0:00
Good afternoon.
0:01
Thank you very much for coming to my talk and thank you to Biocrates for the opportunity to share our work that we've done together.
0:10
I work at Cambridge University in the laboratory of Professor Sabine Bahn, the Cambridge Centre for Neuropsychiatric Research, where we work on different psychiatric conditions.
0:21
I have a disclosure to make.
0:23
I also consult for a company called Psyomics, which is a spin out from our laboratory and I have received licencing fees from that company.
0:33
We work on different conditions in our laboratory and one of them is bipolar disorder, which is a debilitating psychiatric condition, and it affects about 1% of the population.
0:48
It is characterised by extreme mood swings between high or manic or and low depressed phases.
0:58
The central problem in diagnosing bipolar disorder is that it frequently fluctuates between normal and depressive states.
1:12
During this depressive state, symptoms are all but identical to those of bipolar disorder and this combined with the fact that patients tend to seek medical help during depressive episodes and also the subjective nature of psychiatric evaluation, which is based also on patient reported symptoms, leads to high misdiagnosis.
1:35
And it is estimated that about 40% of patients with bipolar disorder are initially misdiagnosed as having major depression.
1:44
This leads also to a delay in obtaining a correct diagnosis and treatment of bipolar disorder with incorrect medication.
1:52
With antidepressants, instead of mood stabilisers, which can make the condition worse, they can induce rapid cycling between episodes or induced mania and also diminished response to the appropriate medication.
2:09
We aim to address this high misdiagnosis rates in bipolar disorder in a study we did together with Biocrates and Psyomics, where we were looking for a robust signature of bipolar disorder using metabolomic and digital questionnaire data.
2:32
The study was done in five stages.
2:35
First, we recruited patients online through Facebook, our website, e-mail from previous studies.
2:46
We checked eligibility, so participants were eligible if they were between 18 and 45 years old, if they had current depressive symptoms measured using the Patient Health questionnaire, they were not suicidal and also not pregnant or breastfeeding.
3:05
Participants who qualified were requested to complete this extensive online questionnaire developed by Psyomics, which collect holistic information on patient demographics and symptoms.
3:18
It was developed together with our lab with experienced psychiatrists and also users.
3:24
So patients.
3:27
It consists of 635 adaptive questions, which means patients on the answer questions that are relevant to them.
3:35
And on average this is 283 questions and takes 46 minutes to complete.
3:43
And the different sections or different kinds of information we collect are demographics, psychiatric symptoms, comorbidities, and personality traits.
3:55
This was one part of the data we collected, and the second came from dry blood spot samples, which we collected remotely.
4:05
We posted blood spot collection kits to patients.
4:09
They collected dry blood spots themselves and returned it to us by post.
4:16
Here in the middle we can see how the dry blood spots return to us look like from the collection card.
4:23
We cut this tiny 3 millimetre discs that we analysed with two mass spectrometry based methods.
4:30
One was in our laboratory where we focused on 120 different proteins measured using selective reaction monitoring.
4:40
And the second part was what we've done with Biocrates, where we measured 630 lipids and metabolites from 26 biochemical classes using the MxP Quant 500 assay.
4:55
What was very important for us is that Biocrates ISO 13485 certified, thinking about prospective application of the method in clinics. And finally to know what the true diagnosis is, we conducted a composite international diagnostic interview with patients by telephone by certified interviewers.
5:23
This is a structured and validated diagnostic interview for mood disorders developed by the World Health Organisation and from this we knew what the true actual diagnosis is.
5:38
Finally, we collected a follow up information after 6- and 12-months concerning changes in diagnosis, medication and well-being.
5:51
The study had several objectives, but we focus here on the primary objective and for this reason, we limited the analysis to participants who were previously diagnosed with major depressive disorder within the previous five years, so quite recently.
6:10
And this is because this is when misdiagnosis of bipolar disorder happens most frequently.
6:19
Those that did not have a previous diagnosis of bipolar disorder or schizophrenia and those that the diagnostic interview confirmed as having major depressive disorder or bipolar disorder, and specifically type 1 bipolar disorder, which is the most clinically distinct type.
6:42
And we have two objectives in this analysis.
6:45
First, to find out whether there is a robust biomarker signal and the second is whether the biomarker signal adds an extra information to what the questionnaires collect.
6:59
And we've done it using the different kinds of collected data and statistical and machine learning tools.
7:07
So first, to identify the biomarker panel, we analysed the proteomic data that unfortunately did not yield much signal.
7:16
The area under the curve was 0.51, so almost random, and this was quite disappointing to see.
7:24
The questionnaire data resulted in a strong signal AUROC of 0.86.
7:30
And as expected, when we combine the proteomic biomarker data with digital data, this did not lead to much improvement in the performance.
7:41
So this was quite disappointing.
7:43
But in contrast, analysing the metabolomics data that we obtained from Biocrates yielded a robust and significant signal.
