Tuesday, 29 July 2014

Trees or the Wood: Symptom based or diagnosis based approaches in Mental Health

The Trees or the Wood: Is a symptoms based approach better than a diagnosis based approach in Mental Health?

Thanks to Dr Ivani Fulli who requested this blog.
The short answer is:  it depends. People who confidently assert that a symptoms based approach or diagnostic based approach is better are omitting to say “for the type of work/ research that I do”. The different approaches have different strengths and weaknesses, of which I will outline some below.

What are the differences?

Symptoms can be thought of as individual complaints or deviations from normal in individual fields. (I’m not going to spend ages with exact definitions as it will suck my enthusiasm).  Doctors sometimes focus on individual symptoms as an important clinical concern and focus of treatment e.g. pain. Sometimes individual symptoms are regarded as particularly important even when using a diagnostic-based approach, often because they are associated with risk or a particularly significant clinical outcome (examples include suicidal/homicidal thoughts).
Now a patient or client (I will stick to patient rather than use both terms) will have their own symptoms.
A diagnosis is based on “archetypes” and in mental health with little validating lab tests etc  based on “vignettes” or “clinical pictures” of what a typical patient will appear to the treating professional. These archetypes are often described with typical symptoms and other typical features and also what features would disqualify them from being given that diagnosis. DSM3 and onwards used specific lists of symptoms and other features which patients had to have before being given this diagnosis.
The diagnostic criteria in psychiatry are called polythetic which in practice means two different patients with the same diagnosis can have different symptoms (and even with the same type of symptom can vary a lot e.g. people with persecutory delusions often have different content e.g. one may be being pursued by the IRA and another may be being attacked by the illuminati).
So a worry people have is that if people have different symptoms this creates uncertainty as to how similar they really are and if you can translate research from this into the patient they see in front of them. It depends on course how good the individual diagnosis is at bringing together people with problems into a category that will predict treatment responses and clinical outcomes.
A good example is depression where people don’t even have to have depression (so long as they anhedonia for DSM or anhedonia and fatigue for ICD10) or panic attacks which can take many forms (but few psychologists object to the use of panic attack as a concept). However polythetic criteria exist in general medicine too such as heart attacks where people don’t have to present with typical crushing chest pain so long as they have enzyme and ECG changes. There also polythetic criteria for conditions in medicine with no objective lab tests to confirm e.g. migraine, chronic fatigue syndrome/ME, polymyalgic rheumatica.
Nevertheless, some people want to stick to symptom only approaches as they feel it’s more rigorous than a diagnostic approach as it’s based on what’s there not what is assumed to be a common condition as assumed by diagnosis.
However it’s rarely a pure symptom based approach. Often it’s symptom in context as management varies. How you treat anxiety brought on only by alcohol withdrawal is different from anxiety in the context of going to the supermarket only. Hallucinations in the context of LSD intoxication are treated differently from hallucinations with no clear cause in clear consciousness. By using these contexts and rules you are effectively making a “diagnosis by proxy” : the same symptom with different causes/ contexts is treated differently and has different prognosis . You aren’t just looking at symptoms but placing them in a context with other data. Sometimes they are also organised into groups of symptoms such as “depressive symptoms” or “panic attacks”.  Some people will organise into “dimensions” but that is a separate topic.
Think of trees and wood, the wood (diagnosis) is comprised of trees (symptoms and other clinical factors such as symptoms only caused by intoxication or withdrawal).  How you conceptualise the problem in terms of research and management depends on whether you target the trees or the wood. Now the trees are there and have an existence outside of the human mind. The wood is an abstract conceptualisation of this group of trees and exists in human minds. In fact for all illnesses  in medicine this applies. The pathology, the symptoms, the disability are all real but how we conceptualise them are abstract concepts.
Now as noted above symptoms based approaches often involve some context and grouping of figures, perhaps a copse rather than a wood.

Research Differences

Apart from the problem of people with the same diagnosis having different symptoms, for some research it may be more fruitful to examine things at the symptom level. It may be easier to track the fluctuations of individual symptoms. It may be more scientifically rigorous just to measure individual symptoms rather than set semi-arbitrary cut off points for wellness or relapse or recovery (e.g. 50% reduction in symptoms, why not 49% or 51%?).
This can give you a nice, scientific set of numbers. However it can be a bit overwhelming as most patients I see have many symptoms. So people often group them together ; “diagnosis by proxy”. And for many treatments you want to know how similar the patients in research trials are to your own patients. This often means setting  various rules e.g. psychotic symptoms when not intoxicated with substances and again getting close to diagnosis if not using diagnosis outright.
Diagnosis based research often also uses measures of symptoms as well such as HAMD for depression or BPRS or PANSS for psychosis symptoms. So this incorporates the best of both worlds.
Of course for certain types of interventions that target specific symptoms then you need research to evaluate this. This can include painkillers for pain, hypnotic drugs for insomnia or behavioural experiments for symptoms in CBT. However these still need context e.g. if behavioural experiment for insomnia then it may differ if alcohol is the cause.
In summary, symptom based research can be regarded as more scientifically rigourous but still needs to be organised at a higher level often for it to be useful clinically. Useful research for interventions can be done if targeted only at symptoms but diagnosis based research often also incorporates symptoms data.


