Profile cover photo
Profile photo
Joshua Wiley
Health Researcher & Statistical Analyst
Health Researcher & Statistical Analyst

Joshua Wiley's posts

Post has attachment
Chrome is seriously pushy. When opening Chrome with Edge set as default, it pops up a tab trying to make me change Google Chrome to be the default browser, and when opening Gmail in Edge, it has a banner asking to switch to Chrome.


2 Photos - View album

Post has attachment
I am currently recruiting for a brief study on how people cope with stress. The study consists of an online survey that takes 10-15 minutes to complete.

Results from this study will help inform some of my future work with individuals diagnosed with cancer and hopefully ultimately the development and tailoring of interventions.

To find out more, you can read the full explanatory statement and then decide if you'd like to participate, here:

As a small thank you, participants who complete the study can enter into a raffle to win a $20 AUD gift voucher for Boost or Subway (may not work outside of Australia, sorry!).

Thanks in advance!

Post has shared content
Wonderful example of research and application from some very talented, passionate people!
Still pinching ourselves after winning an award last night for the Times Higher Education Innovation and Technology prize at a very enjoyable and glitzy ceremony in London. Would not have been possible without a wonderful team of researchers (post-docs who do all the hard work) and working with some amazingly talented people in NHS Wales who make it all happen by translating the research into practice.
4 Photos - View album

On my upteenth read of Elements of Statistical Learning the L1 and L2 penalties for regularization really made sense. I wonder whether it could be useful for some exploratory work.

Post has attachment
Anti-depressants, socioeconomic status, and risk for coronary heart disease mortality

A new study in Psychosomatic Medicine (online first, not in print yet) reports on socioeconomic status (SES), anti-depressant use, and mortality from coronary heart disease in a representative sample from Finland (full citation at the end).

When reading articles, it is important to evaluate the quality of the evidence presented.  Sometimes very interesting, unexpected results are found but for many reasons the quality of it may not be very good.  At the same time, no single study is perfect, and just because the quality is not perfect does not mean the results can ignored or provide no useful information at all.

There are several reasons this study provides quite good quality evidence.  First of all  First of all, it is prospective, which means the predictors (SES and anti-depressants) were measured before (1999) the outcome (2000 to 2007), not at the same time.  Secondly, the predictors are fairly objective.  Anti-depressant use was collected from data on prescription medication purchases which are all tracked in Finland.  SES was measured by education, income, and job type.  Finally, the sample was a random sample of the Finnish adult (40 to 79 years old) population, so this is not a special group like only university students or only construction workers where results may not apply to other groups.

All that is to say that with a large (over 300,000 adults), representative sample, using fairly objective measures of anti-depressants and socioeconomic status, and cause of death coded from a death registry, we can be fairly confident in the findings from this study.  So what did they actually find?

First, statistically taking age and sex into account, the more years an adult purchased anti-depressants from 1995 to 1999 (none, 1-2 years, 3-4 years, or all 5 years) the greater their likelihood of dying.  This is expressed as a statistic called the hazard ratio, or HR, which is a multiplier of your risk.  For example, if a 65 year old adult female who did not purchase any anti-depressants had a 1% chance of dying from coronary heart disease, then a 65 year old adult female who purchased anti-depressants for 1-2 years would have a 1.49% chance of dying from coronary heart disease.   Another example is if a 85 year old adult female who did not purchase any anti-depressants had a 2% chance of dying from coronary heart disease, then a 85 year old adult female who purchased anti-depressants for 1-2 years would have a 2.98% chance of dying from coronary heart disease.  The important thing is that the hazard ratio is a multiplier of risk, it doesn't mean your absolute risk.  If the absolute risk is only .01% chance, 1.5 times a .01% chance is still very low risk.

Another finding, although many other studies have reported similar results, is that people with lower socioeconomic status are at much greater risk of dying from coronary heart disease than are their age and sex equivalent higher socioeconomic status peers, and this is in Finland where there is tax-funded universal health care, so this finding does not only represent not being able to afford health care.