7:53
The AUROC was 0.71.
7:57
And what's important, it validated when we looked at patients followed up for one year and those are patients who did not have a mood disorder diagnosis when we collected samples, but during the follow up period, were diagnosed with depression or bipolar disorder.
8:16
This is where the biomarker panel in samples collected at baseline could differentiate and reproduced with a similar value to the original cohort.
8:31
Now that we knew what the biomarker panel works, we wondered what does it consist of?
8:37
And here we can see the 17 biomarkers that the panel included or that were selected for the final panel.
8:48
One of them is standing out and it's the ceramide, which is almost twice as important as any other biomarker.
8:56
And ceramide is an interesting lipid because it's a sphingolipid, it's involved in brain function.
9:03
It forms the myelin sheet of neurons, for example, and has a known role in neurodegeneration.
9:11
The other biomarkers they are also important for neuronal function through a range of pathways such as they are implicated in neurotransmitter synthesis, Kynurenine pathway, so serotonin and tryptophan related insulin resistance and energy metabolism.
9:34
Because these molecules there were some indications that they may be involved in brain function, we wondered whether they correlate with psychiatric symptoms.
9:45
So in this analysis, we correlated each of the 17 biomarkers with different subsections of the questionnaire.
9:55
And what it shows is that the biomarkers correlated most strongly with the with symptoms specific to bipolar disorder.
10:04
So manic and hypomanic symptoms and also a psychiatric history.
10:13
So this analysis showed that there is a robust and reproducible biomarker signal.
10:19
In the second part of analysis, we evaluated whether it adds an extra information to the symptom data collected directly from patients and for this we analyse different questionnaires included in the online platform with and without biomarkers, marked here with different colours.
10:41
And what this showed is that the area under the curve.
10:46
So the diagnostic performance improves with inclusion of biomarkers and especially in several cases where it became statistically significant.
10:58
And this is when we don't know any symptoms, when we know only the depressive symptoms from patient health questionnaire, when we know patient demographics, and also based on the mood disorder questionnaire, which is used for screening for bipolar disorder.
11:16
Another important thing in this analysis is that it showed that by combining biomarker and digital data, we can get the AUROCC up to 0.96 in the last case, which was extremely high.
11:30
And this is also the objective of the whole study to use the combined digital and biomarker data to detect bipolar disorder.
11:41
Here we look at the same models, but we look at composition.
11:46
So how much do biomarkers contribute to the predictive value of each of the models?
11:51
And we can see in yellow that biomarker contribute most information or predictive value in most of the models except the ones where we predict bipolar disorder from the symptoms.
12:07
Although even there they contribute even about 20% in case of symptoms of mania which is still quite substantial contribution and which adds some objective data.
12:23
So this all looked very promising, at least on paper, but we wondered what does it mean in practical terms?
12:30
And for this, we did an analysis called decision curve analysis, where we look at the actual prevalence and false positives, true positives, and we ask how many patients can we actually detect additionally with this method.
12:48
And it varies between the different kind of questionnaires used, but overall biomarkers can help us detect up to 30% more patients with bipolar disorder compared to when we don't know biomarkers.
13:02
In two cases, this improvement was significant.
13:06
And especially biomarkers are helpful when we don't know much about symptoms of patients.
13:12
This is where they can compensate for dismissing information and also looking at individual patients, individual predictions.
13:22
We can see that where biomarkers help is when the diagnosis based on symptoms alone is not certain.
13:28
So it's 50/50 and this is when biomarkers can help move it one way or another.
13:38
In conclusion, we've identified a biomarker panel, metabolomic biomarker panel for distinguishing bipolar disorder from major depression, where metabolomic biomarkers perform better than protein measurements, achieving AUROC of point 0.71 0.73 and improving when combined with digital data to 0.96.
14:02
They were especially helpful when we did know did not know symptoms and also in uncertain cases.
14:09
They helped us identify up to 30% of misdiagnosed bipolar patients.
14:15
Extra, they were validated in a prospective cohort.
14:19
They point to the function of ceramides or dysfunction of ceramides in bipolar disorder.
14:26
And all of this was done only from a drop of blood and a digital questionnaire, which we hope now to deliver to patients.
14:38
And this is what we are working on.
14:41
We have validated these findings in a study done last year in our group at Cambridge and we are planning together with Psyomics a large scale international clinical validation diagnostic accuracy study for the combined biomarker panel and digital questionnaire together with measuring samples at Biocrates.
15:07
I would like to thank everyone involved in this work and especially the group of professor Sabine Bahn at Cambridge, the team at Cyomics, also Biocrates for the great collaboration and excellent data and support they provide.
15:23
Alice is here with us and we've worked closely also with Denisw and Carlos, all blood donors, everyone involved and also our sponsors.
15:36
Thank you very much for your attention.