In health you often work with other people such as GPs, CPNS, social workers etc. People don’t like the comorbidity in mental health, people often have multiple diagnoses. However imagine the length of letters and other communications if you just listed all symptoms the patient had and how they had changed. It is simpler to refer to diagnosis and any important changes or symptoms (such as suicidality). Of course you could group the symptoms but you are using diagnosis by proxy in this case.
With patients sometimes it is easier to talk about diagnosis but depends on how good the diagnosis is (reliable, how predictive of treatment and prognosis, level of validity) rather than individual symptoms. On the other hand, patients are often interested in WHY they have problems and in may find it easier to link individual symptoms to past experience.

Treatment Differences

Differences in symptom or diagnosis based approaches can be seen in their management and treatment. Doctors may use medications to treat specific symptoms (painkillers, hypnotics for insomnia) but they often use medications to treat groups of symptoms simultaneously. So  if someone had a diagnosis of depression then treating with antidepressants or schizophrenia with antipsychotics or mania with lithium (or antipsychotics).
Now these drugs often target certain symptoms within that diagnosis better than others e.g. antipsychotics are better at treating hallucinations or delusions than they are treating primary negative symptoms (not caused by e.g. mood or hallucinations/delusions) of schizophrenia.  As a rule however, often drugs treat multiple symptoms at a time (sometimes with different levels of success and at different rates of time).  You still measure or enquire about the symptoms because this is how you can tell if the treatment is working (even in general medicine with its’ lab tests you still have to ask about the patient’s symptoms as you treat the patient, not the lab test as no test is 100% accurate).
In some types of psychotherapy e.g. CBT the focus is more on symptoms. The formulation of the patient will identify what are the more core problems driving the rest. The therapist and patient will decide together what symptom to work on first. There will be a mixture of cognitive and behavioural techniques applied to the first symptom. Once this has been tackled, the therapist and patient will move onto the next symptom/problem and hopefully the patient will have learned transferable skills to make tackling the next symptoms easier/quicker. (“I’ve taught you how to use the saw on the first tree, now saw the next tree down”). These steps are repeated until hopefully the symptoms are reduced (or easier to cope with) and/or the patient can on their own tackle the remaining symptoms or problems.
It’s a bit more complicated than that and there are obviously strong effects from the therapist-patient relationship that is having positive effects on the symptoms but that is a crude summary of how someone might use a symptoms based approach to treat someone.
So depending on the type of intervention you offer, either a diagnosis based approach or a symptom based approach will be more fruitful.   A symptom based approach requires more intensive time input than a diagnostic based approach.


Symptom based approaches are rarely purely symptom based as they have to incorporate contextual data (e.g. in context only of substance intoxication or not) and are often grouped together with other symptoms in a diagnosis by proxy approach. Diagnosis based approaches involve making assumptions.
Symptoms may be more scientifically rigorous when measuring than purely diagnosis based approaches but often need to have context and grouping applied similar to diagnosis. Diagnosis based research often also measure symptoms anyway.
In communicating with fellow healthcare workers, diagnosis plus other important info is quicker than laboriously listing all the symptoms. Some patients may find it easier to link specific symptoms to past events.
Some treatment approaches treat several symptoms simultaneously so a diagnosis based approach is helpful here. A symptom based approach may be more helpful where interventions such as CBT often target a single symptom at a time.
Both diagnosis and symptom based approaches can be used and have different usefulness depending often on how the professional helps people with mental health problems.  Symptom based approaches as noted above are rarely purely symptom based but often involve contextual factors and grouping of symptoms.

Friday, 25 July 2014

Utility without validity

Utility and validity

A frequent concern is, if psychiatric disorders don’t actually describe “natural kinds” i.e. clearly separate illnesses such as differentiating pneumonia from a heart attack then this lack of validity means they have no usefulness in treatment or prognosis (utility).

Think of colours

Western and Japanese culture recognises 7 main colours (excluding black and white) other cultures recognise smaller number e.g.  some in Africa recognise 2 main colours. When tested all people from all cultures can recognise differences in tone/ colour when tested. (So clearly separate different colours can be reliably recognised).

Differences between colours are not absolute e.g. orange eventually becomes yellow but the borderline is indistinct. Differences between neighbour colours is more difficult e.g. orange/yellow or indigo/ violet than differences between clearly separate colours e.g. orange/violet.

So this is a spectrum phenomenon (but with an external validated criterion of light frequency/wavelength) that we categorise into separate colours, not by the external criterion but how it appears to us (so the wavelength is assigned to a subjectively decided colour). The number of categories is culturally determined and we can reliably differentiate between colours/ tones if on clearly different points of spectrum but not if close together on spectrum.

