March 2009

Written by ~WunderBunny~

Best wishes to our GAMSAT candidates We would like to wish this year's GAMSAT candidates all the best for the big day. Be there early, have your ticket, photo identification, water bottle, calculator, pencils and eraser with you. Make sure you have fresh batteries for your calculator. If you are inclined to feel cold, bring a jacket. Please remember not to dwell for too long on any single question group. ACER includes trial questions, which will not count towards the final result. You won't be able to identify the trial questions on your paper. You will be working hard all day and you will have to keep yourselves nourished. Here is some food advice from a dietitian member: Breakfast: Large bowl of muesli with low fat milk, one piece fruit, coffee. Morning tea: A quick piece of fruit. The break between SI and SII is a strictly maintained 20 minutes. At Sydney last year, you were not permitted to go to the toilet during this break. If they do not let you leave your seat, they may tell you to sacrifice some of your exam time in SI to go to the toilet. Lunch: Sandwich (mulitgrain bread) with avocado, lettuce, cucumber and tuna. One piece of fruit and a tub of low fat yoghurt. Have some nuts / seeds and dried fruit (just in case you are still hungry), Section 3: A packet of jelly beans to supply my brain with as much glucose as possible! Remember to take more food to the exam than you think you will need. The last thing you want is to finish all your food and still be hungry!!!!  After the exam: A nice bottle of red wine and a cheese platter. This advice was sourced from our What to Bring to GAMSAT thread, which has more useful tips. Queensland State Election Due to the state government election being held on the same day, please ensure you have arranged for an absentee vote. It will not be possible for you to duck out to a polling booth during GAMSAT. To find out about absent voters, go to the Queensland electoral commision web site.  Paging Dr members for supporting our GAMSAT candidates with essay reviews, science advice and tips for GAMSAT: antiseptic vitualis catherine stewiegriffin81 sciencegeek ikka Aethiana Morbo pj1969 jicjoc After the exam, we welcome your feedback about the paper. As usual, we would like you to recall all the questions / essay quotes so that we can identify the trial questions. Bond University offers for 2009 Bond University conducted their interviews last month. Candidates will learn the outcome of their applications by the end of this month. The academic year begins in May. Good luck to all Paging Dr members awaiting an offer.  For Queenslanders Some of you may have heard about the Midnight Muster BnS. With two dance floors, live band, catering, camping and a hundreds of health professionals and uni students from Griffith, UQ, Bond and QUT it promises to be the premiere social event to kick off 2009! It is happening on Saturday the 14th of March at Beaudesert Race Club. Camping from 4pm for a 6pm start. All proceeds to the Royal Flying Doctor Service - so come along to support a fantastic cause. Tickets are just $49 and include camping, dinner, recovery breakfast, live band and four drinks. Theme is "Old MacDonald" with tickets selling exclusively online through out secure website: www.hope4health.org.au  For more information and pictures from last year’s ball you can also check out our Facebook event page at:  http://www.facebook.com/home.php#/event.php?eid=49178463770  There will also be heaps of prizes and give aways on the night, including the chance to win 2009 Splendour in the Grass tickets! Yee Ha!  Forum Basics: Resetting forum clock after daylight savings Summer time (in Australia) ends this month. The default time on the forum is Eastern Standard Time or GMT + 10. If you would like to change the time so that it reflects the time in your location: Press "profile" on the menu bar beneath the Paging Dr banner. Click on "Modify Profile" Scroll down to Dates and Times. Set the time zone to your location. Adjust "Currently observing DST" to say "yes" if your location is observing Daylight Savings Time (northern hemisphere) or "no" if you live in the southern hemisphere (see screenshot below). Click on "Modify Profile" button. Stewiegriffin81's Rant #1 Stewiegriffin81 provides a series of rants for the newsletter. P values: What do they mean, and how do we make them? P values are one of the more ubiquitous statistical values that you will come across in the medical literature (although they’re also pretty common in the science literature in general). Unfortunately, if the medical school you go to (or are going to go to) is anything like mine, you’ll spend all of approximately 5 minutes talking about what a p value is before moving onto other things. This is a bit of a shame, as the worth of any given statistical value is dependent on how it is derived, and what the quality of the source data is. Without knowing this, a p value that looks good at face value might actually be worthless.  So what the hell is a p value? A p value is a number that tells you the likelihood of an outcome between two groups if one assumes that only random chance affects the outcome. For example, suppose that I suspect that my gambling friend is hustling people with a loaded coin. I think that my friend’s coin is landing a lot more on heads than it should be, as I notice that it lands on heads 30 times in a row. In order to test my hypothesis that my friend is cheating, I will actually attempt to disprove the opposite. That is, I will test to see what the likelihood is of a normal coin landing on heads 30 times in a row. In general terms, this opposite hypothesis is known as the null hypothesis. Now, we know that coins should normally land on heads 50% of the time, so what is the likelihood of my friend’s coin landing on heads 30 times in a row if it were a truly normal coin? To get the likelihood of a normal coin landing on heads 30 times in a row, we convert our 50% odds of getting a heads into a decimal (0.5), and then multiply that number by itself 30 times:  p = (0.5)30  so p approximately = 0.000000001 This equates to about a one in a billion chance of this event occurring purely by chance. This is such a rare event to normally observe, that it may lead one to suspect my friend of cheating with a loaded coin. In fact, due to the influence of a statistician named Fisher, a p value equal to or less than 0.05 is generally accepted as ‘significant’ in medical journals. Of course, this is simply an arbitrary number, and has no true significance on its own. Following this, due to our phenomenally low p value, we should reject our null hypothesis of the coin being normal, and assume that my friend is cheating. What is very important to note however, is that this test does not tell us whether or not my friend is actually cheating or not. It only tells us the likelihood of the outcome occurring if one assumes random chance alone. In fact, this applies equally to all of the other tests used to generate p values in statistics. Of course, human individuals are much more complicated than coins, and so we cannot use such a simple test to work out whether two groups of humans are different in some way. In particular, when comparing groups of individuals, we have to remember that there will be significant variation between individuals within those groups, as well as between the groups themselves. Suppose for example, I wish to know if a new anti-hypertensive drug lowers blood pressure in hypertensive patients. What will I look at to test this, and how will I test it? Obviously I will need to run a randomised clinical trial, where I randomly separate my hypertensive patients into two groups. One group gets the new drug, while the other gets a placebo. I can then record the blood pressures of all the individuals in both groups, and then see if there is a statistically significant difference (I’ll go into clinical trial methodology in some other rant). In order to see if there is a statistically significant difference, I need (at the very least) the mean blood pressure of both groups and the standard deviation of the blood pressure of both groups. Without going into too much statistical detail, what we are really looking at is whether or not the mean difference between the two groups ‘outweighs’ the natural variation that occurs between each individual. In other words, is the signal (mean difference) much larger than the noise (standard deviation)?  However, how do I work out what the likelihood of the difference occurring purely by chance, if I only have these numbers? In fact, in order for the means and standard deviations of the groups to actually mean anything, we also need to assume that the natural variation that occurs in our human samples occurs according to some predetermined distribution. In most cases, that is the bell curve distribution known as normal/natural/Gaussian distribution. We have a priori evidence that a lot of things (including blood pressure) do indeed follow a normal distribution, and so it’s fair to assume this.  Of course, occasionally we have reason to assume a different distribution, in which case a different test needs to be applied. Without going into too much statistics, the bell curve can be summarised as “normal stuff happens the most, crazy weird outlier stuff happens the least’. For example, human adult height is roughly normally distributed. Most people are of average height, while it gets increasingly rare to find people of great height or of minimal height (and yes, human height is actually normally distributed, it is not bimodal, as some have assumed).  This is where sample size comes into play. With a large enough sample, we can pretty much assume that the variation in our sample is similar to the real variation in the whole population. With that, we can then assume our normal distribution, and then work out where in the bell curve where our single/noise result fits. If it fits in the outliers of the curve such that it covers less than 5% of the curve, then we have a p value of less than 0.05, and it is therefore statistically significant. This is where we need to have a little scepticism when it comes to p values. Basically, p values fall or stand on whether or not we have good reason to believe that the measured variation in our samples is an accurate representation of the variation in the ‘real’ population. We may be suspicious due to some kind of sampling error (for example, some kind of selection bias), or there may be confounding factors, or the sample size may be too small etc. etc. I’d like to harp on sample size for a second here, as I believe that many people do not realise that you actually cannot correct for small sample sizes in an adequate manner. It is important to know that when performing tests for p values, you either have to have a sample size large enough to assume a distribution, or you have to perform additional statistical corrections to make up for the lack of size. However, as the sample size goes down, the reliability of these corrections also goes down.  In much the same way as there is no magical cut off for p values, there is no magical cut off point as to when the sample size is too small. However, roughly speaking the absolute minimum would probably be at least 50 individuals in each group in order to make a viable p value. Note that this cuts out a hell of a lot of trials, particularly the trials one sees in the exercise physiology journals and the alternative medicine journals. Even if the trial methodology is faultless, regardless of how impressive the p value is, you cannot trust the validity of a p value derived from a small sample. It’s best to consider trials of this size as hypothesis generators until larger trials come along. Stewiegriffin81 is a second year medical student at University of Sydney. 2009 Entry Offers Flinders University vincristine University of Melbourne timmelb09 University of Western Australia maggilla04, hiromyhero, Aethiana University of Western Sydney woozy Good luck for GAMSAT ...

Last Updated ( Thursday, 05 March 2009 )

 

< Prev		Next >

Copyright 2010 Paging Dr.

XHTML 1.0 | CSS