Finally, the authors examined whether there was an interaction between anti-depressant purchases and socioeconomic status to see if, for example, the relationship between anti-depressants and mortality would be different in individuals with higher or lower socioeconomic status.  There is some empirical and theoretical reasons to expect there may be differences.  For example, if you experience a stressful event, such as depression, getting a disease, losing a loved one, it may be easier to cope if you have more resources available.  For example, someone with higher socioeconomic status may be more able to take leave from work to deal with the stressor, and so cope more effectively than someone with lower socioeconomic status who is forced to try to juggle both work and the stressful event.  From that perspective, depression may have an even greater relation with mortality in adults with lower socioeconomic status.

However, what the authors actually found in the data was that there was little evidence of any different effects of anti-depressant purchases for people with different levels of socioeconomic status.  Indeed, what limited evidence there was for a differential effect was that anti-depressant purchases were more strongly related to coronary heart disease mortality for adults with more not less education. However, the representativeness of this finding may be questionable because it only occurred after excluding disability and old-age pensioners.

It is interesting to think about why anti-depressant purchases and socioeconomic status may be related to coronary heart disease mortality beyond any age or sex effects.  In fact, even after statistically accounting for socioeconomic status, anti-depressant purchases were still related to mortality and the reverse was true too. Socioeconomic status was related to mortality after statistically accounting for anti-depressant purchases.

Unfortunately, these data could not explore how or why these effects occur.  Large registry studies like this are great as they often have the most representative samples and prospective designs, but a limitation is that not much data are available.  For example, we cannot know whether adults become ill, then because of their illness become depressed, and also are more likely to die from coronary heart disease.  It may also be that people who are depressed have a harder time taking care of themselves, exercise less, do not make it to medical appointments to receive early preventive care, and so are at greater risk of dying.  These sort of questions remain open, although many other studies have explored what reasons may explain the link between socioeconomic status and depression and mortality.

One final caution in interpreting these findings is that evidence from other studies suggests that only around 1 in 4 people with depression actually are treated with anti-depressants.  In this study, anti-depressant purchases were used as a way to measure whether adults had depression, but it is important to realize that many of the adults grouped into the "no years of anti-depressant purchases" may actually have had depression.  If that is the case, this study may have underestimated the effect of depression.

Konttinen, Kilpi, Moustgaard, & Martikainen. (in press). Socioeconomic position and antidepressant use as predictors of coronary heart disease mortality: A population-based registry study of 362,271 Finns. Psychosomatic Medicine. doi: 10.1097/PSY.0000000000000258
3 Photos - View album

Post has attachment
New ahead-of-print article just out in Psychosomatic Medicine examining SNPs that predict depressive symptoms and change in depressive symptoms over one year in adults with type 2 diabetes enrolled in a lifestyle intervention or control group.

The data came from the Look AHEAD trial, which is really a rather interesting trial I could talk about at some point designed to improve diet and exercise and thus glycemic control in adults with diabetes.

Genetic predictors of depressive symptoms are relevant in this population as depression, cardiovascular disease, and diabetes tend to co-occur, with elevated rates of depression in individuals with both diabetes and cardiovascular disease.

One of the reasons the article stood out is that they actually:
1) report an opposite pattern of relations between three SNPs and depressive symptoms to what has previously been found
2) adjusting for multiple comparisons which increase the probability of detecting an association by chance alone, very few results remained
3) the main, significant, finding was that the SNP that did predict depressive symptoms in these adults with type 2 diabetes was in the KCNE1 gene, which codes for proteins that regulate potassium channels, which control electrophysiology in cells and are implicated in cardiovascular disease, particularly atrial fibrillation, a disease characterized by typically rapid and irregular heart rhythms.

Part of the reason I think opposite findings and many null results were published is that I think they did a very nice and careful job methodologically. They examined everything multiple ways, checked and adjusted as needed for assumptions, and were transparent in their analysis. As science continues to grapple with retracted papers and effect sizes that seem to diminish over time (which many suggest may be due to early popularity and selective publishing of "significant" results), it is good to see that non significant and contrary results are getting published too. We need all of these data. We need all the results available to really make informed judgements about the state of knowledge.

Of course it is also great that they did find a novel SNP that is linked to depressive symptoms and also plausibly related to cardiovascular disease.