We categorise colours despite it actually being a spectrum because we find it useful for purposes of description. We categorise based on archetypes e.g. of orange, of yellow, not clearly separated. The archetype is of the greenest green or the bluest blue etc Note there is no assumption of “clustering” i.e. that in nature, light frequencies/wavelengths encountered will be more common either each named archetype/ category or that the prevalence of light frequency/wavelengths will be commoner in the centre of the archetype/categories light frequency/ wavelength optimum value.

Note Pantone system of allocating number and name codes to varying shades of colour is a form of categorisation using archetypes.

Application to mental disorders

If you assume that mental activities/ behaviours/ symptoms exist on a spectrum (multi-dimensional rather than single dimensional) and are not separated dichotomously from each other or from “normal” range (i.e. are not separate “natural kinds” and not valid separate categories)

You can still apply categories to these dimensions in a similar way to colours above.

These categories form “archetypes” which to be useful need to be
  • Reliably distinguished (neighbour categories harder to distinguish than distant categories)
  • Provide useful information in terms of other associated features, prognostic frequencies (not absolute but increased and decreased rates), guide to treatment based on research, use for stats etc
  • We can expect that differences between “neighbouring” archetypes may be less than those with “distant” archetypes
  • That marginal cases may exist that don’t fit closely into these archetypes (c.f. index levels of caseness in S.C.A.N. interview schedule) but that the optimum number of categories is not necessarily fixed but depends on other factors (c.f. different no. of colours in different cultures) see debates between lumpers and splitters

Now how might this work in practise?

Looking at 2 dimensions: psychotic symptoms and presence/absence of mania. Now some cases of psychosis may have mild manic type symptoms intermittently but overall are closer to the psychosis, no mania archetype. Let’s place them at different part of the spectrum (equivalent to red and violet in he rainbow). Let’s look at one aspect: the effectiveness of Lithium and antipsychotics in treating the presenting picture. We can get the following picture (derived from Cochrane Reviews and clinical practise).

Antipsychotic effective?
Lithium Effective?
Mania, no psychosis
Mania with psychosis
Psychosis, no/little mania

I’ve not described some of the other differences e.g. tendency to different ages of onset, different social class prevalence, different levels of and type of cognitive impairment etc.

We can see from the point of view of a prescriber it is important to distinguish presence of mania and psychosis and to allocate to appropriate category (one can argue the important question is manic yes/no? if no, psychotic yes/ no, give appropriate treatment. Note for a prescriber the individual symptom in this case is unimportant, e.g. auditory hallucination, persecutory delusion, insomnia, what is more important is to which class the symptom in context with other symptoms places the patient into. 

Another important distinguishing feature could be outcome/prognosis. Very few general medical illnesses give an absolute prognosis e.g. 100% guaranteed to recur in a specific time frame. They often give a range of probabilities of outcome.

The AESOP study in the UK looked at people with a First Episode psychosis. They found that people who presented with Mania had 3-4 times better chance of symptom and functional recovery at 3 and 10 years than people presenting with Schizophrenia/ Schizoaffective disorder (a mixture of both psychosis and mood disorder). (From a lecture given at the General Adult Psychiatry Conference in Manchester in 2013).

So can you reliably identify different points of the spectrum? After all, if people confuse red for green what’s the use of a colour system (and of course some people with colour blindness do have difficulty differentiating certain colours but the prevalence of this problem is low enough for us to continue using colours for many purposes).

 One way of looking at this is what’s called a kappa concordance figure. In the UK we use ICD10 as our official diagnostic system. Field trials showed the kappa figures for schizophrenia overall as a category were very good 0.77-0.84 and for bipolar, manic disorders to be 0.69-0.77. (kappa figures 0.61-0.8 are regarded as substantial agreement between separate raters). This suggests that different clinicians have a high chance of reliably agreeing between themselves what is a manic state and a psychotic state without mania.

So we can see how even if psychosis is like a multidimensional spectrum disorder and not divided into clearly separate disorders then it can still be useful to recognise different points of the spectrum if they are associated with different treatment responses and outcomes.

In short, utility without validity.

However this utility of making the diagnosis depends on how reliably the diagnosis can be made by different clinicians and whether there is sufficient difference in treatment and/or prognosis/outcome for it to be worthwhile. This needs to be made for each diagnosis.

Also, within this psychosis spectrum many people don’t come close to either archetype (equivalent to yellow or green) so assumptions of treatment response and prognosis are less likely to apply to them.

At this point we start to get “how-cannery”, how can these broadbrush archetypes actually predict treatment response, outcomes etc? How can people with such varying content even within the same symptom (the wide variety of different types of persecutory delusions for example)  let alone multiple types of symptoms have the same range of probabilities of response to treatment and outcome? Setting aside the fact that the brain outputs are by far the most complex of any bodily organ outputs  (compare the heart which basically pumps blood), the fact is that research and clinical practice demonstrates that this is indeed the case. A hypothetical problem is solved by experience in the real world. Of course in psychiatry and general medicine also, people with the same diagnosis have different responses to treatment and outcome.

People not using the medical model often have other ways of describing problems that they find more useful to them when helping people with their problems