors and misconceptions (57, 85, 86). The commonest combination used, as indicated by clinical practice surveys, has been a combination of fluoxetine with amitriptyline (in-a-word —stupid, because of multiple CYP450 interactions and inducing potential toxicity). The contents of this commentary, and the information on my website, highlight the fact that this is one of the more ill-advised combinations that you could dream up.

It is axiomatic that any doctor prescribing a drug must have a sound basic knowledge of its mechanism of action and ill-effects. Such a level of knowledge would make it plain that this was a high-risk combination, and if significant ill-effects eventuated, the doctor would have a poor defence in law in the case of a negligence action brought against them.

Many of the suggestions and recommendations that are widely made and accepted are simply ill-advised or plain wrong (see my other commentaries and my published papers that address these issues in detail). As an expert in this field I have practised and published concerning various drug combinations over many years. Indeed, that is why I am an internationally recognised expert in serotonin toxicity, that is the ultimate drug interaction with antidepressant drugs that psychiatrists still misunderstand, decades after they could/should have mastered it (as exemplified by the incorrect information in some current guidelines). It is about the only way that psychiatrists can kill somebody within 12 hours by giving the wrong drugs e.g. Otte et al. (87). And yet, still, they try.

I have frequent enquiries via the web site concerning ill-advised, and even dangerous, combinations that specialists have contemplated, or tried to initiate. Fortunately, some patients have educated themselves better than their doctors (especially those who have learned from this website and my papers).

Other combinations

Here are one or two other ‘adventurous’ combinations that have no obvious (major) problems and at least some plausible theoretical merit — a little caution re 2D6 interactions may be required, e.g. see here:

Venlafaxine + bupropion (sig. 2D6 inhibitor, and both drugs are slightly seizure promoting).

MAOI + Mirtazapine (instead of NTP) + bupropion

Selegiline* + bupropion + NTP

Moclobemide** + nortriptyline + bupropion (but not Moc + SSRI)

* Selegiline (trans-D patch), difficult to put in a cost-conscious algorithm. It is a weak AD and very expensive. Does not boost DA as much as TCP. For many people, does not justify the time-delay and expense as a pre-TCP treatment: again, ‘critical-path’ considerations aid with such decisions.

**I’m not a fan of this drug, it is a drug with such weak effects that it is of little practical use as monotherapy. However, it clearly does do something because it aggravates the toxicity of overdoses of SSRIs (see section on ST). There is therefore some small theoretical merit in the notion that it might be worth using it in combination, but not with SRIs, because even in therapeutic doses it can lead to toxicity.

Concluding remarks

Guidelines and algorithms have good features, especially, as Bauer et al. highlight, in reducing unnecessary ‘scattershot therapies’ and the untimely (too soon and too late) switching of treatment measures (88). No guide suggests using three, four, or even five SSRIs sequentially: yet if I had a dollar for every poor patient I have heard of, who has been subjected to that pointless parade of pills, I would be a wealthy man.

Algorithms give insufficient weight to systematic evaluation of the clinical severity after each treatment trial, as Bschor et al. recommend (89), especially the assessment of deleterious effects of functional changes: that is where the ‘critical path’ precepts kick in and indicate faster progression to the next step.

There is generally an under-emphasis on the treatment of residual symptoms (90), and that is the ‘other-side-of-the-coin’ of the fact that doctors often settle for incomplete response in clinical practice: i.e. they fail to set and pursue the goal of achieving complete remission.

Comments and pronouncements in relation incorrect prohibitions concerning combining or swapping from one drug to another are frequently incorrect and due to pharmacological illiteracy: they lead to delays in the progression through the treatment algorithm.

MAOIs are much under-used: I have written a deal about that over the years: my recent editorial (2017), review and diet monograph, sum up key information and references (52-54).

It is hard to avoid the conclusion that the training of psychiatrists in therapeutic psychopharmacology still leaves much to be desired. A report-card might read ‘poor standard, consistently near the bottom of the class, must try harder’.

Appendix

Note 1

‘What is the duration of treatment and current dosage?’ Anyone who cannot answer that quickly is unlikely to be able to make a logical decision about the optimal next step.

My first entry in my medical notes, after the date of the appointment, would always be ‘T 2/12, D 3/52; the shorthand for the total time on drug and the time on the present dose (2/12 = two months, 3/52 = three weeks). If you have not got that information in the forefront of your mind every time you see a patient, you are not going to make logical treatment decisions in a timely manner.

The key assessment is whether there is a definite improvement in energy and motivation on the one hand, and ability to get pleasure, enjoyment and satisfaction on the other. That is complemented by a critical assessment of the level of functioning in work, social and leisure activities.

Note 2

There is a too-easy and uncritical acceptance of partial improvement and of continuing in that state with unchanging treatment.

Doctors’ weak and acquiescent acceptance of partial improvement sets the scene for their all-to-frequent avoidant behaviour. By this I mean giving doctors an easy excuse not to progress to treatments that they are less competent and confident to manage. These, they persuade themselves, are higher risk and ‘after all there is a useful degree of improvement, so perhaps I should just settle for that’, ‘do no harm’ they seem to say to themselves. This is doctors treating themselves, not the patient.

