In general medicine, it is said 60-70% of critical decisions are based on the results of a laboratory test. Psychiatry is an exception to this general rule.
Does this recent study (Sarpal et al 2015) herald a new era of laboratory tests being more frequently used to influence clinical decision making? Will psychiatrists get to channel their inner Dr Kildare (showing my age there) and order this test stat?
The paper points out that antipsychotics are prescribed in severe psychotic illnesses like schizophrenia or bipolar disorder on an empirical basis; the medication is prescribed in a patient to see if it is successful in reducing psychotic symptoms in that patient.
If a biological test proved successful in predicting which patients would respond to antipsychotics (defined as a noticeable reduction in psychotic symptoms) this would have clinical utility (usefulness).
Various biological features that are associated with response to antipsychotics are discussed before ending with the researchers’ previous findings that higher striatal functional connectivity is associated with responding to antipsychotics. (The striatum is part of the basal ganglia area of the brain, functional connectivity was measured between the striatal areas and the rest of the brain).
In a nutshell:
- The authors aim to create an index based on striatal connectivity (a “biomarker”) that predicts response to antipsychotics in a first episode schizophrenia spectrum disorders sample (“discovery cohort”)
- Then repeat in a sample of participants with chronic psychosis (broader range of diagnosis including bipolar disorder) who were hospitalised with an acute relapse (“generalizability cohort”), to see how much this index was predictive of responding to antipsychotics in this second cohort.
There is a diagram to illustrate the methodology. The way they performed the scans and generated the index seems reasonable to me, but my level of knowledge of these techniques is limited. The cut-off point for the striatal connectivity index was set just above the score of the highest scoring participant who counted as a responder to antipsychotics in the first episode group (“discovery cohort”) and this cut-off then applied to the chronic psychosis group (“generalizability cohort”) to see if it separated the antipsychotic responders from the non-responders.
- The first episode sample (41 participants) were recruited as part of a double-blind randomised controlled trial comparing aripiprazole with risperidone (both antipsychotics). This suggests that this is a post-hoc study; that is the study was not designed to test the hypothesis of this paper, but the data collected from it was felt to be useful in testing this hypothesis.
- The generalizability cohort of 40 participants was recruited from a hospital after being admitted with a relapse of chronic psychosis. How they were recruited is not stated, e.g. 40 consecutive appropriate participants; this means it is hard to be sure there is no systematic bias in the recruitment of these participants.
- The chronic psychosis participants were treated according to standard clinical practice and prescribed one of several different types of antipsychotics, not in the more formalised protocol driven manner of the first episode participants being treated in the randomised clinical trial with a choice of aripiprazole or risperidone.
- The duration of assessment was longer in the first episode sample (52 weeks) but was usually 4 weeks in the chronic psychosis sample.
Crucially the criteria of response used varied between the first episode group and the chronic psychosis group:
- For the first episode group a participant was defined as responding if there were two consecutive visits with a CGI improvement score of 1 or 2 (much or very much improved) and a rating of 3 (mild) or less on all of the following items of the BPRS-A (18 item version): conceptual disorganization, grandiosity, hallucinatory behaviour, and unusual thought content.
- For the chronic psychosis group the criteria for responding to antipsychotics was a 20% reduction in severity ratings on core psychotic symptoms (on the BPRS 7-item version) within the first two weeks of treatment (as not having this scale of response by 2 weeks is said to predict not responding to antipsychotics).
There was also a secondary analysis looking at the correlation between a score on the striatal connectivity index and length of stay in the hospital for the chronic psychosis group.
- In the first episode group, 24/41 participants met the criteria for response and in the chronic psychosis group 20/40 participants met the criteria for response.
- The cut-off on the striatal connectivity index separated responders from non-responders to a significant level (p=0.00004) in the first episode group. In the chronic psychosis “generalizability cohort” using this cut-off there was a significant difference in scores between responders and non-responders (p=0.003).
Finally there was a significant correlation between scores on the index and length of stay in hospital (R2=0.11, p=0.04). (They also log-transformed the length of stay data and the correlation remained significant, but the numbers vary between the text and the diagram.)
Strengths and limitations
- The authors repeated the analysis in a separate cohort to confirm that the proposed biomarker was useful in more than one sample
- The two cohorts were very different in ways that could potentially confound results:
- First episode in a double blind randomised controlled trial up to 52 weeks
- Second cohort in a hospitalised group that weren’t in a trial and under usual clinical management for up to 4 weeks
- Both cohorts used different rating scales and different criteria used to decide if responding to antipsychotics
- The first episode group had a diagnosis of schizophrenia spectrum disorders but the chronic psychosis group had a broader range of diagnosis including bipolar disorder. The first episode group was prescribed 1 of 2 possible antipsychotics whilst the chronic psychosis group were prescribed a broader range of antipsychotics
- The size of the cohorts was relatively small
So do I get to be George Clooney in ER? (I’ve modernised since the first paragraph.) I typed the figures from the study into www.clinicalutility.co.uk and it classed the usefulness of this proposed biomarker as a diagnostic screening tool (in the sense that it is screening for potential to respond to antipsychotics) as “Fair” (assuming the sensitivity and specificity figures are reliable and accurate).
This isn’t bad, but given the methodological concerns outlined above, the fact is I can’t actually get a functional MRI at my hospital and they are expensive if I ordered them from my regional centre.
More importantly what would a doctor do with a result that predicted a patient wouldn’t respond to an antipsychotic? If the patient was very distressed and/or at risk of serious harm, the doctor would probably still prescribe an antipsychotic as about 20% would still respond to an antipsychotic and to the doctor that is a reasonable chance in a desperate situation.
So it’s premature to say that this is the dawn of an era of useful biomarkers in clinical practice. It may be the start of the path towards that end, or possibly just a blind alley. Time will tell.
Sarpal DK, Argyelan M, Robinson DG, Szeszko PR, Karlsgodt KH, John M, Weissman N, Gallego JA, Kane JM, Lencz T, Malhotra AK. (2015) Baseline Striatal Functional Connectivity as a Predictor of Response to Antipsychotic Drug Treatment. American Journal of Psychiatry in Advance (doi: 10.1176/appi.ajp.2015.14121571) [PubMed abstract]
Cooke, I. (2012) The promise and challenges of biomarkers for mental illnesses. The Conversation UK, 19 Oct 2012.