Apple Music is really starting to annoy me.  I don't have a particularly huge music library --- around 20GB --- and much of it I have purchased from the iTunes store anyway, but some of it was not (e.g., Gregorian chant sung by monks in a small monastery, just not iTunes sort of music).  I use an iPhone, and iTunes on my desktop and laptop.

At first, aside from a handful of issues, matching went okay and my music synced over the cloud.  Then duplicates of songs and playlists started appearing.  Annoying, but after removing those, it seemed okay.  That is until I updated iTunes on my laptop, at which point I went through the duplicate process again.  Joy.

Then on updating to iOS 9, my phone decided I no longer had Apple Music (I paid for the subscription, not just the free trial), and all of the songs I had added to my music and made available offline I discovered were gone when I went out for my run today.  Signing out and in again convinced it that I did indeed subscribe to Apple Music, but still have to download my songs again.  Now I see iTunes has an update to v 12.3 on my desktop, and I can only imagine what joys the 'general stability and performance improvements' will bring.

Post has shared content
This is amazing! I wish such journals were setup in psychology and the health sciences!
There has been a lot of discussion over the last few (in fact, not so few any more) years about problems with academic publishing and what might be done about them. In particular, many mathematicians, including me, are convinced that thanks to the internet there could be a far cheaper system that would give us everything we need from the current system while not bothering with some very expensive aspects that we do not need any more. 

One simple idea that has often been raised is that of an arXiv overlay journal. The main characteristic feature of such a journal is that submissions take the form of arXiv preprints. There is some variety in the way the phrase is interpreted beyond that, because there are many choices that can be made about what the journal provides over and above accepting and rejecting papers. So for the purposes of this post let me define a pure arXiv overlay journal to be one for which "publication" consists in nothing more than declaring that an arXiv preprint has been accepted. In principle, such a journal should be very cheap, since its main function is to provide peer review and editorial decisions, which are normally done for no charge by mathematicians themselves. 

The blog post linked to is announcing an arXiv overlay journal that I and some other mathematicians have set up. Broadly speaking, it will be in areas that are loosely related to additive combinatorics. To put that pseudo-precisely, if you define a graph with vertices representing areas of mathematics and edges representing pairs of areas of mathematics for which it would not be too surprising to find a mathematician who was an expert in both areas, then this journal will cover not just additive combinatorics but the 1-neighbourhood of additive combinatorics in this graph. 

If you are interested in helping the journal to succeed, and perhaps in encouraging the arXiv overlay model to provide a serious challenge to the traditional publishers, then please consider submitting an article (if you work in a relevant area and have a good quality paper) and please spread the word to others who might have a suitable article to submit. 

We are accepting submissions now, to a temporary website hosted by Scholastica, an outfit set up by some University of Chicago graduates to make it very easy to set up electronic journals. Early in 2016, we hope to launch the journal, in the sense of going live with a permanent website and a few articles already accepted. The permanent website will again be provided by Scholastica but it will feel more independent. (For example, it will have a URL that is not subsidiary to Scholastica's main URL.) 

For more details about the journal and its aims, see the post linked to below.

Post has attachment
I haven't posted much lately as work has been quite busy. Two grant applications and an expression of interest are in. Finishing one more app this week and then there will be a little break to focus on getting papers out.

As my health research is heavy on quantitative models, something I've been thinking about a lot is how to apply them? How to take what I think I know/the data show and turn that into an intervention or a way to actually make a difference in people's health?

Aside from conceptual issues, there is a very practical one. There are plenty of companies that let people fill out surveys online, but how many let you run the results through a model and send the results back to the participant or even a nurse delivering an intervention? How to take the analytical models into real time on a small budget and staff trained in research not software engineering?

Looks like +Domino Data Lab may have a solution. There was a talk on them at useR! 2014 and I took a look just this week at what they offer and was quite impressed. Plus had a really positive experience when emailing to ask questions. I love the idea of being able to make Python and R models connected to a web API easily. I'm sure most of their clients are commercial, but as research tries to be more individualized we really need tools to harness our models and individualize our interventions.

R journal article on the sparkTable package by +Matthias Templ and others is officially published.

I think these [spark plots] would be a great addition to the standard "descriptives" table in many empirical journal articles.
Wait while more posts are being loaded