Such thoughts and attitudes are a distortion of the mis-understood aphorism of ‘Primum non nocere’ (first, do no harm). Contrary to popular opinion, that has nothing to do with the ‘Hippocratic oath’ and has a tenuous claim to validity as a moral precept (91).

I make no apology for proffering the view that a proportion of doctors simply do not have ‘therapeutic balls’. In a publication, a little while ago, I described the attitudes to the use of MAOIs exhibited by many doctors as ‘pusillanimous’ (92), and I have never resiled from that view (54).

Note 3: trial methodologies

A comment about the methodologies used in trials, like STAR*D, it is relevant. It is inevitable and undeniable that any patient sample generated for such trials will contain individuals who were incorrectly diagnosed and/or are never going to respond to medication. Therefore, sequential steps will contain a larger and larger proportion of such patients, compared to sample in previous steps. That will make the treatments used in the subsequent steps seem, artefactually, to be effective in a smaller % of the sample. There is a more detailed analysis of this point [here].

Next, a useful methodology for clarifying the effectiveness vs side effects of treatment is a trial of dosage reduction. I did this quite often in clinical practice as a method of ascertaining the most appropriate balance between beneficial effects and side-effects — this would confer a double benefit: if they got worse they could then continue the previous dose in the knowledge that whatever level of side-effects they had was the price they had to pay for wellness, unless they wanted to try a different drug. That constitutes a drug, no-drug, drug, trial which, if a clear drug response occurs on both occasions, is moderately convincing evidence that it is a drug effect and not a ‘placebo’ effect.

One powerful methodology is to use a sample of patients who have already responded decisively to antidepressant treatment, especially ECT. That generates a good sample of ‘biological’ depressive illnesses. Relapse is quite frequent, soon after ceasing a course of ECT (~ 50%), for a recent review see (93). The ability of different drug treatments to prevent, or reverse, that relapse is an instructive index of their effectiveness.

We must also remember lithium’s proven benefits: Prudic et al. (97) ‘monotherapy with NT was distinctly inferior in relapse prevention compared to combination NT-Li’; for recent review see review Rasmussen (98).

A cross-over trial is like a discontinuation trial and, as explained above, if you swap from amitriptyline to clomipramine quite a lot of people get a better improvement. If you do the opposite, most of them get worse.

Many supposed differences or advantages to do with AD drugs are small, or of uncertain meaning or usefulness; however, when various signposts all point in the same direction it may pay to heed them: also, Bayesian logic re-enforces the power of some conclusions, especially, that CMI is more effective than other TCAs, and that the newer SNRIs (venlafaxine etc.) may be less effective than CMI (and also than the NTP + Sert combination).

I strongly advocate an approach which weighs evidence using Bayesian reasoning. I will give an example of that. There is quite a lot of evidence that drugs that boost either serotonin, or noradrenaline, have some effect as ‘antidepressants’. Various trials (and a lot of clinical experience) done with the older TCAs suggested the superiority of clomipramine over other TCAs (99-101). Clomipramine is the only one of the TCAs which significantly boosts both serotonin and noradrenaline (38). If you swap people from amitriptyline to clomipramine quite a lot of them get a better improvement. If you do the opposite, most of them get worse. It adds up, one does not have to be a rocket scientist when the signposts are pointing in the same direction.

On the other hand, consider a drug like trazodone which does not boost either serotonin or noradrenaline. Although, as with moclobemide, it is possible to find ‘meta-analyses’ suggesting it is an effective AD — it is in the guidelines (102, 103), but I do not know any psycho-pharmacologists who think it is good for serious depression. Bayesian reasoning tells us that we would want much stronger evidence that trazodone, or moclobemide, or vitamin A, was effective, because it simply does not do what we think needs to be done to improve depression. Logic dictates that we assign a lower level of confidence to the possibility it is an AD.

Thank you, Thomas Bayes.

Note 4

I remember how often patients coming for a second opinion, or relating the results of some previous treatment, would say how much better they felt. After a few questions, it would be perfectly clear that they were not in fact functioning any better: e.g. they had not got back to work, their social and relationship functioning was no different than previously, and they had not increased participation in the usual hobbies and pastimes and pleasures of their previous life-style.

In response to the reply of ‘how much better they felt’ I would sometimes say ‘It doesn’t matter how you feel’. After allowing time to register being surprised by that comment, from a supposedly caring psychiatrist, I would go on to explain that this subjective impression of how they felt (frequently loaded by situationally driven subjective symptoms such as anxiety) was less important than whether they had tangible improvement in the drive, motivation and energy to do activities that they had ceased doing — but had previously done whilst well — and were also getting a proper degree of fun, pleasure, satisfaction, enjoyment, fulfilment etc. from reading, listening to music, socialising, hobbies and pastimes, all the usual culprits (i.e. anergia and anhedonia).